Addressing Optical, Recombination and Resistive Losses in Crystalline Silicon Solar Cells

AUSTRALIAN NATIONAL UNIVERSITY

Addressing optical, recombination and resistive losses in crystalline silicon solar cells

by Thomas Allen

A thesis submitted for the degree of Doctor of Philosophy of the Australian National University

in the Research School of Engineering College of Engineering and Computer Science

June 2017

Declaration of Authorship

I, Thomas Allen, declare that this thesis titled ‘Addressing optical, recombination and resistive losses in crystalline silicon solar cells’ and the work presented in it are my own unless stated otherwise.

Signed:

Date:

iii Abstract

The performance of any photovoltaic device is determined by its ability to mitigate optical, recombination, and resistive energy losses. This thesis investigates new materials and nascent technologies to address these energy loss mechanisms in crystalline silicon solar cells.

Optical losses, specifically the suppression of energy losses resulting from front surface reflection, are first analysed. The use of reactive ion etched black silicon texturing, a nano-scale surface texture, is assessed with respect to the two conventional texturing processes: isotexture and random pyramids. While nano-scale surface textures offer a means of almost eliminating front surface reflection, relatively poor internal optical properties (i.e. light trapping) compared to both conventional textures can compromise any optical gains realised on the front surface. It is also shown that enhanced recombination losses remains a barrier to the application of black silicon texturing to further improve high performance devices, though this will likely have less of an impact on multi-crystalline silicon cells where bulk recombination dominates.

The suppression of recombination losses at surface defects by gallium oxide (Ga2O3), an alternative to aluminium oxide (Al2O3), is also investigated. It is demonstrated that, as in Al2O3, thin films of amorphous Ga2O3 can passivate surface defects through a direct reduction of recombination active defects and via the establishment of a high negative charge density. Further investigations demonstrate that Ga2O3 is applicable to random pyramid surfaces textures, and is compatible with plasma enhanced chemical vapour deposited silicon nitride (SiNx) capping for anti-reflection purposes. Indeed, the Ga2O3 / SiNx stack is shown to result in enhanced thermal stability and surface passivation properties comparable to state-of-the-art Al2O3 films. In addition, it is also v

shown that Ga2O3 can act as a Ga source in a laser doping process, as demonstrated by a proof-of-concept p-type laser doped partial rear contact solar cell with an eﬃciency of

19.2%.

Finally, the resistive losses associated with metal / silicon contacts are addressed. It is demonstrated that a significant asymmetry in the work function of the electron and hole contact materials is sufficient to induce carrier selectivity without the need for heavy doping. This had recently been demonstrated for hole contacts with the high work function material molybdenum oxide. In this thesis specific attention is given to finding a suitable low work function material for the electron contact. Calcium, a common low work function electrode in organic electronic devices, is shown to act as a low resistance

Ohmic contact to crystalline silicon without the need for heavy doping. Fabrication of n-type solar cells with partial rear calcium contacts resulted in a device eﬃciency of

20.3%, limited largely by recombination at the Ca / Si interface. This limitation to device eﬃciency is shown to be partially alleviated by the application of a passivating titania

(TiOx) interlayer into the cell structure, resulting in an increase in device eﬃciency to

21.8% – the highest reported eﬃciency for a TiOx-based heterojunction solar cell to date.

Contents

Declaration of Authorship iii

Abstract iv

1 Motivation and Outline1 1.1 Conceptualising solar cells...... 1 1.1.1 The deﬁnition of a solar cell...... 4 1.2 Thesis outline...... 5 1.3 References...... 8

2 Black Silicon Texturing9 2.1 Suppressing front surface reﬂection: an introduction...... 9 2.1.1 Thin ﬁlm interference...... 9 2.1.2 Micro-scale surface texturing...... 14 2.2 Reactive ion etched black silicon...... 15 2.2.1 Nano-scale surface texturing...... 15 2.2.2 Fabrication...... 17 2.2.3 Passivation of b-Si surface defects...... 18 2.3 Foreword...... 22 2.4 References...... 23 Article: Reactive ion etched black silicon texturing: a comparative study .... 27 I. Introduction...... 29 II. Experimental Procedure...... 30 III. Results and Discussion...... 31 A. Surface Area Enhancement Factor ...... 31 B. Surface Passivation Analysis ...... 32 C. Optical Analysis ...... 35 IV. Conclusion...... 39 Acknowledgements...... 39 References...... 41 2.5 Afterword: A note on light trapping and further work...... 43 2.5.1 Additional work: Light trapping...... 43 2.5.2 Further work: Optical performance after encapsulation...... 44 2.6 References...... 47

3 Surface Passivation by Gallium Oxide 49

vii Contents viii

3.1 Surface recombination theory...... 49 3.1.1 Recombination through surface defects...... 49 3.1.2 The significance of surface charge...... 54 3.2 Quantifying recombination...... 57 3.3 Literature review of gallium oxide...... 61 3.3.1 Gallium oxide material properties...... 61 3.3.2 Gallium oxide devices...... 63 3.4 Foreword...... 66 3.5 References...... 68 Article: Electronic passivation of silicon surfaces by thin films of atomic layer deposited gallium oxide ...... 75 I. Introduction...... 77 II. Experimental Procedure...... 78 III. Results and Discussion...... 79 IV. Conclusion...... 85 Acknowledgements...... 85 References...... 87 Article: Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: demonstration of surface passivation and negative interfacial charge ...... 89 I. Introduction...... 91 II. Experimental Procedure...... 92 III. Results and Discussion...... 93 IV. Conclusion...... 100 Acknowledgements...... 101 References...... 102 Article: Demonstration of c-Si solar cells with gallium oxide surface passivation and laser-doped gallium p+ regions ...... 105 I. Introduction...... 107 II. Experimental Procedure...... 108 III. Optical and Electronic Properties of Gallium Oxide...... 109 A. Optical Properties ...... 109 B. Annealing Dependence of the Surface Passivation ...... 110 IV. Application to Solar Cells...... 114 A. Laser Doping from Gallium Oxide ...... 114 B. Solar Cell Results ...... 116 V. Conclusion...... 118 Acknowledgements...... 119 References...... 120 Article: Silicon surface passivation by gallium oxide capped with silicon nitride 123 I. Introduction...... 125 II. Experimental Procedure...... 126 III. Results and Discussion...... 128 A. Recombination at undiffused p-type surfaces ...... 128 B. Recombination at boron diffused p+ surfaces ...... 131 C. Recombination at pyramidally textured surfaces ...... 133 D. Firing stability and crystallinity ...... 136 Contents ix

IV. Conclusion...... 140 Acknowledgements...... 140 References...... 141

4 Calcium-Based Electron Contacts 143 4.1 Introduction...... 143 4.2 Carrier selectivity...... 145 4.2.1 Carrier transport at the contacts...... 147 4.2.2 Suppressing contact recombination...... 151 4.3 Foreword...... 158 4.4 References...... 159 Article: Calcium contacts to n-type crystalline silicon solar cells ...... 165 I. Introduction...... 167 II. Contact Resistance...... 169 III. Partial Rear Contact Solar Cells...... 175 IV. Conclusion...... 184 Acknowledgements...... 185 References...... 186 Article: Low resistance TiO2-passivated calcium contacts for crystalline silicon solar cells ...... 189 I. Introduction...... 191 II. Experimental Procedure...... 192 III. Results and Discussion...... 193 IV. Conclusion...... 197 Acknowledgements...... 198 References...... 199 Article: A Low Resistance Calcium / Reduced Titania Passivating Contact for High Eﬃciency Crystalline Silicon Solar Cells ...... 201 I. Introduction...... 203 II. Results and Discussion...... 205 A. Surface passivation ...... 205 B. Contact resistivity ...... 206 C. PRC solar cells ...... 210 D. Structure and composition of the contact ...... 212 III. Conclusion...... 216 IV. Experimental Section...... 217 Acknowledgements...... 219 References...... 220 Supporting Information...... 223 I. Making the case for passivated partial rear contacts...... 223 References...... 228

Chapter 1

Motivation and Outline

1.1 Conceptualising solar cells

The field of crystalline silicon (c-Si) photovoltaics emerged out of one of the 20th cen- tury’s most transformative technological developments: the invention of the solid state transistor at the Bell laboratories in the 1940s. Early incarnations of c-Si solar cells stemmed from the work of Russell Ohl, through his work in the recrystallisation of pu- rified silicon [1]. Due to the different segregation coefficients of electrically active n- and p-type impurities that remained in the silicon, Ohl formed grown-in p-n junctions per- pendicular to the solidification direction within the multi-crystalline ingots. Graphical details of these ‘photo-E.M.F. cells’ can be found in a patent application for a ‘light- sensitive electric device’ filed by Ohl 1941 [2].

By 1954 Bell Laboratories scientists Daryl Chapin, Calvin Fuller, and Gerald Pearson published a letter to the editor in the Journal of Applied Physics demonstrating a new type of photovoltaic device that they called a ‘silicon p-n junction photocell’ (later

popularised as the ‘Bell Solar Battery’) that converted sunlight to electrical energy

1 Chapter 1. Motivation and Outline 2 at an efficiency of 6% [3]. This device, the first crystalline silicon solar cell formally reported in the literature, was a marked improvement on the cells fabricated by Ohl, whose energy conversion efficiencies up until that time were approximately 1% [1]. The authors describe the operation of the device as follows:

Photons of 1.02 electron volts (λ = 1.2 µm) are able to produce electron-

hole pairs in silicon. In the presence of a p-n barrier, these electron-hole

pairs are separated and made to do work in an external circuit...

noting the action of the p-n junction in the selection of charge carriers to enable their spatial separation. The central importance of the p-n junction was later asserted by

Pearson [4], writing on the then nascent development of the c-Si solar cell:

The heart of the new silicon solar cell is the p-n junction formed near

the front surface of a plate of silicon... An exhaustive study of this important

circuit element has been carried on during the past few years in connection

with the development of transistors and other semiconducting devices. The

solar battery is a direct outgrowth of this accumulated store of knowledge.

A consequence of the historical and technological progression of c-Si photovoltaics has been that the conceptualisation of the operation of crystalline silicon solar cells has evolved into a convolution of the operating principles of diodes and transistors on the one hand, and, as the ﬁeld of photovoltaics matured, a PV-speciﬁc dialogue on the other.

This is evidenced by the terms applied to solar cell characteristics - emitter and base; space charge region. It has also resulted in the primacy of the p-n junction or ‘diffused junction’ solar cell design: diffused junction solar cells have been the dominant solar energy conversion technology up until, and including, today, with a current share of the Chapter 1. Motivation and Outline 3 c-Si solar cell market of > 95% (where the c-Si devices captures > 90% of the global PV market, with competing thin-film technologies making up the remainder) [5]. Indeed, the primacy of the p-n junction in c-Si PV technology has led some to the mistaken assumption that the presence of the p-n junction is a fundamental principle upon which solar cells operate, a confounding of correlation as causation [6].

The empirical counterpoint to this traditional diffused junction solar cell architecture is the silicon heterojuction (SHJ) solar cell: a device design that does not feature a diffused junction, and does not rely upon the formation of a space charge region within the absorber material to separate the photo-generated electron-hole pairs. A device based upon this design currently holds the record efficiency for a monocrystalline silicon solar cell measured to be 26.3% [7], with an increase to 26.6% also reported [8].

While the pioneering papers of [3] and [4] fail to satisfactorily describe the fundamental operating principles of PV devices in a general sense, insofar as the p-n junction is not a fundamental device element, their discussion of the loss mechanisms are more universal.

The authors of the 1954 paper predicted a limiting efficiency of 22% for a c-Si solar cell, noting three main energy loss mechanisms that reduced the operating efficiency of their device to 6%: 1) the effects of front surface reflection (and optical losses more generally);

2) the recombination of electron-hole pairs; and 3) resistive losses within the device and at the contacts.

The chapters that follow in this thesis aim at exploring, some 60 years later, novel techniques and materials to overcome these very same loss mechanisms. Indeed, it would not be an exaggeration to say that the majority of research on solar cells that has followed the 1954 publication of Chapin, Fuller, and Pearson, has been aimed at progressively reducing the energy losses arising from such optical, recombination, and Chapter 1. Motivation and Outline 4 resistive losses, as the brief but prescient loss analysis in [3] touched on the fundamental energy loss mechanisms in any photovoltaic device.

1.1.1 The deﬁnition of a solar cell

Since the time of Chapin, Fuller and Pearson, significant understanding of solar cell operation has developed. In recent years, the theoretical device efficiency for a crystalline silicon solar cell has been evaluated to be 29.4% [9] while the record device efficiency for a diffused p-n junction c-Si cell, in principle the same device as fabricated in [2] and

[3], stands at 25% [10], [11]. Generally speaking, a more universal understanding of the operational principles of solar cells has also developed [12].

Fig. 1: Schematic diagram illustrating the operation of a solar cell. An absorbed photon creates an unbound electron-hole pair. The electron and hole must then be spatially separated at their respective contacts, avoiding recombination.

All devices, be they crystalline silicon-based or otherwise, can be conceived to operate as a solar cell only if a certain set of conditions are met. The minimum requirements for a device to operate as a solar cell are as follows: Chapter 1. Motivation and Outline 5

1. the device must absorb photons to create excess, unbound, and mobile electron-

hole pairs.

2. a population of electron-hole pairs has to be established and maintained by sup-

pressing their recombination.

3. the electron-hole pairs must be spatially separated at the charge carrier’s respective

collection point, called the electron or hole contact, where they are able to pass

through an external circuit.

These three device principles are, in effect, the operational analogues of the loss mechanisms of [3]. The extent to which a device performs these functions determines its efficacy as a solar cell, the metric of note being the device’s power conversion efficiency.

Losses in conversion eﬃciency can be traced back to one of the three main functions described above.

1.2 Thesis outline

In this thesis one chapter is dedicated to novel research into each of the fundamental operational principles enumerated above, with the aim of applying new techniques or materials to enact, or limit the losses arising from, the device operation as previously deﬁned. The thesis is a compilation of the conference papers, letters and journal articles that were written over the course of the author’s PhD degree. Each section is prefaced by an introductory essay that identiﬁes the historical context, physical underpinnings and state-of-the-art of the operational principle under investigation, followed by a brief introduction to the published papers, prior to the presentation of the published articles. Chapter 1. Motivation and Outline 6

In Chapter 2 the use of nano-scale texturing, so-called black silicon (b-Si), is evaluated as a means of limiting front surface reﬂection losses by a graded refractive index eﬀect.

This has the potential to allow the silicon absorber to collect more photons than typical textures used on silicon-based PV devices. The performance of b-Si is compared to the standard industrial textures: random pyramid and iso-texture.

Chapter 3 features an in-depth investigation into the passivation of surface defects of crystalline silicon by thin layers of gallium oxide (Ga2O3), a wide bandgap semiconductor, as an alternative to aluminium oxide (Al2O3). It is demonstrated that Ga2O3 is able to suppress the recombination of electron-hole pairs at surface-defects, thereby fulﬁlling requirement number 2) (neglecting recombination through bulk defects).

Finally, in Chapter 4, the carrier selectivity of the cell’s contacts, typically achieved via heavy doping under the solar cell’s metallised regions, is shown to be achieved in two alternative ways: work function selectivity, whereby materials of sufficiently different work functions from silicon induce carrier selectivity (and so spatial separation); and by the traditional heterojunction approach, whereby the valence or conduction band offsets between the absorber material (crystalline silicon) and the collector material differ for one band and align for the other.

Following the work of Battaglia et al. on dopant-free hole contacts formed by the thermal evaporation of the high work function material molybdenum oxide (MoOx) [13], the use of low work function metal calcium (Ca) is explored as a means of forming dopant- free electron contacts on c-Si. In order to increase the performance of the contact, a passivating interlayer of titania (TiOx) is utilised, resulting in an increase in open circuit voltage without compromising the contact resistance. Chapter 1. Motivation and Outline 7

Using this simple understanding of the basic operating principles of photovoltaic devices, it is shown through the course of this thesis that high efficiency devices can, in principle, be produced without the need for conventional processes that have dominated c-Si PV cell design over the past decades - i.e. wet chemical texturing, dopant diffusions, and direct metallisation. This simplified understanding of PV cell operation can open up an array of hitherto un-utilised materials and production methods in PV technologies, having the potential to streamline the cell fabrication process, lower the thermal energy budget, and simplifying fabrication processes. Chapter 1. Motivation and Outline 8

1.3 References

[1] M. A. Green, “Silicon solar cells: evolution, high-eﬃciency design and eﬃciency enhancements,” Semicond. Sci. Technol., vol. 8, no. 1, p. 1, 1993.

[2] R. S. Ohl, “Light-sensitive electric device,” U.S. Patent number 2,402,662, 25 Jun. 1946.

[3] D. M. Chapin, C. S. Fuller, and G. L. Pearson, “A New Silicon p-n Junction Photocell for Converting Solar Radiation into Electrical Power,” J. Appl. Phys., vol. 25, no. 5, p. 676, 1954.

[4] G. L. Pearson, “Conversion of Solar to Electrical Energy,” Am. J. Phys., vol. 25, no. 9, p. 591, 1957.

[5] ITRPV Working Group, “International Technology Roadmap for Photovoltaic (ITRPV) - Seventh Edition.” Mar. 2016.

[6] U. W¨urfel,A. Cuevas, and P. W¨urfel,“Charge Carrier Separation in Solar Cells,” IEEE J. Photovolt., vol. 5, no. 1, pp. 461469, Jan. 2015.

[7] “NEDO:Worlds Highest Conversion Eﬃciency of 26.33% Achieved in a Crystalline Sil- icon Solar Cell.” [Online]. Available: http://www.nedo.go.jp/english/news/AA5en_ 100109.html. [Accessed: 22-Sep-2016].

[8] “NREL eﬃciency chart.” [Online]. Available: http://www.nrel.gov/pv/assets/ images/efficiency-chart.png. [Accessed: 18-Jan-2017].

[9] A. Richter, M. Hermle, and S. W. Glunz, “Reassessment of the Limiting Eﬃciency for Crystalline Silicon Solar Cells,” IEEE J. Photovolt., vol. 3, no. 4, pp. 11841191, Oct. 2013.

[10] J. Zhao, A. Wang, and M. A. Green, “24.5% Eﬃciency silicon PERT cells on MCZ substrates and 24.7% eﬃciency PERL cells on FZ substrates,” Prog. Photovolt. Res. Appl., vol. 7, no. 6, pp. 471474, Nov. 1999.

[11] M. A. Green, “The path to 25% silicon solar cell eﬃciency: History of silicon cell evolution,” Prog. Photovolt. Res. Appl., vol. 17, no. 3, pp. 183189, May 2009.

[12] P. W¨urfel, Physics of Solar Cells: From Principles to New Concepts. John Wiley & Sons, 2008.

[13] C. Battaglia, X. Yin, M. Zheng, I. D. Sharp, T. Chen, S. McDonnell, A. Azcatl, C.

Carraro, B. Ma, R. Maboudian, R. M. Wallance, and A. Javey, “Hole Selective MoOx Contact for Silicon Solar Cells,” Nano Lett., vol. 14, no. 2, pp. 967971, Feb. 2014. Chapter 2

Black Silicon Texturing

2.1 Suppressing front surface reﬂection: an introduction

2.1.1 Thin ﬁlm interference

The reﬂection of photons from the front, sun-facing side of a photovoltaic device is the

first point of possible energy loss in an operating solar cell. Each individual photon that is incident on the silicon surface has a probabilistic chance of being either transmitted through the air / device interface and into the underlying semiconductor absorber, or being reflected and potentially lost. The probability of reflection (R) at the interface is defined by the Fresnel equation, given by:

2 n1 − n2 R = (1) n1 + n2

for normally incident light, where n1 is the wavelength dependent refractive index of

medium 1, and n2 is the wavelength dependent refractive index of medium 2. Since the

9 Chapter 2. Black Silicon Texturing 10 refractive index of silicon is high (n > 3.5) the probability of reﬂection is high (R > 30%).

The reﬂection from a planar silicon wafer is calculated and plotted in Figure 1a).

Fig. 1: Modelling of reﬂection from a) planar and b) textured silicon surfaces with single and double layer ARCs.

Such a high value of R is unacceptable for PV applications as reflection losses will scale linearly with conversion efficiency in a typical device. In order to reduce the reflection losses in a solar cell, two strategies are typically employed: 1) texturing of the front surface; and 2) the application of one or more anti-reflection coatings.

Historically, the second approach was the first to be explored on c-Si solar cells. An anti-reflection coating (ARC) is a thin film of (ideally) transparent material that, when the thickness is tuned with respect to the refractive index, can minimise front surface reflection be means of an interference effect. For a single layer ARC, a single minimum in R can be obtained and tuned with regard to wavelength (λ). This minimum in R occurs when the film thickness (d) is such that the reflection off the top interface and the intermediate interface are 180 degrees out of phase, so that, for example, the reflection from the air / ARC interface cancels out the reflection off the ARC / silicon interface. Chapter 2. Black Silicon Texturing 11

This occurs when the ‘optical thickness’ (that is, the ﬁlm thickness d multiplied by its refractive index n) of the ARC is equal to a quarter of the minimised wavelength:

λ dn = 0 (2) 1 4

Since there is a broad maximum in the spectral irradiance from the AM1.5 solar spectrum between λ = 500 – 700 nm, an ARC on a solar cell will seek to minimise reﬂection within this wavelength range (e.g. at λ0 ∼ 600 nm). The insertion of a quarter wavelength

ARC changes equation1 to:

2 2 n1 − n0n2 R = 2 (3) n1 + n0n2

From equation3 it is apparent that when the refractive index of the ARC is equal to

the geometric mean of the two outlying materials, i.e. when:

√ n1 = n0n2 (4)

the value of the minimum in R will go to zero. Since the refractive index of air is ∼ 1,

and of silicon ∼ 3.9 at λ0 = 600 nm, an optimal ARC would have a refractive index of

n1 ∼ 2. However, since solar cells are encapsulated in a module with sun-facing coatings

of EVA (ethylene-vinyl acetate) and glass (both with refractive indices of n ∼ 1.5) the

optimal value of n1 increases to ∼ 2.4.

Early applications of ARCs to reduce front surface reﬂection losses in c-Si solar cells

featured thin layers of silicon oxides (SiOx; 1 < x < 2) [1],[2]. The advantage of a

thermally grown SiO2 lay in its ability to passivate surface defects, however the refractive Chapter 2. Black Silicon Texturing 12

Fig. 2: Schematic representation of the suppression of front surface reﬂection by a) the application of an anti-reﬂection coating, and b) surface texturing.

index of SiO2 (n ∼ 1.4) is well below the optimum for anti-reﬂection purposes. Thermally evaporated silicon monoxide (SiO) has a higher refractive index, (n ∼ 1.9) closer to the ideal value for unencapsulated devices, but also a higher absorption of UV photons, limiting Jsc. However, since early incarnations of c-Si solar cells typically featured very heavy front side phosphorus (P) doping, and so were limited in UV collection eﬃciency by recombination through P interstitials rather than Si surface defects, SiO became the

ARC of choice.

The COMSAT ‘violet cell’ of 1973 [2], at the time a world record device with an efficiency at η > 15% [3], used a more transparent tantalum pentoxide (Ta2O5) ARC, registering a significant increase in short circuit current compared to previous cells reported in the literature. Globally, the increase in Jsc was also partly due to other major innovations introduced into the violet cell, most notably: 1) an increases in internal quantum efficiencies by reduced recombination in the phosphorus doped electron contact where the Chapter 2. Black Silicon Texturing 13 surface ‘dead layer’ of interstitial P atoms was minimised by lighter doping; and 2) the development of a finer, photolithographically defined front metal grid that also reduced front surface reflection and shading losses [4].

Later ARC advances would include the introduction of a dual-layer ARC (DLARC) to further suppress front reﬂection losses. DLARCs are capable of introducing a second minima in R, by choosing a low n material as the outermost layer and higher n material

as the inner-most layer. Modelling of reﬂection from single and dual-layer ARCs on

planar silicon is shown in Figure 1a). The MINP (metal-insulator n-p junction) [5] and

PESC (passivated emitter solar cells) [6] devices, that took the record cell eﬃciencies

to over 18% and 19% in 1983 and 1985 respectively [7], were solar cells made on planar

silicon wafers with DLARCs of thermally evaporated ZnS and MgF2 over thin thermally grown SiOx passivation layers. Their reasonable short circuit current densities of 35.5 mA/cm2 and 36 mA/cm2 are a testament to the eﬃcacy of this approach to reducing

reﬂection losses. (Note that these devices were measured at the Solar Energy Research

Institute, now NREL, decades prior to the establishment of the current global AM1.5

solar spectrum standard. Due to this change in the spectrum, measurements at SERI

from this time over-estimate Jsc by as much as 4% relative [7]. Nonetheless, it is remarkable that record breaking device efficiencies were achieved with planar front surfaces – a testament, not only to the efficacy of DLARCs in suppressing front reflection losses, but also the refinement of the other device features, like the improved surface passivation, including the passivated electron contact in the case of the MINP cell, and the optimised contact openings in the PESC cell).

Currently, the industrial standard ARC is silicon nitride (SiNx) deposited by plasma

enhanced chemical vapour deposition (PECVD) due to its high and tunable refractive Chapter 2. Black Silicon Texturing 14 index, and its ability to passivate defects and suppress recombination on both undiffused and heavily phosphorus doped surfaces. On standard HIT cells with both sides contacted, the front-side ITO layer acts as the ARC, as well as the lateral conductive layer. Parasitic absorption in the ITO, as well as the underlying a-Si layers, places a limit on the current obtained with this device architecture. Signiﬁcant optical gains have been realised by adopting an interdigitated back contact design (IBC), such as those devices reported by Panasonic and Kaneka [8],[9].

2.1.2 Micro-scale surface texturing

Improvements in cell eﬃciency to over 20% came by applying a surface texture to further

2 increase the optical performance of the PESC cell, increasing Jsc to 38.3 mA/cm [10].

Texturing the front surface increases the probability of absorption of an incident photon by re-directing reflected photons back onto the absorber material, allowing additional chances of transmission into the bulk of the device. This effect on front surface reflection, coupled with single and dual layer ARCs, is displayed in Figure 1b).

Typical surface texturing of mono-crystalline silicon wafers is achieved through immersion of the Si wafers into a dilute, heated alkaline solution of, typically, potassium hydroxide (KOH). The KOH solution anisotropically etches the silicon crystal due to the different etch rates for different crystal orientations, an effect that is enhanced with the addition of isopropyl alcohol (IPA). Other alkaline metal hydroxides (e.g. NaOH,

LiOH, CsOH) are also know to anisotropically etch silicon, so too solutions of TMAH

(tetramethyl ammonium hydroxide; N(CH3)4OH). (Two diﬀerent solutions containing

TMAH will be used in the studies that follow to achieve either a surface texture or a planarised surface to remove saw damage from the wafer). Chapter 2. Black Silicon Texturing 15

During the etching process the h100i plane is etched much faster than the h111i plane due to the different density of Si–Si backbonds and available bonding sites at the silicon surface [11]. The result is an exposure of h111i crystal facets in a random pyramid array, a reflection of the silicon’s crystalline geometry. If the front surface features are defined by photolithography, the same chemical solution can be used and an ordered array of etched structures, like microgrooves [10] and inverted pyramids [12], can be formed.

This texturing procedure is dependent on the wafer orientation being uniformly h100i prior to etching. A different isotropic texturing method has therefore been developed for multi-crystalline silicon (mc-Si) since multiple crystal orientations are present in any given wafer. An industrially standard process see the mc-Si wafers immersed in a cooled solution of nitric and hydrofluoric acids, forming etch pits that are seeded by the initial surface roughness. This ‘isotexture’ has been detailed elsewhere and the resulting influence on front reflection has been thoroughly characterised [13],[14].

The eﬀect of isotexture and random pyramid surface texturing (also referred to as rantex) on front surface reﬂection are detailed in the conference paper that follows.

2.2 Reactive ion etched black silicon

2.2.1 Nano-scale surface texturing

The above examples of front surface texturing to reduce reflection losses is predicated upon the surface features being significantly greater than the wavelength of light. In this regime one can treat the photon ballistically, as though the photon were a dimension-less particle. However, if the feature size of the texture is reduced to the same or smaller scale as the wavelength of the incoming photon, the ballistic, ray-tracing characterisation Chapter 2. Black Silicon Texturing 16 approach fails, and the surface has to be treated as an ‘effective’ optical medium: as a material with optical properties at the surface differing to those in the bulk material.

For tapered etched features that are sub-wavelength in lateral dimension this can have the effect of creating a grade in the refractive index, from low n at the surface, to a higher n in the bulk. If the variation in refractive index changes at the surface such that the difference between the ‘effective’ surface n and the initial medium n is small, and the effective index changes gradually to that of the bulk, the graded refractive index can reduce the front surface reflection to near zero over a broad spectral range

[15]. This is equivalent to a multi-layer anti-reﬂection stack of materials of gradually increasing refractive indices from top to bottom, except where the change in index is continuous with depth and is facilitated by manipulating the porosity of the surface of the underlying absorber material.

An example of using such an approach to reduce front surface reﬂection is displayed in

Figure 3 below. The figure shows modelled data of the front surface reflection off silicon with a planar multilayer stack calculated using the transfer matrix method. Here, the total layer thickness is kept constant at 500 nm, while the number of layers is changed.

The refractive index also changes from the upper-most layer through to the bulk. This grade in n is calculated using the Bruggeman eﬀective medium approximation [16] with the refractive index data for silicon and air, and maintaining a linear change in porosity in the vertical direction.

The modelling demonstrates that after the division of the 500 nm optical layer into 8 to

16 blocks of equal thickness and varying the eﬀective n, the total reﬂection approaches zero across a broad band of wavelengths. Such a grade in porosity can be achieved physically by a combination of physical and chemical plasma etching of the silicon surface, as described below. Chapter 2. Black Silicon Texturing 17

0 . 8

0 . 7 T o t a l l a y e r t h i c k n e s s : 5 0 0 n m b a r e ( 0 l a y e r s ) 0 . 6 2 l a y e r s )

1 4 l a y e r s / ( 0 . 5 8 l a y e r s R

1 6 l a y e r s n o i

t 0 . 4 c e l f 0 . 3 e R 0 . 2

0 . 1

0 . 0 4 0 0 6 0 0 8 0 0 1 0 0 0 1 2 0 0 ( W a v e l e n g t h ، ( n m

Fig. 3: Reﬂection modelled using the Bruggeman eﬀective medium approximation where the porosity – i.e. the percentage composition of air and silicon – of the 500 nm layer varies linearly, decreasing from the air interface to the silicon bulk.

2.2.2 Fabrication

The realisation of sub-wavelength surface features, and the broadband suppression of reﬂection, can be realised on silicon wafers using a number of techniques: laser irradia- tion, metal-assisted etching, and various plasma-based, dry etching approaches. Reviews detailing these diﬀerent etching methods can be found in refs [17] and [18]. A thorough description of the inductively coupled plasma (ICP) reactive ion etching (RIE) process used in the proceeding article can be found in [19]; a brief summary of the process follows.

Reactive ion etching involves the exposure of a material’s surface to reactive radical and ionic species derived from a plasma. By controlling the chemistry of the plasma and tuning the operating conditions, the interplay between the etched material with the Chapter 2. Black Silicon Texturing 18 surface reactions and ion bombardment can be ﬁnely controlled. An inductively coupled plasma source has the advantage over more direct, capacitive plasma sources in that the plasma density can be tuned independently of the ion bombardment energy, allowing for the generation of a high density plasma with varying kinetic ion energies.

The ICP-RIE process used to fabricate the black silicon structures in the following paper utilizes a mixture of SF6 and O2 input gases. The resultant plasma comprises

+ O* and F* radicals and SFx ions. The F* react with the silicon surface, resulting in a volatile SiF4 molecule which actively etches the silicon. The O* reacts with silicon surface forming SiOxFy F* blocking layer that protects the side walls of the structure

+ from further etching. The directional bombardment from the SFx ions sputters the passivating SiOxFy layer, revealing the underlying Si surface for further etching from the ﬂuorine radical. The degree of isotropy, tapering, and depth of the etch can be controlled by optimising the gas ﬂows, ion energy and plasma power.

2.2.3 Passivation of b-Si surface defects

Considerable attention has recently been paid to the use of reactive ion etched black silicon as an alternative to the wet chemical etching processes described above. Various market and technological trends, coupled with the advantages of RIE b-Si texturing, have been the driver behind this interest, notably:

• the use of diamond wire sawing, which results in a smoother wafer surface with

reduced kerf loss. The surface morphology makes it more challenging to texture

by wet chemical processes due to the lack of roughened seeding sites that catalyse

the isotexture etching process [20]. Chapter 2. Black Silicon Texturing 19

• the trend towards thinner wafers, including kerﬂess and epitaxial wafering tech-

niques which do not result in a h100i surface, and are too thin to chemically etch.

• improvements in the quality of mc-Si wafers (with regards to their bulk lifetime)

and the relatively poor performance of isotexture on mc-Si (prior to encapsulation;

see article below).

• the inherently single-side nature of the RIE process.

• the minimisation of wet chemical waste products.

• the increased feasibility of atomic layer deposition in production environments to

apply conformal surface passivation layers on the nano-scale texture.

• the ability to de-couple surface passivation and anti-reﬂection materials.

How well founded these apparent advantages are remains to be seen, but together they are enough of a driver to for the c-Si PV industry to consider RIE texturing as an alternative to wet chemical etching. For example, JA Solar states that it will be including

RIE texturing processes in production of its mc-Si RIECIUM module [21]; and recently

Hanwha Q Cells and 1366 (a kerﬂess wafer manufacturer) announced η > 19.3% mc-Si

PERC cells featuring RIE texturing [22]. In addition, Jinko Solar has recently unveiled a mc-Si module featuring b-Si etched cells that is has called ‘Eagle Black’, with a reported module eﬃciency > 17%, equivalent to its standard iso-textured mc-Si ‘Eagle’ modules

[23],[24].

The application of nano-scale texturing for PV applications have long been studied, however the potential optical gains of the graded refractive index eﬀect had not been realised due to high rates of recombination at the textured surface [25],[26]. Since most of the light transmitted into a c-Si solar cell is absorbed near the front surface, the Chapter 2. Black Silicon Texturing 20 high rate of recombination compromises the optical gains, especially from higher UV absorption, via poor internal quantum eﬃciencies. The high rate of recombination has been assessed as being due to the much increased surface area, owing to the high aspect ratio of the etched structures.

An improvement in the passivation of defects at the b-Si surface was achieved by applying ALD Al2O3 as the passivation layer, attributed largely to the conformality of the passivating layer and its high negative charge density [27],[28], and the subsequent eﬀect on minority carrier densities within the nanostructure [29]. Fuschel et al. and

Otto et al. reported on a RIE etched b-Si structure with an average reﬂection of > 2% and a minority carrier lifetime over 1 ms at an injection level of ∆n = 5 × 1015 cm−3

[27],[30]; around the same time Repo et al. demonstrated a suppression of front surface reﬂection to < 1% as well as minority carrier lifetimes in excess of 1 ms on 3 Ω.cm p-type

CZ wafers [28]. Interestingly, the minority carrier lifetimes reported in all three studies diﬀered only marginally from the planar control samples, despite the considerable increase in surface area. Later results on boron diﬀused p+ surfaces were reported in [31],

−2 demonstrating a J0 of 51 fA.cm . The sheet resistance of the p+ layer was estimated to be 70 Ω/, though no measurements of the sheet resistance were made. J0 control samples on planar and conventionally textured surfaces were also not reported in the study. Interestingly, the authors also noted only a small increase in spectrally weighted reﬂection (from 0.5% to between 0.7 and 1.2%, depending on the processing conditions) after the boron diﬀusion process due to the consumption of silicon in the formation of the boro-silicate glass and subsequent drive-in oxidation steps.

These advances in surface passivation have been applied at the cell level. Until 2013 the highest conversion eﬃciency of a device with black silicon texturing was that of

Oh et al., who reported an eﬃciency of 18.2% using the wet-chemical metal assisted Chapter 2. Black Silicon Texturing 21 etching technique [32]. This device suﬀered from a poor blue spectral response, and

2 a low Jsc (36.4 mA/cm ). Indeed, both the Jsc and eﬃciency of the b-Si device was lower than the random pyramid control (38.3 mA/cm2; 18.6%). Improvement in device

performance have followed the improvements in surface passivation described above.

Repo et al. reported n-type PERC devices with an efficiency of η = 18.7%, though again, the short circuit current and conversion efficiency was less than a control cell with pyramidal texturing (39.3 vs. 39.9 mA.cm−2; 18.9%) due to etching of the black silicon structures during processing [33]. More significant was the interdigitated back contact (IBC) device reported in [34] with an efficiency of η = 22.1%, however this device showed only a marginal improvement in Jsc over the random pyramid reference device (42.2 vs. 42.0 mA/cm2). Nevertheless, the result did conclusively demonstrate the potential application of b-Si to high efficiency cell structures. A thorough summary of devices reported in the literature prior to those described above can be found in [17]. Chapter 2. Black Silicon Texturing 22

2.3 Foreword

Following the work of Otto et al. [27] and Repo et al. [28], the work presented below aimed to evaluate the electrical and optical performance of black silicon texturing, not only with regard to front surface reﬂection and surface recombination, but also light trapping. The paper that follows, ‘Reactive ion etched black silicon texturing: a comparative study’, was presented as a poster at the 2014 IEEE Photovoltaics Specialist

Conference in Denver, Colorado. The results presented identiﬁed that, for low quality wafers, where the lifetime is largely limited by bulk defects and impurities, the eﬀect of the front surface texture is unlikely to be the limitation on device performance.

The front surface reflection from the RIE black silicon texture is shown to be significantly lower than other industrial textures, as had been identified previously. However, the absorption of long wavelength photons, due to ineffective scattering from the front surface, mitigates much of the gain. This was later quantified by Barugkin et al. [35].

The incorporation of both b-Si and random pyramid textures, or the application of rear scattering plasmonic structures are identiﬁed as a potential means of overcoming this limitation, as discussed in the Afterword that follows the paper. Chapter 2. Black Silicon Texturing 23

2.4 References

[1] M. A. Green, Silicon solar cells: advanced principles & practice, Centre for photovoltaic devices and systems, University of New South Wales, 1995.

[2] J. Lindmayer and J. F. Allison, “The violet cell: An improved silicon solar cell,” Comsat Tech. Rev., vol. 3, no. 1, pp. 151–166, 1973.

[3] M. A. Green, “Silicon solar cells: evolution, high-eﬃciency design and eﬃciency enhancements,” Semicond. Sci. Technol., vol. 8, no. 1, p. 1, 1993.

[4] J. Lindmayer, “Fine geometry solar cell,” U.S. Patent number 3,811,954, 21-May- 1974.

[5] M. A. Green, A. W. Blakers, J. Shi, E. M. Keller, and S. R. Wenham, “High-eﬃciency silicon solar cells,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 679–683, May 1984.

[6] M. A. Green, A. W. Blakers, J. Shi, E. M. Keller, and S. R. Wenham, “19.1% eﬃcient silicon solar cell,” Appl. Phys. Lett., vol. 44, no. 12, pp. 1163–1164, Jun. 1984.

[7] M. A. Green, “The path to 25% silicon solar cell eﬃciency: History of silicon cell evolution,” Prog. Photovolt. Res. Appl., vol. 17, no. 3, pp. 183–189, May 2009.

[8] K. Masuko, M. Shigematsu, T. Hashiguchi, D. Fujishima, M. Kai, N. Yoshimura, T. Yamaguchi, Y. Ichihashi, T. Mishima, N. Matsubara, T. Yamanishi, T. Takahama, M. Taguchi, E. Maruyama, and S. Okamoto, “Achievement of More Than 25% Conversion Eﬃciency With Crystalline Silicon Heterojunction Solar Cell,” IEEE J. Photovolt., vol. 4, no. 6, pp. 1433–1435, Nov. 2014.

[9] “NEDO: World’s Highest Conversion Eﬃciency of 26.33% Achieved in a Crystalline Silicon Solar Cell.” [Online]. Available: http://www.nedo.go.jp/english/news/ AA5en_100109.html. [Accessed: 22-Sep-2016].

[10] A. W. Blakers and M. A. Green, ?20% eﬃciency silicon solar cells,? Appl. Phys. Lett., vol. 48, no. 3, pp. 215–217, Jan. 1986.

[11] H. Seidel, L. Csepregi, A. Heuberger, and H. Baumg¨artel,“Anisotropic etching of crystalline silicon in alkaline solutions I. Orientation dependence and behavior of passivation layers,” J. Electrochem. Soc., vol. 137, no. 11, pp. 3612–3626, 1990.

[12] P. Campbell and M. A. Green, “Light trapping properties of pyramidally textured surfaces,” J. Appl. Phys., vol. 62, no. 1, pp. 243–249, Jul. 1987.

[13] S. C. Baker-Finch, K. R. McIntosh, and M. L. Terry, “Isotextured silicon solar cell analysis and modeling 1: Optics,” IEEE J. Photovolt., vol. 2, no. 4, pp. 457–464, 2012.

[14] S. C. Baker-Finch, K. R. McIntosh, D. Inns, and M. L. Terry, “Modelling isotextured silicon solar cells and modules,” in 2012 38th IEEE Photovoltaic Specialists Conference, Austin, USA, pp. 000192–000198, 2012. Chapter 2. Black Silicon Texturing 24

[15] R. B. Stephens and G. D. Cody, “Optical reﬂectance and transmission of a textured surface,” Thin Solid Films, vol. 45, no. 1, pp. 19–29, 1977.

[16] M. Khardani, M. Bouacha, and B. Bessa¨ıs,“Bruggeman eﬀective medium approach for modelling optical properties of porous silicon: comparison with experiment,” Phys. Status Solidi C, vol. 4, no. 6, pp. 1986–1990, May 2007.

[17] X. Liu, P. R. Coxon, M. Peters, B. Hoex, J. M. Cole, and D. J. Fray, “Black silicon: fabrication methods, properties and solar energy applications,” Energy Environ. Sci., vol. 7, no. 10, pp. 3223–3263, 2014.

[18] M. Otto, M. Algasinger, H. Branz, B. Gesemann, T. Gimpel, K. F¨uchsel, T. K¨asebier,S. Kontermann, S. Koynov, X. Li, V. Naumann, J. Oh, A. N. Sprafke, J. Ziegler, M. Zilk, and R. B. Wehrspohn, “Black Silicon Photovoltaics,” Adv. Opt. Mater., vol. 3, no. 2, pp. 147–164, Feb. 2015.

[19] H. Jansen, M. de Boer, R. Legtenberg, and M. Elwenspoek, “The black silicon method: a universal method for determining the parameter setting of a ﬂuorine-based reactive ion etcher in deep silicon trench etching with proﬁle control,” J. Micromechanics Microengineering, vol. 5, no. 2, p. 115, 1995.

[20] F. Cao, K. Chen, J. Zhang, X. Ye, J. Li, S. Zou, and X. Su, “Next-generation multicrystalline silicon solar cells: Diamond-wire sawing, nano-texture and high eﬃciency,” Sol. Energy Mater. Sol. Cells, vol. 141, pp. 132–138, Oct. 2015.

[21] JA Solar [Online]. Available: http://en.jasolar.com/News_corporatedetails/ 244-JA+Solar+Launching+A+New+Series+of+Highly+Efficient+PV+Modules. [Accessed: 12-Feb-2017].

[22] R. Jonczyk, A. Lorenz, A. Ersen, J. Hofstetter, K. Hübener, K. Duncker, J. Scharf, L. Neilbergall, K. Petter, J. Müller,and D. Jeong “Low-cost Kerfless Wafers with Gradient Dopant Concentration Exceeding 19% Cell Efficiency in PERC Production Line,” Energy Procedia, vol. 92, pp. 822–827, Aug. 2016.

[23] Jinko Solar [Online]. Available: http://www.jinkosolar.com/press_detail_ 1118.html. [Accessed: 12-Feb-2017].

[24] Jinko Solar [Online]. Available: http://www.jinkosolar.com/product_detail_ 269.html. [Accessed: 12-Feb-2017].

[25] H.-C. Yuan, V. E. Yost, M. R. Page, P. Stradins, D. L. Meier, and H. M. Branz, “Eﬃcient black silicon solar cell with a density-graded nanoporous surface: Optical properties, performance limitations, and design rules,” Appl. Phys. Lett., vol. 95, no. 12, p. 123501, Sep. 2009.

[26] F. Toor, H. M. Branz, M. R. Page, K. M. Jones, and H.-C. Yuan, “Multi-scale surface texture to improve blue response of nanoporous black silicon solar cells,” Appl. Phys. Lett., vol. 99, no. 10, p. 103501, Sep. 2011. Chapter 2. Black Silicon Texturing 25

[27] M. Otto, M. Kroll, T. K¨asebier,R. Salzer, A. T¨unnermann,and R. B. Wehrspohn, “Extremely low surface recombination velocities in black silicon passivated by atomic layer deposition,” Appl. Phys. Lett., vol. 100, no. 19, p. 191603, May 2012.

[28] P. Repo, A. Haarahiltunen, L. Sainiemi, M. Yli-Koski, H. Talvitie, M. C. Schubert, and H. Savin, “Eﬀective passivation of black silicon surfaces by atomic layer deposition,” IEEE J. Photovolt., vol. 3, no. 1, pp. 90–94, 2013.

[29] G. von Gastrow, R. Alcubilla, P. Ortega, M. Yli-Koski, S. Conesa-Boj, A. Fontcu- berta i Morral, and H. Savin, “Analysis of the Atomic Layer Deposited Al2O3 ﬁeld-eﬀect passivation in black silicon,” Sol. Energy Mater. Sol. Cells, vol. 142, pp. 29–33, Nov. 2015.

[30] K. Füchsel, M. Kroll, T. Käsebier,M. Otto, T. Pertsch, E.-B. Kley, R. B. Wehrspohn, N. Kaiser, and A. Tünnermann, ?Black silicon photovoltaics,? in Proc. SPIE Photonics for Solar Energy Systems IV, vol. 8438, Brussels, Belgium, 2012.

