Signature Redacted January 10, 2020 Certified By: Rafael Jaramillo Assistant Professor of Materials Science and Engineering Thesis Supervisor

Mechanism and Applications of Large and Persistent

Photoconductivity in Cadmium Sulfide

Han Yin

B.Eng. Materials Science and Engineering, Peking University (2014) S.M. Mechanical Engineering, Massachusetts Institute of Technology (2016)

Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Mechanical Engineering

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

February 2020

Signature f A u oj.: Signatureredacted Han Yin Department of Mechanical Engineering Signature redacted January 10, 2020 Certified by: Rafael Jaramillo Assistant Professor of Materials Science and Engineering Thesis Supervisor

Certified by: Signature redacted Nicholas Xuanlai Fang Profes ' aianieering Chair

pted hv- Signature redacted~ AceA o CNSTT1STTT OFTECHNOLOGY Nicolas adjiconstantinou Professdr of Mechanical Engineering ;U FEB 0 5 2020 Chair, Departmental Committee on Graduate Students 0 LIBRARIES [This page is intentionally left blank.]

2 Mechanism and Applications of Large and Persistent Photoconductivity in Cadmium Sulfide

by Han Yin Submitted to the Department of Mechanical Engineering on January 10, 2020 in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Mechanical Engineering

Abstract Photoconductivity is the phenomenon where electrical conductivity changes as a result of photoexcitation of new charge carriers. In some semiconductors, photoconductivity is accompanied with enormous conductivity change and long decay time after photoexcitation is ceased. This effect is called large and persistent photoconductivity (LPPC). LPPC is due to the trapping of photo-generated minority carriers at crystal defects. Theory has suggested that anion vacancies in II-VI semiconductors are responsible for LPPC due to negative-U behavior, whereby two minority carriers become kinetically trapped by lattice relaxation following photo-excitation. By performing a detailed analysis of photoconductivity in CdS, we provide experimental support for this negative-U model. We also show that, by controlling sulfur deficiency in CdS, we can vary the photoconductivity of CdS films over nine orders of magnitude, and vary the LPPC characteristic decay time from seconds to 104 seconds. Sulfur vacancies are deep donors at equilibrium in the dark, but convert to shallow donors in a metastable state under photoexcitation. We demonstrate two-terminal all-electrical thin film resistive switching devices that exploit this defect-level switching (DLS) mechanism as a new way to control conductivity. We introduce a hole injection layer to inject holes into the deep donor levels in CdS and switch CdS into a "photoconductive" state. The device is in low resistance state as fabricated, and shows repeatable resistance switching behavior under electrical bias with no electro-forming. Results from mechanism study rule out switching mechanisms based on mass transport and support our DLS hypothesis. LPPC is pronounced in n-type carrier-selective contact (CSC) materials in thin film solar cells, but its effect is rarely recognized. We numerically model the effect of LPPC in CSC by switching defect levels between deep and shallow donor states. CSC photoconductivity can substantially affect solar cell performance. For instance, the power conversion efficiency of both CIGS and CdTe solar cells can be improved by over 4% (absolute) depending on the photoconductivity of the CdS CSC. The primary underlying cause is the influence of CSC shallow donor density on the junction depletion region. Optimizing CSC photoconductivity may be effective in solar cell engineering across multiple platforms.

Thesis Supervisor: Rafael Jaramillo Title: Assistant Professor of Materials Science and Engineering

3 [This page is intentionally left blank.]

4 Acknowledgments First and foremost, I would like to express my sincere gratitude to Prof. Rafael Jaramillo for his kind guidance in the past five years. Raf is a remarkable instructor and knowledgeable researcher who led me into the field of semiconductor physics and worked together with me through various projects. There are always obstacles during research, but Raf constantly embraces an optimistic mind. His attitude inspired me whenever I was depressed by bad research results, and we together made this thesis possible.

I am obliged to Prof. Nick Fang and Prof. Jeehwan Kim for their willingness to be in my thesis committee and their insightful suggestions during committee meetings which led to great improvements of my thesis.