[31] P. Repo, J. Benick, G. von Gastrow, V. Vähänissi,F. D. Heinz, J. Schön,M. C. Schubert, and H. Savin, “Passivation of black silicon boron emitters with atomic layer deposited aluminum oxide,” Phys. Status Solidi RRL ? Rapid Res. Lett., vol. 7, no. 11, pp. 950–954, Nov. 2013.

[32] J. Oh, H.-C. Yuan, and H. M. Branz, “An 18.2%-eﬃcient black-silicon solar cell achieved through control of carrier recombination in nanostructures,” Nat. Nanotech- nol., vol. 7, no. 11, pp. 743–748, Nov. 2012.

[33] P. Repo, J. Benick, V. Vähänissi,J. Schön,G. von Gastrow, B. Steinhauser, M. C. Schubert, M. Hermle, and H. Savin, “N-type Black Silicon Solar Cells,” Energy Procedia, vol. 38, pp. 866–871, Jan. 2013.

[34] H. Savin, P. Repo, G. von Gastrow, P. Ortega, E. Calle, M. Garin, and R. Alcubilla, “Black silicon solar cells with interdigitated back-contacts achieve 22.1% eﬃciency,” Nat. Nanotechnol., vol. 10, no. 7, pp. 624–628, May 2015.

[35] C. Barugkin, T. Allen, T. K. Chong, T. P. White, K. J. Weber, and K. R. Catchpole, “Light trapping eﬃciency comparison of Si solar cell textures using spectral photolumi- nescence,” Opt. Express, vol. 23, no. 7, pp. A391–A400, Apr. 2015.

Reactive ion etched black silicon texturing: a comparative study

Thomas Allen1, James Bullock1, Andr´esCuevas1, Simeon

Baker-Finch1,2 and Fouad Karouta3

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

2PV Lighthouse, Coledale, NSW, 2515, Australia

3ANFF, Research School of Physics and Engineering,Australian National University,

Canberra, ACT, 0200, Australia

Published in the Proceedings of the 40th IEEE Photovoltaic Specialist

Conference, June 2014.

Abstract — We report on signiﬁcant progress towards the application of reactive ion etched (RIE) black silicon (b-Si) as an alternative to the most commonly applied front-side textures utilized in the crystalline silicon photovoltaics industry – random pyramids and isotexture. The as-etched b-Si surface displays approximately 1% front side reﬂectance weighted across the

27 Chapter 2. Black Silicon Texturing 28 solar spectrum, outperforming both random pyramids (2.83%) and isotexture (6.06%) with optimized anti-reﬂection coatings. The b-Si front surface reﬂectance reduces to below 0.4% after the application of an Al2O3 surface passivation layer. At low injection levels, recombination of charge carriers at the b-Si surface poses no limitation on the minority carrier lifetimes of bulk-limited Cz and multicrystalline samples. At higher injection, or with higher quality substrates, additional recombination at the b-Si surface, char-

−2 acterized by a surface J0s of 20 fA.cm , may play a more signiﬁcant role.

This study provides a rigorous empirical justiﬁcation for recent advances in b-Si textured solar cells and indicates pathways for further eﬃciency gains.

Index Terms — black silicon, texturing, isotexture, pyramids, surface passivation. Chapter 2. Black Silicon Texturing 29

I. Introduction

Textured front surfaces have long been employed to increase light absorption in silicon- based photovoltaic devices by 1) reducing reﬂection from the front, sun-facing surface, and 2) coupling light into the bulk material at oblique angles outside the escape cone, thereby increasing the optical path-length of the light propagating through the device.

Typically though there exists a trade-oﬀ between the enhanced optical performance and increased recombination at the textured surface [1]. This study compares the optical and electrical properties of reactive ion etched (RIE) black silicon (b-Si) with the two most commonly applied texturing morphologies utilized in the crystalline silicon photovoltaics industry – random pyramids (rantex) and isotexture.

The term ‘black silicon’ has come to represent a class of nano-scale surface features that reduce the front surface reﬂection to near zero by a graded refractive index eﬀect.

Such nano-structures can be fabricated in a number of ways, including metal-assisted etching [2], laser etching [3], and reactive ion etching (RIE) [4], none of which require lithographically deﬁned masks to produce the features. In addition to the advantageous optical properties, the formation of black silicon is independent of crystal orientation, making it an ideal candidate for texturing on multicrystalline (mc-Si) and cast-mono solar cells, as well as emerging thin crystalline silicon technologies that do not result in a h100i oriented surface [5]. Indeed, the very low reﬂection from the air – b-Si interface also makes it a potential candidate for applications on high performance monocrystalline silicon solar cells. The RIE fabrication technique has the added advantages of being inherently single-sided, and free of toxic and corrosive wet chemicals.

Regardless of the b-Si etching method, it has frequently been supposed that the advantageous optical effects of the surface features are outweighed by the detrimental electronic Chapter 2. Black Silicon Texturing 30 properties of the much-enhanced surface area [6]. However, recent studies have indicated that b-Si texturing can be compatible with high performance solar cells, with efficiencies over 18% reported by Repo et al. using the RIE method [7], and Oh et al. using the metal-assisted etching method [8]. RIE-based texturing has also been successfully incorporated in other high efficiency devices, most notably the mc-Si solar cell architecture of Schultz et al. [9] which remains the most efficient mc-Si solar cell reported [10].

Although the mc-Si cell of Schultz et al. utilized masked patterning of the features on a scale orders of magnitude larger than those described in this study, it does indicate that the RIE technique, with its inherent ion bombardment of the etched surface, is compatible with the manufacture of high eﬃciency photovoltaic devices.

II. Experimental Procedure

Monocrystalline (p-type, Cz, 2.3 Ω.cm, h100i) and mc-Si (p-type, 1.5 Ω.cm, h100i) silicon wafers were isotextured in an acidic solution containing nitric and hydroﬂuoric acids, resulting in etch pits accurately modelled by hemispherical caps with a characteristic angle ω = 67 ◦ [11]. A sub-set of these isotextured wafers were planarized in a 1:10:1

HF:HNO3:CH3COOH solution. A further sub-set of the planarized samples were double- side b-Si etched in a Versaline LL inductively coupled plasma (ICP) etching system in an SF6 and O2 plasma for 40 minutes. The SF6 and O2 gas ﬂows were both 20 sccm and the chamber pressure was 15 mTorr. The ICP and RF power sources were set to

600 W and 16 W respectively. A backside He ﬂow cooled the wafers to 15 ◦C during etching. Single-side textured optical samples were also fabricated for both texturing morphologies. Chapter 2. Black Silicon Texturing 31

Another set of lifetime samples was prepared using a higher quality n-type FZ material

(3.5 Ω.cm, h100i). These samples were saw-damage etched in TMAH for 15 minutes before undergoing either the same double-side b-Si RIE process described above, or a standard alkaline pyramidal texturing process in TMAH.

The textured and planar lifetime samples were cleaned in the standard RCA process and passivated with plasma-assisted atomic layer deposition (PA-ALD) Al2O3. The resulting

◦ ﬁlms were approximately 20 nm thick. All samples were then annealed at 425 C in N2

for 30 minutes to activate the passivation.

III. Results and Discussion

A. Surface Area Enhancement Factor

The surface area enhancement factor of the black silicon texture was empirically de-

termined by the Brunauer-Emmett-Teller (BET) method which relates the amount of

nitrogen adsorbed on the textured silicon surface with its area. The b-Si wafer was

degassed at 300 ◦C for three days prior to measurement in a TriStar II 3020 surface

area analyzer at 77.35 K. The speciﬁc surface area extracted from the data was 0.0244

2 −1 m .g , in the lower range of sensitivity of the BET technique using N2 as the adsorptive gas. Applying this speciﬁc surface area to the known wafer thickness gives a surface area enhancement factor of 13.9 ±0.71, where the error represents the known variation in the wafer thickness. Greater accuracy in the BET measurement technique may be possible using krypton instead of nitrogen as the adsorptive gas, however this option was unavailable for this experiment. Chapter 2. Black Silicon Texturing 32

Fig. 1: Scanning Electron Microscope (SEM) images of the b-Si nanostructures. The images were taken after passivation with 20nm of PA-ALD Al2O3 and demonstrate the conformality of the dielectric coating. The image on the left is taken from directly above the sample (tilt angle of 0 ◦), while the image on the right is at an angle of 52 ◦.

SEM images of the b-Si nano-structures are shown in Fig.1. The images correspond to an Al2O3 passivated mc-Si sample. The image on the left was taken from directly above the sample, while the image on the right was taken from an angle of 52 ◦ from normal. From the images it is apparent that the ALD Al2O3 is conformal to the etched nano-structures, allowing for eﬀective passivation of the b-Si surface, as shown below.

B. Surface Passivation Analysis

The eﬀective minority carrier lifetime (τeff ) was measured as a function of the excess carrier concentration (∆n) on the symmetrical lifetime structures using a Sinton Instru- ments WCT-120 photoconductance tool operating in the PCD transient mode for the monocrystalline samples and in the QSS generalized mode for the mc-Si samples. The optical constants used in the generalized measurements were calculated from reﬂectance and transmittance data, as per [12]. Chapter 2. Black Silicon Texturing 33

Fig. 2: The colored bars in the SRV plot indicate the range in Seff defined by the Seff,UL and Seff,LL. Error bars in the figures indicate the spread in the data. The single dot in the figure represents the Seff,UL on the planar, FZ sample. The Seff,LL on the FZ b-Si and rantex samples were calculated by substituting the planar control 15 lifetime for τbulk. All values are assessed at an injection level of ∆n = 1 × 10 .

The surface recombination velocity (SRV ) parameter (Seff ) was extracted from the

minority carrier lifetime data. The colored bars in Fig.2 display the upper and lower

limits on Seff . The upper limit of Seff was calculated with an assumed inﬁnite bulk

lifetime while the lower limit on Seff was conservatively assessed using the highest

eﬀective lifetime recorded on phosphorus-gettered control samples for τbulk.

The results indicate that the texturing morphology is not the limiting factor on the

lifetime of minority carriers at ∆n = 1 × 1015 in the Cz and mc-Si samples. Rather,

in low injection, τeff is predominantly determined by the bulk lifetime and not by any

enhanced recombination activity at the surface. This is conﬁrmed by the plots of τeff

against ∆n in Fig.3, where the Cz and mc-Si cases diﬀer in low-injection only by

random variation within the sample set.

In high injection, however, the b-Si curves consistently diverge from the pyramidally Chapter 2. Black Silicon Texturing 34

Fig. 3: Representative injection dependent minority carrier lifetime curves of the textured and planar samples. Chapter 2. Black Silicon Texturing 35

textured, isotextured and planar samples. The topmost curves of Fig.3 plot τeff against ∆n for the b-Si sample and its planar and pyramidally textured equivalents on the higher quality FZ substrate. The uniform decrease in τeff observable in the b-Si sample clearly indicates an increase in carrier recombination at the b-Si surface that is not readily observable in the Cz and mc-Si b-Si samples at lower injection levels due to bulk recombination eﬀects.

In order to understand the consistently divergent behaviour in τeff observable in all of the b-Si samples in high injection compared to the planar, isotextured and pyramidally textured samples, the surface recombination current parameter (J0s) [13] was extracted from the FZ samples by the method outlined in [14]. The measured J0s values

(3 fA.cm−2, 11 fA.cm−2, and 20 fA.cm−2 for the planar, rantex, and b-Si samples respectively) were then used as the only surface recombination inputs to model the injection dependent eﬀective lifetime. The dashed curves in Fig.3 are the modelling results, with the bulk recombination modelled with an asymmetrical defect with densities between

11 11 −14 4.25 × 10 and 5.8 × 10 and capture cross sections (σn and σp) equal to 1.3 × 10

−17 and 7 × 10 respectively. The modelling indicates that the measured J0s accurately represents the enhanced recombination activity at the b-Si surface. However, it is important to note that a J0s behaviour of this magnitude is not observable in the works of

[7] and [15] and so is not inherent to all RIE b-Si processes.

C. Optical Analysis

Single side b-Si etched samples were separately prepared on 300 µm thick wafers in order to empirically derive the optimal Al2O3 thickness. This was determined by evaluating the spectrally weighted reﬂectance (SWR) on samples with diﬀerent Al2O3 thicknesses Chapter 2. Black Silicon Texturing 36 using a PerkinElmer Lambda 1050 integrating sphere spectrophotometer. The results are displayed in Fig.4.

Fig. 4: The spectrally weighted reﬂectance (SWR) and NIR photocurrent (Jph,NIR) plotted as a function of Al2O3 thickness. The SWR is assessed over the wavelength range 280 nm to 900 nm; the NIR photocurrent is calculated over the range 900 nm to 1200 nm.

The minimum in reflectance (≈ 0.4%) corresponds to an Al2O3 thickness of approximately 25 nm. For thicker films the front surface reflectance increases monotonically as the nano-scale features are filled-in and become less defined. The thicker Al2O3 films also result in reduced absorption in the near-infrared (NIR) region (from 900 to 1200 nm), which is likely due to both the increase in reflection losses from the front surface and a reduction in scattering from the silicon nano-structures as the Al2O3 layer fills in the textured features [16].

Figure5 plots the absorption as a function of wavelength for silicon samples with the three textures – isotexture, random pyramids, and black silicon – and a planar control.

The b-Si sample has a 20 nm Al2O3 passivation layer, while the planar, isotexture and Chapter 2. Black Silicon Texturing 37

Fig. 5: Absorption as a function of wavelength for each surface morphology. The black dashed curve represents the perfect ARC and Lambertian light trapping model for a 200 µm thick substrate.

rantex samples have an optimized PECVD deposited SiNx ARC on their front surface.

All samples have a thick SiNx (approximately 200 nm) and evaporated silver stack on the rear to enhance the rear reﬂectance. The random pyramid sample is also double side textured, as this conﬁguration is known to produce near Lambertian light trapping

[17]. The modelled (dashed) curve in the figure assumes perfect anti-reflection properties and Lambertian light trapping and a substrate thickness of 200 µm, which corresponds to the thickness of the isotextured sample. The thickness of the other samples varies between 155-185 µm, rendering the modelled data in the figure slightly incomparable for those samples. For enhanced accuracy, and to provide a direct comparison between the different texturing morphologies, the photocurrents extracted from the spectrophotometry data are given in Table1 as a percentage of the ideal ARC and light trapping model, whereby the thickness of each sample is used as the input to the corresponding calculation. Chapter 2. Black Silicon Texturing 38

Table 1: Summary of Absorption Data

Texture Jph(total) Ideal Jph(280−900nm) Ideal Jph,NIR Ideal (mA.cm−2) (%) (mA.cm−2) (%) (mA.cm−2) (%) isotexture 40.85 93.29 31.70 93.94 9.14 87.81 rantex 43.00 98.35 32.80 97.17 10.21 98.63 b-Si 42.71 98.00 33.54 99.38 9.17 89.81 planar 37.83 86.65 31.03 91.93 6.80 66.17

The parasitic absorption in the measured samples, evidenced by the intersection of the measured data and the modelled data at around 1150 nm, is likely due to absorption in the rear reflector, and artificially inflates the Jph,NIR values in Table1. However it is assumed that the effect is common to all of the measured samples and is otherwise ignored.

As expected, the b-Si sample displays nearly perfect (99.38%) absorption in the UV to visible range of the spectrum but is clearly hampered by poor light trapping compared to both the isotextured and rantex samples. The poor light trapping capacity of the b-Si is evidenced in Fig.5 by the fact that the b-Si sample absorbs less than the isotextured sample in the 1000-1200 nm wavelength range despite a front surface reflectance of just 0.6%, more than 10 times lower than the isotextured sample. This reduction in absorption for the b-Si sample is indicative of a much lower path length enhancement factor for long wavelength photons. In contrast, the double side random pyramid textured sample exhibits near Lambertian light trapping, absorbing up to 98.63% of the spectrum between 900-1200 nm (noting that this number is slightly inflated due to the parasitic absorption in the rear reflector).

Despite the poor NIR light trapping, the b-Si texture outperforms the isotextured sample across the wavelength range as a whole, absorbing 98% of the Lambertian limit, compared to 93.29%, and is comparable to the rantex sample (98.35%), making it an Chapter 2. Black Silicon Texturing 39 ideal candidate for the front side texture on multicrystalline silicon solar cells. If the

NIR absorption of the b-Si texture could be improved, by the addition of rear pyramidal texturing or plasmonic structures for example, the b-Si front side texture could also enhance the overall photon absorption of monocrystalline solar cells.

IV. Conclusion

Reactive ion etched (RIE) black silicon (b-Si) has been shown to provide very low front surface reﬂection (SWR < 0.4%), though poor light trapping compared to both isotexture and random pyramids. Plasma-assisted ALD Al2O3 provides a surface passivation

quality on the nano-structured black silicon surface that can be accurately modelled by a

−2 −2 J0s of 20 fA.cm , compared to 11 fA.cm for the random pyramid textured surface, de-

spite a measured surface area enhancement of approximately 14 times. These properties

indicate that RIE black silicon texturing has the capacity to improve multicrystalline

silicon solar cell eﬃciencies, and could conceivably enhance the photon collection of

monocrystalline silicon solar cells if improvements can be made in enhancing photon

absorption in the near-infrared region of the solar spectrum.

Acknowledgements

The authors acknowledge the ﬁnancial support of the Australian Solar Institute (ASI)

and Australian Renewable Energy Agency (ARENA). The authors would also like to

thank Dr. Li Li of the Australian National Fabrication Facility for assistance with Chapter 2. Black Silicon Texturing 40

SEM imagery, and Dr. Antonio Tricoli and Noushin Nasiri of the ANU Nanotechnology

Research laboratory for the BET measurement. Chapter 2. Black Silicon Texturing 41

References

[1] K. R. McIntosh, and L. P. Johnson, “Recombination at textured silicon surfaces passivated with silicon dioxide,” Journal of Applied Physics, vol. 105, issue 12, pp. 124520-124520-10, Jun. 2009.

[2] H.-C. Yuan, V. E. Yost, M. R. Page, P. Stradins, D. L. Meier, and H. M. Branz, “Eﬃcient black silicon solar cell with a density-graded nanoporous surface: optical properties, performance limitations and design rules,” Applied Physics Letters, vol. 95, no. 12, pp.123501-1-123501-3, Sep. 2009.

[3] T.-H. Her, R. J. Finlay, C. Wu, S. Deliwala, and E. Mazur, “Microstructuring of silicon with femtosecond laser pulses,” Applied Physics Letters, vol. 73, no. 12, pp. 1673-1675, Sep. 1998.

[4] H. Jansen, M. de Boer, R. Legtenberg, and M. Elwenspoek, “The black silicon method: a universal method for determining the parameter setting of a ﬂuorine-based reactive ion etcher in deep silicon trench etching with proﬁle control,” Journal of Mi- cromechanics and Microengineering, vol. 5, no. 2, pp. 115-120, Jun. 1995.

[5] F. J. Henley, “Kerf–free wafering: technology overview and challenges for thin PV manufacturing,” Photovoltaic Specialists Conference (PVSC), 2010 35th IEEE, pp. 001184-001192, Jun. 2010.

[6] B. M. Damiani, R. Ludemann, D. S. Ruby, S. H. Zaidi, and A. Rohatgi, “Devel- opment of RIE-textured silicon solar cells,” Photovoltaic Specialists Conference, 2000. Conference Record of the Twenty-Eighth IEEE, pp. 371-374. IEEE, Sep. 2000.

[7] P. Repo, J. Benick, V. Vhnissi, J. Schn, G. V. Gastrow, B. Steinhauser, M. C. Schubert, M. Hermle, and H. Savin, “N-type Black Silicon Solar Cells,” Energy Procedia, vol. 38, pp. 866-871, 2013.

[8] J. Oh, H.-C. Yuan, and H. M. Branz, “An 18.2%-eﬃcient black-silicon solar cell achieved through control of carrier recombination in nanostructures,” Nature Nanotech- nology, vol. 7, pp. 743-748, Nov. 2012.

[9] O. Schultz, S. W. Glunz, and G. P. Willeke, “Multicrystalline silicon solar cells exceeding 20% eﬃciency,” Progress in Photovoltaics: Research and Applications, vol. 12, issue 7, pp. 553-558, Oct. 2004.

[10] M. A. Green, K. Emery, Y. Hishikawa, W. Warta, and E. D. Dunlop, “Solar cell eﬃciency tables (version 43),” Progress in Photovoltaics: Research and Applications, vol. 22, issue 1, pp. 1-9, Jan. 2014.

[11] S. Baker-Finch, K. R. McIntosh, and M. L. Terry, “Isotextured silicon solar cell analysis and modeling 1: optics,” IEEE Journal of Photovoltaics, vol. 2, no. 4, pp. 457-464, Oct. 2012. Chapter 2. Black Silicon Texturing 42

[12] F. E. Rougieux, P. Zheng, M. Thiboust, J. Tan, N. E. Grant, D. H. Macdonald, and A. Cuevas, “A contactless method for determining the carrier mobility sum in silicon wafers,” IEEE Journal of Photovoltaics, vol. 2, no. 1, pp. 41-46, Jan. 2012.

[13] K. R. McIntosh, P. P. Altermatt, T. J. Ratcliﬀ, K. C. Fong, L. E. Black, S. C. Baker-Finch, and M. D. Abbot, “An examination of three common assumptions used to simulate recombination in heavily doped silicon,” Proceedings of the 28th European Photovoltaic Solar Energy Conference and Exhibition, pp. 1672-1679, Sep. 2013.

[14] D. E. Kane and R. M. Swanson, “Measurement of the emitter saturation current by a contactless photoconductivity decay method,” Proceedings of the 18th IEEE Pho- tovoltaics Specialist Conference, pp. 578-583, 1985.

[15] M. Otto, M. Kroll, T. K¨asebier,R. Salzer, A. T¨unnermann,and R. B. Wehrspohn, “Extremely low surface recombination velocities in black silicon passivated by atomic layer deposition,” Applied Physics Letters, vol. 100, issue 19, pp. 191603191603-4, May 2012.

[16] M. Kroll, M. Otto, T. Käsebier,K. Füchsel, R. B. Wehrspohn, E.-B. Kley, A. Tünnermann,and T. Pertsch, “Black silicon for solar cell applications,” Proceedings of SPIE, vol. 8438, pp. 843817-843817-10, May 2012.

[17] P. Campbell and M. A. Green, “Light trapping properties of pyramidally textured surfaces,” Journal of Applied Physics, vol. 62, issue 1, pp. 243-249, Jul. 1987. Chapter 2. Black Silicon Texturing 43

2.5 Afterword: A note on light trapping and further work

2.5.1 Additional work: Light trapping

Since the publication of the conference paper above, the author has contributed to two other works that provide additional information regarding the assessment of b-Si texturing vs. conventional wet chemical approaches.

Most recently, McIntosh et al. [1] evaluated the internal optics of isotexture for a range of texture morphologies using the wafers of the preceeding conference paper and those of Baker-Finch et al. [2]. The study used spectrophotometry measurements of R and T

on isotextured wafers at various stages of processing to mimic a conventional cell struc-

ture (front and rear thin ﬁlms; rear metallisation) and compared the results with planar

and random pyramid textured control wafers. The results demonstrated that the light

trapping properties of isotextured wafers are largely independent of isotexture morphol-

ogy, and approach that of random pyramids. Interestingly, even for etch depths of 9

µm, where the front reﬂection is similar to a planar wafer (since reﬂected photons strike

the isotextured surface only once), signiﬁcant light trapping, approaching Lambertian

for wavelengths > 1000 nm after rear metallisation, is demonstrated. These results are

consistent with those presented in the preceding conference paper and highlights the

eﬃcacy of isotexture to scatter internally reﬂected light.

In comparison, the work of Barugkin et al. [3], evaluated the eﬃcacy RIE b-Si texture

with regard to light trapping using the photoluminesence spectra from the sample in

[4]. The authors found a substantial improvement in light trapping eﬃciency, calculated

following the approach of Schuster et al. [5], from the b-Si after the application of

a diffuse reflector and silver nanoparticles to the planar rear side of the wafer. The Chapter 2. Black Silicon Texturing 44 combination of front side b-Si and rear side random pyramid texturing was also found to be an equally effective light trapping scheme. Both approaches were shown to be as effective as conventional random pyramid texturing with regard to light trapping, with the advantage of reduced front surface reflection, as outlined above. This work shows that the integration of ‘double scale’ texturing, or light trapping features like that demonstrated by Toor et al. [6], is a credible means of increasing the optical performance of b-Si textures.

Ideally, of course, the as-etched black silicon front surface would promote the propaga- tion of light within the silicon wafer at sufficiently oblique angles to achieve path length enhancements approaching, or even exceeding, that of conventional textures. Draw- ing conclusions from optical modelling in three dimensional reconstructions of RIE b-Si textures, Kroll et al. estimate that an etch depth of > 1 µm is required to achieve significant scattering from the front surface texture [7]. Since the etch depth of the texture presented above is approximately half this value, manipulating the etch procedure to achieve a greater depth may be a route to enhanced light trapping. Further work is required to optimise the light trapping of the as-etched structures, with close attention being paid to any trade-off in surface passivation.

2.5.2 Further work: Optical performance after encapsulation

The conference paper presented above demonstrates the enhanced suppression of front surface reﬂection by RIE b-Si compared to industrial textures with a single layer ARC, with the b-Si texture clearly outperforming both of the conventional textures with regards to front surface reﬂection, most notably the isotexture. However, this study, like Chapter 2. Black Silicon Texturing 45 most investigations into b-Si textures to date, was performed on unencapsulated optical test structures, and so is not representative of the optical conditions of solar cells embedded in modules.

Indeed, it has been noted that the encapsulation of isotextured silicon results in a large fraction of the light being totally internally reflected at the glass–encapsulant interface back into the absorber material, owing to the near-lambertian nature of the reflection off isotexture [2]. Baker-Finch et al. demonstrated that, after encapsulated, photogeneration within cells with isotexture approaches 99% of the photogeneration of random pyramid textured devices [8]. This optical gain after encapsulation may decrease the advantage of b-Si texturing relative to isotexture; with regard to random pyramid texturing, the relative advantage of b-Si has consistently shown to the be minimal, even prior to encapsulation, in instances where credible comparisons have been made [4],[9].

The effect of the optical gain after encapsulation of isotextured silicon is compounded by the fact that the standard encapsulant material, poly-ethylene vinyl acetate (EVA) absorbs strongly in the UV (λ < 400 nm) spectrum. In addition, the heightened recombination at the b-Si textured surface has shown to result in poor quantum efficiencies at short wavelengths [10]. Since this is where b-Si most significantly outperforms isotexture optically, it is not clear if the gains in front surface reflection are as significant even as those presented above. However, for unencapsulated devices, the optical advantage of b-

Si texturing over isotexture is clear. The extension of the assessment of the optical gains from b-Si texturing vs. conventional textures after encapsulation and within completed modules is the subject of further investigation.

While the potential optical beneﬁts of b-Si texturing compared to the conventional wet chemical approaches may be marginal, the adoption of RIE b-Si within industry is Chapter 2. Black Silicon Texturing 46 predicated on a number of inter-related factors, not just optical performance, as noted in the introduction to this chapter. The demonstrations of b-Si texturing in the literature, including the conference paper above, and recently in industry, illustrates that RIE b-Si texturing is competitive with isotexture. Further adoption of RIE b-Si texturing is likely to be dependent on the relative cost of processing compared to wet chemical etching, coupled with the rate of adoption of kerf-less wafers, the proliferation of diamond wire sawing, and the up-take of ALD Al2O3 surface passivation. Chapter 2. Black Silicon Texturing 47

2.6 References

[1] K. R. McIntosh, T. G. Allen, S. C. Baker-Finch, and M. D. Abbott, “Light trapping in isotextured silicon wafers,” IEEE J. Photovolt., vol. 7, no. 1, pp. 110–117, 2017.

[2] S. C. Baker-Finch, K. R. McIntosh, and M. L. Terry, “Isotextured silicon solar cell analysis and modeling 1: Optics,” IEEE J. Photovolt., vol. 2, no. 4, pp. 457–464, 2012.

[3] C. Barugkin, T. Allen, T. K. Chong, T. P. White, K. J. Weber, and K. R. Catchpole, “Light trapping eﬃciency comparison of Si solar cell textures using spectral photolumi- nescence,” Opt. Express, vol. 23, no. 7, pp. A391–A400, Apr. 2015.

[4] T. Allen, J. Bullock, A. Cuevas, S. Baker-Finch, and F. Karouta, “Reactive ion etched black silicon texturing: A comparative study,” in 40th IEEE Photovoltaic Specialist Conference, Denver, USA, pp. 562–566, 2014.

[5] C. S. Schuster, A. Bozzola, L. C. Andreani, and T. F. Krauss, “How to assess light trapping structures versus a Lambertian Scatterer for solar cells?,” Opt. Express, vol. 22, no. 102, pp. A542–A551, Mar. 2014.

[6] F. Toor, H. M. Branz, M. R. Page, K. M. Jones, and H.-C. Yuan, “Multi-scale surface texture to improve blue response of nanoporous black silicon solar cells,” Appl. Phys. Lett., vol. 99, no. 10, p. 103501, Sep. 2011.

[7] M. Kroll, T. K¨asebier,M. Otto, R. Salzer, R. Wehrspohn, E.-B. Kley, A. T¨unnermann, and T. Pertsch, “Optical modeling of needle like silicon surfaces produced by an ICP- RIE process,” in Proc. SPIE 7725, Photonics for Solar Energy Systems III, 772505, May 18, 2010.

[8] S. C. Baker-Finch, K. R. McIntosh, D. Inns, and M. L. Terry, “Modelling isotextured silicon solar cells and modules,” in 38th IEEE Photovoltaic Specialists Conference, Austin, USA, pp. 000192–000198, 2012.

[9] H. Savin, P. Repo, G. von Gastrow, P. Ortega, E. Calle, M. Garin, and R. Alcubilla, “Black silicon solar cells with interdigitated back-contacts achieve 22.1% eﬃciency,” Nat. Nanotechnol., vol. 10, no. 7, pp. 624–628, May 2015.

[10] J. Oh, H.-C. Yuan, and H. M. Branz, “An 18.2%-eﬃcient black-silicon solar cell achieved through control of carrier recombination in nanostructures,” Nat. Nanotech- nol., vol. 7, no. 11, pp. 743–748, Nov. 2012.

Chapter 3

Surface Passivation by Gallium

Oxide

3.1 Surface recombination theory

3.1.1 Recombination through surface defects

The crystalline nature of a solid arises from the three dimensional array in which the crystal’s constituent atoms bind to one another – the repetition of a geometrically defined cluster of atoms, called the unit cell, whose form is dependent on the properties of the atomic species. In a perfect crystalline solid, the unit cell is repeated ad infinitum in the bulk of the crystal, giving rise to long range structural order and a well defined set of chemical and electrical properties. In crystalline silicon, each silicon atom binds covalently to its neighbouring silicon atom. Since silicon has four valence electrons, each

Si atom forms covalent bonds to its four neighbouring Si atoms, satisfying their outer

49 Chapter 3. Surface Passivation by Gallium Oxide 50

3p subshells. This results in the formation of a diamond lattice, equivalent to two inter- penetrating face-centred cubic unit cells, a reﬂection of the tetrahedral coordination of the silicon atom’s valence electrons.

This long range order is, of course, abruptly disrupted at the surfaces of the crystal. At the surface the stable structural order that deﬁnes the bulk of the crystal terminates, and the Si atoms lack neighbouring atoms with which to bond. These unsatisﬁed, or

’dangling’, bonds introduce defect states within the c-Si bandgap that greatly enhance the recombination of electron-hole pairs. The rate at which an electron-hole pair recom- bines via surface defects at a single energy level Et within the bandgap is given by the

Shockley-Read-Hall (SRH) equation [1],[2]:

2 vthDit(nsps − ni ) Us = (3.1) ns+n1 + ps+p1 σp σn

where vth is the thermal velocity, Dit is the interface defect density at the energy level Et, ns and ps are the electron and hole concentrations at the Si surface, ni is the intrinsic carrier concentration, and σn and σp are the capture cross sections for electrons and holes. The terms n1 and p1 relate to the process whereby the charge carriers are at ﬁrst trapped by the defect and then re-emitted back to the conduction or valence bands and are approximated by

E − E E − E n ≡ n exp t i ; p ≡ n exp i T (3.2) 1 i kT 1 i kT

where Ei and ET are the intrinsic Fermi energy and defect energy levels, respectively,

2 and assuming Boltzmann statistics. Note also that, as np = ni by the law of mass

2 action, so too does n1p1 = ni . Chapter 3. Surface Passivation by Gallium Oxide 51

Since there tends to be a distribution of defect states within the bandgap (Dit(E)), and other terms (σn, σp, n1, p1) also have an energy dependence, the total recombination

rate at the surface is then the integration of the recombination rate with respect to

energy, throughout the whole bandgap, i.e., where the lower bound of the integral is the

valence band energy, and the upper bound is the conduction band energy.

Z Ec 2 (nsps − ni ) Us = vth Dit dE (3.3) ns+n1 ps+p1 E + v σp σn

By inspection of equations 3.1, 3.2, and 3.3, a number of signiﬁcant issues regarding sur-

face recombination become apparent. Firstly, and most intuitively, the recombination

rate is directly proportional to the density of defects (Dit); the greater the density of

defects, the higher the rate of recombination. Secondly, the driving force behind SRH

recombination is the deviation in the densities of carriers from their equilibrium value.

Indeed, recombination is the process by which the excited system (the Si crystal) re-

2 stores itself back to equilibrium. This is appreciable in the numerator term nsps − ni ;

2 the further the departure from equilibrium conditions (ni ), the greater the recombi-

nation rate. Finally, with consideration of equation 3.2, defects at an energy level

where Et ≈ Ei (or approximately midgap) will be the most recombination active, as

the re-emmision processes will be minimised. This generalisation assumes roughly equal

capture cross sections for holes as for electrons. Hence why the midgap interface defect

density (Dit,midgap) can often be used as a proxy for the quality of surface passivation.

One other notable aspect of surface recombination relates to the fact that a SRH re-

combination event is equivalent to a chemical reaction that necessarily involves the

interaction of two sub-atomic species (an electron and a hole). It follows then that the

reaction rate (i.e. the recombination rate Us) must be at a maximum, as in all chemical Chapter 3. Surface Passivation by Gallium Oxide 52

reactions, when the two reacting species are equal in value, that is when ns = ps [3].

This is simply a literal application of the law of mass action.

It follows from this analysis that there are two means by which surface recombination can be suppressed:

1. by the direct suppression of the interfacial defect density Dit (especially those

defects at or around midgap), and

2. by the manipulation of the density of charge carriers at the surface, that is, by

introducing asymmetries in ns and ps.

The implementation of the ﬁrst means of suppressing Us, that is, by reducing the number of defects that are able to act as recombination sites, is known as ‘chemical passivation’, as it is directly attributable to the chemical bonding of the silicon surface atoms with another chemical species to satisfy their valence electrons (or ‘dangling bonds’). Materials are typically either grown (e.g. thermal SiO2) on the surface or physically deposited on it

(e.g. PECVD, ALD, and APCVD deposited materials like a-Si, SiNx, Al2O3, etc.). The role of hydrogen within these passivating materials, and hydrogen’s ability to passivate interfacial defects has been extensively noted [4]–[9]. Indeed, some of the most eﬀective materials used to suppress Us are acids like hydroﬂuoric acid (HF) [10],[11], sulfuric acid

(H2SO4) [10] and, recently bis(triﬂuoromethane)sulfonimide (TFSI; (CF3SO2)2NH) [12], given an acid’s inherent ability to donate dissociated hydrogen ions which then bond to the silicon surface atoms.

The second point listed above – the manipulation of carrier densities at the surface – is frequently employed in conjunction with chemical passivation, intentionally or otherwise.

The surface concentration of charge carriers can be manipulated in one of three ways: Chapter 3. Surface Passivation by Gallium Oxide 53

1) the diﬀusion of dopant atoms, commonly phosphorus and boron, into the surface region; 2) the deposition of materials which contain or develop a high charge density when applied to the silicon surface, like SiNx or Al2O3; and 3) by applying a material

on the silicon surface with a diﬀerent work function to that of silicon.

Each of these means of manipulating ns and ps works, fundamentally, in the same

way. In each case the equilibrium carrier concentrations are made to be (often) vastly

asymmetric, and this asymmetry persists under excitation, thereby reducing Us [3]. Take

the example of the diﬀusion of dopant atoms. Phosphorus, a common donor impurity,

and boron, a common acceptor impurity, are routinely thermally diﬀused into the silicon

subsurface region to form low resistance, Ohmic contacts to silicon, at concentrations

often exceeding 1×1019 electrically active dopant atoms per cubic cm. For the purposes

of contacting silicon this has the eﬀect of narrowing the barrier width that impedes

current ﬂow at the silicon / metal interface (see Chapter 4). But the diﬀusion of dopant

atoms into the silicon surface also results in a dramatic reduction in the concentration of

the non-diﬀused charge carrier. For example, the diﬀusion of 1 × 1019 cm−3 phosphorus

19 −3 atoms (therefore the equilibrium electron concentration n0 = 1×10 cm free electrons,

assuming 100% ionisation of the dopant atoms) into the silicon surface results, by the law

2 of mass action (n0p0 = ni ), in an equilibrium hole (minority carrier) concentration p0 on

the order of merely ∼ 10 holes per cubic cm. Under excitation conditions, e.g. after the

application of a bias voltage, or exposure to sunlight, the charge carrier concentrations

increase by

E − E (n + ∆n)(p + ∆p) = n2 exp fn fp (3.4) 0 0 i kT

where Efn − Efp is the diﬀerence in quasi-Fermi energies for electrons and holes. This Chapter 3. Surface Passivation by Gallium Oxide 54

value is equivalent to an electrochemical potential within the Si material (V = (Efn −

Efp)/q). For the example of the heavily phosphorus diﬀused surface given above, the term n0 +∆n ≡ ND, while p0 +∆p ≡ ∆p for any reasonable value of Efn −Efp. Looking at equation 3.4 another way, for any given value of Efn − Efp (i.e. for any given level of excitation, like sunlight exposure), a lower value of ∆p will be required to sustain the same degree of quasi-Fermi level splitting for the doped case compared to an undoped case because n0 + ∆n and hence Efn starts, in equilibrium, at a very high value. A lower minority carrier density translates to a lower recombination rate Us.

This is a simpliﬁed, but illustrative, description of how asymmetries in n and p can lower

Us. The same principle applies, not just to doped regions or surfaces, but also to the manipulation of charge carrier densities by surface band bending which can be achieved either by the deposition of charged materials, or materials with diﬀerent work functions to that of silicon (the former case is explored in this chapter; a discussion of the latter case follows in Chapter 4). Indeed, the above example of using surface dopant atoms to introduce asymmetries in ns and ps ignores the concomitant drivers that increase the recombination rate that the introduction of dopant atoms necessitates, most notably

Auger recombination [13], but also the potential addition of defect states in the diﬀused region from phosphorus precipitates [14], boron-oxygen complexes [14],[15], grown-in defects [17],[18], or the addition of unwanted impurities during the thermal diﬀusion process [19],[20].

3.1.2 The signiﬁcance of surface charge

The presence of ﬁxed charges (Qf ) in a surface passivation layer has a similar eﬀect to surface doping on Us, as noted above. When a charged material is deposited on silicon, the charge in that material Qf is screened by an equal but opposite amount Chapter 3. Surface Passivation by Gallium Oxide 55

of charge in the silicon surface region, Qs, such that Qf + Qs = 0, thereby satisfying

charge neutrality. This charge screening in the semiconductor is equivalent to a bending

of the conduction and valence bands relative to the Fermi level, extending into the bulk

until the semiconductor is in a quasi-neutral condition (i.e. n0 = ND). The induced

surface charge Qs can be made up of mobile electron and holes, or ﬁxed charges from

substitutional (and so ionised) acceptor or donor dopant atoms. Depending on the

silicon doping and the polarity of the ﬁxed charges in the passivation layer, the surface

may be in either an accumulated state (an increase in majority carriers at the surface

compared to the quasi-neutral bulk), a depleted state (a decrease in majority carriers at

the surface compared to the bulk) or in an inverted state (when the majority carriers at

the surface diﬀer from those in the bulk).

An extensive analysis of the eﬀect of Qf on Us for n and p-type silicon surfaces passivated

11 −2 by SiO2, a positively charge material (Qf ∼ 1 × 10 cm ), was given by Aberle et al.

[21]. In that contribution, the authors demonstrated that Us is minimised in accumula-

tion and inversion conditions (i.e. for highly asymmetric charge carrier concentrations

at the surface), and maximised in depletion (where the surface charge carrier concen-

trations are roughly equal). The authors also demonstrated that at increasing injection

levels, the eﬀect of Qf on Us decreases as the surface concentrations of both electrons

and holes approaches ∆n.

A more interesting case is that of SiNx, a material typically with a higher charge den-

sity than SiO2, that has become the industry standard ARC on silicon solar cells. The

11 −2 high charge density of SiNx (Qf > 5 × 10 cm ) has been shown to lead to inversion

conditions at the surface of moderately doped p-type wafers, typically used to fabricate

c-Si solar cells [22]. The formation of the inversion layer leads to an injection depen-

dence in the carrier lifetime τeff as well as parasitic shunting, whereby minority carriers Chapter 3. Surface Passivation by Gallium Oxide 56

(electrons) are eﬀectively transported to recombination active sites, like cell edges, grain boundries, and metallised contact regions [23]. This limits the application of SiNx to n-type and n+ surfaces of c-Si solar cells, but was advantageous in the fabrication of

MIS (metal–insulator–semiconductor) solar cells (see Chapter 4).

This posed a serious problem as p-type cell architectures moved toward the implementation of local rear contacts, replacing conventional full-area aluminium alloyed contacts.

For this application a negatively charged material, which would accumulate majority carrier holes on the p-type material, would be required to eﬀectively passivate the boron doped p-type, and boron diﬀused p+ surfaces on such partial rear contact cell designs.

In addition, the expanding interest in solar cells fabricated on n-type wafers, to avoid device degradation associated with the boron-oxygen defect, made the passivation of p+ regions additionally signiﬁcant.

Such a negatively charged material was found in aluminium oxide Al2O3. Indeed, the negative charge associated with Al2O3 has been known to the PV research community since the work of Hezel and Jaeger in the late 80s [24], however, Al2O3 only gained significant attention after the work of Agostinelli et al. [25] and Hoex et al. [26]. Since that time, the nature of the material and its interface with c-Si has been extensively studied (see [27],[28]). The interface defect density and fixed charge have been quantified for a variety of deposition methods, conditions, and post-deposition thermal treatments

[8],[29]–[32]. The thermal stability of the passivation has been established, and the compatibility of Al2O3 with SiNx capping (i.e., the formation of an Al2O3 / SiNx stack for front side anti-reﬂection purposes), has also been demonstrated [33],[34]. In addition

Al2O3 has been demonstrated as a suitable Al dopant source in laser-ﬁring processes to form localised holes contacts, though the recombination rates at these contacts remains Chapter 3. Surface Passivation by Gallium Oxide 57

high, limiting device performance [35]–[37]. These developments have made Al2O3 the

industry standard passivation layer for p-type and p+ surfaces.

3.2 Quantifying recombination

In reality, the quantiﬁcation of the recombination rate at the surface (Us) is complicated

by the relative complexity of the input parameters of equation 3.3. The determination

of the electron and hole concentrations at the surface is non-trival, especially when a

charged passivation layer is applied, and the energy dependent parameters Dit, are also

challenging to probe, especially for passivation layers that are conductive and permit the

ﬂow of carriers across the silicon interface. Recombination mechanisms in the bulk of

the wafer add to the complexity. In the papers that follow the eﬀective minority carrier

lifetime (τeff ) is often used as a proxy for the recombination rate at the surface, where

τeff is deﬁned as

∆n ∆n ∆n τeff ≡ ≡ + (3.5) Utotal Us Ubulk

where Ubulk is the recombination rate in the bulk of the silicon wafer and is given by

Ubulk = Urad + UAuger + USRH (3.6)

The terms Urad and UAuger represent the intrinsic recombination rates related to radiative and Auger recombination processes, respectively. Parameterisations of these two intrinsic recombination mechanisms can be found in [13],[38],[39]. The USRH refers to the recombination rate in the silicon wafer through bulk, as opposed to surface, defects. Chapter 3. Surface Passivation by Gallium Oxide 58

The physical model that is used to quantify USRH is identical to that of equation 3.3 except that the spatial parameters are now associated with a three dimensional volume

(i.e. the semiconductor bulk) rather than a two dimensional plane (the semiconductor surface).

The term τeff is quite easily quantiﬁed and is typically measured as a function of the minority carrier injection level ∆n (where ∆n = ∆p) by monitoring the conductance of the silicon wafer under diﬀering illumination intensities. A detailed examination of this photoconductance method for determining τeff is given in [40].

Using this approach, the eﬀective lifetime can be considered as a combination of all recombination mechanisms acting in parallel, each with their own contribution to τeff , such that

1 1 1 = + (3.7) τeff τbulk τs

where τs is the contribution to τeff from the surface. Considering a symmetrically passivated wafer

W ∆n τs = (3.8) 2 Us where W is the wafer thickness. Here we can deﬁne an eﬀective surface recombination velocity term Seff

U S = s (3.9) eff ∆n Chapter 3. Surface Passivation by Gallium Oxide 59 such that equation 3.7 becomes

W τs = (3.10) 2Seff

This surface recombination parameter Seff is here defined with respect to a virtual surface, not the real surface where a determination of the carrier concentrations is non- trivial (hence effective SRV). The virtual surface is taken to be at the edge of quasi- neutral region of the semiconductor sub-surface, i.e. where band bending at the surface stops, flat-band conditions prevail, and the assumption of the excess carrier concentration being equal to ∆n is more likely to hold. With regard to the effective lifetime, the effective surface recombination velocity becomes

W 1 1 Seff = − (3.11) 2 τeff τbulk

where τbulk can be approximated by the intrinsic carrier lifetime parameterisation of

Richter et al. [13], ignoring the role of bulk defects, giving rise to an upper limit surface

recombination velocity Seff,UL. A lower limit SRV (Seff,LL) can be determined by

measuring the eﬀective lifetime on a suitable control sample using a surface passivation

technique known to eﬀectively eliminate the contribution of τs on τeff like PECVD SiNx,

PEALD Al2O3, or immersion in HF, and using this value for τbulk in equation 3.11.

These methods for the determination of Seff , Seff,UL and Seff,LL are used throughout

the thesis.

More recently the term surface recombination current density J0s has been used as an

alternative to Seff . McIntosh and Black introduced this metric to account for situations Chapter 3. Surface Passivation by Gallium Oxide 60

when band bending at the surface complicates the interpretation of Seff due to discrep- ancies between the excess carrier concentration at the edge of the surface space-charge region and the surface itself [41]. The authors describe J0s an undiffused analogue to the so-called ‘emitter saturation current density’ J0e, a reflection of the analogous means in which surface doping and band bending from overlying charges influence Us. The term can be thought of as a flow of minority carrier charge (i.e. a current) that is drawn to the surface to recombine:

qUs = Jrec (3.12)

Since the recombination rate is proportional to the extent the divergence of the carrier concentrations of electrons and holes from equilibrium, it follows that the recombination current density can also be expressed in terms of the splitting of the quasi-Fermi levels, and hence the implied, or internal voltage, built up in the semiconductor under illumination or excitation:

Efn − Efp V Jrec = J0sexp = J0sexp (3.13) kT VT

The surface saturation current density has the advantage over Seff in that, for most circumstances over which surface recombination is studied on undiﬀused surfaces (high charge densities on moderately doped wafers), it is independent of injection level and wafer doping [41]. The surface saturation current density J0s can be extracted from injection dependent lifetime measurements by applying the method of Kane and Swanson

[42]. Chapter 3. Surface Passivation by Gallium Oxide 61

3.3 Literature review of gallium oxide

3.3.1 Gallium oxide material properties

The advent of passivation by Al2O3 has prompted research into other negatively charged materials. At the time of writing the journal articles that follow, few negatively charged materials that passivated silicon surface defects had been demonstrated. Since their publication, surface passivation by a number of other negatively charged materials has been shown. Tantalum pentoxide, used in the COMSAT violet cell as an ARC, has recently been shown to passivate surface defects when capped with SiNx [43],[44]; so too

TiO2 [45],[46] and aluminium nitride (AlN) [47], both high refractive index anti-reﬂection

materials.

The motivation behind such research is largely related to the area of passivated contacts

(see Chapter 4). Aluminium oxide is a wide bandgap, electrical insulator. This, in

one respect, is considered beneﬁcial, insofar as it ensures transparency to photons of

interest in PV applications. However, at the contact, the non-conductive nature of

Al2O3 renders the fabrication of a heterojunction impossible, while the wide bandgap

of Al2O3 ensures a large band oﬀset (and so energy barrier) to both electrons and holes,

limiting passivating tunnelling contacts to Al2O3 thicknesses < 0.5 nm even when used

in conjunction with heavy surface doping [48]. However, such thin layers compromise

surface passivation; tunnelling layers can also be diﬃcult to fabricate on large areas

due to spatial inhomogeneities in the ﬁlm thickness. Similar arguments can be made

against the use of SiO2, though some researchers report on spatial inhomgeneities in

SiOx interlayers as being critical to device performance [49]. Chapter 3. Surface Passivation by Gallium Oxide 62

Unlike Al2O3, gallium oxide is a wide bandgap (Eg > 4.2 eV) semiconductor [50]–[53].

Ga2O3 is electrically insulative in its stoichiometric form [54], but can be doped with silicon or tin for n-type conductivity [55],[56]. Most studies have focused on crystalline

β-phase Ga2O3, in which typically higher conductivities are reported. Suzuki et al. report a resistivity of 4.27×10−2 Ω.cm at a carrier density of 2.26×1018 cm−3, with a reduction in the transmittance to > 85% (closer to 100% for undoped crystals) for Sn- doped β-Ga2O3 FZ single crystals, with electron mobilities in the range of 49.3 – 87.5 cm2/Vs [57]. Villora et al. report a higher mobility (µ ∼ 100 cm2/Vs), independent of

16 18 free carrier concentration in the range 10 – 10 , in Si-doped β-Ga2O3 with reported resistivities varying from 33 to 0.2 Ω.cm [55].

By controlling the partial pressure of oxygen in the crystal growth ambient, Ueda et

9 −2 al. were able to vary the resistivity of β-Ga2O3 from greater that 10 to 2.6×10

Ω.cm, indicative of conductivity in un-doped material through oxygen vacancies [54].

This conduction mechanism has been challenged by Varley et al. who calculated, using density functional theory, that oxygen vacancy defects in β-Ga2O3 to form deep level defects with ionisations energies greater that 1 eV and so cannot contribute to the conductivity [58]. Instead, they propose that unintentional impurity doping is the cause of the conductivity, identifying H, Cl, F, Sn, Ge, and Si as shallow donors. Si, often a background impurity in Ga2O3 source powders, as well as crucibles used in many crystal growth methods, may be a likely candidate, in agreement with the work of Villora et al. While the work of Varley et al. does not satisfactorily explain the correlation of increasing conductivity with decreasing crystal ambient oxygen partial pressure by

Ueda et al., it does identify a range of potential shallow donor impurities in β-Ga2O3.

The amorphous phase of GaOx has proved less conductive, even for Sn concentrations

> 1 × 1020 cm−3, due to suppressed dopant activation [59]. Chapter 3. Surface Passivation by Gallium Oxide 63

In addition to the conductivity, reported values of the Ga2O3 electron aﬃnity are in the

range of 2.7 eV [60] to 3.7 eV [59], translating to a potential oﬀset between the Ga2O3

and Si conduction bands of as little as 0.35 eV. However, as noted by Heinemann et al.,

Ga2O3 has been used as the electron collector in both CIGS and Cu2O devices which

themselves have widely varying electron aﬃnities (4.5 eV and 3.2 – 4 eV, respectively),

leading the authors to reintroduce the importance of oxygen vacancy defect band, this

time in a-Ga2O3, in the case of the CIGS contact [61].

Finally, gallium, like aluminium, is also a p-type dopant in crystalline silicon, and so can

be used to form Ga doped p+ regions, as described in the articles that follow. Taken as

a whole, Ga2O3 has the potential to be used as a passivated electron contact, a p-type

dopant source for hole contact formation, and as a lateral charge transport layer. The

search for versatile, multifunctional materials is the ultimate driver behind the research

presented in this, and the following, chapter. To be eﬀective in these roles however, the

passivation of surface defects by Ga2O3 ﬁrst has to be demonstrated, and hence is the

primary focus of this chapter.

3.3.2 Gallium oxide devices

Prior to the articles that follow, Ga2O3 had received little attention as a functional

layer of any kind in silicon based electronics. Gallium oxide had been shown to develop

a signiﬁcant charge density at the interface with strained silicon (a silicon/germanium

alloy) of 6 × 1012 cm−2 [62], however it wasn’t until the independent observations of

capacitance–voltage measurements by Altuntas et al. and this author that a high nega-

tive charge density was reported on metal-insulator-semiconductor (MIS) test structures

fabricated with ALD deposited Ga2O3 on silicon substrates [50],[63],[64]. A reduction of Chapter 3. Surface Passivation by Gallium Oxide 64 the density of interface defects was also ﬁrst reported by the author in [50],[64], together providing further insight into the surface passivation results reported in [65],[66].

In recent years the electrical properties of Ga2O3 – it’s wide bandgap, UV transparency, high electrical ﬁeld strength, oxygen vacancy conductivity, and ability to be doped for n- type conductivity – have inspired investigations into the use of for the fabrication of thin

ﬁlm transistors [60], high temperature gas sensors [67],[68], and solar-blind photodetector applications [69],[70]. Investigations into the use of Ga2O3 in PV applications on cuprous oxide [52],[71], CIGS [72], and dye sensitised absorber materials [73] have also been performed. The articles that follow are the ﬁrst investigations carried out in the area of c-Si solar cells.

One of the ﬁrst PV applications of Ga2O3 was that of Chandiran et al., who used an ALD deposited Ga2O3 tunnel layer as a means of blocking the back reaction of photo-excited electrons between a porous TiO2 scaﬀold and the absorbing dye [73]. The application of

Ga2O3 resulted in a remarkable increase in device open circuit voltage from 690 mV to

1.1 V. The authors later demonstrated that it is the favourable band structure of Ga2O3 and conduction band oﬀset with the TiO2 layer that were of particular signiﬁcance in this context [74].

The application of Ga2O3 to cuprous oxide (Cu2O) followed; the interest stemming from a favourable band alignment of electron collection. In the work of Minami et al., a Ga2O3 electron collecting layer up to 80 – 90 nm thick was shown to provide optimal performance (η ∼ 5.4%) indicative of bulk conduction and a favourable conduction band alignment [75]. Lee et al. would later measure a Voc of 1.2 V on a Ga2O3 passivated

Cu2O device, though, despite a favourable conduction band alignment indicated by Chapter 3. Surface Passivation by Gallium Oxide 65

XPS measurements, the cell suffered from very high series resistance losses measuring an efficiency of < 4% and a fill factor of only 44.7% [52].

Most recently, Ga2O3 was applied to CIGS (Cu(In,Ga)Se2) devices as a cadmium-free

alternative to the state-of-the-art electron collector, CdS. Koida et al. demonstrate

comparable open circuit voltages using up to 80 nm of Ga2O3 in such an arrangement,

though reduced ﬁll factors, a result of larger barrier heights at the CIGS/Ga2O3 interface

compared to CdS, and lower shunt resistances, limited the eﬃciency in this case too [72]. Chapter 3. Surface Passivation by Gallium Oxide 66

3.4 Foreword

The journal papers that follow represent the first attempts to investigate the application of Ga2O3 to c-Si PV applications, with a focus on using Ga2O3 to passivate c-Si surface defects. The first paper, a letter published in Applied Physics Letters, is the first to demonstrate surface passivation by Ga2O3. The investigation used trimethylgallium

(TMG; the gallium analogue of the commonly used Al2O3 precursor, TMA) and ozone

(O3) as reactants in a thermal ALD process. Thermal activation of the passivation, and subsequent de-activation, are studied.

The subsequent papers detail investigations into surface passivation by plasma ALD, owing to the higher deposition rate and improved passivation properties when replacing

O3 with O2 plasma as the oxygen source. From high frequency capacitance–voltage measurements, the surface passivation is shown to be a result of a reduction in the density of surface defects (midgap defect density < 1011 eV−1.cm−2), and the formation of a high density of negative charge in the film (> 1012 cm−2) after annealing, similar to state-of-the-art Al2O3. This is shown to result in an equivalence of surface recombination parameters for Ga2O3 passivated silicon surface, compared to Al2O3, on both undiffused and boron diffused p+ surfaces.

In the third paper presented below the Ga2O3 film is shown to act as a suitable Ga doping source in a laser doping processes. Proof-of-concept p-type cells with Ga-doped partial rear contacts are fabricated with rear Ga2O3 surface passivation. The resulting device efficiency of 19.2% was hampered largely by a low fill factor 74.5% which is attributed to unoptimised laser doping conditions. The open circuit voltage of the device (658 mV) is commensurate with Al laser-doped devices from Al2O3 sources reported in the literature. Chapter 3. Surface Passivation by Gallium Oxide 67

The demonstration of surface passivation is also extended to random pyramid textured surfaces and PECVD silicon nitride capping for applications to the front side of n-type solar cells, for example. Capping with PECVD SiNx is shown to both improve surface

passivation and thermal stability, with initial investigations indicating that the Ga2O3

/ SiNx stack is firing stable. From x-ray diffraction data the SiNx capping layer is proposed to inhibit the crystallisation of Ga2O3 during the firing process. Chapter 3. Surface Passivation by Gallium Oxide 68

3.5 References

[1] W. Shockley and W. T. Read Jr, “Statistics of the recombinations of holes and electrons,” Phys. Rev., vol. 87, no. 5, p. 835, 1952.

[2] R. N. Hall, “Electron-Hole Recombination in Germanium,” Phys. Rev., vol. 87, no. 2, pp. 387–387, Jul. 1952.

[3] A. Cuevas, T. Allen, J. Bullock, Y. Wan, Di, and X. Zhang, “Skin care for healthy silicon solar cells,” in 2015 IEEE 42nd Photovoltaic Specialist Conference (PVSC), New Orleans, USA, 2015.

[4] J. I. Pankove, M. A. Lampert, and M. L. Tarng, “Hydrogenation and dehydrogenation of amorphous and crystalline silicon,” Appl. Phys. Lett., vol. 32, no. 7, pp. 439–441, Apr. 1978.

[5] J. I. Pankove and M. L. Tarng, “Amorphous silicon as a passivant for crystalline silicon,” Appl. Phys. Lett., vol. 34, no. 2, pp. 156–157, Jan. 1979.

[6] E. Cartier, J. H. Stathis, and D. A. Buchanan, “Passivation and depassivation of silicon dangling bonds at the Si/SiO2 interface by atomic hydrogen,” Appl. Phys. Lett., vol. 63, no. 11, pp. 1510–1512, Sep. 1993.

[7] A. G. Aberle, “Overview on SiN surface passivation of crystalline silicon solar cells,” Sol. Energy Mater. Sol. Cells, vol. 65, no. 1–4, pp. 239–248, Jan. 2001.

[8] A. Richter, J. Benick, M. Hermle, and S. W. Glunz, “Reaction kinetics during the thermal activation of the silicon surface passivation with atomic layer deposited Al2O3,” Appl. Phys. Lett., vol. 104, no. 6, p. 61606, Feb. 2014.

[9] G. Dingemans, F. Einsele, W. Beyer, M. C. M. van de Sanden, and W. M. M. Kessels, “Inﬂuence of annealing and Al2O3 properties on the hydrogen-induced passivation of the Si/SiO2 interface,” J. Appl. Phys., vol. 111, no. 9, p. 93713, May 2012.

[10] E. Yablonovitch, D. L. Allara, C. C. Chang, T. Gmitter, and T. B. Bright, “Unusu- ally Low Surface-Recombination Velocity on Silicon and Germanium Surfaces,” Phys. Rev. Lett., vol. 57, no. 2, pp. 249–252, Jul. 1986.

[11] N. E. Grant, K. R. McIntosh, and J. T. Tan, “Evaluation of the Bulk Lifetime of Silicon Wafers by Immersion in Hydroﬂuoric Acid and Illumination,” ECS J. Solid State Sci. Technol., vol. 1, no. 2, pp. P55–P61, Jan. 2012.

[12] J. Bullock, D. Kiriya, N. Grant, A. Azcatl, M. Hettick, T. Kho, P. Phang, H. C. Sio, D. Yan, D. Macdonald, M. A. Quevedo-Lopez, R. M. Wallace, A. Cuevas, and A. Javey, “Superacid Passivation of Crystalline Silicon Surfaces,” ACS Appl. Mater. Interfaces, vol. 8, no. 36, pp. 24205–24211, Sep. 2016.

[13] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, and A. Cuevas, “Improved quanti- tative description of Auger recombination in crystalline silicon,” Phys. Rev. B, vol. 86, no. 16, Oct. 2012. Chapter 3. Surface Passivation by Gallium Oxide 69

[14] J. Lindmayer and J. F. Allison, “The violet cell: An improved silicon solar cell,” Comsat Tech. Rev., vol. 3, no. 1, pp. 151–166, 1973.

[15] J. Schmidt and A. Cuevas, “Electronic properties of light-induced recombination centers in boron-doped Czochralski silicon,” J. Appl. Phys., vol. 86, no. 6, pp. 3175– 3180, Aug. 1999.

[16] K. Bothe and J. Schmidt, “Electronically activated boron-oxygen-related recombination centers in crystalline silicon,” J. Appl. Phys., vol. 99, no. 1, p. 13701, Jan. 2006.

[17] N. E. Grant, F. E. Rougieux, D. Macdonald, J. Bullock, and Y. Wan, “Grown-in defects limiting the bulk lifetime of p-type ﬂoat-zone silicon wafers,” J. Appl. Phys., vol. 117, no. 5, p. 55711, Feb. 2015.

[18] N. E. Grant, V. P. Markevich, J. Mullins, A. R. Peaker, F. Rougieux, and D. Mac- donald, “Thermal activation and deactivation of grown-in defects limiting the lifetime of ﬂoat-zone silicon,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 10, no. 6, pp. 443–447, Jun. 2016.

[19] D. MacDonald, A. Cuevas, K. McIntosh, L. Barbosa, and D. De Ceuster, “Impact of Cr, Fe, Ni, Ti and W surface contamination on diﬀused and oxidised a-type crystalline silicon wafers,” in the 20th European Photovoltaic Solar Energy Conference, Barcelona, Spain, 2005.

[20] J. Schmidt, B. Lim, D. Walter, K. Bothe, S. Gatz, T. Dullweber, and P. P. Altermatt, “Impurity-Related Limitations of Next-Generation Industrial Silicon Solar Cells,” IEEE J. Photovolt., vol. 3, no. 1, pp. 114–118, Jan. 2013.

[21] A. G. Aberle, S. Glunz, and W. Warta, “Impact of illumination level and oxide

parameters on Shockley-Read-Hall recombination at the Si-SiO2 interface,” J. Appl. Phys., vol. 71, no. 9, pp. 4422–4431, May 1992.

[22] R. Hezel, “Plasma Si nitride–A promising dielectric to achieve high-quality silicon MIS/IL solar cells,” J. Appl. Phys., vol. 52, no. 4, p. 3076, 1981.

[23] S. Dauwe, L. Mittelst¨adt,A. Metz, and R. Hezel, “Experimental evidence of parasitic shunting in silicon nitride rear surface passivated solar cells,” Prog. Photovolt. Res. Appl., vol. 10, no. 4, pp. 271–278, Jun. 2002.

[24] R. Hezel and K. Jaeger, “Low-temperature surface passivation of silicon for solar cells,” J. Electrochem. Soc., vol. 136, no. 2, pp. 518–523, 1989.

[25] G. Agostinelli, A. Delabie, P. Vitanov, Z. Alexieva, H. F. W. Dekkers, S. De Wolf, and G. Beaucarne, “Very low surface recombination velocities on p-type silicon wafers passivated with a dielectric with ﬁxed negative charge,” Sol. Energy Mater. Sol. Cells, vol. 90, no. 18–19, pp. 3438–3443, Nov. 2006.

[26] B. Hoex, S. B. S. Heil, E. Langereis, M. C. M. van de Sanden, and W. M. M. Kessels, “Ultralow surface recombination of c-Si substrates passivated by plasma-assisted atomic

layer deposited Al2O3,” Appl. Phys. Lett., vol. 89, no. 4, p. 42112, Jul. 2006. Chapter 3. Surface Passivation by Gallium Oxide 70

[27] G. Dingemans and W. M. M. Kessels, “Status and prospects of Al2O3-based surface passivation schemes for silicon solar cells,” J. Vac. Sci. Technol. Vac. Surf. Films, vol. 30, no. 4, p. 40802, Jul. 2012.

[28] L. E. Black, “New Perspectives on Surface Passivation: Understanding the Si-Al2O3 Interface,” Springer, 2016.

[29] L. E. Black and K. R. McIntosh, “Modeling Recombination at the Si-Al2O3 Inter- face,” IEEE J. Photovolt., vol. 3, no. 3, pp. 936–943, 2013.

[30] L. E. Black, T. Allen, K. R. McIntosh, and A. Cuevas, “Eﬀect of boron concentration on recombination at the p-Si–Al2O3 interface,” J. Appl. Phys., vol. 115, no. 9, p. 93707, Mar. 2014.

[31] L. E. Black, T. Allen, A. Cuevas, K. R. McIntosh, B. Veith, and J. Schmidt, “Ther- mal stability of silicon surface passivation by APCVD Al2O3,” Sol. Energy Mater. Sol. Cells, vol. 120, pp. 339–345, 2014.

[32] P. Saint-Cast, D. Kania, M. Hofmann, J. Benick, J. Rentsch, and R. Preu, “Very low surface recombination velocity on p-type c-Si by high-rate plasma-deposited aluminum oxide,” Appl. Phys. Lett., vol. 95, no. 15, p. 151502, Oct. 2009.

[33] A. Richter, S. Henneck, J. Benick, M. H¨orteis,M. Hermle, and S. W. Glunz, “Fir- ing stable Al2O3/SiNx layer stack passivation for the front side boron emitter of n-type silicon solar cells,” in Proceedings of the 25th European Photovoltaic Solar Energy Con- ference and Exhibition, Valencia, Spain, 2010.

[34] J. Schmidt, B. Veith, and R. Brendel, “Eﬀective surface passivation of crystalline silicon using ultrathin Al2O3 ﬁlms and Al2O3/SiNx stacks,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 3, no. 9, pp. 287–289, Nov. 2009.

[35] A. Fell, E. Franklin, D. Walters, D. C. Suh, and K. Weber, “Laser doping from Al2O3 layers,” in Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, Frankfurt, Germany, 2012.

[36] A. Fell, D. Walter, S. Kluska, E. Franklin, and K. Weber, “Determination of Injection Dependent Recombination Properties of Locally Processed Surface Regions,” Energy Procedia, vol. 38, pp. 22–31, Jan. 2013.

[37] E. Cornagliotti, A. Uruena, B. Hallam, L. Tous, R. Russell, F. Duerinckx, and J. Szlufcik, “Large area p-type PERL cells featuring local p+ BSF formed by laser processing of ALD Al2O3 layers,” Sol. Energy Mater. Sol. Cells, vol. 138, pp. 72–79, Jul. 2015.

[38] T. Trupke, M. A. Green, P. W¨urfel,P. P. Altermatt, A. Wang, J. Zhao, and R. Corkish, “Temperature dependence of the radiative recombination coeﬃcient of intrinsic crystalline silicon,” J. Appl. Phys., vol. 94, no. 8, p. 4930, 2003.

[39] H. T. Nguyen, S. C. Baker-Finch, and D. Macdonald, “Temperature dependence of the radiative recombination coeﬃcient in crystalline silicon from spectral photolumines- cence,” Appl. Phys. Lett., vol. 104, no. 11, p. 112105, Mar. 2014. Chapter 3. Surface Passivation by Gallium Oxide 71

[40] A. Cuevas and D. Macdonald, “Measuring and interpreting the lifetime of silicon wafers,” Sol. Energy, vol. 76, no. 1–3, pp. 255–262, Jan. 2004.

[41] K. R. McIntosh and L. E. Black, “On eﬀective surface recombination parameters,” J. Appl. Phys., vol. 116, no. 1, p. 14503, Jul. 2014.

[42] D. E. Kane and R. M. Swanson, “Measurement of the emitter saturation current by a contactless photoconductivity decay method,” in the 18th IEEE Photovoltaic Specialists Conference, Las Vegas, USA, 1985.

[43] Y. Wan, J. Bullock, and A. Cuevas, “Passivation of c-Si surfaces by ALD tantalum oxide capped with PECVD silicon nitride,” Sol. Energy Mater. Sol. Cells, vol. 142, pp. 42–46, Nov. 2015.

[44] Y. Wan, J. Bullock, and A. Cuevas, “Tantalum oxide/silicon nitride: A negatively charged surface passivation stack for silicon solar cells,” Appl. Phys. Lett., vol. 106, no. 20, p. 201601, May 2015.

[45] B. Liao, B. Hoex, A. G. Aberle, D. Chi, and C. S. Bhatia, “Excellent c-Si surface passivation by low-temperature atomic layer deposited titanium oxide,” Appl. Phys. Lett., vol. 104, no. 25, p. 253903, Jun. 2014.

[46] J. Cui, T. G. Allen, Y. Wan, J. Mckeon, C. Samundsett, D. Yan, X. Zhang, Y. Cui, Y. Chen, P. Verlinden, and A. Cuevas, “Titanium oxide: A re-emerging optical and passivating material for silicon solar cells,” Sol. Energy Mater. Sol. Cells, vol. 158, pp. 115–121, 2016.

[47] G. Krugel, A. Sharma, W. Wolke, J. Rentsch, and R. Preu, “Study of hydrogenated AlN as an anti-reﬂective coating and for the eﬀective surface passivation of silicon,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 7, no. 7, pp. 457–460, Jul. 2013.

[48] D. Zielke, J. H. Petermann, F. Werner, B. Veith, R. Brendel, and J. Schmidt, “Contact passivation in silicon solar cells using atomic-layer-deposited aluminum oxide layers,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 5, no. 8, pp. 298–300, Aug. 2011.

[49] R. Peibst, U. Römer,Y. Larionova, M. Rienäcker, A. Merkle, N. Folchert, S. Reiter, M. Turcu, B. Min, J. Krügener,D. Tetzlaff, E. Bugiel, T. Wietler, and R. Brendel, “Working principle of carrier selective poly-Si/c-Si junctions: Is tunnelling the whole story?,” Sol. Energy Mater. Sol. Cells, vol. 158, pp. 60–67, Dec. 2016.

[50] T. G. Allen, M. Ernst, C. Samundsett, and A. Cuevas, “Demonstration of c-Si Solar Cells With Gallium Oxide Surface Passivation and Laser-Doped Gallium p+ Regions,” IEEE J. Photovolt., vol. 5, no. 6, pp. 1586–1590, Nov. 2015.

[51] F. K. Shan, G. X. Liu, W. J. Lee, G. H. Lee, I. S. Kim, and B. C. Shin, “Structural, electrical, and optical properties of transparent gallium oxide thin ﬁlms grown by plasma- enhanced atomic layer deposition,” J. Appl. Phys., vol. 98, no. 2, p. 23504, Jul. 2005.

[52] Y. S. Lee, D. Chua, R. E. Brandt, S. C. Siah, J. V. Li, J. P. Mailoa, S. W. Lee, R. G. Gordon, and T. Buonassisi, “Atomic Layer Deposited Gallium Oxide Buﬀer Layer Chapter 3. Surface Passivation by Gallium Oxide 72

Enables 1.2 V Open-Circuit Voltage in Cuprous Oxide Solar Cells,” Adv. Mater., vol. 26, no. 27, pp. 4704–4710, Jul. 2014.

[53] D. J. Comstock and J. W. Elam, “Atomic Layer Deposition of Ga2O3 Films Using Trimethylgallium and Ozone,” Chem. Mater., vol. 24, no. 21, pp. 4011–4018, Nov. 2012.

[54] N. Ueda, H. Hosono, R. Waseda, and H. Kawazoe, “Synthesis and control of conductivity of ultraviolet transmitting β-Ga2O3 single crystals,” Appl. Phys. Lett., vol. 70, no. 26, p. 3561, 1997.

[55] E. G. Villora, K. Shimamura, Y. Yoshikawa, T. Ujiie, and K. Aoki, “Electrical conductivity and carrier concentration control in β-Ga2O3 by Si doping,” Appl. Phys. Lett., vol. 92, no. 20, p. 202120, May 2008.

[56] M. Orita, H. Hiramatsu, H. Ohta, M. Hirano, and H. Hosono, “Preparation of highly conductive, deep ultraviolet transparent β-Ga2O3 thin ﬁlm at low deposition temperatures,” Thin Solid Films, vol. 411, no. 1, pp. 134–139, May 2002.

[57] N. Suzuki, S. Ohira, M. Tanaka, T. Sugawara, K. Nakajima, and T. Shishido,

“Fabrication and characterization of transparent conductive Sn-doped β-Ga2O3 single crystal,” Phys. Status Solidi C, vol. 4, no. 7, pp. 2310–2313, Jun. 2007.

[58] J. B. Varley, J. R. Weber, A. Janotti, and C. G. Van de Walle, “Oxygen vacancies and donor impurities in β-Ga2O3,” Appl. Phys. Lett., vol. 97, no. 14, p. 142106, Oct. 2010.

[59] S. C. Siah, R. E. Brandt, K. Lim, L. T. Schelhas, R. Jaramillo, M. D. Heinemann, D. Chua, J. Wright, J. D. Perkins, C. U. Segre, R. G. Gordon, and T. Buonassisi, “Dopant activation in Sn-doped Ga2O3 investigated by X-ray absorption spectroscopy,” Appl. Phys. Lett., vol. 107, no. 25, p. 252103, Dec. 2015.

[60] S. R. Thomas, G. Adamopoulos, Y.-H. Lin, H. Faber, L. Sygellou, E. Stratakis, N. Pliatsikas, P. A. Patsalas, and T. D. Anthopoulos, “High electron mobility thin-

ﬁlm transistors based on Ga2O3 grown by atmospheric ultrasonic spray pyrolysis at low temperatures,” Appl. Phys. Lett., vol. 105, no. 9, p. 92105, Sep. 2014.

[61] M. D. Heinemann, J. Berry, G. Teeter, T. Unold, and D. Ginley, “Oxygen deﬁciency and Sn doping of amorphous Ga2O3,” Appl. Phys. Lett., vol. 108, no. 2, p. 22107, Jan. 2016.

[62] S. Pal, S. K. Ray, B. R. Chakraborty, S. K. Lahiri, and D. N. Bose, “Gd2O3, Ga2O3(Gd2O3), Y2O3, and Ga2O3, as high-k gate dielectrics on SiGe: A comparative study,” J. Appl. Phys., vol. 90, no. 8, pp. 4103–4107, Oct. 2001.

[63] H. Altuntas, I. Donmez, C. Ozgit-Akgun, and N. Biyikli, “Eﬀect of postdeposition annealing on the electrical properties of β-Ga2O3 thin ﬁlms grown on p-Si by plasma- enhanced atomic layer deposition,” J. Vac. Sci. Technol. Vac. Surf. Films, vol. 32, no. 4, p. 41504, Jul. 2014. Chapter 3. Surface Passivation by Gallium Oxide 73

[64] T. G. Allen and A. Cuevas, “Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: demonstration of surface passivation and negative interfacial charge,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 9, no. 4, pp. 220–224, Apr. 2015.

[65] T. G. Allen and A. Cuevas, “Electronic passivation of silicon surfaces by thin ﬁlms of atomic layer deposited gallium oxide,” Appl. Phys. Lett., vol. 105, no. 3, p. 31601, Jul. 2014.

[66] T. G. Allen, Y. Wan, and A. Cuevas, “Silicon Surface Passivation by Gallium Oxide Capped With Silicon Nitride,” IEEE J. Photovolt., vol. 6, no. 4, pp. 900–905, 2016.

[67] M. Fleischer and H. Meixner, “Gallium oxide thin ﬁlms: A new material for high- temperature oxygen sensors,” Sens. Actuators B Chem., vol. 4, no. 3, pp. 437–441, Jun. 1991.

[68] M. Ogita, N. Saika, Y. Nakanishi, and Y. Hatanaka, “Ga2O3 thin ﬁlms for high- temperature gas sensors,” Appl. Surf. Sci., vol. 142, no. 1–4, pp. 188–191, Apr. 1999.

[69] D. Guo, Z. Wu, P. Li, Y. An, H. Liu, X. Guo, H. Yan, G. Wang, C. Sun, L. Li, and

W. Tang, “Fabrication of β-Ga2O3 thin ﬁlms and solar-blind photodetectors by laser MBE technology,” Opt. Mater. Express, vol. 4, no. 5, pp. 1067–1076, May 2014.

[70] Y. Kokubun, K. Miura, F. Endo, and S. Nakagomi, “Sol-gel prepared β-Ga2O3 thin ﬁlms for ultraviolet photodetectors,” Appl. Phys. Lett., vol. 90, no. 3, p. 31912, Jan. 2007.

[71] R. E. Brandt, M. Young, H. H. Park, A. Dameron, D. Chua, Y. S. Lee, G. Teeter, R. G. Gordon, and T. Buonassisi, “Band oﬀsets of n-type electron-selective contacts on

cuprous oxide (Cu2O) for photovoltaics,” Appl. Phys. Lett., vol. 105, no. 26, p. 263901, Dec. 2014.

[72] T. Koida, Y. Kamikawa-Shimizu, A. Yamada, H. Shibata, and S. Niki, “Cu(In,Ga)Se2 Solar Cells With Amorphous Oxide Semiconducting Buﬀer Layers,” IEEE J. Photovolt., vol. 5, no. 3, pp. 956–961, 2015.

[73] A. K. Chandiran, N. Tetreault, R. Humphry-Baker, F. Kessler, E. Baranoﬀ, C.

Yi, M. K. Nazeeruddin, and M. Gr¨atzel,“Subnanometer Ga2O3 Tunnelling Layer by Atomic Layer Deposition to Achieve 1.1 V Open-Circuit Potential in Dye-Sensitized Solar Cells,” Nano Lett., vol. 12, no. 8, pp. 3941–3947, Aug. 2012.

[74] A. K. Chandiran, M. K. Nazeeruddin, and M. Gr¨atzel,“The Role of Insulating Oxides in Blocking the Charge Carrier Recombination in Dye-Sensitized Solar Cells,” Adv. Funct. Mater., vol. 24, no. 11, pp. 1615–1623, Mar. 2014.

[75] T. Minami, Y. Nishi, and T. Miyata, “High-eﬃciency Cu2O-based heterojunction solar cells fabricated using a Ga2O3 thin ﬁlm as n-type layer,” Appl. Phys. Express, vol. 6, no. 4, p. 44101, 2013.

Electronic passivation of silicon surfaces by thin ﬁlms of atomic layer deposited gallium oxide

Thomas Allen1 and Andr´esCuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

Published in Applied Physics Letters, volume 105, issue 3, 2014.

Abstract — This paper proposes the application of gallium oxide (Ga2O3) thin films to crystalline silicon solar cells. Effective passivation of n- and p-type crystalline silicon surfaces has been achieved by the application of very thin Ga2O3 films prepared by atomic layer deposition (ALD) using trimethylgallium (TMGa) and ozone (O3) as the reactants. Surface recombination velocities as low as 6.1 cm/s have been recorded with films less than

4.5 nm thick. A range of deposition parameters has been explored, with growth rates of approximately 0.2 A/cycle˚ providing optimum passivation.

75 Chapter 3. Surface Passivation by Gallium Oxide 76

The thermal activation energy for passivation of the Si–Ga2O3 interface has been found to be approximately 0.5 eV. Depassivation of the interface was observed for prolonged annealing at increased temperatures. The activation energy for depassivation was measured to be 1.9 eV.

Index Terms — surface passivation, Ga2O3, ozone, atomic layer deposition, activation energy. Chapter 3. Surface Passivation by Gallium Oxide 77

I. Introduction

Recombination of charge carriers at the silicon surface, especially beneath the metallised contact regions, remains one of the biggest loss mechanisms in crystalline silicon (c-Si) solar cells [1]. An array of dielectric materials have been utilised for their capacity to suppress such defect-assisted Shockley-Read-Hall recombination, most notably silicon dioxide (SiO2), silicon nitride, hydrogenated amorphous silicon (a-Si:H) and aluminium oxide (Al2O3). Other dielectrics, like hafnium oxide and amorphous silicon carbide have also been shown to provide chemical and charge-induced passivation of recombination active surface states.

Gallium oxide (Ga2O3) is a wide bandgap (Eg ≈ 4.23–5.24 eV), optically transparent semiconductor [2]. Though typically non-conductive in its stoichiometric form, Ga2O3

can be doped with either silicon or tin for n–type conductivity [3]. Gallium, like alu-

minium, is also a p–type dopant in silicon, and so Ga2O3 has the potential to be used as a p–type dopant source via thermal diﬀusion or laser processing. Additionally, the energy band structure of Ga2O3 suggests that it could be utilised as a carrier-selective contact to crystalline silicon [4].

These properties have prompted research into the application of Ga2O3 to photovoltaics.

Chandiran et al. [5] have applied Ga2O3 to a dye-sensitized solar cell as an electron tun-

nelling interlayer between the porous, electron-collecting TiO2 layer and the absorbing

dye. The application of Ga2O3 reduced the electron back reaction into the electrolyte,

resulting in an increase in the open circuit voltage (Voc) of the device. Recently both Mi-

nami et al. [6] and Lee et al. [7] used Ga2O3 as a buﬀer layer between the p–type Cu2O

and electron collecting ZnO:Al in cupric–based heterojunction thin ﬁlm solar cells. The Chapter 3. Surface Passivation by Gallium Oxide 78

application of the Ga2O3 buﬀer layer to the cupric–AZO interface reduced the recombination of charge carriers while permitting electron transfer to the negative terminal of the device, thereby enhancing Voc.

Despite these demonstrated qualities, and gallium oxide’s known ability to passivate

GaAs [8] and reduce the density of surface defects in strained silicon [9], no attempt has yet been made to study this material on c-Si for photovoltaic applications. In this Letter we provide evidence of surface passivation of crystalline silicon by Ga2O3 prepared by atomic layer deposition (ALD).

II. Experimental Procedure

Ga2O3 thin ﬁlms were prepared by ALD on 4–5 Ω.cm n-type and 2–3 Ω.cm p-type

FZ h100i silicon in a Beneq TFS 200 ALD reactor with trimethylgallium (TMGa) and ozone (O3) used as the reactants. A 200 µm diameter oriﬁce was installed between the precursor and reactor to limit the precursor dose. All of the silicon samples were saw-damage etched in TMAH, cleaned in the standard RCA process, and immersed in a dilute HF solution prior to deposition. Variable angle spectroscopic ellipsometry (J.A.

Woolam M2000D) was performed on polished samples that were also immersed in HF prior to Ga2O3 deposition in order to measure the thickness of the deposited ﬁlms. A

Tauc-Lorentz model was used to ﬁt the ellipsometry data.

The eﬀective minority carrier lifetime (τeff ) was measured as a function of the excess carrier concentration (∆n) on symmetrically deposited lifetime structures using a Sinton

Instruments WCT-120 photoconductance tool operating in the photoconductance decay

(PCD) transient mode for the annealed samples, and in the quasi-steady-state (QSS) generalized mode for the as-deposited samples. All lifetime samples have been annealed Chapter 3. Surface Passivation by Gallium Oxide 79

in a forming gas ambient composed of 5% H2 / 95% Ar after initial comparisons with

N2 annealing resulted in a considerable improvement in τeff .

III. Results and Discussion

◦ Figure1 plots the as-deposited Ga 2O3 ﬁlm thickness and post-anneal (400 C for 15

minutes in forming gas) τeff against deposition temperature. All depositions comprised

200 ALD cycles. The range in deposition rate (0.13–0.62 A/cycle)˚ is slightly greater

than that reported by Comstock and Elam (0.2–0.5 A/cycle)˚ over the same temperature

range [10]. From Figure 1, an observable optimum deposition temperature window for

surface passivation exists between 200 and 250 ◦C with the maximum in minority carrier

lifetime corresponding to Ga2O3 ﬁlms as thin as 2.85 nm. The deposition rate in this

window ranges from 0.14 to 0.22 A/cycle,˚ approximately a quarter of the deposition

◦ rate at 400 C (0.63 A/cycle)˚ and much lower than ALD Al2O3 using TMA at similar

temperatures (approximately 1–1.5 A/cycle).˚ Donmez et al. have reported a steadier

growth rate using TMGa and O2 plasma with a temperature-independent growth rate

of 0.53 A/cycle˚ reported in the deposition window of 100–400 ◦C [11]. Dezlah et al. also

report on atomic layer deposited Ga2O3 with Ga2(NMe2)6 and H2O with growth rates

in excess of 1 A/cycle˚ for deposition temperatures between 150 and 200 ◦C: similar to

TMA-H2O Al2O3 thermal ALD deposition rates [12].

A deposition temperature of 250 ◦C was used for further study of the annealing param-

eters because of its favourable initial passivation results and growth rate. Float zone,

4–5 Ω.cm n-type samples deposited at 250 ◦C were annealed in forming gas at a range of

temperatures between 300 and 450 ◦C in order to derive the optimum anneal conditions.

The results are plotted in Figure2. Chapter 3. Surface Passivation by Gallium Oxide 80

1 6 0 0 1 4 τ n - t y p e , F Z , 4 - 5 Ω. c m , < 1 0 0 > e f f f i l m t h i c k n e s s 1 4 0 0

)

1 2 s µ ( )

f f m e ؔ n

( 1 2 0 0

s 1 0 m s i t e e n f i k l

c 1 0 0 0 i r h 8 e i t

r 3 r a O c 2

8 0 0 y t i

G 6 r o n i 4 6 0 0 M

2 4 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0 4 0 0 D e p o s i t i o n t e m p e r a t u r e ( o C )

Fig. 1: As-deposited Ga2O3 ﬁlm thickness and post-anneal minority carrier lifetime (τeff ) on 4–5 Ω.cm n-type silicon plotted as a function of deposition temperature. All depositions comprised 200 ALD cycles. The lines serve as guides for the eyes.

2 5 0 0 n - t y p e , F Z , 4 - 5 Ω. c m , < 1 0 0 > 3 0 0 o C

) o

s 3 5 0 C µ

( 2 0 0 0 o

f 3 7 5 C e o ؔ 4 0 0 C e o m i

t 1 5 0 0 4 2 5 C e

f o i l

4 5 0 C r e i r

r 1 0 0 0 a c

y t i r

o 5 0 0 n i M

0 0 1 0 2 1 0 3 1 0 4 C u m u l a t i v e a n n e a l t i m e t ( s ) a n

Fig. 2: Minority carrier lifetime as a function of cumulative annealing time. The dashed curves represent the modelled data using Eq. (1). The as-deposited data is shown at tan = 0. Chapter 3. Surface Passivation by Gallium Oxide 81

The x-axis in Figure2 denotes the cumulative anneal time tan for a double-side deposited

15 sample; the y-axis represents the τeff measured at an injection level of ∆n = 1 × 10 cm−3. The optimum annealing condition — 180 min at 350 ◦C — resulted in a minority carrier lifetime of approximately 2.5 ms and an upper limit to the surface recombination velocity (SRV ) of 6.1 cm/s, where the Auger parameterization of Richter et al. [13] was used to calculate the bulk lifetime. This compares to a τeff of 9.5 ms and an SRV of

1.5 cm/s for a similar sample passivated with 20 nm of plasma-assisted ALD Al2O3 and

◦ annealed in forming gas for 30 minutes at 425 C. Spatially resolved τeff maps for both n- and p-type Ga2O3 passivated samples are displayed in Figure3.

From Figure2, it is evident that the rate of activation of the passivation is increasing with temperature. The optimum annealing time (tan) for each annealing temperature

(Tan) also results in an increase in the maximum observed τeff as Tan decreases, with

◦ the exception of Tan = 300 C where a maximum in τeff had not been reached after

over 9 hours of annealing. A similar eﬀect is observable for the reverse, depassivation

reaction for temperatures greater than 300 ◦C: as the annealing temperature increases,

the rate at which the interface depassivates also increases. After the initial activation

of the passivation, the lifetime decreases to an equilibrium value (approximately 100

µs), greater than the as-deposited lifetime (approximately 10 µs), indicating that some

passivation of the interface is retained.

An exponential model has been applied to ﬁt the τeff (tan,Tan) data to further analyse

the passivation and depassivation reactions. The model is based on that of Mitchell et

al., who applied an exponential ﬁt to temperature and time dependent τeff data from a-

Si:H-passivated silicon surfaces [14], and Dingemans et al., who applied the same model

to SiO2–Al2O3 stacks on c-Si [15]. The exponential model has been extended to account

for both the passivation and depassivation of the Si–Ga2O3 interface Chapter 3. Surface Passivation by Gallium Oxide 82

Fig. 3: Spatially resolved minority carrier lifetime maps from PL images of a) Ga2O3- passivated 4–5 Ω.cm n-type and b) 2–3 Ω.cm p-type silicon, both annealed at 375 ◦C for 50 min in forming gas. The images reveal a uniform passivation across the wafer surface. Note the diﬀerence in scale between the two images.

1 = [a + b exp(−tan/Rpass)] + [c − d exp(−tan/Rdepass)] (1) τeff where the ﬁrst bracketed term represents the exponential ﬁtting to the passivation data, and the second bracketed term represents the exponential decay associated with the depassivation data. The parameters Rpass and Rdepass are the time constants of the passivation and depassivation reactions. These reaction time constants are then used to extract the activation energies for both the passivation and depassivation reactions according to the Arrhenius equation

EA R(de)pass = A exp(− ) (2) kBTan

The dashed curves in Figure2 represent the summed exponential model of Eq. (1) applied to the measured values of τeff . The divergence of the model from the data in

Figure2 can be ascribed to a number of factors. The use of a single sample for each

Tan study introduces some uncertainty in τeff , resulting in an artiﬁcial stretching of the Chapter 3. Surface Passivation by Gallium Oxide 83 curves along the x-axis and compression in the y-axis. Also, if the mechanism for passivation involves the interaction with another chemical species (hydrogen, for example, is commonly identiﬁed as a passivating mechanism for surface defects in crystalline silicon

[14-17]), then the availability of that species is not accounted for in the model. The results, for example, are consistent with the progressive eﬀusion of hydrogen from the

Si–Ga2O3 interface with increasing annealing temperature and time after the observed peak in τeff . This hypothesis of a reaction mechanism that involves a limited and di- minishing source of one species may also account for the better ﬁtting of the model to the data for annealing times after the peak in passivation is achieved. Despite these possible limitations, the model does provide a reasonable ﬁt to the lifetime data.

1 0 0 2 . 5 p a s s i v a t i o n - 1 2 . 0 1 0 d e p a s s i v a t i o n ) V

e 1 . 5 ( )

1 - A 1 . 0 s

- 2 E (

1 0 s

s 0 . 5 a p )

e 0 . 0

d - 3 ( 1 0 R

e t

a - 4 r

1 0 n o i t

c - 5

a 1 0 e R 1 0 - 6

1 0 - 7 1 6 1 8 2 0 2 2 1 / k T ( e V - 1 ) B a n

Fig. 4: Arrhenius plot of the reaction rate (R(de)pass) extracted from Eq. (1). The error bars in R represent the upper and lower bounds in reaction rate values that give a credible ﬁt to the data in Figure2. The error bars in the extracted activation energies are the values of EA extracted from the bounds in possible R values.