It has also been great time working and discussing technical problems with our illustrious group members. Our discussions can often lead to new ideas and innovative ways to solve research problems. I also want to thank my collaborators for their constructive suggestions and assistance in performing measurements which could be done at MIT.

I am grateful to my parents for their constant support and love through my life. They always encourage me to pursue my goals and really appreciate this freedom during my growth.

Finally, I want to thank my friends for their company during the years at MIT. Many thanks to Huifeng Du, who is also a PhD student at the Mechanical Engineering department at MIT, for his friendship since undergraduate. I also want to thank my partner Siyu for meeting me at the best time, for her accompany and care, and for being the person to share my feelings with.

5 [This page is intentionally left blank.]

6 Table of Contents

Chapter 1. Introduction ...... 13 1.1 Photoconductivity in semiconductors ...... 13 1.1.1 Photoconductivity process ...... 13 1.1.2 Recombination processes...... 17 1.1.3 Recombination models...... 19 1.2 Cadmium sulfide as a photoconductor...... 21 1.2.1 Thin film deposition techniques...... 22 1.2.2 Optoelectronic applications ...... 24 1.3 Persistent photo-effects in semiconductors...... 25 1.3.1 Observation of persistent photoconductivity ...... 25 1.3.2 Theory of persistent photoconductivity ...... 26 1.4 M otivation and main findings for this study...... 29 Chapter 2. Mechanism of Large and Persistent Photoconductivity in CdS...... 33 2.1 Introduction ...... 33 2.2 Experimental Section...... 35 2.2.1 Synthesis of CdS thin films...... 35 2.2.2 Structural and morphology characterizations ...... 36 2.2.3 Optical measurements...... 37 2.2.4 Compositional measurements ...... 38 2.2.5 Electrical measurements ...... 38 2.3 Results...... 40 2.3.1 CdS structure and morphology ...... 40 2.3.2 Basic optical and electrical measurements ...... 43 2.3.3 The relationship between photoconductivity and synthesis parameters...... 44 2.3.4 M odeling photoconductivity ...... 52 2.4 Discussion...... 63 2.4.1 Modeling large and persistent photoconductivity in CdS...... 63 2.4.2 Energy level diagram ...... 64 2.5 Conclusion ...... 66 2.6 Appendix...... 70 2.6.1 Conditions for transient maxim um photoconductivity ...... 70 2.6.2 Analyzing numerical simulations of the standard model...... 72

7 Chapter 3. Two-Terminal Resistive Switches Based on Defect-Level Switching in CdS74 3.1 Introduction...... 74 3.1.1 Two-term inal electronic devices...... 74 3.1.2 Point defects in sem iconductors...... 75 3.1.3 Defect-level switching device concept ...... 76 3.2 Experim ental m ethods ...... 77 3.2.1 Device fabrication...... 77 3.2.2 Device characterizations ...... 78 3.2.3 Sim ulation methods ...... 78 3.3 Results...... 80 3.3.1 Device concept...... 80 3.3.2 Current-voltage characteristics ...... 81 3.3.3 Switching m echanism ...... 82 3.3.4 Pulsed voltage m easurem ents ...... 91 3.3.5 Device endurance...... 92 3.4 Discussion...... 93 3.5 Conclusion ...... 94 Chapter 4. Effect of Contact Layer Photoconductivity on Solar Cell Performance ..... 95 4.1 Introduction...... 95 4.1.1 The role of carrier-selective contacts in solar cells...... 95 4.1.2 Photoconductivity in carrier-selective contact layer materials ...... 95 4.1.3 Contribution of CSC photoconductivity to light soaking-effects in PV..... 97 4.2 M ethods...... 98 4.2.1 Solar cell m odeling ...... 98 4.2.2 Surveying CSC layer materials...... 99 4.2.3. Simulation for non-contact measurements of photoconductivity ...... 101 4.3 Results...... 103 4.3.1 The effect of CSC photoconductivity on CIGS and CdTe solar cell performance ...... 103 4.3.2 Photoconductive response of common n-type CSC materials...... 108 4.4 Discussion...... 111 4.4.1 Opportunities for solar cell engineering: CSC photodoping and material processing ...... 111 4.4.2 M easurem ent challenges...... 112 4.5 Conclusion ...... 114