The Arrhenius plots of Figure4 display the extracted reaction time constants of Eq.

(1) as a function of inverse energy. The plotted symbols in Figure4 give the best fits to both the τeff data and the Arrhenius model of Eq. (2). The error bars in the plot Chapter 3. Surface Passivation by Gallium Oxide 84 represent the upper and lower bounds in reaction rate values that still give a credible fit to the data in Figure2, and so they are not directly representative of the error in the reaction rate, which is difficult to quantify given the reason outlined in the discussion above.

The application of Eq. (2) to the reaction rates of Eq. (1) indicate an activation energy

EA of approximately 0.5 eV for the passivation reaction and 1.9 eV for the depassivation reaction. The error bars in the inset graph of EA in Figure4 correspond to the activation energies extracted from the error bar values for the reaction rates. Therefore, they are not indicative of the error in the ﬁts of Eq. (2) to the plotted data in the ﬁgure, which is small, but rather represent a range of possible, albeit less likely, values for EA.

The approximate activation energy for passivation extracted from the data in Figure4

(0.5 eV) is similar to the values reported by Mitchell et al. [14] (0.67 eV) and de Wolf and Kondo [16] (0.6 eV) for the activation of a-Si:H; and outside the range reported by Richter et al. [17] (1.4-1.5 eV) and Saint-Cast [18] (1.5 eV) for Al2O3 ﬁlms, and

Dingemans et al. [15] (0.9 and 1.2 eV) for Al2O3–SiO2 stacks. Mitchell et al. suggest that the lower activation energies for a-Si:H indicate an interfacial hydrogen dependency on passivation, rather than a bulk hydrogen eﬀusion process [14]. Conversely, Richter et al. claim that the higher activation energy for Al2O3 passivation relates to both the existence of an interfacial SiO2 layer, citing the similar activation energy of Stesmans

[19] for the Si–SiO2 interface, and the diﬀusion of hydrogen from the bulk of the Al2O3

film. The diffusion of bulk hydrogen from Al2O3 films to passivate the Si–SiO2 interface is further demonstrated in Ref. 15. Richter et al. further attribute the passivation of the interface to structural changes in the Al2O3 film which could be a factor in the annealing behaviour of the Ga2O3 films [17]. Chapter 3. Surface Passivation by Gallium Oxide 85

For the Ga2O3 passivation studied in this work, uncertainty in EA and the as-yet un- known properties of the material itself, compared to the other ﬁlms studied using this technique, does not support such strong conclusions regarding the nature of the passivation mechanism. What can be concluded from the data is that the diﬀerence in the passivation and depassivation activation energies is not as great as for a-Si:H, Al2O3

or SiO2 passivation dielectrics. This could have implications relating to the stability of

the passivation, and for the potential thermal processing windows if Ga2O3 were to be

incorporated into a photovoltaic device.

IV. Conclusion

Very thin ﬁlms of ALD deposited gallium oxide (Ga2O3) have been shown to eﬀectively

passivate n- and p-type silicon surfaces, with an optimum post-deposition annealing

condition (350 ◦C for 180 min) resulting in a minority carrier lifetime of 2.5 ms on 4–5

Ω.cm n-type silicon. Photoluminscence (PL) imaging of the passivated wafers show a

uniformity in passivation across the 4 in. wafer surface. The annealing temperature

and time dependence of τeff has been modelled using a sum of two exponentials in

order to calculate the activation energies for the passivation and depassivation reactions:

approximately 0.5 and 1.9 eV, respectively.

Acknowledgements

The authors would like to thank Dr D. Macdonald for his many suggestions and help in

preparing the manuscript. This work has been supported by the Australian government Chapter 3. Surface Passivation by Gallium Oxide 86 through the Australian Renewable Energy Agency (ARENA). Ellipsometer facilities at the Australian National Fabrication Facility were used in this work. Chapter 3. Surface Passivation by Gallium Oxide 87

References

[1] R. M. Swanson in Proceedings of the 31st IEEE Photovoltaic Specialists Conference, 2005.

[2] F. K. Shan, G. X. Liu, W. J. Lee, G. H. Lee, I. S. Kim, and B. C. Shin, J. Appl. Phys. 98, 023504 (2005).

[3] S. Ohira, N. Suzuki, N. Arai, M. Tanaka, T. Sugawara, K. Nakajima, and T. Shishido, Thin Solid Films 516, 5763 (2008).

[4] J. Robertson, Thin Solid Films 516, 1419 (2008).

[5] A. K. Chandiran, N. Tetreault, R. Humphry-Baker, F. Kessler, E. Baranoﬀ, C. Yi, M. K. Nazeeruddin, and M. Grtzel, Nano Letters 12, 3941 (2012).

[6] T. Minami, T. Miyata, and Y. Nishi, Solar Energy 105, 206 (2014).

[7] Y. S. Lee, D. Chua, R. E. Brandt, S. C. Siah, J. V. Li, J. P. Mailoa, S. W. Lee, R. G. Gordon, and T. Buonassisi, Adv. Mat. 26(27), 4704?4710 (2014).

[8] M. Passlack, M. Hong, E. F. Schubert, J. R. Kwo, J. P. Mannaerts, S. N. G. Chu, N. Moriya, and F. A. Thiel, Appl. Phys. Lett. 66, 625 (1995).

[9] S. Pal, S. K. Ray, B. R. Chakraborty, S. K. Lahiri, and D. N. Bose, J. Appl. Phys. 90, 4103 (2001).

[10] D. J. Comstock and J. W. Elam, Chem. Mater. 24, 4011 (2012).

[11] I. Donmez, C. Ozgit-Akgun, and N. Biyikli, J. Vac. Sci. Technol., A 31, 01A110 (2013).

[12] C. L. Dezlah IV, J. Niinisto, K. Arstila, L. Niinisto, and C. H. Winter, Chem. Mater. 18, 471 (2006).

[13] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, and A. Cuevas, Phys. Rev. B 86, 165202 (2012).

[14] J. Mitchell, D. Macdonald, and A. Cuevas, Appl. Phys. Lett. 94, 162102 (2009).

[15] G. Dingemans, F. Einsele, W. Beyer, M. C. M. van de Sanden, and W. M. M. Kessels, J. Appl. Phys. 111, 093713 (2012).

[16] S. de Wolf and M. Kondo, Appl. Phys. Lett. 90, 042111 (2007).

[17] A. Richter, J. Benick, M. Hermle, and S. W. Glunz, Appl. Phys. Lett. 104, 061606 (2014).

[18] P. Saint-Cast, Ph.D. dissertation, University of Konstanz, 2012.

[19] A. Stesmans, Appl. Phys. Lett. 68, 2076 (1996).

Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: demonstration of surface passivation and negative interfacial charge

Thomas Allen1 and Andr´esCuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

Published in Physica Status Solidi – Rapid Research Letters, volume 9,

issue 4, 2015.

Abstract — Herein we report on the passivation of crystalline silicon by gallium oxide (Ga2O3) using oxygen plasma as the oxidizing reactant in an atomic layer deposition (ALD) process. Excess carrier lifetimes of 2.1 ms 89 Chapter 3. Surface Passivation by Gallium Oxide 90 have been measured on 1.75 Ω.cm p-type silicon, from which a surface re-

−2 combination current density J0 of 7 fA.cm is extracted. From high frequency capacitance-voltage (HF CV) measurements it is shown that, as in the case of Al2O3, the presence of a high negative charge density Qtot/q of up to −6.2 × 1012 cm−2 is one factor contributing to the passivation of silicon by

11 −1 −2 Ga2O3. Defect densities at midgap on the order of ≈ 5 × 10 eV .cm are extracted from the HF CV data on samples annealed at 300 ◦C for 30 minutes in a H2 / Ar ambient, representing an order of magnitude reduction in the defect density compared to pre-anneal data. Passivation of a boron-diﬀused

+ −2 p surface (96 Ω/) is also demonstrated, resulting in a J0 of 52 fA.cm .

Index Terms — surface passivation, Ga2O3, photovoltaics, silicon, atomic layer deposition. Chapter 3. Surface Passivation by Gallium Oxide 91

I. Introduction

Since the ﬁrst authoritative demonstration of the passivation of defects at the crystalline silicon (c-Si) surface by aluminium oxide (Al2O3) by Hezel and Jaeger [1], the application

of Al2O3 in c-Si based photovoltaic devices to passivate defects on p-type silicon and

boron-diﬀused p+ silicon surfaces has been extensive [2]. Prior to the work of Hezel

and Jaeger (and indeed for some years after it) the passivation of p-type silicon surfaces

was typically achieved using materials with a negligible interfacial charge density, like

SiO2. The advent of surface passivation by Al2O3, with its high density of negative

ﬁxed charge, resulted in both the formation of an induced hole-accumulation layer at

the p-type silicon surface and a low density of interfacial defects.

+ To date few alternatives to Al2O3 have been proposed to passivate p-type and p crys-

talline silicon surfaces due to a paucity of materials that provide both a high negative

interfacial charge density and a signiﬁcant reduction in the interfacial defect density

when applied to c-Si. In this study we report for the ﬁrst time on the passivation of

+ p-type and p crystalline silicon surfaces by gallium oxide (Ga2O3) deposited by plasma

enhanced atomic layer deposition (PE-ALD). In addition, we provide the ﬁrst evidence of

the presence of a negative charge density Qtot/q at the amorphous Ga2O3–Si interface of

up to −6.2×1012 cm−2 and a reduction in the density of interfacial defects to ≈ 5×1011

−1 −2 eV .cm at midgap. As a consequence, Ga2O3 is shown to eﬀectively passivate base

+ resistivity p-type silicon and boron-diﬀused p silicon surfaces, indicating that Ga2O3 is

an alternative passivation material to Al2O3 in crystalline silicon photovoltaic devices.

In addition to a high negative ﬁxed charge density, Ga2O3, unlike Al2O3, can be inten-

tionally doped with a variety of impurities for n-type conductivity. Tin, for example, has

been shown to increase the conductivity of doped Ga2O3 to a range spanning from less Chapter 3. Surface Passivation by Gallium Oxide 92 than 0.1 S.cm−1 to over 8.2 S.cm−1, with a strong dependence on annealing temperature and crystal structure [3]. As such, Ga2O3 has the potential to replace other conductive layers used in photovoltaic devices, like doped amorphous silicon and tin-doped indium oxide (ITO). Gallium oxide is also a wide bandgap, optically transparent material with a reported range in bandgap energies of 4.23 eV to 5.24 eV [4]. The magnitude of the bandgap is advantageous for optoelectronic devices like solar cells or photodetectors that interface with the UV–visible spectrum. Indeed Ga2O3 has recently been utilised as a transparent buﬀer layer in dye sensitised solar cells [5], Cu2O/ZnO heterojunction cells [6, 7], and a cadmium-free alternative to CdS in CIGS devices [8], resulting in an increase in open-circuit voltage (Voc) in each case.

II. Experimental Procedure

Gallium oxide ﬁlms (∼20 nm) were deposited on p-type, 1.75 Ω.cm, ﬂoat zone (FZ), h100i silicon substrates in a Beneq TFS 200 ALD reactor, with trimethyl-gallium (TMGa) and oxygen plasma as the reactants. The excess carrier lifetime τeff of samples symmetrically passivated with Ga2O3 was measured as a function of excess carrier concentration ∆n using a Sinton Instruments WCT-120 tool operating in the photoconductance decay

(PCD) transient mode. All lifetime samples were gettered of bulk contaminants in a phosphorus diﬀusion process before being surface damage etched in TMAH, cleaned in the standard RCA process and immersed in HF prior to Ga2O3 deposition. After deposition in the ALD reactor, all samples were annealed in a quartz tube furnace in a forming gas ambient composed of 5% H2 in Ar. Fourier transform infrared spectroscopy

(FTIR, Bruker Vertex 80 V) was performed on separately prepared polished samples at a resolution of 7 cm−1 to analyse the ﬁlm composition and structure. Chapter 3. Surface Passivation by Gallium Oxide 93

High frequency capacitance-voltage (HF CV) measurements were carried out on metal- insulator-semiconductor (MIS) samples prepared on single side polished Ω.cm, p-type,

FZ h100i silicon. The top (polished side) contact of the CV samples was formed by thermally evaporating 100 nm of aluminium directly on the Ga2O3 coated silicon surface, while a Ga/In eutectic paste formed the rear contact with the silicon base. AC parallel capacitance measurements were made using a HP 4284A Precision LCR Meter at a frequency of 1 MHz. Low frequency quasi-static CV measurements were attempted on the same samples but could not be taken due to high leakage currents through the Ga2O3

at lower frequencies.

III. Results and Discussion

The use of oxygen plasma in lieu of ozone in the ALD process resulted in an increase

in the deposition rate from a previously reported maximum of 0.2 A/cycle˚ [9] to a

temperature independent deposition rate of ∼0.7 A/cycle˚ in the temperature range

studied (50–150 ◦C), higher than that of Donmez et al. (0.53 A/cycle)˚ using identical

precursors [10]. Even higher rates of deposition (∼1 A/cycle)˚ have been reported by

Dezlah et al. using a dialkylamido-based precursor in an ALD process [11]. An optical

bandgap of ∼4.6 eV was determined by ellipsometry for the PE-ALD deposited ﬁlms.

◦ FTIR spectra of a Ga2O3 sample deposited at 75 C, before and after annealing in

forming gas, are shown in Fig.1. They feature a broad peak between 400 cm −1 and

800 cm−1 (labelled (a) in the ﬁgure), characteristic of amorphous metal oxide ﬁlms, and

similar to that reported by Oritz et al. for amorphous Ga2O3 [12]. After annealing, the

relative intensity of the peak increases and broadens, indicating a structural change in

the Ga–O bonding arrangements within the ﬁlm while remaining representative of an Chapter 3. Surface Passivation by Gallium Oxide 94

◦ Fig. 1: Infrared absorption spectra of a PE-ALD Ga2O3 sample deposited at 75 C before and after annealing in forming gas at 300 ◦C for 30 minutes. All data have been smoothed by applying the Savitsky-Golay algorithm over 6 adjacent data points. amorphous structure, lacking the well-deﬁned peaks at 450 cm−1 and 670 cm−1 indicative of the crystalline β-phase Ga2O3 [12]. Indeed, there is no data in the literature that indicates the formation of a crystalline phase of Ga2O3 at an annealing temperature of 300 ◦C, though phase transitions have been shown to occur at temperatures as low as ∼380 ◦C [13]. A smaller peak at 1000 cm−1 (labelled (b) in the ﬁgure) in both the pre- and post-anneal data indicates the presence of a sub-stoichiometric SiOx interfacial layer [14], similar to other metal oxides deposited by ALD [15, 16]. The absence of any discernible features for both the pre- and post-anneal data between 1400 cm−1 and

1700 cm−1 (band (c) in the ﬁgure) suggests that there is little residual carbon in the

film. Carbon contamination is common for Al2O3 ALD films, and is a remanent from the organic methyl (CH3) ligands common to the trimethylaluminium and trimethylgallium precursors [17]. Altuntas et al. also report negligible residual carbon in similarly prepared PE-ALD Ga2O3 films, as determined by sputter depth profilometry [18]. A Chapter 3. Surface Passivation by Gallium Oxide 95 small, broad feature in the pre-annealed sample in the range 2600 cm−1 to 3700 cm−1

(band (d) in the ﬁgure) indicates the possible incorporation of hydrogen in the ﬁlm, in

the form of O–H bonds [17]. After annealing this feature ﬂattens out, suggesting the

hydrogen is released from the bulk material, which can contribute to passivating active

bonding sites at the silicon surface.

Fig. 2: Injection-dependent excess carrier lifetime of PE-ALD Ga2O3 and Al2O3 samples after annealing in forming gas. The intrinsic lifetime of Ref. [19] is also displayed in the ﬁgure.

Injection dependent excess carrier lifetime data are displayed in Fig.2. The pre-anneal

lifetime data (not shown) was orders of magnitude less than the post-anneal data (τeff <

10µs). The results indicate the critical eﬀect of deposition temperature on the excess

carrier lifetime, with the optimum τeff corresponding to PE-ALD Ga2O3 deposited at

75 ◦C. For reference, a similarly gettered sample from the same ingot has been passivated

◦ with PE-ALD Al2O3, and annealed at 425 C for 30 minutes in the same forming gas

ambient. The τeff vs. ∆n data for this sample, which has been included in Fig.2,

indicates that Ga2O3 achieves comparable passivation of p-type silicon surfaces to the Chapter 3. Surface Passivation by Gallium Oxide 96

state-of-the-art PE-ALD Al2O3. There is, nevertheless, room for further improvements in passivation by Ga2O3 by optimising both deposition and annealing conditions.

The surface recombination current density J0 has been extracted from the excess carrier lifetime data of Fig.2 by subtracting the intrinsic lifetime, following the parameterisation of Richter et al. [19], and applying the method of Kane and Swanson [20], as per McIntosh and Black [21]. The intrinsic carrier density ni value of Misiakos and

9 −3 Tsamakis at a temperature of 300 K (ni = 9.696 × 10 cm ) was used in the calculation of J0 [22]. A summary of the passivation results, including a calculation of implied open-circuit voltage iVoc is tabulated below [23]. Note that the ni value used in the

◦ calculation of iVoc has been evaluated at T = 25 C for the purposes of consistency with the calculation embedded in the Sinton Instruments tool.

Compared to thermal ALD, PE-ALD Ga2O3 realises a signiﬁcant improvement in surface

−2 −2 passivation with the J0 decreasing from 30 fA.cm for thermal ALD to 7 fA.cm for

PE-ALD. As a consequence, the iVoc has similarly improved to 721 mV for the PE-ALD

Ga2O3 sample, from a value below 690 mV for the thermal ALD Ga2O3 passivated sample [9]. Expressed in terms of recombination current or implied voltage, the best results achieved with PE-ALD Ga2O3 approach those with PE-ALD Al2O3, for which

−2 we measured J0 = 4 fA.cm and iVoc = 729 mV.

The total charge Qtot, i.e. the charge associated with the bulk Ga2O3 material and the interface it forms with silicon, was calculated from the HF CV measurements with the data corrected for series resistance eﬀects [24]. Representative HF CV data for one of the samples studied is displayed in Fig.3. The accumulation capacitance was extrapolated from the series resistance corrected data [25], while the ﬂatband voltage VFB was taken Chapter 3. Surface Passivation by Gallium Oxide 97

Fig. 3: Representative CV data of the MIS structures analysed in this study. The data have been corrected for series resistance eﬀects. The shift in the data to higher positive voltages is indicative of a negative charge. from the gate bias corresponding to the calculated high frequency ﬂatband capacitance

[24].

This method for calculating Qtot erroneously assumes a negligible contribution to the

capacitance from charge carriers trapped in states within the bandgap. This source of

error could be especially pronounced for the pre-anneal sample where the interfacial

defect density is high relative to the annealed samples. The error in VFB due to Dit has

therefore been conservatively estimated by calculating the contribution of the charged

traps to VFB using the method of Ref. [24]. Uncertainty in the doping density NA

also contributes to uncertainty in Qtot. This source of error is small in comparison to

the eﬀect of Dit and has been accounted for by comparing the eﬀect of the NA values

determined independently by the Sinton Instruments PCD tool and the slope of the

2 1/C vs. V curve on Qtot. The compound eﬀects of these two sources of error on the

calculation of Qtot is represented in the error margins ascribed to the values of Qtot/q Chapter 3. Surface Passivation by Gallium Oxide 98 in Table 1.

Fig. 4: Interfacial defect distributions for the PE-ALD Ga2O3 passivated samples plotted as a function of the energy level within the bandgap relative to the intrinsic Fermi level Ei. The position of the valence and conduction bands, labelled VB and CB respectively, are noted in the ﬁgure.

The interface defect density Dit has also been extracted from the HF CV data by the

Terman method [24]. The resulting plot of defect density as a function of energy within the bandgap relative to the intrinsic Fermi level (ET − Ei) is represented in Fig.4. Also indicated are the positions of the silicon valence and conduction bands (labelled VB and

CB).

Due to the inherent inaccuracies in the Terman method, as discussed in Ref. [24], the conclusions that can be drawn from the Dit(E) data are limited. What is apparent

12 from the CV data is an order of magnitude approximation to Dit at midgap of ∼ 10 eV−1.cm−2 for the pre-anneal sample, and ∼ 1011 eV−1.cm−2 for the post-anneal sam-

12 −2 12 −2 ples, and a doubling of Qtot/q from 3 × 10 cm to 6.2 × 10 cm for the samples

◦ deposited at 150 C. These changes in the properties of the Ga2O3–Si interface correlate Chapter 3. Surface Passivation by Gallium Oxide 99 well to the excess charge carrier lifetime measurements, though the CV data present no trend with τeff as a function of deposition temperature Tdep.

Table 1: Summary of the transient PCD and CV data

Sample Qtot/q midgap Dit τeﬀ J0 iVoc (1012 cm−2) (1011 eV−1.cm−2)(µs)a (fA.cm−2) (mV) ◦ Ga2O3 Tdep = 150 C -3.0±1.6 ∼50 <10 - - (pre-anneal) ◦ Ga2O3 Tdep = 150 C -6.2±0.3 - 734 25±2 697 ◦ Ga2O3 Tdep = 75 C -5.2±0.2 ∼5 2147 7±1 721 ◦ Ga2O3 Tdep = 50 C -4.5±0.2 ∼5 1716 9±1 717 b b Al2O3 -2.7 0.54 3720 4±1 729 Ga2O3 thermal - - 2370 30±3 687 ALDc

a 15 −3 τeff data are quoted at ∆n = 1 × 10 cm . b Values of Qtot/q and Dit for the Al2O3 sample are taken from Ref. 26 and are presented as a reference only. c Data for the Ga2O3 thermal ALD sample has been taken from Ref. 9.

The values extracted from the HF CV data are similar to the properties of the PE-

ALD Al2O3–Si interface reported in the literature. Liang et al. measured values for

12 −2 10 −1 −2 Qtot/q and midgap Dit of −2.7 × 10 cm and 5.4 × 10 eV .cm , respectively for

Al2O3 ﬁlms prepared in the same manner as the Al2O3 reference samples reported in

this study [26]. Other studies report diﬀering values of post-anneal Qtot and Dit for the

Al2O3–Si interface, though they typically fall within the same order of magnitude of Ref.

[26] and result in similar levels of surface passivation that approach the intrinsic carrier

lifetime parameterisation of Refs. [19], [17, 27, 28]. Indeed, the values (and polarity)

of the charge reported in this study for the Ga2O3–Si interface are similar to those

recently reported by Altuntas et al. for crystalline (β-phase) Ga2O3 on silicon formed

by annealing PE-ALD ﬁlms at elevated temperatures (≥ 700 ◦C), though the authors

in that study were unable to obtain CV data for un-annealed, amorphous Ga2O3–Si

structures [18].

The combination of a high density of negative charge and a signiﬁcant reduction in Chapter 3. Surface Passivation by Gallium Oxide 100

the defect density at the Ga2O3–Si interface after annealing indicates that Ga2O3 is an ideal candidate for the passivation of boron-diﬀused p+ silicon surfaces. To investigate this, high resistivity (>100 Ω.cm) n-type FZ silicon samples were saw damaged etched, cleaned in the standard RCA process and diﬀused with boron in a quartz tube furnace using BBr3 as the boron source, as per Ref. [29]. The sheet resistance of the resulting

+ + boron-diﬀused p silicon surface was ∼96 Ω/. The p silicon surface was then passi-

◦ ◦ vated by PE-ALD Ga2O3 deposited at 150 C and annealed in forming gas at 300 C for

30 minutes.

−2 The J0 extracted from the Ga2O3 passivated sample was 52 fA.cm . This is compared

−2 to 27 fA.cm for the PE-ALD Al2O3 passivated sample reported by Bullock et al. on a sample that underwent an identical diﬀusion process that resulted in a sheet resistance

◦ of 110 Ω/ [29]. Note that for a deposition temperature of 150 C the undiﬀused p-type

−2 silicon sample passivated with Ga2O3 has a J0 of 25 fA.cm (see Table 1). Further optimisation of the passivation, including applying the 75 ◦C deposition temperature condition, is expected to lead to a further reduction in the Ga2O3 passivated J0 value on p+ surfaces. Furthermore, while all samples in this study were annealed at 300 ◦C, and detrimental eﬀects such as blistering were not observed at any stage of processing, the stability of the Ga2O3 ﬁlms at higher temperatures, such as those used for screen printing metallisation of solar cells, is yet to be studied.

IV. Conclusion

PE-ALD Ga2O3 has been shown to passivate base resistivity p-type and boron-diﬀused p+ silicon surfaces by a simultaneous reduction in the post-anneal interfacial defect density (to ∼ 5×1011 eV−1.cm−2) and the formation of a large density of negative charge Chapter 3. Surface Passivation by Gallium Oxide 101

(up to −6.2 × 1012 cm−2). This accounts for measured recombination current densities

−2 down to 7 fA.cm for Ga2O3-passivated 1.75 Ω.cm p-type silicon samples, similar to

−2 −2 state-of-the-art PE-ALD Al2O3 (4 fA.cm ). An unoptimised J0 of 52 fA.cm was measured on a moderately doped (96 Ω/) boron diﬀusion. Both as-deposited and post-anneal samples have been shown to be amorphous, with FTIR spectra indicating the presence of an interfacial SiOx layer. The IR spectra also show little residual carbon contamination in the ﬁlms and suggest that hydrogen may be released from the Ga2O3 during annealing at 300 ◦C. These results indicate that gallium oxide is high quality passivating material for p-type surfaces in crystalline silicon photovoltaic devices.

Acknowledgements

The authors would like to thank Lachlan Black for his help in analysing the CV data, and James Bullock for providing the boron-diﬀused samples. This work has been supported by the Australian government through the Australian Renewable Energy Agency

(ARENA). Ellipsometry equipment at the Australian National Fabrication Facility was used in this work. Chapter 3. Surface Passivation by Gallium Oxide 102

References

[1] R. Hezel and K. Jaeger, J. Electrochem. Soc. 136(2), 518 (1989).

[2] G. Dingemans and W. M. M. Kessels, J. Vac. Sci. Technol. A 30(4), 040802-1 (2012).

[3] M. Orita, H. Hiramatsu, H. Ohta, M. Hirano, and H. Hosono, Thin Solid Films 411, 134 (2002).

[4] F. K. Shan, G. X. Liu, W. J. Lee, G. H. Lee, I. S. Kim, and B. C. Shin, J. Appl. Phys. 98(2), 023504 (2005).

[5] A. K. Chandiran, N. Tetreault, R. Humphry-Baker, F. Kessler, E. Baranoﬀ, C. Yi, M. K. Nazeeruddin, and M. Gr¨atzel, Nano Lett. 12, 3941 (2012).

[6] T. Minami, T. Miyata, and Y. Nishi, Sol. Energy 105, 206 (2014).

[7] Y. S. Lee, D. Chua, R. E. Brandt, S. C. Siah, J. V. Li, J. P. Mailoa, S. W. Lee, R. G. Gordon, and T. Buonassisi, Adv. Mater. 26(27), 4704 (2014).

[8] T. Koida, Y. Kamikawa-Shimizu, A. Yamada, H. Shibata, and S. Niki, IEEE J. Photovolt. 99, 1 (2015).

[9] T. G. Allen and A. Cuevas, Appl. Phys. Lett. 105(3), 031601 (2014).

[10] I. Donmez, C. Ozgit-Akgun, and N. Biyiklia, J. Vac. Sci. Technol. A 31(1), 01A110 (2013).

[11] C. L. Dezlah IV, J. Niinisto, K. Arstila, L. Niinisto, and C. H. Winter, Chem. Mater. 18(2), 471 (2006).

[12] A. Ortiz, J. C. Alonso, E. Andrade, and C. Urbiola, J. Electrochem. Soc. 148(2), F26 (2001).

[13] M. Orita, H. Hiramatsu, H. Ohta, M. Hirano, and H. Hosono, Thin Solid Films 411, 134 (2002).

[14] P. G. Pai, S. S. Chao, Y. Takagi, and G. Lucovsky, J. Vac. Sci. Technol. A 4(3), 689 (1986).

[15] M.-T. Ho, Y. Wang, R. T. Brewer, L. S. Wielunski, Y. J. Chabal, N. Moumen, and M. Boleslawski, Appl. Phys. Lett. 87(13), 133103 (2005).

[16] F. Werner, B. Veith, D. Zielke, L. Khnemund, C. Tegenkamp, M. Seibt, R. Brendel, and J. Schmidt, J. Appl. Phys. 109(11), 113701 (2011).

[17] A. Richter, S. Henneck, J. Benick, M. Hrteis, M. Hermle, and S. W. Glunz, in: Proc. 25th European Photovoltaic Solar Energy Conference, Valencia, Spain, 2010.

[18] H. Altuntas, I. Donmez, C. Ozgti-Akgun, and N. Biyikli, J. Vac. Sci. Technol. A 32(4), 041504 (2014). Chapter 3. Surface Passivation by Gallium Oxide 103

[19] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, and A. Cuevas, Phys. Rev. B 86(16), 165202 (2012).

[20] D. E. Kane and R. M. Swanson, in: Proc. 18th IEEE Photovoltaic Specialists Conference, Las Vegas, USA, 1985.

[21] K. R. McIntosh and L. E. Black, J. Appl. Phys. 116(1), 014503 (2014).

[22] K. Misiakos and D. Tsamakis, J. Appl. Phys. 74(5), 3293 (1993).

[23] M. J. Kerr, A. Cuevas, and R. A. Sinton, J. Appl. Phys. 91(1), 399 (2002).

[24] E. H. Nicollian and J. R. Brews, MOS (Metal Oxide Semiconductor) Physics and Technology (Wiley, New York, Wiley, 1982).

[25] S. V. Walstra and C.-T. Sah, Solid-State Electron. 42(4), 671 (1998).

[26] W. Liang, K. J. Weber, D. Suh, S. P. Phang, J. Yu, A. K. McAuley, and B. R. Legg, IEEE J. Photovolt. 3(2), 678 (2013).

[27] L. E. Black and K. R. McIntosh, IEEE J. Photovoltaics 3(3), 936 (2013).

[28] J. Benick, A. Richter, T.-T. A. Li, N. E. Grant, K. R. McIntosh, Y. Ren, K. J. Weber, M. Hermle1, and S. W. Glunz, in: Proc. 35th IEEE Photovoltaic Specialists Conference, Honolulu, USA, 2010.

[29] J. Bullock, D. Yan, Y. Wan, A. Cuevas, B. Demaurex, A. Hessler-Wyser, and S. De Wolf, J. Appl. Phys. 115(16), 163703 (2014).

Demonstration of c-Si solar cells with gallium oxide surface passivation and laser-doped gallium p+ regions

Thomas Allen1, Marco Ernst1, Christian Samundsett1 and Andr´es

Cuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

Published in IEEE Journal of Photovoltaics, volume 5, issue 6, 2015.

Abstract — Gallium oxide (Ga2O3) deposited by plasma enhanced atomic layer deposition (PEALD) is shown to passivate crystalline silicon surfaces via a combination of a high negative charge and a reduction in the density of surface defects to below 1×1011 eV−1.cm−2 at midgap. The passivation, as determined by the injection-dependent excess carrier lifetime, is demonstrated 105 Chapter 3. Surface Passivation by Gallium Oxide 106

to be commensurate to that of PEALD aluminium oxide. In addition, Ga2O3 is used as a gallium source in a laser-doping process, resulting in a device eﬃciency of 19.2% and an open-circuit voltage of 658 mV in a partial rear contact p-type cell design.

Index Terms — Al2O3, gallium oxide, Ga2O3, laser doping, surface passivation. Chapter 3. Surface Passivation by Gallium Oxide 107

I. Introduction

The industrial production of crystalline silicon (c-Si) photovoltaic (PV) devices is re- liant upon robust, multifunctional processes in order to simultaneously reduce fabrication complexity and increase device efficiency and yield. Among the most useful, and hence widely adopted, of these processes is the deposition of plasma-enhanced chemical vapor deposition (PECVD) silicon nitride (SiNx), which has beneficial optical properties, excellent surface defect passivation qualities, and the ability to passivate bulk defects via hydrogenation. Other fabrication steps offer other synergies: the thermal diffusion of phosphorus to form n+ doped regions also getters bulk impurities like iron; and screen-printed metal pastes, which offer low resistance pathways for the transport of photogenerated current, also negate the need for photolithographic definition of contact openings. One of the most significant material advances in industrial high-efficiency c-Si

PV devices has been the application of aluminium oxide (Al2O3) to passivate p-type

and p+ surfaces [1], as well as, in some device structures, to create p+ regions via laser

doping [2]–[5]. However, Al2O3 oﬀers few other such synergies in the manufacturing

process.

+ Gallium oxide (Ga2O3), like Al2O3, has been shown to passivate p-type and p silicon

[6], [7]. Unlike Al2O3, Ga2O3 is a wide bandgap semiconductor and is believed to exhibit

n-type conductivity through carrier transport via oxygen vacancies[8], [9]. Other studies

have demonstrated enhanced conductivity of Ga2O3 from the incorporation of shallow

donor impurities like silicon [10] and tin [11]; hydrogen [12] and carbon [13] have also

been proposed as other possible donors in Ga2O3. Therefore, Ga2O3 could potentially

be used for the transport and selection of carriers in c-Si solar cells. Chapter 3. Surface Passivation by Gallium Oxide 108

In this study, we present gallium oxide as an alternative material to Al2O3. We assess the suitability of Ga2O3 for c-Si PV applications in terms of its optical properties, Ga2O3–

Si interface properties (i.e., surface passivation), and its ability to be used as a p-type dopant source in a laser-doping process. As a proof of concept, we then fabricate p-type partial rear contact (PRC) devices with a gallium oxide-passivated rear and gallium- doped p+ regions.

II. Experimental Procedure

Gallium oxide ﬁlms were deposited in a Beneq TFS 200 atomic layer deposition (ALD) reactor, with trimethyl-gallium (TMGa) and oxygen plasma as the reactants. All depositions in this study were performed at a temperature of 75 ◦C. The excess carrier lifetime

(τeff ) of the passivated samples was measured as a function of the excess carrier density

(∆n) using a Sinton Instruments WCT-120 tool operating in the photoconductance decay transient mode. The low-resistivity (FZ, 1.6 Ω.cm, h100i) p-type silicon wafers used in this study were gettered of bulk impurities in a phosphorus diﬀusion process prior to

TMAH etching and RCA cleaning.

High-frequency (HF) capacitance?voltage (C–V ) measurements were carried out on single-side Ga2O3?deposited (∼30 nm) samples prepared on polished 3 Ω.cm, p-type, FZ h100i silicon. The top contact of the C–V samples was formed by thermally evaporating aluminium dots directly on the Ga2O3-passivated surface, while a Ga/In eutectic paste formed the rear contact with the silicon base. AC parallel capacitance measurements were made using an HP 4284A Precision LCR Meter at a frequency of 1 MHz. The samples were mounted on a temperature-controlled chuck with measurements performed at Chapter 3. Surface Passivation by Gallium Oxide 109

300 K. High leakage currents through the Ga2O3 at lower frequencies prevented the acquisition of low-frequency C–V data.

In the analysis of the C–V and τeff data, the intrinsic carrier concentration of Misiakos

and Tsamakis at 300 K (9.696 × 109 cm−3) has been applied [14]. Where the intrinsic

carrier lifetime is required in the calculations, the parameterization of Richter et al., has

been used [15]. All lifetime and surface recombination velocity data are quoted at an

injection level of ∆n = 1 × 1015 cm−3.

III. Optical and Electronic Properties of Gallium Oxide

A. Optical Properties

The optical bandgap of the plasma-enhanced atomic layer deposition (PEALD) Ga2O3

ﬁlms was assessed by optical transmission measurements using a PerkinElmer Lambda

1050 spectrophotometer ﬁtted with an integrating sphere. Gallium oxide was deposited

on a quartz glass slide, and transmission measurements were performed. The trans-

mittance data are shown in Fig. 1. The data were normalized to accommodate for

reﬂection losses before applying the Tauc method to calculate the bandgap of the ﬁlm.

The bandgap (Eg) was estimated to be ∼ 5.2 eV (assuming a direct transition) which is within the upper range of Eg quoted in the literature [16]. The wide bandgap of gallium oxide is advantageous for ensuring optical transparency and avoiding parasitic optical losses. Further transmission measurements were conducted on the sample after annealing for 2 h at 300 ◦C in a quartz tube furnace. There was no signiﬁcant diﬀerence in

the transmittance spectrum and bandgap extraction, and so the data are not presented

here. Chapter 3. Surface Passivation by Gallium Oxide 110

Fig. 1: Spectrophotometry (blue) and ellipsometry (red) data for the as-deposited PEALD Ga2O3 ﬁlms. The optical bandgap was calculated from the normalized transmittance data, as represented in the inset graph.

The refractive index of the PEALD Ga2O3 deposited on a polished silicon sample was determined by variable-angle spectroscopic ellipsometry (J.A. Woolam M2000D). The ellipsometry data were fitted with a Tauc-Lorentz model, and the refractive index (n) is presented in Fig.1. The refractive index of the film is ∼ 1.8 at λ = 632 nm, which is slightly higher than the refractive index of Al2O3 (∼ 1.6) but significantly less than that of antireflection coatings like SiNx and TiO2 (n > 2). Consequently, Ga2O3 is unlikely to be used as a front antireflection coating in industrial applications without an additional high n capping layer but is suitable for application on the rear-side metallic reflector provided the material is suitably thick [17].

B. Annealing Dependence of the Surface Passivation

The dependence of τeff on annealing conditions of the PEALD-deposited Ga2O3 has been investigated with respect to both annealing temperature and time. As shown in Chapter 3. Surface Passivation by Gallium Oxide 111

Fig.2, the annealing behaviour of the PEALD-deposited ﬁlms is similar to that in [6] for ozone-deposited Ga2O3-passivated samples: An increase in the carrier lifetime is observed upon annealing, with the peak in τeff occurring faster for higher temperatures.

There is also a signiﬁcant inverse dependence of the peak lifetime on annealing temperature. All passivation samples (except perhaps for those annealed at 250 ◦C) degrade with extended annealing time, and again, as per [6], the degradation occurs faster for higher annealing temperatures, with the general trends for both the passivation and depassivation reactions following an exponential behaviour with time.

Fig. 2: Annealing dependence of the excess carrier lifetime for a range of temperatures. Each dataset represents the cumulative anneal time of a single Ga2O3-passivated sample. All anneals were performed in a quartz tube furnace in a forming gas ambient. The lifetime values are quoted at ∆n = 1 × 1015 cm−3. The lines are guides for the eyes. The three data points represented by stars were separate samples annealed for 120 minutes at their respective temperatures.

However, for the PEALD results of Fig.2, both the forward (passivation) and reverse

(depassivation) reactions occur faster at lower temperatures compared with the thermal

ALD ﬁlms of [6]. Assuming that the passivation reaction is in part due to hydrogenation

of interfacial defects (where the hydrogen is a remnant of the organometallic precursor), Chapter 3. Surface Passivation by Gallium Oxide 112 this result is consistent with the PEALD ﬁlms having both a higher content and mobility of hydrogen.

The optimum annealing condition from Fig.2, an extended anneal (360 minutes) at

250 ◦C, is the best reported passivation result for gallium oxide on p-type silicon with a carrier lifetime of 2.9 ms on a 1.6 Ω.cm wafer. The upper limit to the surface recombination velocity (Seff,UL) has been calculated in the usual manner [15], assuming no Shockley–Read–Hall recombination in the bulk of the silicon wafer. The surface recombination current J0 was also extracted from the inverse lifetime data (corrected for intrinsic recombination) of these samples using the method of Kane and Swanson [18] as per McIntosh and Black [19]. The recombination parameters Seff,UL and J0 for this sample are 2.4 cm.s−1 and 5.5 ± 0.6 fA.cm−2, respectively. This compares with a carrier

−1 −2 lifetime of 3.3 ms, Seff,UL of 1.9 cm.s , and J0 of 3.6 ± 0.3 fA.cm for a PEALD

◦ Al2O3-passivated control sample after annealing in forming gas (425 C, 30 min) on a substrate from the same ingot.

Separate Ga2O3-passivated samples from the same Si ingot were prepared in the manner described above and annealed at 300, 350, and 400 ◦C for 120 minutes in forming gas in order to validate the cumulative anneal results (indicated in Fig.2 with a star). They conform to the trend of the cumulative anneal data, conﬁrming that the passivation and subsequent degradation in the lifetime is due to the annealing conditions (and not, for example, surface damage from handling the samples between anneals). Capacitance– voltage samples were also prepared alongside the lifetime samples.

The C–V data of Fig.3 indicate that the high negative charge density Q/q forms after thermal treatment and increases with annealing temperature, saturating at 350 ◦C to a maximum value of approximately −7 × 1012 cm−2. Unlike the as-deposited result of Chapter 3. Surface Passivation by Gallium Oxide 113

Fig. 3: HF C–V data from the starred samples of Fig.2. As no midgap Dit data were able to be extracted from the as deposited sample, the value of 1 × 1013 eV−1.cm−2 was chosen as being indicative of an unpassivated h100i surface. The error bars on the Q/q data represent the variation of the extracted value of the charge over multiple measurements on the same sample. The error in Dit is estimated from the Dit extracted either side of midgap and is not inclusive of errors related to the extraction technique.

[7] (Q/q = −3 × 1012 cm−2), which was deposited at a higher temperature (150 ◦C),

the negative charge density in this study is one order of magnitude lower (∼ −7 ± 1011

cm−2) prior to annealing.

The midgap Dit was also extracted from the C–V data via the Terman method and

generally trends with the inverse of the lifetime, as expected. However, it is important

to note that the accuracy of the Dit extraction is limited by the technique used, as

discussed at length in other publications [7], [20]. Despite this, it is apparent from the

C–V data that the change in the recombination rate between the annealed samples is

largely determined by the change in the surface defect density. Generally, it is possible

◦ 11 to conclude that the Dit at midgap for the 300 C sample is likely to be below 1 × 10

−1 −2 eV .cm , while the midgap Dit for the samples annealed at higher temperatures is Chapter 3. Surface Passivation by Gallium Oxide 114

11 −1 −2 over 1 × 10 eV .cm . The midgap Dit value of the unannealed sample could not be extracted from the HF C–V data. A value of 1×1013 ±5×1012 eV−1.cm−2 is represented in the ﬁgure as being indicative of an unpassivated surface, given the lifetime of the sample was less than 10 µs.

IV. Application to Solar Cells

A. Laser Doping from Gallium Oxide

Elemental gallium, like aluminum and boron, is a shallow acceptor impurity in crystalline silicon. Bulk Si ingots doped with Ga have been shown to present high minority carrier lifetimes [21], in part because Ga does not form recombination-active pairs with oxygen and is, therefore, compatible with Czochralski growth [22]. As such, gallium has been identified as an alternative bulk p-type dopant to boron as a means of circumvent- ing the light-induced degradation associated with B–O pairs. However, gallium has a low segregation coefficient, and this produces a significant variation in the dopant concentration along a Si ingot [22]. Gallium-doped Si has also been shown to be sensitive to iron contamination through recombination via Ga–Fe pairs [23].

While considerable attention has been paid to gallium as a bulk dopant, few studies have investigated heavily doped gallium p+ regions. Gallium is a fast diffuser in Si, which offers potential advantages to create p+ regions at lower temperatures and shorter times compared with a boron thermal diffusion process. Unfortunately, dopant sources for Ga diffusion are not as readily available as for boron. Globally, the potential of Ga doping of both the bulk material and p+ regions is clear, but it has not been sufficiently explored. Chapter 3. Surface Passivation by Gallium Oxide 115

A particularly attractive option is to use gallium oxide as both a passivating layer and as a dopant source in a laser-ﬁring process, in an analogous process to that already developed using Al2O3 [2]–[5]. Whether Ga doping can deliver advantages over Al doping in a laser-ﬁring process will need more detailed investigation since recombination in the doped region can be dominated by structural defects that result from the recrystallization process, and not the dopant species [24].

Fig. 4: Electrically active dopant concentrationmeasured by ECV. The dopant concentration has been scaled from the raw data to match the sheet resistance measurement. The inset graph is a measure of the current–voltage behaviour of the laser-doped contact formed by evaporated aluminium over multiple laser-doped regions.

+ To demonstrate the viability of using Ga2O3 in a laser doping process, Ga-doped p

regions were formed on polished low-resistivity n-type (1 Ω.cm) silicon deposited with

∼30 nm of PEALD Ga2O3 by exposing the surface to three pulses from a 355 nm (UV)

nanosecond pulse length laser at a ﬂuence of 3 J.cm−2. A 1 cm2 p+ area was formed by

overlapping laser pulses, resulting in a resistivity of 160 Ω/, as measured by a four-

point probe. The dopant proﬁle was subsequently measured using the electrochemical Chapter 3. Surface Passivation by Gallium Oxide 116 capacitance–voltage (ECV) technique, as shown in Fig.4. The measured proﬁle was integrated and scaled to match the measured sheet resistance, assuming the same mobility for holes as for boron doped silicon.

The formation of Ohmic contact to the Ga-doped region was conﬁrmed by measuring the current–voltage behavior between two contact pads formed by thermally evaporated aluminium over 20 small-area (∼ 30 × 30 µm2) laser-doped regions per contact pad on

1.6 Ω.cm p-type silicon. The wafer was etched in TMAH before RCA cleaning, PEALD deposition, and laser processing. The I–V characteristics of the contact are shown in the

20 −3 inset of Fig.4. The high surface concentration of Ga (N A > 10 cm ) undoubtedly contributes to the formation of the good Ohmic contact represented in the inset graph.