8 Chapter 5. Sum mary and Outlook ...... 115 5.1 Sum m ary ...... 115 5.2 Outlook ...... 117 References ...... 119

9 List of Figures

Fig. 1.1. Band diagram showing different photoconductivity transition processes..... 14 Fig. 1.2. Energy band diagram showing the Shockley-Read-Hall trap-assisted recom bination process...... 20 Fig. 2.1. Illustration of lattice distortions around VS in the ground state and non- equilibrium state...... 35 Fig. 2.2. I-V curves of the electrometer measured with different resistances on a log-log scale...... 39 Fig. 2.3. Typical S-shaped growth characteristics for CdS thin films deposited at different tem peratures...... 42 Fig. 2.4. Structural characterization of CdS thin films...... 43 Fig. 2.5. Band gap determined from optical measurements and its relationship with film thickness...... 44 Fig. 2.6. Relationship between PPC and deposition time...... 46 Fig. 2.7. Variation of photoconductivity sensitivity and decay rate with CBD bath chemistry using AM1.5 illumination ...... 47 Fig. 2.8. Relationship between PPC and various deposition parameters...... 50 Fig 2.9. Relationship between photosensitivity and Cd/S ratio in CdS films...... 51 Fig. 2.10. Fine-tuning photocondcutivity by varying the parameters of the chemical bath to achieve the strongest PPC effect...... 52 Fig. 2.11 Photoconductivity excitation spectra for a representative sample with large photoresponse...... 54 Fig. 2.12. Analysis of photoconductivity decay for a representative sample with large photoresponse...... 56 Fig. 2.13. The power-law relationship between the steady-state photoconductivity and light intensity measured at different temperatures for a representative sample with large photoresponse...... 59 Fig. 2.14. Evidence for thermally-activated recombination in the temperature-dependence of photoconductivity for a representative sample with large photoresponse...... 63 Figure 2.15: Energy transition level diagrams...... 66 Fig. 2.16. Comparison of the structure changes due to redox reactions at intrinsic defects in ZnO as calculated by Lany and Zunger, and in CdS as calculated by Nishidate et al. 68 Fig. 2.17. Rates of trapping, emission, and recombination at low and high temperatures...... 7 3 Fig. 2.18. Ratio of concentrations of trapped and free electrons during photoconductivity d ecay ...... 73 Fig. 3.1. Two-terminal resistive switching device design concept...... 81 Fig. 3.2. Measured I-V and R-V characteristics of the resistive switching device...... 82 Fig. 3.3. Tests to rule out conductive filaments based switching mechanisms...... 83 Fig. 3.4. Size-dependence of equilibrium-ON device test results...... 85 Fig. 3.5. Testing the DLS hypothesis by advanced characterization...... 87 Fig. 3.6. Simulation results on the switching mechanism...... 89 Fig. 3.7. Simulation results on the switching mechanism with lower electron affinity of the hole injection layer...... 91

10 Fig. 3.8. Pulsed voltage measurements of the resistive switching device...... 92 Fig. 3.9. Endurance test for device...... 93 Fig. 4.1. Simulation results on the effect of photoconductivity in CdS layer on CIGS solar cell performance under AM1.5 illumination...... 105 Fig. 4.2. Simulation results on the effect of photoconductivity in CdS layer on CdTe solar cell perform ance...... 106 Fig. 4.3. Simulation results for CIGS and CdTe solar cells with absorber/buffer interface defects at a concentration N = 101 cm 2 ...... 108 Fig. 4.4. Estimated steady-state carrier concentrations in the dark and under illumination for various and common CSC materials...... 110 Fig. 4.5. Microwave photoconductivity experiment design...... 113