B. Solar Cell Results

Small-area (2×2 cm2) PRC cells were fabricated on ∼260 µm thick, 1.5 Ω.cm, p-type FZ wafers. The front side was etched in a TMAH solution to form a random pyramid texture before a thermal phosphorus diffusion (Rsheet ∼ 120Ω/) in a quartz tube furnace. The rear side of the cell was then roughly planarized by etching in TMAH, locally removing the phosphorus diffusion. PEALD gallium oxide (∼30 nm) was deposited on the rear surface and annealed in forming gas at 300 ◦C for 30 min. The n+ textured front side was then passivated with PECVD SiNx. The gallium oxide passivated rear side was then laser processed in the manner described above, forming 30 × 30µm2 local gallium- doped p+ regions at a pitch of 400 µm. Rear-side metalization was made via thermally evaporated aluminium, while the front-side metal fingers and busbar were first defined by photolithography and then formed by a thermally evaporated Cr/Pd/Ag stack that was later thickened by electroplated Ag. Chapter 3. Surface Passivation by Gallium Oxide 117

Fig. 4: Current–voltage characteristics of the gallium-doped PRC cell fabricated on a p-type base wafer. The dashed data are the pseudo light J–V characteristics of the cell, free of series resistance eﬀects, as determined by Suns–Voc measurement. A comparison of the two curves was used to determine the series resistance of the cell.

The 1-sun J–V characteristics of the device after annealing for 10 min at 400 ◦C are shown in Fig.4. Prior to annealing the Suns– Voc measurement of the cell showed a strong

Schottky contact behaviour [25], likely a result of poor metal–semiconductor contact in the small openings created by the laser. This is indicative that the laser-doped regions on the cell were not identical to those on the test structures, as the Ohmic behaviour of the contact in Fig.4 was obtained without the need for any thermal treatment. It is likely that the surface roughness of the rear side of the cell, due to the presence of truncated pyramids, aﬀected the interaction between the laser radiation and the Ga2O3- coated silicon. This may have resulted in a diﬀerent level of p+ doping and possibly even an incomplete ablation of the Ga2O3 in some of the contact points made on the cell. It is also possible that a thin native oxide formed on the p+ regions of the cell, and not the contact test structure, prior to metallization, thereby inhibiting the contact formation prior to annealing. These factors may also explain the relatively high series resistance Chapter 3. Surface Passivation by Gallium Oxide 118 of the cell (∼ 0.9Ω.cm2), as determined by a comparison of the 1-sun J–V and pseudo

J–V curves at maximum power [25], [26].

The measured Voc (658 mV) of the cell is comparable with other laser-doped p-type PRC cells reported in the literature based on Al2O3 passivation and Al doping [5], although the cell architecture (rear pitch and contact fraction) may not be the same. Nevertheless, the Voc of the cell is deﬁnitive evidence of well-passivated surfaces, justifying the lifetime and C–V study and demonstrating the suitability of Ga2O3 as a passivating material.

The ﬁll factor of the cell is relatively low (74.5%), possibly due to non-uniformities in the formation of the p+ regions due to surface roughness, or the formation of a native oxide, as mentioned above. Further gains in eﬃciency are likely to be achieved by optimizing the laser processing and controlling the surface topology of the rear laser-doped side.

V. Conclusion

Gallium oxide has been shown to oﬀer similar surface passivation qualities as state-of-

12 −2 the-art Al2O3: a high density of negative charge (−Q/q > 10 cm ), and a low defect

11 −1 −2 density after annealing (midgap Dit < 10 eV .cm ). The application of Ga2O3 as a surface passivation material has translated to a Voc of 658 mV in a PRC cell structure, comparable with open-circuit voltages of similar cell designs using Al2O3. This is the

ﬁrst demonstration of silicon surface passivation by gallium oxide at the device level and also the ﬁrst laser-doped cell using gallium as the p+ dopant. Chapter 3. Surface Passivation by Gallium Oxide 119

Acknowledgements

The authors would like to thank X. Zhang, S. P. Phang, D. Yan, T. Ratcliﬀ, and J.

McKeon for experimental support, and A. Fell for fruitful discussions. Chapter 3. Surface Passivation by Gallium Oxide 120

References

[1] G. Dingemans and W. M. M. Kessels, “Status and prospects of Al2O3-based surface passivation schemes for silicon solar cells,” J. Vacuum Sci. Technol. A, vol. 30, pp. 040802-1–040802-27, 2012.

[2] F. E. Franklin, D. Walter, D. Suh, and K. J. Weber, “Laser doping from Al2O3 Layers,” in Proc. 27th Eur. Photovoltaic Sol. Energy Conf., Munich, Germany, 2012, pp. 706–708.

[3] P. Ortega, I. Martin, G. Lopez, M. Colina, A. Orpella, C. Voz, and R. Alcubilla,

“p-type c-Si solar cells based on rear side laser processing of Al2O3/SiCx stacks,” Sol. Energy Mater. Sol. Cells, vol. 106, pp. 80–83, 2012.

[4] N.-P. Harder, Y. Larionova, and R. Brendel, “Al+ doping of Si by laser ablation of

Al2O3/SiN passivation,” Phys. Status Solidi A, vol. 9, pp. 1871–1873, 2013.

[5] E. Cornagliotti, A. Uruena, B. Hallam, L. Tous, R. Russell, F. Duerinckx, and J. Szlufcik, “Large area p-type PERL cells featuring local p+ BSF formed by laser processing of ALD Al2O3 layers,” Sol. Energy Mater. Sol. Cells, vol. 138, pp. 72–79, 2015.

[6] T. G. Allen and A. Cuevas, “Electronic passivation of silicon surfaces by thin ﬁlms of atomic layer deposited gallium oxide,” Appl. Phys. Lett., vol. 105, pp. 031601-1– 031601-4, 2014.

[7] T. G. Allen and A. Cuevas, “Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: Demonstration of surface passivation and negative interfacial charge,” Phys. Status Solidi RRL, vol. 9, pp. 220–224, 2015.

[8] M. R. Lorenz, J. F. Woods, and R. J. Gambino, “Some electrical properties of the semiconductor β-Ga2O3,” J. Phys. Chem. Solids, vol. 28, pp. 403–404, 1967.

[9] N. Ueda, H. Hosono, R.Waseda, and H. Kawazoe, “Synthesis and control of conductivity of ultraviolet transmitting β-Ga2O3 single crystals,” Appl. Phys. Lett., vol. 70, pp. 3561–3563, 1997.

[10] E. G. Villora, K. Shimamura, Y. Yoshikawa, T. Ujiie, and K. Aoki, “Electrical conductivity and carrier concentration control in β-Ga2O3 by Si doping,” Appl. Phys. Lett., vol. 92, pp. 202120-1–202120-3, 1997.

[11] M. Orita, H. Hiramatsu, H. Ohta, M. Hirano, and H. Hosono, “Preparation of highly conductive, deep ultraviolet transparent β-Ga2O3 thin ﬁlm at low deposition temperatures,” Thin Solid Films, vol. 411, pp. 134–139, 2002.

[12] J. B. Varley, H. Peelaers, A. Janotti, and C. G. Van De Walle, “Hydrogen cation vacancies in semiconducting oxides,” J. Phys. Condensed Matter, vol. 23, pp. 1–9, 2011. Chapter 3. Surface Passivation by Gallium Oxide 121

[13] J. L. Lyons, D. Steiauf, A. Janotti, and C. G. Van De Walle, “Carbon as a shallow donor in transparent conducting oxides,” Phys. Rev. Appl., vol. 2, pp. 064005-1– 064005-8, 2014.

[14] K. Misiakos and D. Tsamakis, “Accurate measurements of the silicon intrinsic carrier density from 78 to 340 K,” J. Appl. Phys., vol. 74, pp. 3293–3297, 1993.

[15] A. Richter, S.W. Glunz, F.Werner, J. Schmidt, and A. Cuevas, “Improved quanti- tative description of auger recombination in crystalline silicon,” Phys. Rev. B, vol. 86, pp. 165202-1–165202-14, 2012.

[16] F. K. Shan, G. X. Liu, W. J. Lee, G. H. Lee, I. S. Kim, and B. C. Shin, “Structural, electrical, and optical properties of transparent gallium oxide thin ﬁlms grown by plasma- enhanced atomic layer deposition,” J. Appl. Phys., vol. 98, pp. 023504-1–023504-6, 2005.

[17] Z. Holman, M. Filipic, B. Lipovsek, S. De Wolf, F. Smole, M. Topic, and C. Ballif, “Parasitic absorption in the rear reflector of a silicon solar cell: Simulation and measurement of the sub-bandgap reflectance for common dielectric/metal reflectors,” Sol. Energy Mater. Sol. Cells, vol. 120, pp. 426–430, 2014.

[18] D. E. Kane and R. M. Swanson, “Measurement of the emitter saturation current by a contactless photoconductivity decay method,” in Proc. 18th IEEE Photovoltaic Spec. Conf., Las Vegas, NV, USA, 1985, pp. 578–583.

[19] K. R. McIntosh and L. E. Black, “On eﬀective surface recombination parameters,” J. Appl. Phys., vol. 116, pp. 014503-1–014503-10, 2014.

[20] E. H. Nicollian and J. R. Brews, MOS (Metal Oxide Semiconductor) Physics and Technology. New York, NY, USA: Wiley, 1982.

[21] T. F. Ciszek and T. H. Wang, “Silicon defect and impurity studies using controlled samples,” in 14th Eur. Photovoltaic Sol. Energy Conf., Barcelona, Spain, 1997, pp. 396–399.

[22] S.W. Glunz, S. Rein, J. Knobloch, W.Wettling, and T. Abe, “Comparison of boron- and gallium-doped p-type czochralski silicon for photovoltaic application,” Prog. Pho- tovoltaics, Res. Appl., vol. 7, pp. 463–469, 1999.

[23] J. Schmidt and D. Macdonald, “Recombination activity of iron–gallium and iron– indium pairs in silicon,” J. Appl. Phys., vol. 97, pp. 113712-1–113712-9, 2005.

[24] Z. Hameiri, T. Puzzer, L. Mai, A. B. Sproul, and S. R. Wenham, “Laser induced defects in laser doped solar cells,” Prog. Photovoltaics, Res. Appl., vol. 19, pp. 391–405, 2011.

[25] R. A. Sinton and A. Cuevas, “A quasi-steady-state open-circuit voltage method for solar cell characterization,” in Proc. 16th Eur. Photovoltaic Sol. Energy Conf., Glasgow, U.K., 2000, pp. 1–4. Chapter 3. Surface Passivation by Gallium Oxide 122

[26] A. G. Aberle, S. R. Wenham, and M. A. Green, “A new method for accurate measurements of the lumped series resistance of solar cells,” in Proc. 23rd IEEE Photovoltaic Spec. Conf., Louisville, KY, USA, 1993, pp. 133–139. Silicon surface passivation by gallium oxide capped with silicon nitride

Thomas Allen1, Yimao Wan1 and Andr´esCuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

Published in IEEE Journal of Photovoltaics, volume 6, issue 4, 2016.

Abstract — Advances in the passivation of p-type and p+ surfaces have been one of the main developments in crystalline silicon solar cell technology in recent years, enabling significant progress in p-type solar cells with partial rear contacts, and n-type solar cells with front-side boron diffusions. In this contribution, we demonstrate improvements in the passivation of p-type and boron diffused p+ surfaces with plasma-enhanced atomic layer deposition (PEALD) gallium oxide (Ga2O3) with the addition of plasma-enhanced chemical vapor deposition (PECVD) silicon nitride (SiNx). On 1.6 Ω.cm

123 Chapter 3. Surface Passivation by Gallium Oxide 124 p-type wafers, we measure an improvement in the upper limit surface re-

−1 combination velocity (Seff,UL) from 2.5 to 1.4 cm.s on optimized Ga2O3 passivated samples before and after SiNx capping. We also show an improvement in the passivation of boron diﬀused p+ surfaces over previously

−2 reported data, measuring a recombination parameter (J0) of 26 fA.cm on a Ga2O3 passivated 85 Ω/ boron diﬀusion, approaching the Auger limit of

∼ 21 fA.cm−2 for this diﬀusion. In addition, we show that initial studies on the thermal stability of the Ga2O3 / SiNx stack indicate that it is compatible with conventional screen-printed metallization ﬁring procedures.

Index Terms — Gallium oxide, silicon nitride, surface passivation, thermal stability. Chapter 3. Surface Passivation by Gallium Oxide 125

I. Introduction

Improvements in the passivation of surface defects have been one of the main drivers of the increase in eﬃciency of industrial crystalline silicon (c-Si) solar cells in recent years.

As p-type cell architectures move toward the implementation of local rear contacts, replacing conventional full-area aluminium alloyed contacts, the passivation of boron- doped p-type and boron diﬀused p+ surfaces has taken on an increasingly important role in c-Si solar cell design. Similarly, with an expanding market share of cells fabricated on n-type wafers to avoid device degradation associated with the boron-oxygen defect, as well as other metallic impurities, the importance of the passivation of p+ regions is further compounded [1].

These technological advances have been facilitated by the development of surface passivation by the negatively charged material aluminium oxide (Al2O3). However, due to

its moderate refractive index (n ∼ 1.6), Al2O3 has to be capped with a higher index

material in order to be suitable for front-side passivation and antireﬂection schemes in

encapsulated PV modules. The higher refractive index of plasma-enhanced chemical

vapor deposition (PECVD) silicon nitride (SiNx; n = 1.9 – 2.7) is well matched to the

geometric mean of the refractive indices of the encapsulant materials, glass and EVA

(n ∼ 1.5), and silicon, resulting in a minimum reﬂection that can be optimized with

respect to the incident photon energy by using the appropriate capping layer thickness.

The advantageous optical properties of SiNx, coupled with its ability to hydrogenate

bulk and interfacial defects, have made it the most common antireﬂection coating for

c-Si solar cell applications [2]. However, SiNx is positively charged and so forms in-

version or depletion conditions in the silicon sub-surface region when applied to p-type

and p+ material, typically resulting in enhanced surface recombination [3] and parasitic Chapter 3. Surface Passivation by Gallium Oxide 126

shunting [4]. The application of SiNx on top of negatively charged Al2O3 overcomes these limitations with respect to solar cells with a front-side boron diﬀusion [5], typical of n-type solar cell designs.

Recently, TiO2, with an appropriately high refractive index, has been shown to passivate p-type and p+ surfaces, eliminating the need for an optical capping layer [6],[7]. However, the thermal instability of the surface passivation aﬀorded by TiO2, attributed to the formation of the crystalline anatase phase at elevated temperatures, is such that it cannot withstand the high temperature processing needed to form electrical contacts with standard screen-printed metal paste technologies [6].

The authors have previously reported on the passivation of c-Si surface defects by atomic layer deposited (ALD) gallium oxide (Ga2O3) [8],[9]. In this paper, we go on to show that surface passivation by gallium oxide is compatible with, and indeed improved by,

PECVD silicon nitride capping, a significant finding for applications to the front side of solar cells given that the refractive index of Ga2O3 is similar to that of Al2O3 [10]. We also report on an improvement of the passivation of p+ surfaces by gallium oxide, and extend the experimental investigation to pyramidally textured c-Si surfaces. Finally, we comment on the firing stability of the passivation of Ga2O3 films with and without SiNx capping.

II. Experimental Procedure

Thin ﬁlms of Ga2O3 and Al2O3 (∼20 nm, unless stated otherwise) were deposited by a plasma-enhanced atomic layer deposition (PEALD) process (Beneq TFS 200), at temperatures of 75 and 175 ◦C, respectively, using the precursors trimethylgallium (TMGa) and trimethylaluminum (TMAl) with oxygen plasma. All subsequent annealing steps of Chapter 3. Surface Passivation by Gallium Oxide 127

both Ga2O3 and Al2O3 ﬁlms were performed in a quartz tube furnace in a forming gas ambient comprised of 2% hydrogen in argon. The SiNx ﬁlms (∼85 nm) were prepared by

PECVD (Roth and Rau AK400) at a deposition temperature of 300 ◦C and a processing

time of 30 min.

Symmetrically passivated lifetime samples were prepared on planar saw-damage etched

and RCA cleaned silicon substrates (∼260 µm thick, 1.6 Ω.cm p-type, unless stated oth-

erwise). Random upright pyramidally textured samples were formed by etching h100i

oriented silicon wafers in a solution of dilute tetramethylammonium hydroxide, isopropyl

alcohol, and dissolved silicon at 85 ◦C for 60 min. The textured wafers were subsequently

immersed in dilute hydroﬂuoric acid (HF) before RCA cleaning prior to further process-

ing.

Heavily doped p+ surfaces were formed by thermal boron diﬀusion carried out in a quartz

◦ tube furnace using BBr3 as the dopant source at a temperature of 950 C, resulting in

a sheet resistance Rsh of between 85 and 90 Ω/. The symmetrically diﬀused samples

were stripped of the borosilicate glass in dilute HF, RCA cleaned and passivated in the

same manner as described above.

The eﬀective lifetime (τeff ) of the passivated samples was measured as a function of mi-

nority carrier injection level (∆n) on a Sinton Instruments WCT-120 photoconductance

tester. Subsequent analysis of the lifetime data utilized the intrinsic lifetime parame-

terization of Richter et al. [11], and the intrinsic carrier density parameterization of

Misiakos and Tsamakis at a temperature of 300 K [12]. Chapter 3. Surface Passivation by Gallium Oxide 128

III. Results and Discussion

A. Recombination at undiﬀused p-type surfaces

In previous studies, PEALD Ga2O3 has been shown to passivate silicon surface defects

12 −2 in a similar fashion to Al2O3: through a high negative charge density (Qf > 10 cm )

11 and a considerable reduction in the density of interfacial defects (midgap Dit < 10 eV−1.cm−2) [10]. The injection-dependent lifetime curve of Fig. 1(a) directly compares the passivation of a 1.6 Ω.cm p-type wafer by PEALD Al2O3 with the best result achieved

◦ by Ga2O3 [10]. The Al2O3 passivated sample was annealed at 425 C for 30 min, while

◦ the Ga2O3 sample was annealed for 360 min at 250 C. The similar recombination behaviour of the two materials is reﬂected in the injection-dependent lifetime curves and subsequent extraction of the surface recombination velocity Seff,UL:

W 1 1 Seff,UL = − (1) 2 τeff τint

where τint is the intrinsic bulk lifetime of the silicon samples, as parameterized in [11]. At

15 −3 −1 an injection level of ∆n = 1×10 cm , the Seff,UL for the Al2O3 sample is 1.9 cm.s

−1 compared with 2.5 cm.s for the Ga2O3 sample. While the inverse Auger-corrected lifetime data, typically used to extract the surface recombination parameter J0s [13],

[14], showed some nonlinearities with respect to the injection level ∆n, we estimate the

−2 J0s of the samples to be 3 ± 1 fA.cm for the Al2O3 passivated sample, and 4.5 ± 1

−2 fA.cm for the Ga2O3 passivated sample, based on a ﬁtting of the lifetime data to the

J0 recombination model in [15].

Fig. 1(b) compares the injection-dependent lifetime data of an Al2O3 / SiNx stack before and after annealing with a Ga2O3 / SiNx stack prior to any thermal treatment and after Chapter 3. Surface Passivation by Gallium Oxide 129

b ) w i t h S i N c a p p i n g a ) w i t h o u t S i N x c a p p i n g x ) s m (

f f e ؔ

e Ω m p - t y p e , 1 . 6 c m , F Z , < 1 0 0 > i t o e

f A l O 4 2 5 C F G A , 3 0 m i n s i 2 3 l

r 1 o e i G a 2 O 3 2 5 0 C F G A , 3 6 0 m i n s r r

a A l 2 O 3 / S i N x a s d e p o s i t e d c

s G a O / S i N a s d e p o s i t e d s 2 3 x e o c

x A l 2 O 3 / S i N x 4 2 5 C F G A , 3 0 m i n s E o G a 2 O 3 / S i N x 4 0 0 C F G A , 1 8 0 m i n s I n t r i n s i c l i f e t i m e ( R i c h t e r e t a l . )

1 0 1 4 1 0 1 5 1 0 1 6 1 0 1 4 1 0 1 5 1 0 1 6 ( n ( c m - 3 פ n ( c m - 3 ) E x c e s s c a r r i e r d e n s i t y פ E x c e s s c a r r i e r d e n s i t y

Fig. 1: Exemplary injection-dependent lifetime curves of 1.6 Ω.cm p-type silicon passivated with (a) Al2O3 and Ga2O3 after annealing (solid squares); and (b) Al2O3 and Ga2O3 capped with PECVD SiNx before (open circles) and after (solid circles) annealing. Note that for the Ga2O3 / SiNx passivated samples, no thermal processing was required to activate the passivation, unlike the Al2O3 / SiNx passivated sample. The arrows in (b) indicate the progression of the samples after annealing in forming gas.

◦ a 180 min anneal at 400 C. Both the Al2O3 and Ga2O3 layers were deposited in an

identical manner to those in Fig. 1(a). As shown in Fig. 1(b), the lifetime of both SiNx

capped samples improved to values approaching the intrinsic limit in high injection,

compared with the uncapped samples of Fig. 1(a). This improvement in passivation

−1 corresponds to a reduction in the Seff,UL and J0s of the samples to 1.4 cm.s and

−2 −1 −2 2.3 ± 0.5 fA.cm , and 1.2 cm.s and 1.3 ± 0.5 fA.cm , for the Ga2O3 / SiNx and

Al2O3 / SiNx passivated samples, respectively. Signiﬁcantly, unlike the Al2O3 / SiNx

passivated sample, the Ga2O3 / SiNx passivated sample did not require any additional thermal treatments to activate the passivation.

In the case of the as-deposited Ga2O3 / SiNx stack, the resulting τeff data of Fig. 1(b)

are consistent with the improvement in passivation being achieved via an increase in

Qf and a decrease in Dit of the Ga2O3–Si interface after an extended low-temperature

anneal evidenced in [10]. This is a result of the thermal budget of the PECVD process Chapter 3. Surface Passivation by Gallium Oxide 130

(> 60 min total processing time at 300 ◦C for double-side processing) being in the vicinity of the ideal thermal treatment for the PEALD Ga2O3 films. This result, and the poor passivation of the as-deposited Al2O3 / SiNx stack, is consistent with the difference in activation energies for the passivation of single-layer Al2O3 and Ga2O3 films, measured to be ∼1.4 [16] and ∼0.5 eV [8], respectively. The PECVD SiNx deposition is also likely to have resulted in a further reduction of the defect density due to hydrogenation of the Ga2O3–Si interface, as has been demonstrated for other SiNx capped layers [17].

Capacitance–voltage analysis of the Ga2O3 / SiNx stack was attempted, although no

Qf or Dit data could be extracted due to a very strong hysteresis eﬀect, commonly associated with mobile charge within the ﬁlms.

p - t y p e , 1 . 6 Ωc m , F Z , < 1 0 0 > 3 8 0 0 G a 2 O 3 / S i N x )

s o

µ 4 0 0 C a n n e a l (

f 3 6 0 0

e a f t e r 3 0 f l a s h e s ؔ o e 3 0 0 C a n n e a l m

i 3 4 0 0 t a f t e r 3 0 f l a s h e s e f i l

e 3 2 0 0 i r r a c

s 3 0 0 0 s e c x 2 8 0 0 E

2 6 0 0

0 5 0 1 0 0 1 5 0 C u m u l a t i v e a n n e a l t i m e t ( m i n s )

15 −3 Fig. 2: Excess carrier lifetime at an injection level of ∆n = 1 × 10 cm for Ga2O3 / SiNx passivated samples as a function of cumulative anneal time. Of signiﬁcance is the slower rate of degradation in the lifetime with annealing compared with uncapped Ga2O3 samples reported in [10], and the slight recovery of the lifetime with illumination. The lines in the ﬁgure are guides for the eyes.

The annealing behaviour of the Ga2O3 / SiNx stack was investigated by annealing Ga2O3

◦ / SiNx passivated samples in forming gas at 300 and 400 C. The excess carrier lifetime Chapter 3. Surface Passivation by Gallium Oxide 131 of the samples, reported at an injection level of ∆n = 1 × 1015 cm−3, is plotted in Fig.

2 as a function of cumulative annealing time. From the ﬁgure, we observe that the lifetime of the Ga2O3 / SiNx stack was highest prior to annealing with the degradation of the passivation occurring at a much slower rate compared with the uncapped Ga2O3

passivated data of [10]. This is indicative of enhanced thermal stability of the SiNx

capped Ga2O3 layer, a critical consideration with respect to industrial processing and

reliability. Furthermore, a slight recovery in the lifetime after exposure to 30 ﬂashes

from the light source in the Sinton WCT-120 lifetime tester was observed, as indicated

in the ﬁgure. The increase in the lifetime observed after the light exposure was not

stable, and relaxed back to the original value soon after the measurement was taken and

the external excitation ceased.

B. Recombination at boron diﬀused p+ surfaces

The data plotted in Fig.3 show the recombination parameter J0 extracted from boron

+ diﬀused p surfaces as a function of Ga2O3 thickness for samples with and without

SiNx capping layers. In this instance, the uncapped samples have been annealed for 30

◦ min at 300 C in forming gas, while the SiNx capped samples are as-deposited. Also

displayed in the ﬁgure is data taken from an Al2O3 passivated control sample on the

same diﬀusion, after annealing at 425 ◦C for 30 min in forming gas. While both datasets

show a trend of decreasing J0 with increasing Ga2O3 thickness, the J0 extracted from

SiNx capped samples indicates that a thinner Ga2O3 layer is permissible to achieve

−2 suﬃciently low surface recombination, with the J0 saturating to values ∼30 fA.cm

at thicknesses of only 4 nm, compared with 20 nm for the uncapped samples. For

greater Ga2O3 thicknesses, little impact is observed on the J0 as the presence of the

boron diﬀusion mitigates the eﬀect of residual surface defects, and Auger recombination Chapter 3. Surface Passivation by Gallium Oxide 132

dominates. Interestingly, the SiNx capping has a demonstrably negative eﬀect on the thinnest Ga2O3 sample (∼2 nm), with the J0 more than double that of the uncapped sample. In this instance, we expect the thin Ga2O3 layer is failing to completely screen the positive charge in the thicker nitride ﬁlm, measured to be ∼ 5.6 × 1011 cm−2 [18], resulting in a higher minority carrier (electron) concentration at the surface, and hence a higher recombination rate. ) 3 - m c (

) A 1 9 2

N 1 0 1 0 0 n m o i t c a / r t n A e f c ( n

o 0 c 1 8 J n 1 0

o r r o B e t e 0 . 0 0 . 2 0 . 4 0 . 6 m

a D e p t h x ( µm ) r a p

n o i t a n i b Ω Ω m p / p + , 1 0 0 c m , 9 0 / s q , F Z , < 1 0 0 > o c G a 2 O 3 A l 2 O 3 e R G a 2 O 3 / S i N x

0 5 1 0 1 5 2 0 2 5 3 0

G a 2 O 3 t h i c k n e s s t ( n m )

Fig. 3: Recombination current densities measured on boron diffused surfaces (Rsh ∼ 90 Ω/) as a function of Ga2O3 film thickness, with and without SiNx capping layers. The diffusion profile, measured by ECV, is displayed in the inset graph. Note that the SiNx capped samples were not annealed, while the Ga2O3 samples were annealed for 30 min ◦ ◦ at 300 C and the Al2O3 sample for 30 min at 425 C, both in forming gas. The error bars are indicative of the spread in extracted J0 values taken from the linear region of the inverse lifetime data.

−2 −2 The best results reported in the ﬁgure: 23 fA.cm for the Al2O3 and 26 fA.cm for the

Ga2O3 passivated sample are approaching the Auger limit for this diffusion, calculated to be 21 fA.cm−2 using the diffusion profile measured by ECV (see inset of Fig.3) as the input to the software EDNA 2 [19], assuming no Shockley–Read–Hall recombination in the diffused region. The modelling of the diffusion in EDNA 2 also validates the J0s Chapter 3. Surface Passivation by Gallium Oxide 133

data of Fig.3: the data of Fig.1 can be modelled in EDNA 2 with J0s values of 2.5 and

−2 5.5 fA.cm for the Al2O3 and Ga2O3 samples, respectively, within the range of values extracted from the data in Fig.1(a).

C. Recombination at pyramidally textured surfaces

The investigation of the passivation on nonplanar surface morphologies is critical for device integration as silicon solar cells are routinely manufactured with textured surfaces to increase optical coupling into, and within, the device. Pyramidally textured samples are the industry standard texture for c-Si solar cells manufactured on monocrystalline wafers and are formed by etching h100i oriented single crystal wafers to expose h111i

orientated facets.

The exposure of a h111i orientated surface, as well as an increase in surface area by

a factor of ∼1.7 resulting from the textured surface topology, can result in a diﬀerent

recombination behaviour on textured surfaces compared with planar h100i surfaces. So

too, potentially, the presence of edges, the presence of edges and vertices associated

with pyramidal textures [20], [17].By analyzing the surface recombination behavior of

Ga2O3 on planar h100i and h111i, as well as textured surfaces, we attempt to separate

the inﬂuence of each individual eﬀect on the overall recombination behaviour.

The data of Fig.4 compare the ratio of Seff,UL measured on planar h111i and h100i

wafers, as well as the ratio between the surface area corrected Seff,UL extracted from

random pyramid textured samples with the planar h111i data, taken from a variety of

passivating materials in a previous investigation [17]. The data in the far left of the

ﬁgure are representative of the same ratios taken from the J0s of samples passivated

◦ with ∼20 nm of Ga2O3 and annealed in forming gas for 30 min at 300 C, as it is a more Chapter 3. Surface Passivation by Gallium Oxide 134

o 1 0 i t { 1 1 1 } / { 1 0 0 } a * t h i s w o r k r

c T e x t u r e / { 1 1 1 } i r t e m

n o i t a n i b

m 1 o c e r

d e t c e r r o c - a e r

A 0 . 1 x 2 3 ) ) i * N O O 9 3 3 i i l . . - S O / S S 2 1 2 2 A a a 5 A = = G O G D n n a 2 F L ( ( T A x i N i N x S S

Fig. 4: Area-corrected recombination metric ratio for a variety of materials on un- diﬀused c-Si surfaces. The Ga2O3 data from this paper were calculated from the J0s values measured on 1 Ω.cm n-type wafers. The data to the right of the dotted line (from [17]) represent the ratio of Seff,UL for the diﬀerent passivating schemes. consistent recombination metric, given the dopant density of the n-type wafers used in

14 15 −3 this study (ND ≈ 5 × 10 – 5 × 10 cm ) and the high negative charge density at the

Ga2O3 – Si interface [14].

The h111i/h100i ratio of 2.6 for the Ga2O3 data in Fig.4 indicates a strong eﬀect of crystal orientation on the recombination at the Ga2O3–Si interface, similar to that of

Al2O3. However, the ratio of the random pyramid texture to the planar h111i J0s (∼1.4 after correcting for surface area) is indicative of further non-negligible recombination eﬀects from the topology, i.e., the edges and vertices of the textured surface. While this ratio is similar to most of the other materials in Fig.4, it is slightly higher than the value for Al2O3 (∼0.9). It is expected, therefore, that textured surfaces passivated with

Ga2O3 will consistently have slightly higher rates of recombination compared with those passivated with Al2O3. Chapter 3. Surface Passivation by Gallium Oxide 135 ) s / Ω

m p - t y p e , 1 . 6 c m , F Z , < 1 0 0 > c (

p l a n a r G a O L 2 3 U , f f e p l a n a r G a 2 O 3 / S i N x S

y r a n t e x G a O t 2 3 i

c 1 0 0

o r a n t e x G a O / S i N l 2 3 x e v

n o i t a n i b m o

c 1 0 e r

e c a f r u S 0 5 1 0 1 5 2 0 2 5 3 0

G a 2 O 3 t h i c k n e s s t ( n m )

Fig. 5: Seff,UL as a function of Ga2O3 thickness for planar and textured samples, with and without SiNx capping. The lines in the ﬁgure are guides for the eyes.

Fig.5 plots Seff,UL as a function of Ga2O3 thickness for both planar and random pyramid textured p-type samples, with and without SiNx capping layers, prepared in the same manner as the samples of Fig.3. From the data of Fig.5, we see that for the thinnest Ga2O3 samples, similar to the J0 data on diﬀused surfaces of Fig.3, the

SRV increases after SiNx capping. However, at a Ga2O3 thickness of only ∼4 nm, the

addition of the SiNx capping layer results in a halving of the SRV compared with the

uncapped samples, as the thicker Ga2O3 layer screens the eﬀect of the positively charged

silicon nitride. Furthermore, while the SRV data of Fig.5 are consistent with the ratios

of Fig.4 for the uncapped samples, the data are also indicative that the SiN x capped

samples exhibit slightly higher ratios than the uncapped samples, as the lower SRV s on

planar surfaces do not translate proportionally to random pyramid textured surfaces.

This is evidenced by the ratio of the texture to planar Seff,UL data in Fig.5 taken at

a Ga2O3 thickness of 30 nm: for the uncapped samples, this is 2.5, compared with 3 for

the SiNx capped samples, after correcting for surface area. Chapter 3. Surface Passivation by Gallium Oxide 136

p + / n , 1 0 0 Ωc m , 8 5 - 9 0 Ω/ s q , F Z

) p l a n a r G a O

1 2 3 - s (

p l a n a r A l O

r 2 3 r o c 4 0 0 0

r a n t e x G a O ؔ

2 3 e r a n t e x A l O 2 m 2 3 i

t 9 4 f A / c m e f i

l 2 5 6 f A / c m d e t c e r r 2 0 0 0 o 2 c

4 4 f A / c m e s r e v 2 n

I 3 3 . 5 f A / c m

0 5 . 0 x 1 0 1 5 1 . 0 x 1 0 1 6 ( n ( c m - 3 פ E x c e s s c a r r i e r d e n s i t y

Fig. 6: J0 extraction on planar and random pyramid textured samples passivated with Ga2O3 and Al2O3.

We also extended the analysis of orientation and surface topology to boron diﬀused p+ on pyramidally textured and planar h100i surfaces, where the diﬀusion process was identical to that of the samples in Fig.3( Rsh ≈ 85 – 90 Ω/). The results are presented in Fig.

6. While the J0 values are consistently higher for both the planar Al2O3 and Ga2O3 passivated samples compared with Fig.3, the data are consistent with the Ga 2O3 passivated textured sample having a disproportionately higher J0 compared with the

Al2O3, as expected from the data of Fig.4.

D. Firing stability and crystallinity

Thermal stability of passivation is a critical aspect of a material?s suitability to industrial processing as typical screen-printed metal pastes used in the manufacturing of silicon solar cells are ﬁred at high temperatures (> 700 ◦C). While other lower temperature metallization technologies do exist, including low temperature screen printed pastes and Chapter 3. Surface Passivation by Gallium Oxide 137 electroplating, the majority of industrial production is performed with a short high- temperature processing step.

Previous experiments have demonstrated that the passivation by Ga2O3 is thermally

sensitive, with depassivation of the Ga2O3 – Si interface occurring at moderate annealing

temperatures after extended annealing times [8], [10]. In this experiment, we subject the

Ga2O3 and Ga2O3 / SiNx ﬁlms to the thermal budget of an industrial ﬁring process by

passing the samples through an inline infrared belt furnace (Schmidt 7K9-70C69-SLIR)

at a nominal peak temperature of 800 ◦C for approximately 2–3 s.

1 0 1 ) s m (

f 0 f

e 1 0 ؔ

e m i t e f i l

- 1 r 1 0 e

i Ω

r p - t y p e , 1 . 6 c m , < 1 0 0 > , F Z r

a G a O / S i N a s d e p o s i t e d

c 2 3 x

s o s G a 2 O 3 3 0 0 C , 1 2 0 m i n F G A e - 2 c 1 0 x G a 2 O 3 / S i N x f i r e d E

G a 2 O 3 f i r e d i n t r i n s i c l i f e t i m e ( R i c h t e r e t a l . )

1 0 1 4 1 0 1 5 1 0 1 6 ( n ( c m - 3 פ E x c e s s c a r r i e r d e n s i t y

Fig. 7: Injection-dependent lifetime data of Ga2O3 and Ga2O3 / SiNx passivated samples before (closed squares) and after (open squares) ﬁring in an industrial belt furnace.

The injection-dependent lifetime of a Ga2O3 / SiNx passivated sample (prepared in

the same manner as those samples in Fig.1) before and after ﬁring is shown in Fig.7.

Unlike the uncapped sample that was subjected to the same ﬁring process, whose lifetime

degraded below 200 µs, the silicon nitride capped samples degraded only marginally after

firing, confirming that SiNx capping improves the thermal stability of the Ga2O3 films, Chapter 3. Surface Passivation by Gallium Oxide 138

as indicated by the data in Fig.2. The Seff,UL of the Ga2O3 / SiNx passivated sample degraded from 1.4 to only 3.3 cm.s−1 after ﬁring, corresponding to an increase in the

−2 J0s to approximately 6 fA.cm .

To further investigate the impact of annealing on the Ga2O3 ﬁlms, grazing incidence

X-ray diffraction (GIXRD) measurements (PANalytical X’Pert PRO) were carried out on Ga2O3 deposited on polished h100i silicon samples. A subset of the samples were annealed for 2 h in forming gas at temperatures between 300 and 600 ◦C at 100 ◦C increments; another sample was annealed in a rapid thermal annealing (RTA) process for 2 min at a peak temperature of 800 ◦C; while the final sample was subjected to the same firing conditions in the belt furnace described above. The resulting XRD spectra are displayed in Fig.8.

) o

s 8 0 0 C b e l t f u r n a c e t i n u o y

r 8 0 0 C R T A a r t i b r

a o (

6 0 0 C I

y t i

s o n 5 0 0 C e t n i

e a s d e p o s i t e d s i l a m r

o s i l i c o n r e f e r e n c e N

2 0 3 0 4 0 5 0 6 0 7 0 8 0 ( A n g l e 2 ؑ ( d e g

Fig. 8: GIXRD of Ga2O3 ﬁlms after diﬀerent thermal treatments. The data indicate that the onset of crystallization occurs between 500 and 600 ◦C. The features between 50 ◦ and 60 ◦ are an artefact from the c-Si substrate.

The XRD data conﬁrm that the ﬁlms are amorphous as deposited, supporting the conclusion drawn from the infrared spectroscopy data of [9]. From Fig.8, it is evident Chapter 3. Surface Passivation by Gallium Oxide 139

that the onset of crystallization of the PEALD Ga2O3 ﬁlms occurs at between 500 and

600 ◦C (note that data from the 2 h annealing at 300 and 400 ◦C were indistinguishable

from the 500 ◦C data and so are not displayed), and the 800 ◦C RTA sample also shows

the evidence of crystallization. The XRD data also indicate that, while the ﬁlms an-

nealed at temperatures below 600 ◦C consistently display a small feature in the X-ray

spectra between 30 ◦ and 40 ◦, there is no evidence of structural changes in the ﬁlm

that correlate with the degradation of the lifetime evident in previous studies [8], [10].

Signiﬁcantly, the sample that passed through the belt furnace also shows no evidence of

crystallization, which is likely to have a detrimental eﬀect on the passivation quality due

to the formation of defects via strain and lattice mismatch at the Ga2O3 – Si interface.

The lack of crystalline features in the XRD data for the belt furnace ﬁred sample is a

possible indication that: 1) the actual temperature realized by the sample in contact

with the belt is lower than the set point temperature; and 2) the smaller thermal budget

of the belt furnace ﬁring procedure (that is, the shorter time at the peak temperature) is

suﬃcient to prevent the formation of a crystalline phase in the ﬁlms. The data of Fig.8

are also indicative that the degradation in lifetime of the uncapped samples was unlikely

a result of structural changes in the Ga2O3 ﬁlm, and rather further evidence of depassi-

vation of the Ga2O3 – Si interface, as demonstrated in other thermal annealing studies

[8], [10]. Evidently, the SiNx capping layer, perhaps through further hydrogenation of

interface states, was able to oﬀset or reduce the extent of the degradation experienced

by the uncapped samples. This is again consistent with the thermal annealing results of

Fig.2, which demonstrate that the Ga 2O3 / SiNx stack is more thermally stable than

uncapped Ga2O3. Chapter 3. Surface Passivation by Gallium Oxide 140

IV. Conclusion

We have demonstrated an improvement in the surface passivation of c-Si by Ga2O3 with the addition of a PECVD SiNx capping layer, with excess carrier lifetimes approaching the intrinsic Auger limit without the need for a post-deposition anneal. We have also shown that, on planar surfaces, both Ga2O3 and Ga2O3 / SiNx ﬁlms are able to pas-

+ sivate boron diﬀused p surfaces in a commensurate fashion to Al2O3. However, the passivation of textured silicon is somewhat compromised by poorer passivation of h111i compared with h100i surfaces, and by evidence of contributions to recombination by edges and vertices, with little improvement after SiNx capping. Initial studies into the

firing stability of the films indicate that the Ga2O3 / SiNx stack is firing stable. These advancements are critical to module integration vis-à-vis the electrical and optical performance of passivation and antireflection schemes on front-side boron diffused n-type solar cells and rear-side passivating schemes on partial rear contact p-type devices.

Acknowledgements

The authors would like to thank Dr. Z. Hameiri for his assistance in operating the ﬁring furnace, and Dr. P. Phang for his assistance with the PECVD depositions. Chapter 3. Surface Passivation by Gallium Oxide 141

References

[1] “International Technology Roadmap for Photovoltaic (ITRPV).” [Online]. Available: http://www.itrpv.net/Home/. Accessed on: Feb. 4, 2016.

[2] A. G. Aberle, “Overview on SiN surface passivation of crystalline silicon solar cells,” Solar Energy Mater. Solar Cells, vol. 65, nos. 1–4, pp. 239–248, Jan. 2001.

[3] M. J. Kerr, “Surface, emitter and bulk recombination in silicon and development of silicon nitride passivated solar cells,” Ph.D. dissertation, Australian Nat. Univ., Canberra, ACT, Australia, Jun. 2002.

[4] S. Dauwe, L. Mittelst¨adt,A. Metz, and R. Hezel, “Experimental evidence of parasitic shunting in silicon nitride rear surface passivated solar cells,” Prog. Photovoltaics Res. Appl., vol. 10, no. 4, pp. 271–278, Jun. 2002.

[5] A. Richter, J. Benick, and M. Hermle, “Boron emitter passivation with Al2O3 and Al2O3 / SiNx stacks using ALD Al2O3,” IEEE J. Photovoltaics, vol. 3, no. 1, pp. 236–245, Jan. 2013.

[6] B. Liao, B. Hoex, A. G. Aberle, D. Chi, and C. S. Bhatia, “Excellent c-Si surface passivation by low-temperature atomic layer deposited titanium oxide,” Appl. Phys. Lett., vol. 104, no. 25, Jun. 2014, Art. no. 253903.

[7] B. Liao, B. Hoex, K. D. Shetty, P. K. Basu, and C. S.Bhatia, “Passivation of boron- doped industrial silicon emitters by thermal atomic layer deposited titanium oxide,” IEEE J. Photovoltaics, vol. 5, no. 4, pp. 1062–1066, Jul. 2015.

[8] T. G. Allen and A. Cuevas, “Electronic passivation of silicon surfaces by thin ﬁlms of atomic layer deposited gallium oxide,” Appl. Phys. Lett., vol. 105, no. 3, Jul. 2014, Art. no. 031601.

[9] T. G. Allen and A. Cuevas, “Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: Demonstration of surface passivation and negative interfacial charge,” Physica Status Solidi Rapid Res. Lett., vol. 9, no. 4, pp. 220–224, Apr. 2015.

[10] T. G. Allen, M. Ernst, C. Samundsett, and A. Cuevas, “Demonstration of c-Si solar cells with gallium oxide surface passivation and laser-doped gallium p+ regions,” IEEE J. Photovoltaics, vol. 5, no. 6, pp. 1586–1590, Nov. 2015.

[11] A. Richter, S.W. Glunz, F.Werner, J. Schmidt, and A. Cuevas, “Improved quanti- tative description of Auger recombination in crystalline silicon,” Phys. Rev. B, vol. 86, no. 16, Oct. 2012.

[12] K. Misiakos and D. Tsamakis, “Accurate measurements of the silicon intrinsic carrier density from 78 to 340 K,” J. Appl. Phys., vol. 74, no. 5, pp. 3293–3297, Sep. 1993.

[13] D. E. Kane and R. M. Swanson, “Measurement of the emitter saturation current by a contactless photoconductivity decay method,” in Proc. IEEE Photovoltaic Spec. Conf., 1985, vol. 18, pp. 578–583. Chapter 3. Surface Passivation by Gallium Oxide 142

[14] K. R. McIntosh and L. E. Black, “On eﬀective surface recombination parameters,” J. Appl. Phys., vol. 116, no. 1, Jul. 2014.

[15] A. Cuevas, “The recombination parameter J0,” Energy Procedia, vol. 55, pp. 53–62, 2014.

[16] A. Richter, J. Benick, M. Hermle, and S. W. Glunz, “Reaction kinetics during the thermal activation of the silicon surface passivation with atomic layer deposited Al2O3,” Appl. Phys. Lett., vol. 104, no. 6, Feb. 2014.

[17] Y. Wan, J. Bullock, and A. Cuevas, “Passivation of c-Si surfaces by ALD tantalum oxide capped with PECVD silicon nitride,” Solar Energy Mater. Solar Cells, vol. 142, pp. 42–46, Nov. 2015.

[18] Y. Wan, K. R. McIntosh, A. F. Thomson, and A. Cuevas, “Low surface recombination velocity by low-absorption silicon nitride on c-Si,” IEEE J. Photovoltaics, vol. 3, no. 1, pp. 554–559, Jan. 2013.

[19] K. R. McIntosh and P. P. Altermatt, “A freeware 1D emitter model for silicon solar cells,” in Proc. IEEE 35th Photovoltaic Spec. Conf., 2010, pp. 002188–002193.

[20] S. C. Baker-Finch and K. R. McIntosh, “The contribution of planes, vertices, and edges to recombination at pyramidally textured surfaces,” IEEE J. Photovoltaics, vol. 1, no. 1, pp. 59–65, Jul. 2011. Chapter 4

Calcium-Based Electron Contacts

4.1 Introduction

Recent years have seen significant improvements in the energy conversion efficiency of crystalline silicon photovoltaic devices at both the research and industrial scale. Prior to 2014 the record energy conversion efficiency for c-Si solar cells was 25% which was attained by a 4 cm2 device fabricated in 1999 - the so-called PERL (passivated emitter rear locally diffused) cell [1],[2]. The PERL cell itself was invented at UNSW in 1989, with a reported efficiency of 23.2% [2]. Throughout the course of the 1990s, all solar cell efficiency records came from sequential improvements in the PERL cell design at UNSW, resulting in 25 years of dominance and an increase of 1.8% absolute energy conversion efficiency over this entire time period [2],[3].

Since 2014, the UNSW PERL cell eﬃciency record was ﬁnally surpassed by solar cell manufacturers on devices fabricated on large area (> 150 cm2), industrially relevant

wafers: ﬁrst by Sunpower (25%; now 25.2%) [4],[5], and then by Kaneka (25.1%) [6] and

Panasonic (25.6%) [7]. On the laboratory scale, researchers at Fraunhofer ISE fabricated

143 Chapter 4. Calcium-Based Electron Contacts 144 a small area device (4 cm2) with an efficiency of 25.1% [8]. More recently, Kaneka, after moving to an all back contact cell architecture to minimise optical reflection losses from the front metal grid, reported a device with a confirmed efficiency of 26.6% – the current high watermark for c-Si solar cells [3],[9].

Put into historical context, in just the past two years c-Si solar cell research has yielded an additional 1.6% improvement in record device efficiency, compared to just 1.8% over the preceding 25 years (from 1989 to 2014). This is even more remarkable considering the theoretical efficiency limit for silicon solar cells is only 29.4% [10] – it is intuitive to recognise that the closer the maximum efficiency limit, the harder the incremental efficiency improvements become.

Tellingly, all of the devices to surpass the 25% eﬃciency record have come from solar cells with passivated contacts. Indeed, the most successful devices have both electron and hole contacts passivated. However it is quite remarkable that these high performance devices have come from a variety of cell architectures: front and back side contacted; all back contacted; full area contacts; partial area contacts. It is clear that the common path to achieving high performance devices is to passivate the contact regions to minimise energy losses associated with the recombination of charge carriers.

This Chapter is an in depth look at the energy losses that occur at the metal contacts of silicon solar cells, and ways in which they are minimised. It begins with a deﬁnition of the contact, its contribution to the functionality of a solar cell, and its eﬃcacy with reference to the notion of carrier selectivity. The methods by which carrier selectivity is achieved in c-Si solar cells is reviewed, then a novel contact structure based on the low work function metal calcium, and later calcium/titania, is explored in detail. Chapter 4. Calcium-Based Electron Contacts 145

4.2 Carrier selectivity

To operate as a solar cell, electrical contact has to be made to a light-absorbing semiconductor in order to extract the photo-excited electrons and holes and drive an external circuit. It follows that there must be two spatially separated contacts, one for electrons and another for holes, to create the negative and positive terminal of the device. These contacts work by simultaneously permitting the ﬂow of one charge carrier at one contact, while blocking the other, and vice versa at the contact of opposite polarity [11].

Taking this approach, it follows that an electrical contact works by maximising the conductivity of one carrier (the selected carrier) towards the contact, while suppressing the conductivity of the other [12]. In the vicinity of the electron contact, for example, the conductivity of electrons is maximised, while the conductivity for holes is minimised.

The opposite is true for the hole contact.

The asymmetry in conductivities can be achieved in a variety of ways; heavy doping underneath the metal contact region being by far the most common. Other methods include inducing band bending underneath the contact (as per the passivation affect outlined in the preceding chapter), or through the alignment of the energy bands of the contact materials with the absorber material (the heterojunction approach). Generally though, each method has a complex dependence on the contact materials used, and the effect they have on the absorber interface and sub-surface properties. A general definition of the contact must therefore pay consideration, not just to the absorber surface properties at the contact interface, but also to the absorber material’s sub-surface region where the asymmetry in conductivities prevails.

Energy losses associated with the formation of the electrical contacts occur and take two probeable forms: that of an electrical resistance that impedes the ﬂow of collected Chapter 4. Calcium-Based Electron Contacts 146 carriers across the semiconductor/contact interface; and an additional recombination loss through defects at the semiconductor/contact interface. These losses are characterised

2 by a contact resistivity ρc (Ω.cm ) and a contact recombination current density J0c (an analogous contact-area speciﬁc metric to the surface recombination current density J0s introduced in the previous chapter). The eﬃcacy of an electrical contact is therefore governed by the extent to which it simultaneously suppresses resistive losses (minimises

ρc) and recombination losses (minimises J0c) [13],[14]. With respect to the contact deﬁnition provided above, this can be thought of as minimising the resistance (ρc) posed to the charge carrier that passes through the contact (i.e. the selected carrier; the carrier that is to be extracted from the contact) and maximising the resistance posed to the non-selected carrier (insofar as minimising J0c implies maximising a ‘recombination resistance’).

Since high efficiency c-Si solar cell designs have been known to be limited by losses at the contacts [15], the interplay between ρc, J0c, and the areal contact fraction fc, has become an area of increased research interest. In order to quantify the quality (or selectivity) of a given contact structure, and thus compare different contacts reported in the literature, Brendel and Peibst recently introduced a ‘selectivity parameter’ S, where S = Vth/(J0cρc) [14]. Here Vth is the thermal voltage (kT/q) which depends only on temperature, while ρc and J0c are the material-dependent contact parameters defined above. The value of S, and hence ρc and J0c, can be directly related to an upper limit cell efficiency. Being able to accurately quantify ρc and J0c is therefore critical to understanding the impact of a given contact structure on device performance. Chapter 4. Calcium-Based Electron Contacts 147

4.2.1 Carrier transport at the contacts

The direct metallisation of silicon surfaces to form both electron and hole contacts would, ideally, be possible by the choice of diﬀerent metals of appropriate work function. Ac- cording to Schottky-Mott theory, the work function of a metal, when in contact with a semiconductor, will result in the formation of an energy barrier ΦB at the metal/semiconductor interface that follows the relationship:

ΦBn = φM − χS (4.1)

ΦBp = (Eg + χS) − φM

for an n-type and p-type semiconductor, respectively. Here φM is the metal work func-

tion and χS is the electron aﬃnity of the semiconductor. This model for barrier height

formation implies that the barrier height (i.e. the extent of band bending at the semi-

conductor surface), and hence the selectivity of a contact, can be directly and simply

modulated by choosing metals with a reasonable asymmetry in their work functions.

For c-Si, this would imply that Ti would create a barrier of approximately 0 eV for

electrons, and Pt approximately 0 eV for holes. Metals with even greater asymmetry in

their work functions would then bend bands further into accumulation of the collected

carrier to the betterment of both ρc and J0c.

The ability to use metal work functions to enable carrier selectivity in this way, however,

is seldom realised in practice. In reality, when metals are directly applied to the surface

of silicon the relationship between barrier height and work function is greatly divergent

from that of equation 4.1. This is shown in Figures 4.1 and 4.2, where the empirically Chapter 4. Calcium-Based Electron Contacts 148 determined barrier height is plotted as a function of metal work function for n- and p- type silicon (note the data for the figures has been taken from Ref. [16]). For reference, the Schottky-Mott relationship of equation 4.1 is also displayed. The linear fit to the data shows that the metal work function has far less influence on the barrier height formation than equation 4.1 would predict. It is also demonstrated that, empirically, barrier heights for common metals tend to be larger on n-type silicon than on p-type, indicated by the fact that the barrier height data tend to lie above the value predicted by Schottky-Mott theory on n-type silicon, while for barriers to holes on p-type silicon the opposite is true. This is due to a pheonomenon called Fermi level pinning, whereby defect states at the metal–silicon interface hold the Fermi level at the surface to values typically below the intrinsic Fermi level for a wide range of φM [17].

In the absence of heavy doping (or heavily accumulated contact interfaces, more generally) where the collected carrier concentration at the surface is < 1 × 1017 cm−3, the current transport over the energy barriers at the metal/silicon contacts is dictated by thermionic emission. The relationship between barrier height and contact resistivity for the thermionic emission process is given by:

k qΦ ρ (TE) = exp B (4.2) c qA∗∗T kT

Where A∗∗ is the reduced eﬀective Richardson constant, k is Boltzmann’s constant, and

T temperature. Inputting the linear parameterisation of the data of Figure 4.1 (i.e.

ΦB(E)), into the thermionic emission model, one can predict the contact resistivity for any given metal work function. This is plotted in Figure 4.3 for electron contacts on n-type silicon. Chapter 4. Calcium-Based Electron Contacts 149

Fig. 1: Empirical values of barrier height for a range of metals with diﬀerent work functions deposited on n-type silicon. The blue dashed line-of-best-ﬁt is indicative of the deviation from the ideal Schottky-Mott behaviour.

Note that the data of Figure 4.3 represents the room temperature contact resistivity

(as derived from the paramterisation of ΦB(E)) for an electron contact on an n-type silicon surface. From the ﬁgure, if a metal work function of < 3 eV is targeted, contact resistivities below 10 mΩ.cm2 could be achievable. Note this is working within the conﬁnes of Fermi level pinning, at least as far as the linear parameterisation of Figure

4.1 holds.

One such low work function metal is calcium (Ca; φ ∼ 2.9 eV), which is commonly used as the electrode in organic electronic devices. The journal articles that follow document the application of Ca contacts on silicon, including the formation of a heterocontact using a TiO2 passivating interlayer. Chapter 4. Calcium-Based Electron Contacts 150

Fig. 2: Empirical values of barrier height for a range of metals with diﬀerent work functions deposited on p-type silicon. The blue dashed line-of-best-ﬁt is indicative of the deviation from the ideal Schottky-Mott behaviour. Note that many of the values for barrier height lie below ideal. The data from the previous graph is left for comparison.

When the concentration of collected carriers exceeds ∼ 1×1017 cm−3 transport across the interfacial energy barrier becomes dominated by tunnelling processes: ﬁrst thermionic-

ﬁeld emission, and then ﬁeld emission. In this instance, the barrier height formed at the metal–silicon interface does not change markedly from those values represented in

Figures 4.1 and 4.2 (ignoring the effect of image force lowering), but rather the width of the barrier narrows [16]. In the case of thermionic-field emission, the barrier width narrows to the extent that only thermally excited carriers can tunnel through the barrier width, unlike in field emission where non-thermally excited carriers are able to pass through the barrier. These carrier transport processes, which are dominant in all diffused junction solar cells, allow for very low contact resistivities even in the presence of large Chapter 4. Calcium-Based Electron Contacts 151

Fig. 3: Paramterisation of the barrier height and thermionic emission contact resistivity vs. metal work function for n-type silicon, using the line-of-best-ﬁt data from ﬁgure 1. interfacial energy barriers.

4.2.2 Suppressing contact recombination

The conceptual understanding of the suppression of recombination has been discussed in

Chapter 3, and those concepts for minimising J0s hold for minimising J0c. However the need for carrier transport through the contact places severe constraints on the means by which the recombination rate at the contact/silicon interface can be reduced. Either the passivation material has to be made suﬃciently thin for tunnelling processes to occur

(< 2 nm; which typically compromises the suppression of J0c), or thicker but conductive, forming a hetero-contact to the silicon material. Chapter 4. Calcium-Based Electron Contacts 152

In diﬀused junction solar cells, the addition of dopants to form the electron and hole contacts suppresses the surface recombination rate by establishing a very low minority carrier concentration in the vicinity of the unpassivated surface defects. This approach

−2 leads to an Auger-limited minimum in J0c at the metallized contact of > 300 fA.cm

[18]. Since the contact resistivity of heavily doped contacts can be very low (on the order of 1 µΩ.cm2 for controlled metallization procedures like thermal evaporation [19]), the strategy toward high efficiencies has long been to make the contact fraction (fc) as small as possible, and to passivate the remaining un-metallized areas. Early examples are shown in the c-Si thermo-photovoltaic cell architecture of Swanson [20] and the original all-back-contact concept of Schwartz and Lammert [21]. This idea of minimising fc and passivating the remaining surface was developed over the course of the 1970s and 80s, culminating in the former world record efficiency PERL (passivated emitter, rear locally diffused) cell from UNSW (η = 25%) [1], [2]. Since that time, devices exceeding 25% efficiency values have come from n-type cells with chemically passivated contacts [22].

Early attempts to chemically passivate the metal-silicon interface took the form of tunnelling SiOx layers thermally grown into the silicon surface prior to metallisation. These metal-insulator-semiconductor (MIS) cells were noteworthy for their improvements in open circuit voltage, as highlighted by Godfrey and Green who reported a Voc of 655 mV in 1979, 20 mV higher than any previously reported c-Si device, demonstrating the potential of the passivated contact approach to cell design [23]. Early MIS (also known as MIS-IL, or metal-insulator-semiconductor inversion layer) cells relied on the passivating thermal oxide, silicon monoxide ARC and overlying Al at the local front contacts to form an induced junction on p-type substrates. (Following Figure 4.1, more eﬀec- tive induced junctions, in principle, should be possible using a metal with a lower work function that Al to act as the electron contact). The work of Hazel et al. demonstrated Chapter 4. Calcium-Based Electron Contacts 153 the application of PECVD silicon nitride, replacing the SiO ARC in [23], to increase the overlying positive charge aiding in the formation of the induced junction due to the higher density of positive charge in the SiNx layer [24],[25]. Hezel and co-workers would

+ later go on to add Cs ions to the SiNx layer to further increase the ﬁxed positive charge,

resulting in an increase in Voc and reduction in series resistance at the cell level [26],[27].

The MINP (metal-insulator n-p junction) cell concept evolved out of the MIS inver-

sion layer cell design. MINP cells replaced the inversion layer with a more conductive

phosphorus diﬀused front surface, while retaining the thin tunnel oxide to passivate the

surface defects within the contact structure [28],[29]. This cell design was the ﬁrst to

achieve eﬃciencies over 18% [2]. The overlying front metal grid was also composed of Ti,

a metal with a lower work function than Al, in order to aid in the extraction of electrons

at the front contact. Inspection of the early patents describing the MINP cell indicate

that the authors were clearly interested in using low work function materials in order to

promote more favourable band bending at the front electron contact [29]. Indeed, some

interations of MIS cells incorporated magnesium as the electron contact metal due to

its low work function (φ = 3.7 eV), and so subsequently low energy barrier presented to

electrons at the silicon surface [30]. Wan et al. would later incorporate Mg contacts on an full area rear a-Si passivated contact, achieving an eﬃciency of 19% [31].

Perhaps the most signiﬁcant advancement in passivated contact technologies occurred in the 1980s when semi-insulating polycrystalline silicon (SIPOS) with an interfacial tunnel oxide layer was ﬁrst demonstrated on c-Si for solar cell applications in passivation and contact formation [32],[33]. This conceptual breakthrough removed the dopants from the absorber material, placing them within the contact structure itself, thereby reducing Auger recombination losses. Initial demonstrations indicated remarkable surface

−2 passivation of c-Si by an n-type SIPOS/SiOx/c-Si structure with J0 values ∼10 fA.cm Chapter 4. Calcium-Based Electron Contacts 154 and open circuit voltages of 720 mV as early as 1985 [33]. There has recently been a revival in the interest of polysilicon contacts to c-Si solar cells with current state-of- the-art polysilicon contacts exhibiting contact recombination and resistivity values of

−2 2 0.1 < J0c < 10 fA.cm and ρc < 1 mΩ.cm for both n-type and p-type polysilicon contacts [34]. A 25.1% device using a full area n-type polysilicon rear contact with a boron diﬀused front hole contact has been demonstrated [8]. All polysilicon contacted

IBC devices have currently been demonstrated with an eﬃciency of 24.25% [34].

In a similar progression to that of the doped polysilicon contacts, the silicon heterojunction (SHJ) cell, commercialised by Panasonic as the HIT cell (heterojunction with intrinsic thin layer), replaces the tunnel oxide of the polysilicon-based contacts with a passivating a-Si layer, and also replaces the doped polysilicon with n- and p-doped a-Si layers [35]. Conceptually though, the polysilicon and a-Si based approaches are identical. Mixtures and variants of these approaches have also taken root: using doped micro- or nano-crystalline silicon [36][39] instead of a-Si, or combining SiOx [40],[41] or a-SiCx

[42] passivating layers with doped a-Si, for example.

More recently, keen interest has developed in applying materials from organic PV to c-Si, incentivised by the low cost of ownership, ease of manufacture and low cost of fabrication. These silicon-organic heterostructures have focused primarily on the hole contact, with organic polymers like PEDOT:PSS [43][46], P3HT [47], and spiro-OMeTAD [48].

The most successful iteration of this approach to carrier selection had been that of

Yu et al. whose device featured nano-scale front surface texturing with a spin-coated

PEDOT:PSS-based front hole contact on an n-type wafer. This device measured an efficiencies of 13% [49]. Within recent years, the work of Zielke et al. demonstrated a more competitive efficiency of 20.6% by optimising the surface passivation of the organic material by incorporating a room temperature native oxide layer, and by applying the Chapter 4. Calcium-Based Electron Contacts 155 material to an optimised p-type cell structure with the PEDOT:PSS layer on the planar full rear surface [50]. The same research group have demonstrated efficiencies on full area 6” devices of 19.5% [51].

Transition metal oxides (TMOs), again commonly utilised in organic PV and dye- sensitised solar cells, have also recently been successfully transferred to c-Si devices.

Sub-stoichiometric molybedenum oxide (MoOx, x < 3) has been demonstrated to form an eﬀective dopant-free hole contact when applied directly on to c-Si [52],[53]. The high work function of MoOx (φ ∼ 6.6 eV) and the defect band that arises from the reduced stoichiometry are thought to be the most salient features enabling the extraction of holes

[52]. Bullock et al. characterised the contact resistivity and recombination of the MoOx

2 −2 / c-Si contact to be ρc = 1 mΩ.cm and J0c ∼ 200 fA.cm [53]. Interestingly, the authors also showed the formation of an inversion layer at the MoOx / n-type silicon sub surface, evidence of a level of band bending indicative of a de-pinning of the Fermi level owing to the passivation of surface defects aﬀorded by the high work function MoOx.

Coupled with a passivating a-Si interlayer, the MoOx contact has been successfully integrated into a SHJ cell with an eﬃciency of 22.5% [54]. Other high work function TMOs

– tungsten and vanadium oxides – are also being investigated owing to their similar properties to MoOx, that is, their high work function and conductivity after thermal evaporation [55]–[57].

At the electron contact, recent developments have demonstrated that titanium dioxide

(TiO2), long a mainstay of dye-sensitised solar cell device architectures, can act as an electron contact to crystalline silicon [58]. This development has come despite a long history in the use of TiO2 in c-Si research and manufacturing as an ARC [59]. Prior to the publication of the journal articles that follow, the most successful implementation of the a TiO2 heterocontact at the solar cell level had been that of Yang et al., which Chapter 4. Calcium-Based Electron Contacts 156 achieved a power conversion eﬃciency of 19.8%, increasing to 21.6% with the addition of a SiO2 interlayer [60],[61]. These devices feature full area TiO2 contacts on the planarised rear of the solar cells due to the relatively high contact resistivities of the Al

/ TiO2 structure.

In addition to the use of TMOs and organic semiconducting layers for electron and hole contacts, low work function materials, common to OPV, OLED, and perovskite cell designs, have also gained interest in the c-Si community. The most notable example is that of the aluminium / lithium ﬂuoride (LiF) electron contact. When thermally evaporated, the LiF stoichiometry, like that of thermally evaporated molybdenum oxide, reduces to

LiFx, where x < 1 [62]. Coupling the LiFx with an overlying Al capping layer results in a low work function (φ ∼ 2.9 eV) localised at the Al / LiFx / Si interface. A corre-

2 spondingly low contact resistivity (ρc = 2 mΩ.cm ) results on undiffused 1 Ω.cm n-type silicon. This effect has been exploited to create an n-type partial rear contact cell without doping with an efficiency of 20.6% [63]. Thermally evaporated caesium carbonate

(Cs2CO3), which decomposes to Cs-rich CsOx during the evaporation procedure, has also been explored owing to its similarly low work function (φ ∼ 2.2 eV) [64]. Caesium carbonate oﬀers the added advantage over LiF as it retains its low work function after capping with Ag, a more highly reﬂective and conductive metal than Al.

Combinations of the low work function materials with passivating interlayers has followed as natural evolution of this technology. The most prominent example being the dopant- free asymetric heterocontact cell of Bullock et al., in which the high and low work function materials MoOx and LiFx feature as the hole and electron contact materials respectively. Both of these contacts were passivated by an intrinsic a-Si layer, resulting in a Voc of 712 mV. Series resistance issues appeared to limit the fill factor to 73% and the efficiency to 19.4%, however this result represents a marked improvement in similar Chapter 4. Calcium-Based Electron Contacts 157 cell designs reported in the literature [62]. In addition, recent publications have followed this dopant-free approach with thermally evaporated contacts to simplified IBC cells however currently efficiencies reported in the literature remain below 20% [65], [66]. Chapter 4. Calcium-Based Electron Contacts 158

4.3 Foreword

The papers that follow are the ﬁrst to report on calcium contacts to crystalline silicon solar cells. By using calcium as the contact metal, it is demonstrated that low resistivity electron contacts (on the mΩcm2 scale) can be achieved without heavy doping.

This enabled the fabrication of the directly metallised partial rear contact (PRC) cells reported in [67]. It is shown that this due to the low work function metal establishing a low energy barrier for electrons at the contact interface.

After demonstrating the utility of the low work function metal Ca in forming low resis- itivity electron contacts, the ﬁnal two papers demonstrate a ﬁrst-of-its-kind passivated n-type PRC cell using a similar cell structure to the directly metallised Ca PRC cell.

The major diﬀerence lies in the insertion of a TiOx interlayer between the c-Si and the

Ca contact.

The second paper presented in this chapter demonstrates the compatibility of with regards to the retention of low contact resistivities. The third and ﬁnal paper of this thesis demonstrates on the cell level the improvement over the directly metallised PRC cell.

This paper also looks in more detail at the contact using analytical TEM, demonstrating a reduction in the TiOx layer from x = 2 to x ∼ 1 after the application of the overlying

Ca metal, as well as the diﬀusion of Ca through the passivating TiOx layer. Both aﬀects are believed to be contributing to the low contact resistivity, in addition to the low work function of the metallic layer. Chapter 4. Calcium-Based Electron Contacts 159

4.4 References

[1] J. Zhao, A. Wang, and M. A. Green, “24.5% Eﬃciency silicon PERT cells on MCZ substrates and 24.7% eﬃciency PERL cells on FZ substrates,” Prog. Photovolt. Res. Appl., vol. 7, no. 6, pp. 471–474, Nov. 1999.

[2] M. A. Green, “The path to 25% silicon solar cell eﬃciency: History of silicon cell evolution,” Prog. Photovolt. Res. Appl., vol. 17, no. 3, pp. 183–189, May 2009.

[3] NREL eﬃciency chart. [Online]. Available: http://www.nrel.gov/pv/assets/ images/efficiency-chart.png. [Accessed: 18-Jan-2017].

[4] D. D. Smith, P. Cousins, S. Westerberg, R. D. Jesus-Tabajonda, G. Aniero, and Y. C. Shen, “Toward the Practical Limits of Silicon Solar Cells,” IEEE J. Photovolt., vol. 4, no. 6, pp. 1465–1469, Nov. 2014.

[5] D. D. Smith, G. Reich, M. Baldrias, M. Reich, N. Boitnott, and G. Bunea, “Silicon solar cells with total area eﬃciency above 25%,” in IEEE 43rd Photovoltaic Specialists Conference (PVSC), Portland, USA, 2016.

[6] D. Adachi, J. L. Hernandez, and K. Yamamoto, “Impact of carrier recombination on ﬁll factor for large area heterojunction crystalline silicon solar cell with 25.1% eﬃciency,” Appl. Phys. Lett., vol. 107, no. 23, p. 233506, Dec. 2015.

[7] K. Masuko, M. Shigematsu, T. Hashiguchi, D. Fujishima, M. Kai, N. Yoshimura, T. Yamaguchi, Y. Ichihashi, T. Mishima, N. Matsubara, T. Yamanishi, T. Takahama, M. Taguchi, E. Maruyama, and S. Okamoto, “Achievement of More Than 25% Conversion Eﬃciency With Crystalline Silicon Heterojunction Solar Cell,” IEEE J. Photovolt., vol. 4, no. 6, pp. 1433–1435, Nov. 2014.

[8] S. W. Glunz, F. Feldmann, A. Richter, M. Bivour, C. Reichel, H. Steinkemper, J. Benick, and M. Hermle, “The irresistible charm of a simple current ﬂow pattern — 25% with a solar cell featuring a full-area back contact,” in 31st European Photovoltaic Solar Energy Conference and Exhibition, Hamburg, Germany, 2015.

[9] NEDO: “Worlds Highest Conversion Eﬃciency of 26.33% Achieved in a Crystalline Silicon Solar Cell.” [Online]. Available: http://www.nedo.go.jp/english/news/ AA5en_100109.html. [Accessed: 22-Sep-2016].

[10] A. Richter, M. Hermle, and S. W. Glunz, “Reassessment of the Limiting Eﬃciency for Crystalline Silicon Solar Cells,” IEEE J. Photovolt., vol. 3, no. 4, pp. 1184–1191, Oct. 2013.

[11] P. W¨urfel, Physics of Solar Cells: From Principles to New Concepts, John Wiley & Sons, 2008.

[12] U. W¨urfel,A. Cuevas, and P. W¨urfel, “Charge Carrier Separation in Solar Cells,” IEEE J. Photovolt., 2014. Chapter 4. Calcium-Based Electron Contacts 160

[13] J. Bullock, “Advanced contacts for silicon solar cells,” PhD thesis, Australian Na- tional University, Canberra, ACT, 2016.

[14] R. Brendel and R. Peibst, “Contact Selectivity and Eﬃciency in Crystalline Silicon Photovoltaics,” IEEE J. Photovolt., vol. 6, no. 6, pp. 1413–1420, Nov. 2016.

[15] R. M. Swanson, “Approaching the 29% limit eﬃciency of silicon solar cells,” in Conference Record of the 31st IEEE Photovoltaic Specialists Conference, pp. 889–894, 2005.

[16] D. K. Schroder and D. L. Meier, “Solar cell contact resistance: A review,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 637–647, May 1984.

[17] R. T. Tung, “The physics and chemistry of the Schottky barrier height,” Appl. Phys. Rev., vol. 1, no. 1, p. 11304, Mar. 2014.

[18] D. Yan and A. Cuevas, “Empirical determination of the energy band gap narrowing in highly doped n+ silicon,” J. Appl. Phys., vol. 114, no. 4, p. 44508, Jul. 2013.

[19] K. C. Fong, T. C. Kho, A. Fell, E. Franklin, N. Zin, A. W. Blakers, K. R. McIntosh, T. Ratcliﬀ, M. Stocks, J. Bullock, and E. C. Wang, “Contact Resistivity of Evaporated Al Contacts for Silicon Solar Cells,” IEEE J. Photovolt., vol. 5, no. 5, pp. 1304–1309, Sep. 2015.

[20] R. M. Swanson, “Silicon photovoltaic cells in thermophotovoltaic energy conversion,” in International Electron Devices Meeting, vol. 24, pp. 70–73, 1978.

[21] R. J. Schwartz and M. D. Lammert, “Silicon solar cells for high concentration applications,” in International Electron Devices Meeting, vol. 21, pp. 350–352, 1975.

[22] C. Battaglia, A. Cuevas, and S. De Wolf, “High-eﬃciency crystalline silicon solar cells: status and perspectives,” Energy Env. Sci, vol. 9, no. 5, pp. 1552–1576, 2016.

[23] R. B. Godfrey and M. A. Green, “655 mV opencircuit voltage, 17.6% eﬃcient silicon MIS solar cells,” Appl. Phys. Lett., vol. 34, no. 11, pp. 790–793, Jun. 1979.

[24] R. Hezel, “Plasma Si nitride—A promising dielectric to achieve high-quality silicon MIS/IL solar cells,” J. Appl. Phys., vol. 52, no. 4, p. 3076, 1981.

[25] R. Hezel, “High charge densities in Si-nitride and their eﬀect on the inversion layer mobility of silicon MIS/IL solar cells,” in Conf. Rec. IEEE Photovoltaic Spec. Conf., 1982.

[26] K. Jager and R. Hezel, “Optical stability of silicon nitride MIS inversion layer solar cells,” IEEE Trans. Electron Devices, vol. 32, no. 9, pp. 1824–1829, Sep. 1985.

[27] A. Metz, R. Meyer, B. Kuhlmann, M. Grauvogl, and R. Hezel, “18.5% eﬃcient ﬁrst- generation MIS inversion-layer silicon solar cells,” in Conference Record of the Twenty Sixth IEEE Photovoltaic Specialists Conference, pp. 31–34, 1997. Chapter 4. Calcium-Based Electron Contacts 161

[28] M. A. Green, A. W. Blakers, J. Shi, E. M. Keller, and S. R. Wenham, “High- eﬃciency silicon solar cells,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 679–683, May 1984.

[29] M. A. Green and A. W. Blakers, patent, “High eﬃciency solar cell structure”, 1981.

[30] Y. W. Lam, M. A. Green, and L. W. Davies, “Electrostatic eﬀects in inversion-layer metal-insulator-semiconductor solar cells,” Appl. Phys. Lett., vol. 37, no. 12, p. 1087, 1980.

[31] Y. Wan, C. Samundsett, D. Yan, T. Allen, J. Peng, J. Cui, X. Zhang, J. Bullock, and A. Cuevas, “A magnesium/amorphous silicon passivating contact for n-type crystalline silicon solar cells,” Appl. Phys. Lett., vol. 109, no. 11, p. 113901, Sep. 2016.

[32] Y. H. Kwark, R. Sinton, and R. M. Swanson, “SIPOS Heterojunction contacts to silicon,” in International Electron Devices Meeting, vol. 30, pp. 742–745, 1984.

[33] E. Yablonovitch, T. Gmitter, R. M. Swanson, and Y. H. Kwark, “A 720 mV open circuit voltage SiOx:cSi:SiOx double heterostructure solar cell,” Appl. Phys. Lett., vol. 47, no. 11, pp. 1211–1213, Dec. 1985.

[34] M. Rienäcker, M. Bossmeyer, A. Merkle, U. Römer,F. Haase, J. Krügener,R. Brendel, and R. Peibst, “Junction Resistivity of Carrier-Selective Polysilicon on Oxide Junctions and Its Impact on Solar Cell Performance,” IEEE J. Photovolt., in press, 2016.

[35] M. Tanaka, M. Taguchi, T. Matsuyama, T. Sawada, S. Tsuda, S. Nakano, H. Hana- fusa, and Y. Kuwano, “Development of New a-Si/c-Si Heterojunction Solar Cells: ACJ- HIT (Artiﬁcially Constructed Junction-Heterojunction with Intrinsic Thin-Layer),” Jpn. J. Appl. Phys., vol. 31, no. 11R, p. 3518, Nov. 1992.

[36] H. Yamamoto, Y. Takaba, Y. Komatsu, M.-J. Yang, T. Hayakawa, M. Shimizu, and H. Takiguchi, “High-eﬃciency c-Si/c-Si heterojunction solar cells,” Sol. Energy Mater. Sol. Cells, vol. 74, no. 1–4, pp. 525–531, Oct. 2002.

[37] M. W. M. van Cleef, J. K. Rath, F. A. Rubinelli, C. H. M. van der Werf, R. E. I. Schropp, and W. F. van der Weg, “Performance of heterojunction p+ microcrystalline silicon n crystalline silicon solar cells,” J. Appl. Phys., vol. 82, no. 12, pp. 6089–6095, Dec. 1997.

[38] G. Nogay, J. P. Seif, Y. Riesen, A. Tomasi, Q. Jeangros, N. Wyrsch, F. J. Haug, S. De Wolf, and C. Ballif, “Nanocrystalline Silicon Carrier Collectors for Silicon Hetero- junction Solar Cells and Impact on Low-Temperature Device Characteristics,” IEEE J. Photovolt., vol. 6, no. 6, pp. 1654–1662, Nov. 2016.

[39] O. Madani Ghahfarokhi, K. von Maydell, and C. Agert, “Enhanced passivation at amorphous/crystalline silicon interface and suppressed Schottky barrier by deposition of microcrystalline silicon emitter layer in silicon heterojunction solar cells,” Appl. Phys. Lett., vol. 104, no. 11, p. 113901, Mar. 2014. Chapter 4. Calcium-Based Electron Contacts 162

[40] J. P. Seif, A. Descoeudres, M. Filipi˘c,F. Smole, M. Topi˘c,Z. C. Holman, S. De Wolf, and C. Ballif, “Amorphous silicon oxide window layers for high-eﬃciency silicon heterojunction solar cells,” J. Appl. Phys., vol. 115, no. 2, p. 24502, 2014.

[41] J. B. Heng, J. Fu, B. Kong, Y. Chae, W. Wang, Z. Xie, A. Reddy, K. Lam, C. Beitel, C. Liao, C. Erben, Z. Huang, and Z. Xu, “High-Eﬃciency Tunnel Oxide Junction Bifacial Solar Cell With Electroplated Cu Gridlines,” IEEE J. Photovolt., vol. 5, no. 1, pp. 82–86, Jan. 2015.

[42] M. Boccard and Z. C. Holman, “Amorphous silicon carbide passivating layers for crystalline-silicon-based heterojunction solar cells,” J. Appl. Phys., vol. 118, no. 6, p. 65704, Aug. 2015.

[43] T.-G. Chen, B.-Y. Huang, E.-C. Chen, P. Yu, and H.-F. Meng, “Micro-textured conductive polymer/silicon heterojunction photovoltaic devices with high eﬃciency,” Appl. Phys. Lett., vol. 101, no. 3, p. 33301, Jul. 2012.

[44] K. A. Nagamatsu, S. Avasthi, J. Jhaveri, and J. C. Sturm, “A 12% eﬃcient silicon/PEDOT:PSS heterojunction solar cell fabricated at < 100 ◦C,” IEEE J. Photovolt., vol. 4, no. 1, pp. 260–264, 2014.

[45] L. He, C. Jiang, H. Wang, D. Lai, and Rusli, “High eﬃciency planar Si/organic heterojunction hybrid solar cells,” Appl. Phys. Lett., vol. 100, no. 7, p. 73503, Feb. 2012.

[46] S. Jeong, E. C. Garnett, S. Wang, Z. Yu, S. Fan, M. L. Brongersma, M. D. McGehee, and Y. Cui, “Hybrid Silicon Nanocone–Polymer Solar Cells,” Nano Lett., vol. 12, no. 6, pp. 2971–2976, Jun. 2012.

[47] S. Avasthi, S. Lee, Y.-L. Loo, and J. C. Sturm, “Role of Majority and Minority Carrier Barriers Silicon/Organic Hybrid Heterojunction Solar Cells,” Adv. Mater., vol. 23, no. 48, pp. 5762–5766, Dec. 2011.

[48] X. Shen, B. Sun, D. Liu, and S.-T. Lee, “Hybrid Heterojunction Solar Cell Based on Organic–Inorganic Silicon Nanowire Array Architecture,” J. Am. Chem. Soc., vol. 133, no. 48, pp. 19408–19415, Dec. 2011.

[49] P. Yu, C.-Y. Tsai, J.-K. Chang, C.-C. Lai, P.-H. Chen, Y.-C. Lai, P.-T. Tsai, M.-C. Li, H.-T. Pan, Y.-Y. Huang, C.-I Wu, Y.-L. Chueh, S.-W. Chen, C.-H. Du, S.-F Horng, and H.-F. Meng, “13% Eﬃciency Hybrid Organic/Silicon-Nanowire Heterojunction Solar Cell via Interface Engineering,” ACS Nano, vol. 7, no. 12, pp. 10780–10787, Dec. 2013.

[50] D. Zielke, C. Niehaves, W. Lövenich, A. Elschner, M. Hörteis,and J. Schmidt, “Organic-silicon Solar Cells Exceeding 20% Efficiency,” Energy Procedia, vol. 77, pp. 331–339, Aug. 2015.

[51] J. Schmidt, D. Zielke, R. Gogolin, R. Sauer, and W. L¨ovenich, “Recent advances in polymer/silicon heterjunction solar cells,” presented at the 32nd European Photovoltaic Solar Energy Conference, Munich, Germany, 2016. Chapter 4. Calcium-Based Electron Contacts 163

[52] C. Battaglia, X. Yin, M. Zheng, I. D. Sharp, T. Chen, S. McDonnell, A. Azcatl,

C. Carrio, B. Ma, R. Maboudian, R. M. Wallace, and A. Javey, “Hole Selective MoOx Contact for Silicon Solar Cells,” Nano Lett., vol. 14, no. 2, pp. 967–971, Feb. 2014.

[53] J. Bullock, A. Cuevas, T. Allen, and C. Battaglia, “Molybdenum oxide MoOx:A versatile hole contact for silicon solar cells,” Appl. Phys. Lett., vol. 105, no. 23, p. 232109, Dec. 2014.

[54] J. Geissb¨uhler, J. Werner, S. Martin de Nicolas, L. Barraud, A. Hessler-Wyser, M. Despeisse, S. Nicolay, A. Tomasi, B. Niesen, S. De Wolf, and C. Ballif, “22.5% eﬃcient silicon heterojunction solar cell with molybdenum oxide hole collector,” Appl. Phys. Lett., vol. 107, no. 8, p. 81601, Aug. 2015.

[55] M. Bivour, J. Temmler, F. Z¨ahringer,S. Glunz, and M. Hermle, “High work function metal oxides for the hole contact of silicon solar Cells,” in IEEE 43rd Photovoltaic Specialists Conference (PVSC), Portland, OR, USA, 2016.

[56] M. Bivour, J. Temmler, H. Steinkemper, and M. Hermle, “Molybdenum and tungsten oxide: High work function wide band gap contact materials for hole selective contacts of silicon solar cells,” Sol. Energy Mater. Sol. Cells, vol. 142, pp. 34–41, Nov. 2015.

[57] L. G. Gerling, S. Mahato, A. Morales-Vilches, G. Masmitja, P. Ortega, C. Voz, R. Alcubilla, and J. Puigdollers, “Transition metal oxides as hole-selective contacts in silicon heterojunctions solar cells,” Sol. Energy Mater. Sol. Cells, vol. 145, pp. 109– 115, Feb. 2016.

[58] S. Avasthi, W. E. McClain, G. Man, A. Kahn, J. Schwartz, and J. C. Sturm, “Hole- blocking titanium-oxide/silicon heterojunction and its application to photovoltaics,” Appl. Phys. Lett., vol. 102, no. 20, p. 203901, 2013.

[59] B. Richards, “Novel uses of titanium dioxide for silicon solar cells,” PhD Thesis, University of New South Wales, Sydney, Australia, 2002.

[60] X. Yang, P. Zheng, Q. Bi, and K. Weber, “Silicon heterojunction solar cells with

electron selective TiOx contact,” Sol. Energy Mater. Sol. Cells, vol. 150, pp. 32–38, Jun. 2016.

[61] X. Yang, Q. Bi, H. Ali, K. Davis, W. V. Schoenfeld, and K. Weber, “High-

Performance TiO2-Based Electron-Selective Contacts for Crystalline Silicon Solar Cells,” Adv. Mater., vol. 28, no. 28, pp. 5891–5897, Jul. 2016.

[62] J. Bullock, M. Hettick, J. Geissb¨uhler,A. J. Ong, T. Allen, C. M. Sutter-Fella, T. Chen, H. Ota, E. W. Schaler, S. De Wolf, C. Ballif, A. Cuevas, and A. Javey, “Eﬃcient silicon solar cells with dopant-free asymmetric heterocontacts,” Nat. Energy, vol. 1, no. 3, p. 15031, Jan. 2016.

[63] J. Bullock, P. Zheng, Q. Jeangros, M. Tosun, M. Hettick, C. M. Sutter-Fella, Y. Wan, T. Allen, D. Yan, D. Macdonald, S. De Wolf, A. Hessler-Wyser, A. Cuevas, and A. Chapter 4. Calcium-Based Electron Contacts 164

Javey, “Lithium Fluoride Based Electron Contacts for High Eﬃciency n-Type Crystalline Silicon Solar Cells,” Adv. Energy Mater., vol. 6, no. 14, May 2016.

[64] J. Huang, Z. Xu, and Y. Yang, “Low-Work-Function Surface Formed by Solution- Processed and Thermally Deposited Nanoscale Layers of Cesium Carbonate,” Adv. Funct. Mater., vol. 17, no. 12, pp. 1966–1973, Aug. 2007.

[65] W. Wu, J. Bao, X. Jia, Z. Liu, L. Cai, B. Liu, J. Song, and H. Shen, “Dopant- free back contact silicon heterojunction solar cells employing transition metal oxide emitters,” Phys. Status Solidi RRL – Rapid Res. Lett., vol. 10, no. 9, pp. 662–667, Sep. 2016.

[66] H.-D. Um, N. Kim, K. Lee, I. Hwang, J. H. Seo, and K. Seo, “Dopant-Free All-

Back-Contact Si Nanohole Solar Cells Using MoOx and LiF Films,” Nano Lett., vol. 16, no. 2, pp. 981–987, Feb. 2016.

[67] T. G. Allen, J. Bullock, P. Zheng, B, Vaughan, M. Barr, Y. Wan, C. Samundsett, D. Walter, A. Javey, and A. Cuevas, “Calcium contacts to n-type crystalline silicon solar cells,” Prog. Photovolt. Res. Appl., in print, 2016. Calcium contacts to n-type crystalline silicon solar cells

Thomas Allen1, James Bullock2,3, Peiting Zheng1, Ben Vaughan4,

Matthew Barr4, Yimao Wan1, Christian Samundsett1, Daniel

Walter1, Ali Javey2,3 and Andr´esCuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

2Department of Electrical Engineering and Computer Sciences, University of

California, Berkeley, CA, 94720, USA

3Materials Science Division, Lawrence Berkeley National Laboratory, Berkeley, CA,

94720, USA

4Centre for Organic Electronics, University of Newcastle, Callaghan, NSW, 2308,

Australia

Published in Progress in Photovoltaics: Research and Applications, 32nd

EU PVSEC Special Issue, 2016.

165 Chapter 4. Calcium-Based Electron Contacts 166

Abstract — Direct metallization of lightly doped n-type crystalline silicon (c-

Si) is known to routinely produce non-Ohmic (rectifying) contact behaviour.

This has inhibited the development of n-type c-Si solar cells with partial rear contacts (PRC), an increasingly popular cell design for high performance p- type c-Si solar cells. In this contribution we demonstrate that low resistance

Ohmic contact to n-type c-Si wafers can be achieved by incorporating a thin layer of the low work function metal calcium (φ ∼ 2.9 eV) between the silicon surface and an overlying aluminium capping layer. Using this

2 approach contact resistivities of ρc ∼ 2 Ω.cm can be realised on undiﬀused n- type silicon wafers, thus enabling PRC cell designs on n-type silicon without the need for a phosphorus diﬀusion. Integrating the Ca / Al stack into a partial rear contact solar cell architecture fabricated on a lightly doped

14 −3 (ND = 4.5 × 10 cm ) n-type wafer resulted in a device eﬃciency of η =

17.6% where the Ca / Al contact comprised only ∼1.26% of the rear surface.

We demonstrate an improvement in this cell structure to an eﬃciency of

η = 20.3% by simply increasing the wafer doping by an order of magnitude

15 −3 to ND = 5.4 × 10 cm .

Index Terms — partial rear contacts, dopant-free contacts, calcium, work function, barrier height, contact resistance. Chapter 4. Calcium-Based Electron Contacts 167

I. Introduction

The formation of carrier selective contacts for both electrons and holes is critical to the functionality of any photovoltaic device. In crystalline silicon (c-Si) solar cells this is typically achieved by thermally diﬀusing or physically depositing acceptor and donor doped layers within the contact structure. This approach to contact formation can lead to slow and complex fabrication procedures, typically performed at high temperatures, and necessarily results in unwanted energy loss mechanisms like free carrier absorption

[1] and Auger recombination [2]. Balancing the trade-oﬀ between the fundamental losses and fabrication processes is a complicated and often dynamic engineering challenge.

Progress has recently been made in overcoming these challenges by the development of dopant-free contact structures on crystalline silicon. Advances have been particularly striking at the hole contact, with the fabrication of high eﬃciency devices incorporating

PEDOT:PSS [3] and sub-stoichiometric MoOx [4], [5] hole-selective contacts. The lower

2 resistivity of the direct p-type c-Si / MoOx contact (ρc ∼ 1 mΩ.cm ) [6] compared to other hole-selective materials has even made MoOx applicable to p-type cell architectures with partial rear contacts (PRC) [7]. The fabrication of dopant-free PRC solar cells on phosphorus doped n-type silicon, considered a more robust material for solar cell devices due to its relative insensitivity to common impurities like oxygen and iron, has, until recently, not been possible due to the technological challenges associated with contacting lowly doped n-type silicon.

Historically it is well known that direct metal contacts consistently exhibit non-Ohmic

(rectifying) I–V behaviour when applied to lowly doped n-type silicon for the range of metals commonly used to contact silicon solar cells. This is due to the Fermi level pinning phenomenon, whereby the Fermi level at the silicon surface is consistently held Chapter 4. Calcium-Based Electron Contacts 168 at values below the intrinsic Fermi level for metals like aluminium, copper, chromium and silver [8]. In practice, this means that larger energy barriers for electrons tend to form at the n-type Si / metal interface, while conversely, smaller energy barriers tend to form for holes at the p-type Si / metal interface. However, the work of Bullock et al., in which a thin layer of the work function modiﬁer LiF was incorporated into a c-Si / LiF

/ Al contact [9], demonstrates that by signiﬁcantly lowering the metal work function, dopant free electron contacts are realisable on moderately doped n-type silicon wafers, leading to the ﬁrst demonstration of an n-type PRC cell without the use of heavy doping

[10].

Ideally though the direct metallization of a lowly doped silicon wafer would achieve selectivity of either electrons or holes by virtue of its work function. Herein we document our progress in overcoming the limitations of typical n-type silicon / metal contacts by utilizing calcium, a low work function (φ ∼ 2.9 eV) metal commonly used as the electrode

2 in organic photovoltaic devices [11]. We demonstrate low resistivity (ρc ∼ 2 Ω.cm )

Ohmic contact to moderately doped n-type silicon by a thermally evaporated Ca / Al contact structure. We go on to apply the directly metallized Ca / Al contact structure to n-type partial rear contact (PRC) solar cells with contact fractions fc = 0.32% and

1.26%. To highlight the eﬃcacy of the Ca / Al contact we fabricated the cells on lightly

14 −3 phosphorus doped (ND = 4.5 × 10 cm ) n-type wafers, measuring a device eﬃciency

15 −3 of η = 17.6%. Applying the same cell architecture to a ND = 5.4×10 cm phosphorus doped wafer resulted in an increase in the eﬃciency to η = 20.3%. Chapter 4. Calcium-Based Electron Contacts 169

II. Contact Resistance

Thin, ∼30 nm, layers of calcium were thermally evaporated at low vacuum pressures

(between 1 × 10−7 and 1 × 10−6 Torr) from a solid source in a glovebox-integrated metal evaporator. An initial ∼150 nm capping layer of Al was sequentially evaporated in the same chamber without breaking vacuum. A further ∼150 nm of Al was subsequently evaporated after breaking vacuum to thicken the capping layer.

Contact resistance measurements were performed using Cox and Strack [12] and transfer length method (TLM) [13] test structures. All contact resistance structures were fabricated using shadow masks to deﬁne the contact geometry. The rear contact of the Cox and Strack test structure was formed by evaporating Al (> 500 nm) on a phosphorus

diﬀused n+ surface in order to increase the sensitivity of the measurement. The contact

resistivity was extracted from the I–V data obtained from a Keithley 2420 source meter.

Figure 1 shows the total measured resistance Rtotal of the Cox and Strack structure as a function of the top contact diameter. The inset graph features a subset of the I–V curves used to determine the data in the main figure for the Ca / Al contact, as well as an I–V curve of a direct Al contact to the same 1.2 Ω.cm n-type wafer, demonstrating the typical rectifying behavior of Al contacts to undiffused n-type surfaces. Since the contact structure features front and rear side contacts and bulk transport through the whole wafer thickness, the sensitivity of the measurement is limited by contributions to Rtotal by both the rear contact resistance and the bulk resistance (a function of wafer thickness and doping). To illustrate this, we have plotted the intrinsic spreading resistance in the figure, which assumes no resistive contribution from the front Ca / Al

+ or rear n / Al contacts. From the Rtotal measurements of Figure 1 we estimate the

2 contact resistivity of the Ca / Al contact to be ρc ∼ 2 Ω.cm . Note, however, that many Chapter 4. Calcium-Based Electron Contacts 170

Fig. 1: Contact resistance data used to extract the contact resistivity for the Ca/Al structure. Inset: a subset of the I–V curves (blue lines) used to extract the data set in the main ﬁgure (blue squares); and typical rectifying contact behaviour of direct aluminium contact measured on the same n-type wafer (purple line). data points lie below the modelled value and are indistinguishable from the intrinsic resistance of the bulk wafer, indicating the contact resistivity extraction is within the sensitivity of the technique and, therefore, the extracted value represents an upper limit to ρc.

To investigate the stability of the contact resistance the same test structure was remea- sured 14 months after the initial measurement, during which time the sample was stored unencapsulated in ambient conditions. The resulting change in the contact resistance is represented by the purple crosses in Figure 1. The data demonstrate that the contact resistivity has remained relatively stable over this period with any potential increase in the contact resistivity being within the error of the measurement.

Subsequent Ca / Al evaporations were performed at higher pressures (∼ 1 × 10−6 Torr) Chapter 4. Calcium-Based Electron Contacts 171 than those of Figure 1. The resulting contact resistance, as determined by the TLM test structures are plotted in Figure 2 as a function of wafer doping after subtracting the series resistance of the measurement setup, determined from a pad-to-pad resistance measurement. The results plotted in Figure 2 indicate a strong trend of increasing contact resistivity with decreasing wafer doping, which is similar to the results of Bullock et al. for the LiF / Al contact structure [10]. This trend is consistent, for data at

15 −3 ND ∼ 5 × 10 cm , with a low barrier height forming at the n-type Si / Ca interface, determined by Crowell et al. to be ∼0.4 eV for the n-type Si / Ca contact [14]. Note that for a metal with such a low work function (Ca φ ∼ 2.9 eV), Schottky-Mott theory would predict a negative barrier height, that is, an accumulation of electrons at the contact interface. Clearly Fermi-level pinning is still signiﬁcant in this contact structure, though the work function of calcium is low enough in this instance to contain its eﬀect on ρc to suitably low values.

Interestingly, Ohmic contact is maintained over the current density range investigated,

13 −3 even on the very lowly doped wafer (ND ∼ 4.3 × 10 cm ; ρ ∼ 100 Ω.cm) though the value of the contact resistivity is extremely high — a result that is likely attributable to extraneous bulk resistance effects. The analyses of Berger [15] and Woelk et al. [16] indicate that the sensitivity of the transmission line model (a precursor to the transfer length method) can be limited by contributions to the total resistance by spreading resistance into the semiconductor bulk material, leading to an overestimation of ρc when the semiconductor sheet resistance is high and when the sheet thickness departs significantly from the 1D assumption (i.e. for a thick semiconductor layer). They restrict the applicability of the TLM technique to structures in which the ratio of vertical resistances of the interface and semiconductor layer per unit area (η), defined as η = ρc/(ρs · h), are much greater than 0.19, where ρs is the semiconductor resistivity and h the thickness. Chapter 4. Calcium-Based Electron Contacts 172

Fig. 2: Contact resistivity as a function of wafer doping concentration extracted from TLM measurements. The purple triangle represents the ρc value extracted from the data of Figure 1. Also represented are the TLM data after applying the bulk resistance corrections outlined in [14] and [15], as well as modelled values of thermionic emission ρc. Inset: the I–V behaviour of the Ca/Al contact applied to 2 Ω.cm p-type silicon.

For values of η lower than this, the spreading resistance dominates over the contact resistance. For values of 0.19 < η < 2 Berger introduced the extended transmission line model (ETLM) model to account for errors associated with bulk spreading resistance

15 contributions. Applying the ETLM correction of Berger to the data at ND = 5.3 × 10 cm−3 (η ∼ 0.3) reduces the extracted apparent contact resistivity from 5.9 mΩ.cm2 to

2.1 mΩ.cm2, a value in agreement with the value extracted from the Cox and Strack data of Figure 1. For the data of Figure 2 at lower dopant densities the value of η falls below the range of applicability of both the TLM technique and Bergers correction (i.e.,

η < 0.19) and so the apparent increase in ρc may be an artefact of the uncorrected TLM measurement. Chapter 4. Calcium-Based Electron Contacts 173

Recently, Eidelloth and Brendel [17] introduced a bulk resistance correction to the TLM technique valid over a wider range of wafer dopant densities and thicknesses compared to that of Berger in order to determine ρc on test structures that more accurately reﬂect the rear contact of silicon solar cells. We have also applied the correction of Eidelloth and

Brendel to the data of Figure 2. For high dopant densities a solution to the equations presented in [17] could not be found and so no corrected data is displayed. For the data

15 −3 point at ND = 5.3 × 10 cm the corrected value of ρc matches the value extracted from the Cox and Strack data of Figure 1, consistent with the Berger correction. More

15 −3 strikingly, at ND = 1.1 × 10 cm the correction of [17] reduces the raw data by

2 approximately an order of magnitude to ρc ∼ 1 mΩ.cm , while at lower dopant densities

the correction of [17] has only a marginal eﬀect on the value of ρc. This may be because

the correction of Eidelloth and Brendel does not take into account the 3D eﬀect of lateral

spreading resistance, a feature of all of the TLM samples measured in this study.

Indeed, thermionic emission (i.e., current transport over a ﬁxed energy barrier) is inde-

pendent of surface doping (ignoring eﬀects of image force lowering). The relationship

governing the contact resistivity associated with thermionic emission (TE) over a ﬁxed

energy barrier is given by the equation [18]:

k qΦ ρ (TE) = exp B (4.1.1) c qA∗∗T kT

Where A∗∗ is the reduced eﬀective Richardson constant, k is Boltzmann’s constant, and

ΦB is the energy barrier at the contact. Inputting the value of the barrier height given by

Crowell et al. and using the value of A∗∗ = 110 Acm−2K−2 from [19] results in a dopant-

independent thermionic emission contact resistivity of ∼ 14 mΩ.cm2, slightly higher

15 −3 than the measured data at ND = 5.3 × 10 cm , a suﬃciently low dopant density for Chapter 4. Calcium-Based Electron Contacts 174

thermionic emission to dominate. Reducing the value of the barrier height to ΦB = 0.35

2 eV results in a reduction of the modelled ρc(TE) to ∼ 2 mΩ.cm . It is likely therefore that the value for the barrier height at the interface of the samples measured here lies somewhere between 0.35 – 0.4 eV, and the contact resistivity values on the lowly doped material actually fall within the values of ρc associated with these values for the barrier height. As will be shown in Section 3, results at the solar cell level are consistent with a ρc for lowly doped silicon much lower that the values in Figure 2, that is, in better agreement with the thermionic emission model. We conclude, therefore that the error induced by the bulk and lateral spreading resistances compromises the resolution of the TLM extraction, and its correction, especially at low dopant densities. This inaccuracy is likely

17 to contribute to at least some of the apparent doping dependence below ND = 1 × 10 cm−3 where current transport by thermionic emission dominates. This same artefact would may have also eﬀected the c results for the LiF / Al contact in [10] at low doping densities. Of course, as the wafer or surface dopant density increases beyond

17 −3 ND = 1 × 10 cm , the contact resistivity becomes dominated by, ﬁrst thermionic

ﬁeld emission, then ﬁeld emission processes, and so results in the trend of decreasing ρc seen in Figure 2.

The inset of Figure 2 plots the I–V behaviour of the Ca / Al contact on a 2 Ω.cm p-type

Cox and Strack structure, demonstrating rectiﬁcation that is indicative of the formation of a much larger potential barrier for holes at the p-type Si / Ca interface, consistent with the generally symmetrical trend of barrier height formation versus metal work function on p- compared to n-type silicon [8]. For reference, the value of ρc extracted from the data of Figure 1 is represented in Figure 2 (purple triangle). Chapter 4. Calcium-Based Electron Contacts 175

III. Partial Rear Contact Solar Cells

Partial rear contact (PRC) solar cells (∼155 µm thick; ∼2×2 cm2, isolated by a front-

14 −3 side mesa etch) were fabricated on both high (ρ = 9.9 Ωcm; ND = 4.5 × 10 cm )

15 −3 and low resistivity (ρ = 0.9 Ωcm; ND = 5.4 × 10 cm ) n-type ﬂoat zone silicon wafers. The cells feature a front-side boron diﬀusion (Rsheet ∼ 120 Ω/) on random

pyramid texturing, passivated by an Al2O3 / SiNx stack. The planarised rear-sides of

the cells were passivated with PECVD SiNx. The front contact openings were deﬁned

by photolithography and formed by a thermally evaporated Cr / Pd / Ag stack that was

later thickened with additional Ag by electroplating. The rear-side contacts were also

deﬁned by photolithography prior to the Ca / Al metal evaporation procedure outlined

above. Two sets of cells with diﬀering rear contact geometries have been fabricated in

this fashion: cells 1 and 3 feature 38 µm diameter dots at a pitch of 300 µm in a square

geometry (fc = 1.26%), while cells 2 and 4 feature the same dot size and pattern at

a pitch of 600 µm (fc = 0.32%). The pitch and the dot sizes were measured with an

optical microscope. Details of the cell architecture are represented in the schematic of

Figure 3.

Fig. 3: Schematic diagram of the partial rear contact cell structure. Cells of diﬀering rear contact pitch and resistivity have been fabricated.

The current-voltage (J–V ) characteristics of the cells were measured using both a xenon lamp solar simulator under standard 1-sun conditions (l000 mW/cm2, AM l.5 spectrum, Chapter 4. Calcium-Based Electron Contacts 176

25 ◦C; Figure 4) and a Sinton Instruments FCT-450 flash tester (Figure 5), both of which were calibrated using a certified reference cell from Franhaufer ISE CalLab. Suns–Voc measurements were taken using a Sinton WCT 110 (Figure 4) in addition to the flash tester (Figure 5). The quantum efficiency of cell 3 was measured using a Protoflex Corpo- ration QE measurement system (QE-1400-03). Reflection measurements were performed using a PerkinElmer Lambda 1050 UV/VIS/NIR spectrophotometer with an integrating sphere attachment.

Fig. 4: One sun J–V curves and pseudo J–V curves (from Suns-Voc) Fig. 5: One sun J–V curves and of the Ca/Al partial rear contact so- pseudo J–V curves (from Suns-Voc) of lar cells fabricated on lightly doped the Ca/Al partial rear contact solar 14 −3 (ND = 4.5×10 cm ) n-type wafers. cells fabricated on more heavily doped 15 −3 Cell 1 (blue) and cell 2 (purple) diﬀer (ND = 5.4×10 cm ) n-type wafers. only in their contact pitch (300 vs. 600 Cell 3 (blue) and cell 4 (purple) are µm) and contact fraction (∼1.26% vs identical in structure to cells 1 and 2 ∼0.32%). of Figure 4.

Figure 4 shows the standard one sun light J–V and Suns–Voc pseudo J–V measurements for the PRC cells fabricated on the 9.9 Ω.cm wafers. The cells underwent the Ca / Al evaporation in the same condition of the Cox and Strack test structures of Figure 1. The

J–V curves show that, impressively, the Ca / Al contact has been successfully implemented on the cells despite the order of magnitude decrease in wafer doping compared Chapter 4. Calcium-Based Electron Contacts 177 to the Cox and Strack test structure. However, it is also apparent that both cells 1 and

2 suffer from large series resistance losses, whose values, calculated from the comparison of the light J–V maximum power point (MPP) and the corresponding pseudo J–V data point [20] are given in the inset of Figure 4. The high series resistance of the cells is both a result of a high contact resistivity at the Si / Ca contact for the lower wafer doping (as per Figure 2) and due to current crowding effects within the cell as the carriers migrate towards the point contacts. These two effects are exacerbated in cell 2 because of its larger pitch and lower contact fraction, resulting in a high series resistance of 4.2 Ω.cm2.

While we applied the Ca / Al contact structure to a lightly doped wafer to accentuate the

eﬀect of the contact resistance, a consequence of using a 9.9 Ω.cm wafer is an increased

sensitivity to local recombination, particularly at the Si / Ca interface (see Figure 8).

The combination of the light base doping and high recombination at the point contacts

limits the Voc of the cells to 612 mV for cell 1, and 630 mV for cell 2; values that scale

appropriately with contact fraction. These simultaneous eﬀects of internal and contact

series resistance losses, as well as recombination losses can be, in part, overcome by

increasing the phosphorus doping concentration of the base region of the cells.

Figure 5 plots the same one sun light J–V and Suns–Voc pseudo J–V measurements of

the cells fabricated on the 0.9 Ω.cm material. As expected, the eﬃciency of both cells 3

and 4 increases relative to cells 1 and 2, from η = 15.1% to 19.7% for the cells with the

600 µm pitch, and from η = 17.6% to 20.3% for the cells with the 300 µm pitch. All cell

parameters improved signiﬁcantly, including the Jsc, which increased slightly due to an

improvement in the photolithograpic deﬁnition of the front metal grid. The integration

of the external quantum eﬃciency (EQE; measured on cell 3) after multiplication by the

2 AM1.5 spectrum yields a short circuit current of Jscint ∼ 39.1 mA/cm , within tolerable

2 agreement with the light J–V data (Jsc = 39.6 mA/cm ) given the uncertainty in the Chapter 4. Calcium-Based Electron Contacts 178 measurement resulting from, for example, mismatch between illumination areas in the

EQE and reﬂection measurements. We attribute the non-ideal Jsc to a small additional optical loss deriving from the relative reduction in reﬂection from the SiNx / Ca rear mirror (R ∼ 91% at λ = 1250 nm), compared to SiNx / Al (R ∼ 95%) and SiNx /

Ag (R ∼ 98%) rear reﬂectors. By inputting these reﬂection values into the generation current model of Basore [21] we calculate an upper limit to the short circuit current

2 of these Ca PRC devices of Jscmax ∼ 40 mA/cm , compared to approximately 40.3 mA/cm2 and 40.5 mA/cm2 for Al and Ag rear reﬂectors, respectively.

Fig. 6: Quantum eﬃciency and reﬂection data of cell 3.

To estimate the contribution that the rear Ca / Al contact makes to the total recombination, test structures were fabricated to measure the recombination occurring in diﬀerent areas of the PRC cells. Recombination factors J0 for the front boron diﬀused region and Chapter 4. Calcium-Based Electron Contacts 179

for the rear surface passivated with SiNx were extracted from the excess carrier dependent lifetime τeff (∆n) of control test structures measured by photoconductance decay

(PCD), as reported in [10]. These values, given in Table I, were used in conjunction with the geometrical parameters detailed in Table II to simulate the performance of the various PRC devices, exploring diﬀerent values of the rear contact surface recombination velocity (SRV). A quasi-analytical, iterative model of the three-dimensional device geometry was used to model the output parameters of the solar cell (Voc, Jsc, FF and

η) [22]. A comparison between the measured parameters of the four diﬀerent solar cells

4 and the simulated results using a lower bound of the rear contact SRV of Sc = 5 × 10

cm/s is given in Table II.

Fig. 7: Rear reﬂection measurements comparing Ca/Al to Ag and Al rear mirrors. The structures are planar crystalline silicon with a 75 nm SiNx layer deposited on both sides of the wafer prior to evaporation of the rear side metal.

Figure 8 plots the modelled Voc as a function of Sc, with the individual data points representing the measured Voc values from the light J–V data of Figures 4 and 5. The Chapter 4. Calcium-Based Electron Contacts 180

ﬁgure clearly shows that the Voc is very sensitive to the surface recombination velocity at the contact, particularly for the high resistivity solar cells. From the modelling a

4 lower limit to the contact SRV of Sc > 5 × 10 cm/s can be determined, as indicated by the shaded portion of the plot. This value is one order of magnitude higher than the value determined for LiF / Al contacts reported in [10]. It is diﬃcult to be more speciﬁc about the value of Sc given the uncertainties in the input parameters for the model, particularly the exact size of the openings in the rear SiNx layer, slight variations in which can greatly impact the contact fraction and, as indicated in Figure 8, the Voc.

The modelling results of Figure 8 also indicate that, for the low resistivity cells, Voc is

4 very weakly dependent on Sc for values above 5 × 10 cm/s. It is also evident from the modelling that recombination at the rear Ca / Al contact remains the main loss mechanism in open-circuit, even for the low resistivity cells.

Fig. 9: Modelled values of the ﬁll fac- Fig. 8: Results from the modelling tor plotted as a function of rear con- of the cell structures fabricated in this tact resistivity for the cell structures study. The shaded region of the graph fabricated in this study. The shaded represents possible values of the sur- region of the graph represents the pos- face recombination velocity (SRV) at sible values of ρc given the measured the Si/Ca contact, given the cells re- values of the ﬁll factor (FF ) for the sults of Figures 4 and 5. cells.

The small contact fraction of the cells also ampliﬁes the consequences of non-zero rear Chapter 4. Calcium-Based Electron Contacts 181 contact resistivity as demonstrated in Figure 9 which plots the ﬁll factor (FF ) versus rear contact resistivity. It can be noted that, for these cell structures, the FF decreases in an approximately linear fashion with the rear contact resistivity ρc. The dependence is exacerbated when the contact fraction is reduced (from 1.26% to 0.32%), leading to

2 extreme degradation in FF when ρc > 10 mΩ.cm . The experimental fill factors for the four different cells, also plotted in Figure 9, are in good agreement with the modelled trends. A good fit can be obtained for the low resistivity cells (3 and 4) using a contact resistivity of 3.5 mΩ.cm2, a value that lies between the measured and corrected values of

ρc for the same wafer resistivity (see Figure 2). For the high resistivity wafers, agreement between the modelling and the cell results of Figure 4 could only be achieved by using a contact resistivity of 7 mΩ.cm2, an order of magnitude lower than the measured 80

2 mΩ.cm value of Figure 2. This is further evidence that the ρc values in Figure 2 for

15 −3 ND < 5 × 10 cm have been largely impacted by bulk resistance contributions, and, as thermionic emission theory would predict, the contact resistivity of the Ca contacts in this study is only weakly dependent on doping concentration.

Table 1: Parameters Utilised in the PRC Cell Simulations

Device property Parameter Value Doping Base doping 4.5 × 1014/5.4 × 1015 cm−3 Boron diﬀusion sheet resistance 120 Ω/ Dimension Wafer thickness 155 µm Rear dot contact diameter 38 µm Rear contact pitch 300/600 µm Recombination SRV at rear contact 5 × 104 cm/s −2 Passivated rear J0 3 fA.cm −2 Front J0 (total) 72 fA.cm Optics Front surface shading 2–6%

Front ARC on texture ∼75 nm SiNx Parasitic resistance Series resistance (metal grid) 0.3 Ω.cm2 Shunt resistance > 103 Ω.cm2

A reasonable agreement between measured and modelled Rs of the cells is also obtained. Chapter 4. Calcium-Based Electron Contacts 182

Table 2: Comparison Between Measured and Modelled Cell Performance

ρcell fc ρc Jsc Voc FFRs η (Ωcm) (%) (mΩcm2) (mA/cm2) (mV) (%) (Ωcm2) (%) measured measured measured measured measured (modelled) (modelled) (modelled) (modelled) (modelled) Cell 1 9.9 1.26 7 38.9 (38.9) 612 (612) 74 (74.1) 1.5 (1.3) 17.5 (17.6) Cell 2 9.9 0.32 7 38.8 (38.8) 630 (652) 61 (63.1) 4.2 (3.6) 15.1 (16.0) Cell 3 0.9 1.26 3.5 39.6 (39.6) 652 (654) 78.6 (79.9) 0.8 (0.6) 20.3 (20.7) Cell 4 0.9 0.32 3.5 39 (39) 664 (673) 75.9 (75.3) 1.4 (1.6) 19.7 (19.8)

In addition to the series resistance associated with the rear contact, the total Rs also includes components due to current crowding towards the partial rear contacts, as well as an additional estimated resistance of 0.3 Ω.cm2, representing losses in the front metal grid. Figure 10 shows those resistive contributions to the total Rs for the four cells in this study, as determined by modelling. It can be seen that the contact resistance dominates the series resistance losses for the cells with the small contact fraction (cells

2 and 4), and is approximately the largest contributor to the resistive losses in the cells with the largest contact fraction (cells 1 and 3). The data of Figure 10 demonstrates that while the resistance at the rear contacts decreases when reducing the wafer resistivity

(commensurate with the decrease in contact resistivity), it is the internal crowding resistance that decreases by the largest proportion of the total Rs. This is represented diagrammatically in Figure 9, where the two dashed lines (that is, the 0.9 Ω.cm cells)

2 converge when ρc = 0 Ω.cm . Such a convergence does not occur for the high resistivity wafers due to the magnitude of the crowding eﬀects, much greater for the case of fc =

0.32% (cell 2) than for fc = 1.26% (cell 1).

An optimisation of the rear contact fraction is shown in Figure 11, together with the measured conversion eﬃciencies for the four devices in this study. The modelled results

(assuming an identical 3% metal grid shading for all the cells) indicate that, for the speciﬁc set of parameters assumed here, the optimum pitch between the point contacts Chapter 4. Calcium-Based Electron Contacts 183

Fig. 10: Breakdown of series resistance components, as determined by modelling. A constant series resistance of 0.3 Ωcm2 from the front metal grid is included in the modelled value.

should be slightly larger (about 400 µm; fc = 0.71%) for the low resistivity wafers than for the high resistivity ones, where the optimum pitch is very close to the 300 µm (fc =

1.26%) used to fabricate cell 1. A detailed examination of Voc and FF as a function of the pitch conﬁrms earlier observations that the optimum in conversion eﬃciency occurs when the increase of Voc with increasing pitch is negated by the corresponding drop in

FF [22].

The large diﬀerence in achievable eﬃciency between the low and high resistivity wafers is due to the relatively high surface recombination velocity and contact resistivity of the

Ca / Al contacts implemented here. Such diﬀerences would decrease if a low resistance, passivated contact could be developed, possibly via the insertion of an interlayer between the silicon and calcium layer. Chapter 4. Calcium-Based Electron Contacts 184

Fig. 11: Optimisation of the contact pitch of the solar cells fabricated in this study.

IV. Conclusion

We have demonstrated that Ohmic contact can be formed on n-type silicon over the dop-

13 −3 20 −3 ing range from ND = 4.3×10 cm to 1.7×10 cm by direct calcium metallization.

2 The resistivity of the n-type Si / Ca / Al contacts is suﬃciently low (ρc ∼ 2 mΩ.cm ) that solar cells with partial rear contacts, where the Ca contact fraction comprises ∼

1% of the rear surface, can be successfully fabricated without the need for a thermal phosphorus diﬀusion. The implementation of this PRC cell architecture resulted in a device eﬃciency of η = 17.6% when applied to a 9.9 Ω.cm wafer. Simply by increasing

15 −3 the dopant density of the base region of the solar cells to ND = 5.4 × 10 cm (ρ =

0.9 Ω.cm) resulted in a significant reduction in the series resistance losses, as well as the recombination losses suffered at the Ca / Si interface, translating to an increase in device efficiency to η = 20.3%. We identify that recombination losses at the Si / Ca Chapter 4. Calcium-Based Electron Contacts 185 interface is limiting device performance, losses that could potentially be mitigated with the addition of a passivating interlayer between the silicon and calcium.

Acknowledgements

This work was performed in part at the Materials node of the Australian National

Fabrication Facility, which is a company established under the National Collaborative

Research Infrastructure Strategy to provide nano- and microfabrication facilities for

Australias researchers. This work has been supported by the Australian government through the Australian Renewable Energy Agency (ARENA). Work at the University of

California, Berkeley was supported by the Bay Area Photovoltaic Consortium (BAPVC). Chapter 4. Calcium-Based Electron Contacts 186

References

[1] S. C. Baker-Finch, K. R. McIntosh, D. Yan, K. C. Fong, and T. C. Kho, “Near- infrared free carrier absorption in heavily doped silicon,” J. Appl. Phys., vol. 116, no. 6, p. 063106, Aug. 2014.

[2] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, and A. Cuevas, “Improved quanti- tative description of Auger recombination in crystalline silicon,” Phys. Rev. B, vol. 86, no. 16, Oct. 2012.

[3] D. Zielke, C. Niehaves, W. Lvenich, A. Elschner, M. H¨orteis, and J. Schmidt, “Organic-silicon Solar Cells Exceeding 20% Eﬃciency,” Energy Procedia, vol. 77, pp. 331339, Aug. 2015.

[4] C. Battaglia, X. Yin, M. Zheng, I. D. Sharp, T. Chen, S. McDonnell, A. Azcatl, C.

Carraro, B. Ma, R. Maboudian, R. M. Wallace, and A. Javey, “Hole Selective MoOx Contact for Silicon Solar Cells,” Nano Lett., vol. 14, no. 2, pp. 967–971, Feb. 2014.

[5] J. Geissbhler, J. Werner, S. M. de Nicolas, L. Barraud, A. Hessler-Wyser, M. De- speisse, S. Nicolay, A. Tomasi, B. Niesen, S. D. Wolf, and C. Ballif, “22.5% eﬃcient silicon heterojunction solar cell with molybdenum oxide hole collector,” Appl. Phys. Lett., vol. 107, no. 8, p. 081601, Aug. 2015.

[6] J. Bullock, A. Cuevas, T. Allen, and C. Battaglia, “Molybdenum oxide MoOx:A versatile hole contact for silicon solar cells,” Appl. Phys. Lett., vol. 105, no. 23, p. 232109, Dec. 2014.

[7] J. Bullock, C. Samundsett, A. Cuevas, D. Yan, Y. Wan, and T. Allen, “Proof-of- Concept p-Type Silicon Solar Cells With Molybdenum Oxide Local Rear Contacts,” IEEE J. Photovolt., vol. 5, no. 6, pp. 1591–1594, Nov. 2015.

[8] D. K. Schroder and D. L. Meier, “Solar cell contact resistance: A review,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 637–647, May 1984.

[9] J. Bullock, M. Hettick, J. Geissbhler, A. J. Ong, T. Allen, C. M. Sutter-Fella, T. Chen, H. Ota, E. W. Schaler, S. De Wolf, C. Ballif, A. Cuevas, and A. Javey, “Eﬃcient silicon solar cells with dopant-free asymmetric heterocontacts,” Nat. Energy, vol. 1, no. 3, p. 15031, Jan. 2016.

[10] J. Bullock, P. Zheng, Q. Jeangros, M. Tosun, M. Hettick, C. M. Sutter-Fella, Y. Wan, T. Allen, D. Yan, D. Macdonald, S. De Wolf, A. Hessler-Wyser, A. Cuevas, and A. Javey, “Lithium Fluoride Based Electron Contacts for High Eﬃciency n-Type Crystalline Silicon Solar Cells,” Adv. Energy Mater., vol. 6, no. 14, May 2016.

[11] H. Hoppe and N. S. Sariciftci, “Organic solar cells: An overview,” J. Mater. Res., vol. 19, no. 07, pp. 1924–1945, 2004.

[12] R. H. Cox and H. Strack, “Ohmic contacts for GaAs devices,” Solid-State Electron., vol. 10, no. 12, pp. 1213–1218, Dec. 1967. Chapter 4. Calcium-Based Electron Contacts 187

[13] D. L. Meier and D. K. Schroder, “Contact resistance: Its measurement and relative importance to power loss in a solar cell,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 647–653, May 1984.

[14] C. R. Crowell, H. B. Shore, and E. E. LaBate, “SurfaceState and Interface Eﬀects in Schottky Barriers at nType Silicon Surfaces,” J. Appl. Phys., vol. 36, no. 12, pp. 3843–3850, Dec. 1965.

[15] H. H. Berger, “Models for contacts to planar devices,” Solid-State Electron., vol. 15, no. 2, pp. 145–158, Feb. 1972.

[16] E. G. Woelk, H. Krautle, and H. Beneking, “Measurement of low resistive ohmic contacts on semiconductors,” IEEE Trans. Electron Devices, vol. 33, no. 1, pp. 19–22, Jan. 1986.

[17] S. Eidelloth and R. Brendel, “Analytical Theory for Extracting Speciﬁc Contact Resistances of Thick Samples From the Transmission Line Method,” IEEE Electron Device Lett., vol. 35, no. 1, pp. 9–11, Jan. 2014.

[18] D. K. Schroder, Semiconductor Material and Device Characterization, John Wiley & Sons, 2006.

[19] S. M. Sze and K. K. Ng, Physics of Semiconductor Devices, John Wiley & Sons, 2006.

[20] D. Pysch, A. Mette, and S. W. Glunz, “A review and comparison of diﬀerent methods to determine the series resistance of solar cells,” Sol. Energy Mater. Sol. Cells, vol. 91, no. 18, pp. 1698–1706, Nov. 2007.

[21] P. A. Basore, “Extended spectral analysis of internal quantum eﬃciency,” in Confer- ence Record of the Twenty Third IEEE Photovoltaic Specialists Conference, 1993, 1993, pp. 147–152.

[22] A. Cuevas, “Electrons and holes in solar cells with partial rear contacts,” Prog. Photovolt. Res. Appl., vol. 22, no. 7, pp. 764–774, Jul. 2014.

Low resistance TiO2-passivated

calcium contacts for crystalline

silicon solar cells

Thomas Allen1, Peiting Zheng1, Ben Vaughan2, Matthew Barr2,

Yimao Wan1, Christian Samundsett1, James Bullock1, and Andr´es

Cuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

2Centre for Organic Electronics, University of Newcastle, Callaghan, NSW, 2308,

Australia

Published in the Proceedings of the 43rd IEEE Photovoltaic Specialist

Conference, June 2016.

189 Chapter 4. Calcium-Based Electron Contacts 190

Abstract — It has recently been shown that low resistance Ohmic contact to lightly doped n-type crystalline silicon (c-Si) is possible by direct metallization via a thin layer of the low work function metal calcium (φ ∼ 2.9 eV) and an overlying aluminium capping layer. Using this approach upper limit contact resistivities of < 2 mΩ.cm2 can be realised on undiﬀused n-type surfaces.

However, recombination at the Ca / Si interface limits the application of the

Ca contact to very low contact fractions which leads to non-negligible resistive losses and an increase in device fabrication complexity. Here we show that the low resistance Ohmic contact of the Ca / Al structure is retained after the addition of a TiO2 interlayer, leading the way to the development of a passivated contact device utilizing TiO2 and Ca.

Index Terms — passivated contacts, TiO2, dopant-free contacts, calcium, work function, barrier height, contact resistance. Chapter 4. Calcium-Based Electron Contacts 191

I. Introduction

All photovoltaic devices require low resistance electrical contacts to be made to the absorber material that eﬀectively screen one carrier type at the expense of the other.

Ideally, the direct metallization of a lowly doped silicon wafer would achieve such a contact. However, it is well known that most direct metal contacts consistently exhibit non-Ohmic (rectifying) I–V behaviour, especially when applied to n-type silicon [1]. This places severe constraints on the design of n-type silicon solar cells. Direct silicon – metal contacts also exhibit high rates of recombination, especially when the silicon surface is not heavily doped p- or n-type, compromising their selectivity. Consequently, c-Si solar cells have been typically fabricated with heavily doped, thermally diﬀused regions into the c-Si subsurface – a high temperature step that complicates device fabrication due to the need for cleanliness and thermally stable bulk lifetimes. Heavy doping also introduces unwanted eﬀects that can limit device performance, like Auger recombination

[2] and free carrier absorption [3].

Following the demonstration of low work function enabled electron contacts of Bullock et al. [4],[5], it has recently been shown that the limitations of typical n-type silicon metal contacts can be overcome by utilizing calcium, a low work function (φ ∼ 2.9 eV) metal [6] commonly used as the electrode in organic photovoltaic devices [7]. In ref [6]

2 the authors have demonstrated that low resistivity (ρc < 2 mΩ.cm ) Ohmic contact to lightly doped n-type silicon by a thermally evaporated Ca / Al contact structure. In this

2 addition we further demonstrate that low resistivity Ohmic contact (ρc < 10 mΩ.cm ) is retained with the addition of passivating interlayers of titanium dioxide (TiO2) up to a thickness of ∼8 nm. Similar passivated contact structures using interlayers of Ga2O3

are shown to be considerably more resistive. Chapter 4. Calcium-Based Electron Contacts 192

II. Experimental Procedure

Thin, ∼30 nm, layers of calcium were thermally evaporated at low vacuum pressures

(< 5 × 10−7 Torr) from a solid source in a glovebox-integrated metal evaporator. An initial ∼150 nm capping layer of Al was sequentially evaporated in the same chamber without breaking vacuum. A further ∼150 nm of Al was subsequently evaporated to thicken the capping layer. Passivating interlayers of TiO2 and Ga2O3 were deposited on the silicon surface by atomic layer deposition as per [8] and [9],[10] prior to metal evaporation.

Contact resistance measurements were performed using Cox and Strack [11] and transmission line method (TLM) test structures [1]. All contact resistance structures were fabricated using shadow masks to deﬁne the contact geometry. The rear contact of the Cox and Strack test structures was formed by evaporating Al (> 500 nm) on a phosphorus diﬀused n+ surface in order to increase the sensitivity of the measurement.

The contact resistivity was extracted from the I–V data obtained from a Keithley 2420 source meter.

Symmetrically TiO2 passivated lifetime samples were prepared on planar saw-damage etched and RCA cleaned silicon substrates. The eﬀective lifetime (τeff ) of the passivated samples was measured as a function of minority carrier injection level (∆n) on a Sinton

Instruments WCT120 photoconductance tester in the quasi steady state and transient modes of operation. Chapter 4. Calcium-Based Electron Contacts 193

III. Results and Discussion

Titanium dioxide (TiO2) has recently been shown to passivate silicon surface defects

[8],[12], and act as an electron selective material on c-Si [13], and so is an ideal candidate for a passivating interlayer in the calcium contact structure. To demonstrate the eﬀectiveness of TiO2 as a surface passivation layer, we plot the excess carrier lifetime

(τeff ) as a function of injection level (∆n) of symmetrically passivated TiO2 samples on

1.2 Ω.cm n-type silicon. These samples are measured as-deposited, without any thermal

activation treatment. As can be seen from Figure 1, there is a critical thickness of TiO2,

approximately 6 nm, required for surface passivation, with the lifetime saturating to

approximately 1 ms after this thickness is reached. This corresponds to an upper limit

to the surface recombination velocity of 7.9 cm/s.

Fig. 1: Injection dependent lifetime behaviour of symmetrically TiO2 passivated 1.2 Ω.cm n-type silicon samples of diﬀering thickness. Chapter 4. Calcium-Based Electron Contacts 194

Figure 2 shows the contact resistivity of the Ca / Al contact structure as a function of

TiO2 and Ga2O3 interlayer thickness, measured on the same 1.2 Ω.cm n-type silicon. The data for the TiO2 and bare silicon sample were determined from measurements on Cox and Strack test structures. The spread in the data is indicative of the range of sensible

ρc values extracted from the I–V curves of the individual contact pads. Since this contact structure features front and rear side contacts and bulk transport through the whole wafer thickness, the sensitivity of the measurement is limited by contributions to

Rtotal by both the rear contact resistance and the bulk resistance (a function of the wafer thickness and doping). The lower limit resolution of this contact resistance measurement

2 method is indicated by the dotted line in the ﬁgure (∼1 mΩ.cm ). The Ga2O3 data in the ﬁgure was determined from measurements taken on TLM test structures, a more accurate means of determining ρc.

Fig. 2: Contact resistivity as a function of TiO2 and Ga2O3 interlayer thickness. The dotted line is an estimate of the sensitivity of the Cox and Strack measurement, given the wafer thickness and doping. Chapter 4. Calcium-Based Electron Contacts 195

The data of Figure 2 shows that the contact resistivity of the TiO2 / Ca / Al contact

2 remained low (< 10 mΩ.cm ) for the range of TiO2 thickness investigated, except for the thickest TiO2 layer (∼11 nm; data point not shown) which exhibited rectifying behaviour. The samples with Ga2O3 interlayers, on the other hand, show a much stronger

2 dependence of on thickness, with a ρc of ∼30 mΩ.cm measured on a sample with only a ∼1 nm interlayer.

Fig. 3: Representative I–V curves of 1.2 Ωcm n-type silicon / TiO2 hetero-contacts fabricated on a Cox and Strack test structure with diﬀerent contact metals (Ca, Ti, and Al).

Interestingly, all TiO2 contacts fabricated without calcium consistently showed non-

Ohmic I–V behaviour for all TiO2 thicknesses. Figure 3 shows the I–V behaviour for samples with a 2.2 nm TiO2 interlayer on 1.2 Ω.cm n-type silicon metallized with calcium, aluminium and titanium, demonstrating non-linear behaviour even for such a thin interlayer. Due to the non-linearity, an accurate extraction of ρc could not be made for the Al and Ti metallized samples. Previous studies have attributed a chemical Chapter 4. Calcium-Based Electron Contacts 196

reduction of the TiO2 layer to sub-stoichiometric TiOx (where x < 2) as an essential component to its electron contact functionality [14]. The consistently low ρc values of

Figure 2, and the vast diﬀerence between the behaviour of the TiO2 / Al, TiO2 / Ti, and TiO2 / Ca contacts, may therefore be indicative of a larger degree of reduction in the TiO2 ﬁlm by the overlying Ca layer, compared to the Al and Ti layers.

In order to explore the potential of the TiO2 / Ca / Al passivated contact structure on a device level, we modelled a partial rear contact structure (PRC) with and identical geometry as in ref. [6] in QSCell PRC, a quasi-analytical, iterative model of the three- dimensional device geometry [15]. Using the same input parameters as listed in ref. [6] for a 0.9 Ω.cm n-type silicon base, and a rear contact fraction of 1.26% (38 micron dots at a pitch of 300 micron, in a square arrangement) we explored the effect of fixing the rear contact resistivity to 10 mΩ.cm2 and varying the surface recombination velocity at the rear contact. The modelled device efficiency is plotted in Figure 4 as a function of the rear contact SRV (Sc).

From the modelling it is apparent that the marginal increase in rear contact resistivity

(an increase from ∼3.5 mΩ.cm2 to 10 mΩ.cm2) signiﬁcantly limits the device performance, with the device eﬃciency dropping from a measured values of 20.3% to an ideal

6 value, from Figure 4, of approximately 19.5% for unpassivated (Sc > 10 cm/s) rear contacts. In order to improve upon the unpassivated rear contact device structure of

4 ref. [6], a passivated contact SRV, post-metallization of Sc < 10 cm/s must be realized, given the increase in resistivity losses at the rear contact for this particular device geometry. If the pre-contact SRV of 7.8 cm/s can be retained after the Ca / Al evaporation, the modelling indicates that device eﬃciencies greater that 21.5% can be achieved. Chapter 4. Calcium-Based Electron Contacts 197

Fig. 4: Partial rear contact device eﬃciency potential given the input parameters of ref. [6].

Further optimization of the passivated contact device architecture, for example increasing the rear contact fraction, may be possible once more information about the post- metallization contact SRV is known. It may also be feasible to apply the TiO2 / Ca /

Al contact structure to devices with a full rear contact for TiO2 thickness greater than those explored here (that is > 8 nm) if the trade-oﬀ between the increased resistivity and decreased contact recombination are advantageous.

IV. Conclusion

2 We have demonstrated that low resistivity (ρc < 10 mΩ.cm ) electron hetero-contacts featuring TiO2 interlayers can be made to 1.2 Ω.cm n-type silicon for a wide range of

TiO2 thicknesses. Modelling of the cell structure indicates that an increase in efficiency Chapter 4. Calcium-Based Electron Contacts 198 is possible by simply increasing the wafer doping, which can reduce the internal resistive losses, as well as the recombination losses suffered at the Ca / Si interface. Addition- ally, the insertion of a passivating interlayer can also lead to higher cell voltages and efficiencies. We have identified that TiO2 serves as an ideal candidate for such a role.

Acknowledgements

This work was performed in part at the Materials node of the Australian National

Fabrication Facility, which is a company established under the National Collaborative

Research Infrastructure Strategy to provide nano- and microfabrication facilities for

Australia’s researchers. This work has been supported by the Australian government through the Australian Renewable Energy Agency (ARENA). Chapter 4. Calcium-Based Electron Contacts 199

References

[1] D. K. Schroder and D. L. Meier, “Solar cell contact resistance: A review,” IEEE Trans. Electron Devices, vol. 31, no. 5, pp. 637–647, May 1984.

[2] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, and A. Cuevas, “Improved quanti- tative description of Auger recombination in crystalline silicon,” Phys. Rev. B, vol. 86, no. 16, Oct. 2012.

[3] S. C. Baker-Finch, K. R. McIntosh, D. Yan, K. C. Fong, and T. C. Kho, “Near- infrared free carrier absorption in heavily doped silicon,” J. Appl. Phys., vol. 116, no. 6, p. 063106, Aug. 2014.

[4] J. Bullock, M. Hettick, J. Geissbhler, A. J. Ong, T. Allen, C. M. Sutter-Fella, T. Chen, H. Ota, E. W. Schaler, S. De Wolf, C. Ballif, A. Cuevas, and A. Javey, “Eﬃcient silicon solar cells with dopant-free asymmetric heterocontacts,” Nat. Energy, vol. 1, no. 3, p. 15031, Jan. 2016.

[5] J. Bullock, P. Zheng, Q. Jeangros, M. Tosun, M. Hettick, C. M. Sutter-Fella, Y. Wan, T. Allen, D. Yan, D. Macdonald, S. D. Wolf, A. Hessler-Wyser, A. Cuevas, and A. Javey, “Lithium ﬂuoride based electron contacts for high eﬃciency n-type crystalline silicon solar cells,” Adv. Energy Mater., in press.

[6] T. G. Allen, J. Bullock, P. Zheng, B. Vaughan, M. Barr, Y. Wan, C. Samundsett, D. Yan, A. Javey, and A. Cuevas, “Calcium contacts to high eﬃciency solar cells,” submitted.

[7] H. Hoppe and N. S. Sariciftci, “Organic solar cells: An overview,” J. Mater. Res., vol. 19, no. 07, pp. 1924–1945, 2004.

[8] J. Cui, T. G. Allen, Y. Wan, X. Zhang, and A. Cuevas, “Titanium oxide: a re- emerging optical and passivating material for silicon solar cells,” Submitted to 6th Sili- conPV, Chambery, France, 2016.

[9] T. G. Allen and A. Cuevas, “Plasma enhanced atomic layer deposition of gallium oxide on crystalline silicon: demonstration of surface passivation and negative interfacial charge,” Phys. Status Solidi RRL — Rapid Res. Lett., vol. 9, no. 4, pp. 220–224, Apr. 2015.

[10] T. G. Allen, M. Ernst, C. Samundsett, and A. Cuevas, “Demonstration of c-Si Solar Cells With Gallium Oxide Surface Passivation and Laser-Doped Gallium p+ Regions,” IEEE J. Photovolt., vol. 5, no. 6, pp. 1586–1590, Nov. 2015.

[11] R. H. Cox and H. Strack, “Ohmic contacts for GaAs devices,” Solid-State Electron., vol. 10, no. 12, pp. 1213–1218, Dec. 1967.

[12] B. Liao, B. Hoex, A. G. Aberle, D. Chi, and C. S. Bhatia, “Excellent c-Si surface passivation by low-temperature atomic layer deposited titanium oxide,” Appl. Phys. Lett., vol. 104, no. 25, p. 253903, Jun. 2014. Chapter 4. Calcium-Based Electron Contacts 200

[13] S. Avasthi, W. E. McClain, G. Man, A. Kahn, J. Schwartz, and J. C. Sturm, “Hole- blocking titanium-oxide/silicon heterojunction and its application to photovoltaics,” Appl. Phys. Lett., vol. 102, no. 20, p. 203901, 2013.

[14] A. Agrawal, J. Lin, M. Barth, R. White, B. Zheng, S. Chopra, S. Gupta, K. Wang, J. Gelatos, S. E. Mohney, and S. Datta, “Fermi level depinning and contact resistivity reduction using a reduced titania interlayer in n-silicon metal-insulator-semiconductor ohmic contacts,” Appl. Phys. Lett., vol. 104, no. 11, p. 112101, Mar. 2014.

[15] A. Cuevas, “Electrons and holes in solar cells with partial rear contacts,” Prog. Photovolt. Res. Appl., vol. 22, no. 7, pp. 764–774, Jul. 2014. A Low Resistance Calcium /

Reduced Titania Passivating

Contact for High Eﬃciency

Crystalline Silicon Solar Cells

Thomas Allen1, James Bullock2,3, Quenting Jeangros4, Christian

Samundsett1, Yimao Wan1, Jie Cui1, A¨ıcha Hessler-Wyser4,

Stefaan De Wolf5, Ali Javey2,3 and Andr´esCuevas1

1Research School of Engineering, College of Engineering and Computer Science,

Australian National University, Canberra, ACT, 0200, Australia

2Department of Electrical Engineering and Computer Sciences, University of

California, Berkeley, CA, 94720, USA

3Materials Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA,

94720, USA

4Institute of Micro Engineering, Photovoltaics and Thin-Film Electronic Laboratory,

Ecole Polytechnique Federale de Lausanne, Maladiere 71b, CH-200 Neuchatel,

Switzerland

201 Chapter 4. Calcium-Based Electron Contacts 202

5KAUST Solar Centre, King Abdullah University of Science and Technology, Thuwal,

Saudi Arabia

Published in Advanced Energy Materials, available online, 2017.

Abstract — Recent advances in the eﬃciency of crystalline silicon (c-Si) solar cells have come through the implementation of passivated contacts that simultaneously reduce recombination and resistive losses within the contact structure. In this contribution we demonstrate low resistivity passivated contacts based on reduced titania (TiOx) contacted with the low work function metal, calcium (Ca). By using Ca as the overlying metal in the contact structure we are able to achieve a reduction in the contact resistivity of TiOx passivated contacts of up to two orders of magnitude compared to previously reported data on Al / TiOx contacts, allowing for the application of the Ca

/ TiOx contact to n-type c-Si solar cells with partial rear contacts. Imple- menting this contact structure on the cell level resulted in a power conversion eﬃciency of 21.8% where the Ca / TiOx contact comprised only ∼6% of the rear surface of the solar cell, an increase of 1.5% absolute compared to a similar device fabricated without the TiOx interlayer.

Index Terms — passivated contacts, solar cell, calcium, PERC, TiO2. Chapter 4. Calcium-Based Electron Contacts 203

I. Introduction

Advances in the efficiency of crystalline silicon (c-Si) photovoltaic (PV) devices above the long-held record efficiency value of 25% have all come from solar cell architectures with passivated contacts fabricated on n-type silicon [1]. The most successful devices to date have a silicon heterojunction (SHJ) cell structure, featuring a thin intrinsic amorphous silicon (a-Si) film that passivates c-Si surface defects, effectively separating the solar cell absorber (c-Si) from the remaining contact materials (doped a-Si, transparent conductive oxides, and metals). This separation of absorber and contact materials avoids extraneous

Auger recombination and free carrier absorption losses associated with heavily doped regions, a detrimental feature of diﬀused junction solar cells. However, the amorphous silicon heterocontact structure necessitates the implementation of large contact fractions due to a relatively high contact resistivity. The trade-oﬀ between contact resistivity (ρc),

contact recombination (J0c), and contact fraction (fc) is optimised for these devices by

minimising J0c and combating the increase in ρc by applying the contacts over a large

area. However, optical losses arising from parasitic absorption in the doped and intrinsic

a-Si layers, as well as the transparent conductive oxide, are a limiting factor in this cell

design [2]. Interdigitated back contact structures minimise these losses and have achieved

the highest performance for c-Si solar cells [3], but they still pose challenges for mass

production.

Another strategy to achieve high eﬃciencies is to develop transparent, dopant-free, pas-

sivating heterocontacts that minimise ρc, thereby allowing the application of the contact

structure to devices with low contact fractions, as in the partial rear contact (PRC) ar-

chitectures commonly known as PERC (passivated emitter and rear cell) and PERL

(passivated emitter with rear locally diﬀused) cells [4]. By minimising ρc the constraints Chapter 4. Calcium-Based Electron Contacts 204 on recombination at the contact can be relaxed and the contact can be applied to small areas (fc < 10%), leaving the remaining surfaces to be passivated with materials that have been utilized by the c-Si PV industry for decades, like silicon nitride (SiNx) and aluminium oxide (Al2O3). These materials are known to eﬀectively eliminate Shockley-

Read-Hall (SRH), or defect-assisted, recombination at the silicon surface. Optically this approach can be beneficial, insofar as these materials form non-absorptive and highly reflective optical mirrors when common metals, like Al and Ag, are deposited on top of them. Nevertheless, very few dopant-free electron contacts have been found to date with a sufficiently low ρc to be applied to small contact fractions.

Recently, signiﬁcant progress has been made in the development of directly metalized, dopant-free electron contacts applied to PRC cells fabricated on n-type silicon [5],[6].

These device structures have been made possible by applying low work function materials

2 to facilitate low resistivity (ρc ∼2 mΩ.cm ) Ohmic contacts to undiffused n-type c-Si surfaces. These devices have demonstrated efficiencies of over 20%, but, like the PERC cell architecture, are limited by recombination at the contacts [7],[8]. One promising material choice for the formation of a low resistance, dopant free heterocontact is TiO2, which has been recently demonstrated to form Ohmic contact on undiffused n-type silicon when applied in conjunction with a low work function Al / LiF overlayer, although

2 no solar cells were fabricated due to the high contact resistivity (ρc ∼500 mΩ.cm ) [9]. In addition, the work of Yang et al. has demonstrated full area Al / TiO2 electron contacts on the rear side of 19.6% eﬃcient n-type solar cells. With an additional SiOx interlayer, these devices have shown a remarkable eﬃciency potential of up to 21.6%, however the application of these contacts is limited to large areas due to their relatively high contact

2 resistivities (300 > ρc > 30 mΩ.cm ) even for TiO2 thicknesses less than 3 nm [10],[11].

In this article we show a reduction of up to two orders of magnitude in the contact Chapter 4. Calcium-Based Electron Contacts 205

resistivity of TiO2 heterocontacts by replacing the overlying aluminium metal with the low work function metal calcium. In doing so, we demonstrate the compatibility of TiO2

passivated heterocontacts with n-type c-Si cell designs with partial-area rear contacts,

fabricating a ﬁrst-of-its-kind passivated n-type PRC solar cell with an eﬃciency of 21.8%.

II. Results and Discussion

A. Surface passivation

The passivation of c-Si surface defects by TiO2 deposited by ALD has been demonstrated

previously [12]–[14], with surface recombination velocity (SRV; Seff ) values below 1

cm/s reported in the literature for TiO2 ﬁlms as thick as 15 nm on 10 Ω.cm n-type

silicon [14]. The work of Yang et al. [11] reports SRV values for thinner TiO2 ﬁlms

in the range of 56 cm/s to 11 cm/s on 1 Ω.cm n-type substrates for TiO2 thicknesses

between 2.5 nm and 5.5 nm, respectively. (Note that SRV scales with wafer doping in

low injection conditions). All studies have reported deleterious eﬀects of temperature

on the SRV, even for short annealing times at temperatures over 250 ◦C, a result of a

phase transformation from amorphous to anatase TiO2 [13],[14].

The results of Figure 1 display the eﬀective minority carrier lifetime of silicon wafers

coated with thin TiO2 layers applied in this study after a thermal treatment in a forming

gas ambient at 250 ◦C for 5 minutes, following the annealing optimisations in Refs.

[11] and [13]. The results show an increase in eﬀective minority carrier lifetime with

increasing TiO2 thickness until saturating at 200 cycles. The SRV values, calculated

using the intrinsic lifetime parameterisation of Richter et al. [15], are similar to those Chapter 4. Calcium-Based Electron Contacts 206

1 0 - 3 1 Ω. c m n - t y p e , F Z , < 1 0 0 > ) s (

f f e ؔ

e m i t e

f T i O t h i c k n e s s :

i 2 l

r 1 0 . 5 n m e i

r 7 . 0 n m r

a 5 . 3 n m

c - 4 1 0

s 3 . 5 n m s

e 1 . 8 n m c x E

1 0 1 3 1 0 1 4 1 0 1 5 1 0 1 6 ( n ( c m - 3 פ E x c e s s c a r r i e r d e n s i t y

Fig. 1: Excess carrier lifetime of TiO2 passivated 1 Ω.cm n-type silicon as a function of injection level measured using the photoconductance decay technique. reported previously, ranging from approximately 240 cm/s for the thinnest (1.8 nm) to

15 cm/s for the thickest TiO2 layers (10.5 nm).

B. Contact resistivity

Despite having been a common material for anti-reflection coatings in the c-Si PV industry, and despite its prevalent use in other solar cell devices (e.g. dye sensitised and perovskite solar cells), the capacity for TiO2 to form an electron-selective heterocontact on c-Si has only recently been firmly established [16]. This finding prompted a number of studies into the efficacy of TiO2 heterojunctions at the device level [10],[17], the most successful implementation of the Al / TiO2 contact being that of Yang et al., which Chapter 4. Calcium-Based Electron Contacts 207 achieved a power conversion efficiency of 19.8%, increasing to 21.6% with the addition of a SiO2 interlayer [11]. These devices feature full area TiO2 contacts on the planarised rear of the solar cells as the contact resistivities of those Al / TiO2 structures are not compatible with partial rear contact cell designs.

This limitation of the Al / TiO2 contact is further demonstrated in Figure 2a), where the current densities extracted from Cox and Strack test structures are plotted as a function of voltage. The samples were fabricated on 0.9 Ω.cm n-type Si wafers with a heavily phosphorus diﬀused, aluminium capped rear side to minimise the resistive contribution of the back contact. The undiﬀused front side feature TiO2 layers of 3.5 nm and 5.3 nm contacted with Al; the J −V data from one control sample without TiO2

is also shown in the Figure. It can be seen that without the addition of the TiO2 layer

the direct n-type Al / Si contact exhibits a non-Ohmic, rectifying behaviour typical of

direct metal / undiﬀused n-type silicon contacts. The rectiﬁcation observed is due to

the formation of a large energy barrier for electrons (∼0.7 eV) at the Al / Si interface

that is empirically ascribed to the Fermi-level pinning phenomenon, in which surface

defects limit the metal work functions ability to inﬂuence the magnitude of the barrier

height at the metal-silicon interface [18],[19]. The addition of the 3.5 nm TiO2 interlayer leads to a reduction of the barrier to electrons compared to the directly metallised case, however the J − V data is still representative of a non-Ohmic contact, contrary to the data in [11]. If, however, we restrict the current density in Figure 22) to a range of

−2 ±0.1 A/cm the data appears Ohmic and an approximate extrapolation of ρc ∼150

2 mΩ.cm can tentatively be made. Such is not the case for the thicker TiO2 layer where the contact is rectifying at any reasonable current density range at the resolution of this measurement.

This interpretation of the J–V curves is relevant because the current density at the Chapter 4. Calcium-Based Electron Contacts 208

Fig. 2: J–V measurements (a) of contact resistance test structures showing rectifying behavior of the Al / n-type Si and Al / TiOx / n-type Si contacts, and the Ohmic behavior of the Ca / n-type Si contact; and (b) contact resistivity extracted from TLM test structures for Ca / Si and Ca / TiOx / Si contacts. contacts in a solar cell is proportional to the inverse of the contact fraction (see the inset of Figure 2a)). Very low contact fractions (e.g. fc = 1%) result in large current densities at an individual point contact, on the order of ∼4 A/cm2 for c-Si devices, while the current density at full area contacts mirrors that of the generation current (minus any quantum efficiency losses) and so is typically less than 0.04 A/cm2 at the maximum power point. This has a significant impact when interpreting contact resistance data and the applicability of contact structures to devices with large and small area contacts. This would also be a significant consideration when applying these contacts to concentrator cells which are more sensitive to resistive losses and where the generation current density approximately follows the concentration ratio. Additionally, for non-concentrator cells, the Al / TiO2 contacts shown in Figure 2a) applied to full areas would likely show a signature bending of a Suns-Voc curve typical of rectifying contacts [20], should a suitably high light intensity value be reached [21]. As such, the 3.5 nm Al / TiO2 contact resistivity measured in Figure 2a), is certainly suitable for non-concentrator cells with large area contacts, though the value of ρc is still higher than the value reported in Ref. Chapter 4. Calcium-Based Electron Contacts 209

[11] but less than a prior report of 250 mΩ.cm2 for a similarly prepared contact structure

[10]. The reason for this diﬀerence in reported ρc is unclear, however the assertion in

Refs. [10] and [11] of Ohmic contact for Al / TiO2 test structures for TiO2 thickness up to 5.5 nm may lie in a diﬀerence in the range of current densities explored.

The data of Figure 2a) stands in contrast of that of Figure 2b), in which the low work function metal Ca is used as the contact material, instead of Al. As can be seen in Figure

2b), the addition of 3.5 nm of TiO2 causes a negligible change in contact resistance over

2 the directly Ca-metallised case (ρc ∼5 mΩ.cm ), while increasing the thickness to 5.3 nm resulted in a non-Ohmic I −V response in the as-deposited state. After annealing at

◦ 2 2 250 C, a ρc of 27 mΩ.cm was measured, increasing to 47 mΩ.cm with a further anneal

◦ at 300 C, a trend of increasing ρc also registered for both the directly metallised and thinner TiO2 samples. Using Al as the metallic overlayer, Yang et al. report contact

2 2 resisitivies of 250 mΩ.cm and 750 mΩ.cm for TiO2 layers of comparable thickness (3.5 and 5.5 nm, respectively) [10], indicating a remarkable reduction in contact resistivity of 2 orders of magnitude after replacing the Al layer with the low work function metal

Ca.

While the ρc values measured for the thicker TiO2 sample are not suitable for the application to devices with partial area contacts, and so are not further explored here, it may be that thicker TiO2 layers result in lower J0c values after metallisation and so could be applied to larger contact areas. As the current focus is on passivated PRC cells we leave that optimisation for future work. Chapter 4. Calcium-Based Electron Contacts 210

C. PRC solar cells

The beneficial effect of the insertion of a TiO2 passivating interlayer between the rear point contacts and the silicon absorber material in the PRC cell structure of Ref. [6] is clearly demonstrated in the J − V curves of Figure 3. The Figure compares the data from the cell reported in Ref. [6] with a similarly fabricated device (this work) featuring both a larger rear contact fraction (6.25% vs. 1.26%) and a 3.5 nm TiO2 passivating interlayer between the Ca metal and the underlying c-Si absorber material. The TiO2 passivating interlayer has increased the open circuit voltage of the device by nearly 30 mV to 681 mV, an increase in Voc that is reflected in the increase in device efficiency by

1.5% absolute to 21.8%, making this device the most eﬃcient c-Si solar cell with a TiO2 heterocontact fabricated to date. Remarkably, this increase in eﬃciency is achieved

2 without an increase in series resistance, but rather a decrease in Rs from 0.8 Ω.cm for the directly metallised cell to below 0.5 Ω.cm2. This is indicative of a retention of the

2 low contact resistance (ρc ∼5 mΩ.cm ) despite the addition of the TiO2 interlayer, as shown in the contact resistance data of Figure 2b), and also due to the application of the contact to a larger area. As the thin TiO2 interlayer is transparent to IR wavelengths we measure no change in Jsc, while the increase in ﬁll factor (FF ) is a direct consequence of both the lower Rs losses and higher Voc.

Despite the signiﬁcant increase in Voc measured on the passivated PRC cell, it is apparent from the superposition of the measured eﬃciency data on to the modelled curves of

Figure S1 that the recombination rate at the Si / TiO2 interface has increased markedly after the thermal evaporation of the overlying Ca layer. The minority carrier lifetime data of Figure 1 for a 3.5 nm thick TiO2 passivating layer corresponds to a J0c of

−14 2 approximately 5×10 A/cm . On the cell level the J0c value increases to a value Chapter 4. Calcium-Based Electron Contacts 211

Fig. 3: One sun J–V curves of the PRC solar cells fabricated with and without a passivating TiOx interlayer. Note that the TiOx cell has a higher rear contact fraction (fc = 6.25%) compared to the reference cell without the TiOx (fc = 1.26%).

−11 2 of J0c < 1 × 10 A/cm , an increase of ∼2.5 orders of magnitude. Improving the

J0c value, either by a different choice of passivating interlayer, or through the insertion of a thin SiO2 layer, as per Ref. [11], remains critical to achieving device efficiencies competitive with other passivated contact technologies. Another pathway to higher efficiencies may lie in replacing the Ca layer with a less reactive, low work function material. Nevertheless, the devices presented here represent the first n-type passivated partial rear contact cells and conclusively demonstrate the benefits of such a device architecture. Chapter 4. Calcium-Based Electron Contacts 212

D. Structure and composition of the contact

The composition and stoichiometry of the as-deposited TiO2 ALD layer has been analysed by X-ray photoelectron spectroscopy. The survey scan (not shown) identiﬁed peaks in the spectra associated with Ti, and O elemental species as well as carbon contamination from the organic precursor. Comparing the relative atomic percentages of the

O and Ti components of the TiOx layer indicates a near stoichiometric atomic ratio of x ∼2.09. The Ti 2p core level spectrum, displayed in Figure 4, shows the peak positions of the Ti 2p1/2 (464.5 eV) and Ti 2p3/2 (458.7 eV) electron spin orbitals, a spin-orbit

4+ splitting of 5.8 eV, characteristic of Ti species in TiO2 [22]–[24], further evidence of a stoichiometric as-deposited TiO2 layer.

Fig. 4: XPS of the ALD titania prior to metalisation indicating a near stoichiometric atomic ratio of TiOx, x = 2.09 Chapter 4. Calcium-Based Electron Contacts 213

STEM HAADF imaging coupled to EDX or EELS has been performed to further investigate the contact structure and composition after Al / Ca evaporation. The STEM

HAADF micrograph and corresponding Al, Ca, O, Ti and Si EDX maps shown in Figure

5a highlight the uniform titanium oxide layer separating the silicon absorber from the overlying Al / Ca contact metals. The EDX data also shows an intermixing of the Al /

Ca layers and the accumulation of O in the Ca layer close to the Ca / TiOx interface.

While the surface of the Ca is likely to have oxidised slightly during the transfer from the FIB to the TEM, this eﬀect should give rise to a more uniform O EDX signal at the position of the Ca layer than that observed in Figure 5a. Therefore, it is likely that the ﬂuctuation in the O signal observed in the EDX image of Figure 5a arises from interactions between the TiOx and the Ca layers.

Also observable in Figure 5a is the apparent diﬀusion of Ca through the TiOx layer.

This is supported by the EEL spectra (Figure 5c) recorded across the high-resolution

STEM image shown in Figure 5b which detect Ca in the TiOx layer. The EELS data also highlights some intermixing between Si, Ca and Ti over ∼2 nm (arrowheads in Figure

5c), an observation that might be inﬂuenced by FIB-induced damage or the sample tilt.

In agreement with EDX results, the EELS data demonstrates that O is present mainly on the Ca side of interface with the TiOx layer (shown by the arrow in Figure 5c).

Higher energy resolution EEL spectra of the Ti L2,3 and O K edges of Ca / TiOx and

Al / TiOx contact structures are shown in Figure 5d. In contrast to the stoichiometric

TiO2 composition inferred from the XPS data after deposition, an estimation of the

Ti to O atomic ratio using EELS partial scattering cross sections yields a composition closer to TiO, irrespective of whether it is in contact with Ca or Al. The quantiﬁcation procedure was performed using the software Digital Micrograph (Gatan, Pleasanton,

USA) and involved the computation of a power law background (AE−r, where E is the Chapter 4. Calcium-Based Electron Contacts 214

energy-loss in eV, A and r ﬁtting parameters) in the range 410 to 430 eV (for the Ti L2,3 edges) and 513 to 528 eV (for the O K edge), the extrapolation of this background and its subtraction to the EELS signal in the range 456-476 eV and 532-572 eV, respectively, to yield the integrated Ti and O core loss intensities. These intensities are then related to elemental ratios using scattering cross sections (which have an accuracy on the order of

10%). It should be mentioned that while the shape of the Ti L2,3 edges is indicative of its oxidation state, the amorphous nature of the TiOx layer (as demonstrated in Figure 5b) complicates any fine structure fingerprinting procedure. Indeed, the lack of crystallinity prevents the splitting of the L2 and L3 edges into further peaks irrespective of the oxidation state, meaning that the fine structure of cubic (and presumably amorphous)

TiO appears similar to that of amorphous TiO2 [25],[26]. Overall, these results indicate that the Ca (and Al) uniformly reduces the neighbouring TiO2 layer to a stoichiometry closer to TiOx, where x ∼1. This, in part, can explain the build-up in oxygen at the Ca

/ TiOx interface observed by EDX and EELS in Figure 5: evidence of the spontaneous migration of oxygen from the as-deposited stoichiometric TiO2 to the overlying Ca layer.

The reaction kinetics of metal / TiO2 interfaces have been studied by Fu and Wagner

[27], who found the interfacial reactivity (in this instance, the extent of O2− transfer at the interface) to be dependent on the overlying metal work function. They demonstrated by XPS analysis on thin metal overlayers (∼6 A)˙ the reduction at room temperature of TiO2 by Al, the lowest work function metal studied, which is in agreement with the

EELS data shown in Figure 5d) for the case of a direct contact between TiO2 and Al.

Since the work function of Ca (φ ∼2.9 eV) is considerably lower than that of Al (φ ∼4.2 eV) it follows that the reduction of the TiO2 would occur, and likely to a greater degree.

Furthermore, the preceding interpretation of the EELS data suggesting the reduction of the TiO2 to TiO is supported by the XPS data of Demri et al. [28] in which the Chapter 4. Calcium-Based Electron Contacts 215

(a) (c)

Ca L2,3

5 nm [nm] Si L Ti L2,3 O K (b) 2,3

5 nm

Al Ca O Ti Si 5 (d)

Ti L2,3 I [a.u.] TiOx / Al TiOx / Ca TiOx 0

O K

c-Si -5

Ca L1

440 460 480 500 520 540 560 100 150 200 350 400 450 500 550 E [eV] E [eV]

Fig. 5: STEM HAADF imaging with EDX elemental mapping (a) of the Al / Ca / TiOx / Si contact structure. High resolution STEM (b) with EELS (c) of the same Al / Ca / TiOx / Si contact. The EDX and EELS both show the Ca intermixing with the TiOx layer. Higher energy resolution EELS taken from both the Al / TiOx / Si and Ca / TiOx / Si contacts are shown in (d). Analysis of the high resolution EEL spectra indicate a reduction in the titania after the application of the contact metals.

interface of crystalline TiO2 and Ca was investigated. The authors also report on the formation on TiO and CaO layers due to the transfer of oxygen from the TiO2 to the overlying thermally evaporated Ca layer. The reduction of TiO2 by an overlying metal layer has also been proposed elsewhere as being critical to contact formation in TiOx electron selective heterojunctions on c-Si, where the reduction in the titanium oxide is proposed to enhance electron transport through the layer by increasing its conductivity

[29], and either improving the band alignment with the silicon surface or reducing the barrier height at the metal / TiOx interface [11],[30].

While there is a growing body evidence, including that presented here, suggesting a reduction in the TiOx layer is one factor in enabling electron contact formation with Chapter 4. Calcium-Based Electron Contacts 216 c-Si, the presence of other effects complicates the analysis in this case. Of particular consideration is the difference in the work function of the contact metal used here, as opposed to other TiO2 / c-Si electron contacts reported in the literature. Given the similarities in EEL spectra in Figure 5d, it is apparent that the low work function of the Ca layer is one of the main drivers behind the reduction in ρc for the Ca / TiOx, compared to the Al / TiOx, contact. Additionally, the apparent reduction of the TiO2 passivating layer implies an oxidation of the overlying, and interdiffused, Ca layer (and

Al and Ti layers in Refs. [11] and [30]). This may explain the sensitivity of ρc to TiO2 thickness, though some presence of CaOx is evidently tolerable in forming low contact resistivities. It is likely that both the change in stoichiometry of the TiOx layer and possibly the diffusion of Ca into the TiOx have had an impact on the passivation of the contact structure, explaining the large discrepancy in Seff prior to metallisation and the implied J0 at the contact from Figure S1b. However, the extent to which each of the factors identified above are influencing the J0c and ρc (and hence the device performance) is difficult to distinguish from the data.

III. Conclusion

The use of the low work function metal calcium has enabled the fabrication of a low resistance, dopant-free TiOx passivated electron heterocontact to undiﬀused n-type c-

2 Si substrates. The contact resistivity of the Ca / TiOx / c-Si contact (∼5 mΩ.cm ) represents a reduction in ρc by two orders of magnitude over previously reported data for

TiOx / c-Si heterocontacts. Analytical transmission electron microscopy of the contact has revealed a reduction of TiO2 layer by the overlying Ca metal that is likely to have assisted in the lowering of the contact resistivity but compromised the passivation of c-Si Chapter 4. Calcium-Based Electron Contacts 217

surface defects at the TiOx / Si interface. Nevertheless, the extreme reduction in contact resistance compared to other TiOx-based heterocontacts reported in the literature, as well as the reduction in recombination at the TiOx / Si interface, has enabled the fabrication of a first-of-its-kind passivated partial rear contact n-type silicon solar cell with an efficiency of 21.8%, making this device the most efficient c-Si solar cell with a

TiOx heterocontact fabricated to date.

IV. Experimental Section

The TiO2 layers in this study were deposited by atomic layer deposition (ALD, Beneq

TFS 200) by sequential exposure of titanium isopropoxide (TTIP) and de-ionised water.

The TTIP was heated to a temperature of 40 ◦C, while the deposition chamber was

held at 230 ◦C. Symmetrically passivated lifetime samples were prepared on planar saw-

damage etched, RCA cleaned, silicon substrates that were dipped in dilute HF prior to

further processing. The eﬀective lifetime (τeff ) of the passivated samples was measured

as a function of minority carrier injection level (∆n) on a Sinton Instruments WCT120

photoconductance tester operating in quasi-steady-state and transient modes. X-ray

photoelectron spectroscopy (XPS, Thermo Scientiﬁc ESCALAB 250Xi) survey and core

level scans were performed on the TiO2 layers with a monochromatic Al Kα (1486.68

eV) X-ray source at a vacuum pressure of < 2 × 10−9 mbar.

Thin, ∼30 nm, layers of Ca were thermally evaporated at low vacuum pressures (<

1 × 10−6 Torr) from a solid source in a glovebox-integrated metal evaporator. An initial

∼150 nm capping layer of Al was sequentially evaporated in the same chamber without

breaking vacuum. A further ∼150 nm of Al was subsequently evaporated after breaking

vacuum to thicken the capping layer. Contact resistance measurements were performed Chapter 4. Calcium-Based Electron Contacts 218 using Cox and Strack (with a rear side Al / n+ contact) and transfer length method

(TLM) test structures which were fabricated using shadow masks to deﬁne the contact geometry [31],[32]. The contact resistivity was extracted from the I–V data obtained from a Keithley 2420 source meter.

The Al / Ca / TiO2 and Al / TiO2 contact structures were investigated on polished

Si wafers by transmission electron microscopy (TEM). For this purpose, TEM samples were prepared using the focused ion beam (FIB) lift-out method in a Zeiss NVision

40 workstation and thinned down from the Si substrate side (to curtail possible rede- position of species) up to a ﬁnal thickness of ∼80 nm using a ﬁnal Ga+ voltage of 2 kV to minimise surface damage and Ga implantation. To reduce Ca oxidation, the samples were quickly transferred from the FIB to a monochromated probe- and image-

Cs-corrected FEI Titan Themis microscope, which was operated at 200 kV (air exposure of ∼2 mins). Characterisation then involved scanning TEM (STEM) high-angle annu- lar dark-ﬁeld (HAADF) imaging coupled to either energy-dispersive X-ray spectroscopy

(EDX) or electron energy-loss spectroscopy (EELS) using a probe current of 250 or 50 pA, respectively. EEL spectra were recorded in Dual EELS mode using a Gatan GIF

Quantum ERS high-energy resolution spectrometer and energy ﬁlter with a convergence semi-angle of 20 mrad and a collection semi-angle of 49 mrad. EEL spectra were ac- quired with an energy resolution of either 400 or 1200 meV depending on whether the monochromator was excited or not.

PRC solar cells (∼155 µm thick; ∼ 2 × 2 cm2, isolated by a front-side mesa etch) were fabricated on 0.9 Ω.cm n-type ﬂoat zone silicon wafers. The cells feature a front-side boron diﬀusion (Rsheet ∼120 Ω/) on random pyramid texturing, passivated by a SiNx /

Al2O3 stack. The planarised rear-sides of the cells were passivated with plasma-enhanced chemical vapour deposited (PECVD) SiNx. The front contact openings were deﬁned by Chapter 4. Calcium-Based Electron Contacts 219 photolithography and formed by a thermally evaporated Cr / Pd / Ag stack that was later thickened with additional Ag by electroplating. The rear-side contacts were also deﬁned by photolithography prior to the TiO2 ALD and Ca and Al metal evaporation procedures. The current-voltage (J–V ) characteristics of the cells were measured using a

Sinton Instruments FCT-450 ﬂash tester which was calibrated using a certiﬁed reference cell from Franhaufer ISE CalLab.

Acknowledgements

This work has been supported by the Australian government through the Australian

Renewable Energy Agency (ARENA). Work at the University of California, Berkeley was supported by the Bay Area Photovoltaic Consortium (BAPVC). The authors would like to acknowledge Sorin Lazar for his help with monochromated EELS experiments and the Interdisciplinary Centre For Electron Microscopy of EPFL for the use of their microscope. Chapter 4. Calcium-Based Electron Contacts 220

References

[1] C. Battaglia, A. Cuevas, S. De Wolf, Energy Env. Sci, 2016, 9, 1552.

[2] Z. C. Holman, A. Descoeudres, L. Barraud, F. Z. Fernandez, J. P. Seif, S. De Wolf, C. Ballif, IEEE J. Photovolt., 2012, 2, 7.

[3] NEDO: “Worlds Highest Conversion Eﬃciency of 26.33% Achieved in a Crystalline Silicon Solar Cell.” [Online]. Available: http://www.nedo.go.jp/english/news/ AA5en_100109.html. [Accessed: 22-Sep-2016].

[4] M. A. Green, Sol. Energy Mater. Sol. Cells, 2015, 143, 190.

[5] J. Bullock, P. Zheng, Q. Jeangros, M. Tosun, M. Hettick, C. M. Sutter-Fella, Y. Wan, T. Allen, D. Yan, D. Macdonald, S. De Wolf, A. Hessler-Wyser, A. Cuevas, A. Javey, Adv. Energy Mater., 2016, 6, 1600241.

[6] T. G. Allen, J. Bullock, P. Zheng, B. Vaughan, M. Barr, Y. Wan, C. Samundsett, D. Walter, A. Javey, A. Cuevas, Prog. Photovolt. Res. Appl., 2016.

[7] A. W. Blakers, A. Wang, A. M. Milne, J. Zhao, M. A. Green, Appl. Phys. Lett., 1989, 55, 1363.

[8] M. A. Green, A. W. Blakers, J. Zhao, A. M. Milne, A. Wang, X. Dai, IEEE Trans. Electron Devices, 1990, 37, 331.

[9] J. Bullock, M. Hettick, J. Geissbuhler, A. J. Ong, T. Allen, C. M. Sutter-Fella, T. Chen, H. Ota, E. W. Schaler, S. De Wolf, C. Ballif, A. Cuevas, A. Javey, Nat. Energy, 2016, 1, 15031.

[10] X. Yang, P. Zheng, Q. Bi, K. Weber, Sol. Energy Mater. Sol. Cells, 2016, 150, 32.

[11] X. Yang, Q. Bi, H. Ali, K. Davis, W. V. Schoenfeld, K. Weber, Adv. Mater., 2016, 28, 5891.

[12] A. F. Thomson, K. R. McIntosh, Prog. Photovolt. Res. Appl., 2012, 20, 343.

[13] B. Liao, B. Hoex, A. G. Aberle, D. Chi, C. S. Bhatia, Appl. Phys. Lett., 2014, 104, 253903.

[14] J. Cui, T. Allen, Y. Wan, J. McKeon, C. Samundsett, D. Yan, X. Zhang, Y. Cui, Y. Chen, P. Verlinden, A. Cuevas, Sol. Energy Mater. Sol. Cells, 2016, 158, 115.

[15] A. Richter, S. W. Glunz, F. Werner, J. Schmidt, A. Cuevas, Phys. Rev. B, 2012, 86, 165202.

[16] S. Avasthi, W. E. McClain, G. Man, A. Kahn, J. Schwartz, J. C. Sturm, Appl. Phys. Lett., 2013, 102, 203901.

[17] K. A. Nagamatsu, S. Avasthi, G. Sahasrabudhe, G. Man, J. Jhaveri, A. H. Berg, J. Schwatz, A. Kahn, S. Wagner, J. C. Sturm, Appl. Phys. Lett., 2015, 106, 123906. Chapter 4. Calcium-Based Electron Contacts 221

[18] D. K. Schroder,D. L. Meier, IEEE Trans. Electron Devices, 1984, 31, 637.

[19] R. T. Tung, Appl. Phys. Rev., 2014, 1, 11304.

[20] R. Sinton, A. Cuevas, presented at the 16th European Photovoltaics Solar Energy Conference, Glasgow, Scotland, 2000.

[21] S. W. Glunz, J. Nekarda, H. Mackel, A. Cuevas, presented at the 22nd European Photovoltaics Solar Energy Conference and Exhibition, Milan, Italy, 2007.

[22] A. F. Carley, P. R. Chalker, J. C. Riviere, M. W. Roberts, J. Chem. Soc. Faraday Trans. 1: Phys. Chem. Condens. Phases, 1987, 83, 351.

[23] S. Bartkowski, M. Neumann, E. Z. Fedorenko, S. N. Shamin, V. M. Cherkashenko, S. N. Nemnonov, A. Winiarski, D. C. Rubie, Phys. Rev. B, 1997, 56, 10656.

[24] U. Diebold, Surf. Sci. Spectra, 1996, 4, 227.

[25] G. Bertoni, E. Beyers, J. Verbeeck, M. Mertens, P. Cool, E. F. Vansant, G. Van Tendeloo, Ultramicroscopy, 2006, 106, 630.

[26] E. Stoyanov, F. Langenhorst, G. Steinle-Neumann, Am. Mineral., 2007, 92, 577.

[27] Q. Fu, T. Wagner, J. Phys. Chem. B, 2005, 109, 11697.

[28] B. Demri, M. Hage-Ali, M. Moritz, J. L. Kahn, D. Muster, Appl. Surf. Sci., 1997, 108, 245.

[29] M. T. Greiner, M. G. Helander, W.-M. Tang, Z.-B. Wang, J. Qiu, Z.-H. Lu, Nat. Mater., 2011, 11, 76.

[30] A. Agrawal, J. Lin, M. Barth, R. White, B. Zheng, S. Chopra, S. Gupta, K. Wang, J. Gelatos, S. E. Mohney, S. Datta, Appl. Phys. Lett., 2014, 104, 112101.

[31] R. H. Cox, H. Strack, Solid-State Electron., 1967, 10, 1213.

[32] D. L. Meier, D. K. Schroder, IEEE Trans. Electron Devices, 1984, 31, 647.

Supporting Information

I. Making the case for passivated partial rear contacts

The application of a passivating interlayer to the rear side of a PRC cell design can not only result in higher device performance but can also free up constraints on cell fabrication with regards to wafer resistivity. In Figure S1 the n-type PRC cell structure reported in Ref. [1] is modelled using parameters listed in Table S1, following previously reported empirical data [1],[2]. In the Figure, the modelled device eﬃciency is reported as a function of both the wafer resistivity and rear contact recombination current parameter J0c. The modelling predicts, as was previously demonstrated by the authors

[1], a monotonic increase in eﬃciency with increasing doping for devices with poorly passivated rear contacts, representative of devices with directly metallised rear contacts.

The high sensitivity of this directly metallised device architecture on recombination at the rear contact is due to the fact that as the base doping increases, the minority carrier concentration decreases, and so too the recombination rate. This decrease in the recombination rate also has an impact on internal series resistance due to higher carrier concentrations, and so higher conductivity in the silicon wafer [1].

This trend changes when passivated contacts are introduced. Reducing the recombination losses at the rear point contacts (decreasing J0c) results in a ﬂattening of the curve, until Auger recombination becomes a dominant loss mechanism as the wafer resistivity

223 Chapter 4. Calcium-Based Electron Contacts 224

Fig. S1: Modelling of cell designs reported in Ref. [1] (Figure S1a) and this work (Fig- ure S1b) as a function of wafer resistivity ρc, and rear contact recombination parameter J0c. The three sets of curves in each Figure are representative of the three cases explored: an inﬁnite lifetime in the silicon bulk material (solid curves); the introduction of a defect that limits the bulk lifetime (dashed curves); and an inﬁnite bulk lifetime with a selective emitter front side (dot-dash curves). The stars represent the empirical results from the J–V curves in Figure 3.

Table S1: Cell Modelling Input Parameters

Input parameter Value Wafer type/thickness n-type/155 µm −2 Front J0 72 or 15 fA.cm −6 2 Front ρc 1 × 10 Ω.cm 2 Jgen 40 mA/cm Bulk τ Inﬁnite or bulk limiting −3 2 Rear ρc 5 × 10 Ω.cm

Rear fc 1.26% or 6.25% −2 Rear J0 (non-contacted) 3 fA.cm 2 Rs 0.3 Ω.cm 2 Rsh 10000 Ω.cm

approaches 0.1 Ω.cm. Comparing Figure S1a and S1b, the modelling shows that by in-

creasing the contact fraction fc, the eﬀect of J0c becomes more pronounced as the wafer

resistivity increases. This leads to a higher eﬃciency potential in the lowest J0c case

(as series resistance losses are reduced) and a lower eﬃciency potential for the highest

J0c case (as contact recombination losses dominate), seen in the wider divergence of the modelled data in Figure S1b compared to Figure S1a at high resistivities. At lower resistivities the modelling shows that all J0c conditions for the larger fc case result in Chapter 4. Calcium-Based Electron Contacts 225

a higher eﬃciency compared to the lower fc case: the recombination at the contacts is again being mitigated by the higher wafer doping, even when comparing the data for the highest J0c in Figure S1b with the lowest in Figure S1a, despite the signiﬁcantly

larger area of the recombination active regions.

It is worth noting that the addition of a passivating interlayer between the undiﬀused

contact region and the overlying low work function metal in the PRC cell architecture

modelled here serves much the same function as the addition of localised dopant diﬀu-

sions underneath the metal electrode in the PERL cell vs. the PERC cell. In the PERL

cell design, constraints on base doping are lifted as the heavily doped sub-surface contact

region reduces the SRH recombination rate at the silicon-metal interface at the expense

of introducing additional Auger recombination losses. One point of diﬀerence regard-

ing the calcium based PRC devices modelled here is that, in the case of the PERL vs.

PERC cell structures the heavy doping under the metal contacts also serves to free up

the critical dependence of contact resistivity with base doping, a condition that limited

the fabrication of the PERC cells to wafer resistivities less than 0.5 Ω.cm [3]. As the

low work function enabled contacts in this contribution, as in Ref. [1] and [2], operate

via thermionic emission (and not thermionic ﬁeld emission through a narrowed barrier

width, as is the case for diﬀused and alloyed contacts) this dependence on ρc with wafer

resistivity is already relaxed.

Notably, as recently reported by Steinkemper et al. [4], the PERL cell architecture (and

so too the passivated PRC cell presented here) is sensitive in the eﬃciency response to

wafer resistivity with respect to bulk lifetime, unlike cell architectures that feature full

area rear contacts, like the SHJ and TOPCon cells [5]. This is shown in Figure S1, where

a midgap bulk defect is introduced. The defect parameters are such that the lifetime of

a 1 Ω.cm n-type wafer is limited to 1 ms at an injection level of ∆n = 1 × 1015 cm−3 Chapter 4. Calcium-Based Electron Contacts 226 and, given that the capture cross section ratio has been ﬁxed to unity, little injection dependence in the minority carrier lifetime is introduced, which is known to result in lower ﬁll factors at the device level [6]. This same defect is applied to each wafer resistivity / J0c combination, resulting in the dotted lines in Figure S1.

The decrease in efficiency with increasing wafer resistivity, as in the cells with poorly passivated contacts, and the cells with well passivated contacts but lower bulk lifetime, can be largely attributed to an increase in internal series resistance losses. This is due to the reliance of the PERC and PERL architectures to transfer carriers laterally, as well as vertically, to the rear contacts through the bulk of the device, and not through diffused regions (as in the PERT - passivated emitter, rear totally diffused - cell) or conductive overlayers (as in SHJ cells). Such lateral conductivity issues are of course eliminated with respect to current flows in full rear area contact schemes, as emphasised in Ref. [4] and [5]. This effect could also be mitigated in PERC cell designs if the contact pitch and diameter could be considerably reduced, such that the rear current

flows become quasi-1D [3],[7]. Nevertheless, this dependence of the efficiency on bulk lifetime offers substantial challenges to the manufacturing conditions of cells with partial rear contacts, where sensitivity to metallic surface impurities, even in n-type wafers, is a concern when high temperature processes are performed [8],[9]. It follows that eliminating high temperature processes, like thermal dopant diffusions, and adopting passivated heterocontacts, could increase the yield across a wider range of wafer dopant densities throughout the silicon ingot for solar cells manufactured with dopant-free, partial rear contacts.

Finally, the same devices have been modelled with a reduction in front side recombination, using the data reported in the selective emitter structure reported in Ref. [5]. As can be seen in Figure S1a and S1b, the introduction of a selective emitter to the front Chapter 4. Calcium-Based Electron Contacts 227 side of the cell makes a significant impact on device performance only when the recombination at the rear contacts is significantly suppressed, or when the wafer resistivity is very low. Note that the modelling of the devices with the reduced front recombination has been conducted without the introduction of a bulk defect. Adding the bulk defect creates an identical trend as for the cells modelled with the higher front side J0 and so is not shown. It is also noteworthy that increasing the generation current from 40 mA/cm2 to 42.1 mA/cm2, as reported in Ref. [5], lifts the maximum efficiency of the cells modelled with the best contact passivation and largest contact fraction to over 25%, identifying one limitation of using Ca as the rear metal rather than more highly reflective metals like Ag or even Al [1]. Furthermore, as the modelling here fixes the contact fraction fc and varies other inputs, the cells are not necessarily optimised for the rear contact area; the optimum fc would indeed change as recombination, resistivity and bulk lifetime change. Nevertheless, the modelling indicates numerous potential advantages of the passivated PERC cell architecture: a significantly higher efficiency potential for a wider range of wafer resistivities, with a potential for enhanced manufacturability and yield, compared to PERC cell structure of Ref. [10]. Chapter 4. Calcium-Based Electron Contacts 228

References

[1] T. G. Allen, J. Bullock, P. Zheng, B. Vaughan, M. Barr, Y. Wan, C. Samundsett, D. Walter, A. Javey, A. Cuevas, Prog. Photovolt. Res. Appl., 2016.

[2] J. Bullock, P. Zheng, Q. Jeangros, M. Tosun, M. Hettick, C. M. Sutter-Fella, Y. Wan, T. Allen, D. Yan, D. Macdonald, S. De Wolf, A. Hessler-Wyser, A. Cuevas, A. Javey, Adv. Energy Mater., 2016, 6, 1600241.

[3] M. A. Green, A. W. Blakers, J. Zhao, A. M. Milne, A. Wang, X. Dai, IEEE Trans. Electron Devices, 1990, 37, 331.

[4] H. Steinkemper, M. Hermle, S. W. Glunz, Prog. Photovolt. Res. Appl., 2016, 24, 1319.

[5] S. W. Glunz, F. Feldmann, A. Richter, M. Bivour, C. Reichel, H. Steinkemper, J. Benick, M. Hermle, presented at the 31st European Photovoltaic Solar Energy Conference and Exhibition, Hamburg, Germany, 2015.

[6] D. Macdonald, A. Cuevas, Prog. Photovolt. Res. Appl., 2000, 8, 363.

[7] A. Cuevas, IEEE J. Photovolt., 2012, 2, 485.

[8] D. MacDonald, A. Cuevas, K. McIntosh, L. Barbosa, D. De Ceuster, presented at the 20th European Photovoltaic Solar Energy Conference, Barcelona, Spain, 2005.

[9] J. Schmidt, B. Lim, D. Walter, K. Bothe, S. Gatz, T. Dullweber, P. P. Altermatt, IEEE J. Photovolt., 2013, 3, 114.

[10] A. W. Blakers, A. Wang, A. M. Milne, J. Zhao, M. A. Green, Appl. Phys. Lett., 1989, 55, 1363.