11 List of Tables

Table 2.1. Data table for samples with varying cadmium concentration...... 57 Table 2.2. Data table for samples with varying thiourea concentration...... 57 Table 2.3. Data table for samples with varying ammonia concentration...... 57 Table 3.1. M aterial parameters for simulation...... 79 Table 4.1. Baseline parameters for simulating CIGS and CdTe solar cells...... 100 Table 4.2. Parameters for the simulation of dielectric function of CIGS...... 102 Table 4.3. Summary of photoconductivity ranges for common CSC materials. Results summarized for different illumination conditions: A = AM1.5, B= Hg lamp, C = UV lamp, D = Xe lamp, E = UV LED, F = UV laser, H = Halogen lamp, I = Nd-YAG laser, J= Red LED, K = White LED. The * label indicates filtered or monochromatic UV...... 109

12 Chapter 1. Introduction

1.1 Photoconductivity in semiconductors

Photoconductivity is the phenomenon where electrical conductivity changes as a result of

photoexcitation of new charge carriers. It was first observed in 1873 in selenium leading

to the creation of selenium photocell [1]. Photogeneration of carriers requires the energy of incident photons to be greater than the band gap for intrinsic semiconductors. Electrons and holes transport to opposite electrodes under external electric field and form a current.

Photogenerated carrier density and carrier mobility determine the magnitude of photoconductivity. How long photoconductivity can last after photoexcitation is terminated is affected by many factors, such as carrier lifetimes and trap properties. Therefore, the actual dynamics of various photoconductive systems might be different and complex. We introduce the basic physics of photoconductivity in the following four subsections.

1.1.1 Photoconductivity process

Photoconductivity processes are most conveniently shown in a band diagram, as shown in

Fig. 1.1. The band diagram is electron energy plotted against position. Free electrons in the conduction band and free holes (missing electrons) in the valence band are usually considered to be contributing to electrical conductivity. The conduction band and valence band are separated by the band gap, where only localized states, such as defects or impurities, exist. The basic photoconductivity process involves several components: light absorption, generation of electron-hole pairs, capture and recombination of free carriers, thermal excitation of trapped carriers, and carrier transport.

13 -U p

1)(d)

(a) (c) (e)L(0

Fig. 1.1. Band diagram showing different photoconductivity transition processes. (a) intrinsic band-to-band excitation, (b) and (c) extrinsic excitation from imperfections, (d) and (e) capture and recombination, (f) trapping and detrapping. Arrows indicate direction of electron transitions.

Optical absorption leads to the generation of free photocarriers. Two types of photoexcitation exist: intrinsic and extrinsic excitation. Intrinsic photoexcitation is band- to-band excitation where electrons in the valence band are excited to the conduction band, leaving holes in the valence band (Fig. 1.1(a)). The number of electrons and holes generated in this process are the same. Extrinsic process involves excitation from electron- occupied defect levels or impurities within the band gap to the conduction band, where only electrons are generated in the process as shown in Fig. 1.1(b). It could also be excitation of electrons from valence band to the imperfection levels, in which case only holes are generated in the valence band as in Fig. 1.1(c). The density of photogenerated carriers is quantitatively related to the absorption coefficient a. A larger absorption coefficient leads to more photons being absorbed, such that more carriers are generated.

For intrinsic process, a could be calculated from quantum mechanical perturbation theory using Fermi's golden rule. For extrinsic process, a could be expressed as SortN, where

Sopt is the photon capture cross section, and N is the density of imperfections involved in the photoexcitation process [2].

14 Free electrons and holes can be captured by unoccupied or occupied imperfections,

respectively (Fig. 1.1(d) and (e)). It is quantitatively related to the capture coefficient f,

and proportional to the free carrier concentration n and defect density N. For electrons, the

capture rate R = fnN. The capture coefficient is closely dependent on the properties of the

imperfections, such as the charge states and mechanism of energy dissipation during

recombination. It is usually expressed in the form f = Sv, where S is the capture cross

section and v is the thermal velocity of the free carrier. Two possibilities may happen to a

captured electron: it either recombines with a free hole, or is thermally reexcited to the

nearest band, which is also known as detrapping (Fig. 1.1(f)). Whether an imperfection is a recombination center or a trap is determined by its nature as well as external conditions such as light intensity and temperature.

Another convention to express capture rate is through carrier lifetime. Considerone-carrier effects and electrons being the dominant carrier, carrier lifetime r, = when flnN recombination through imperfections dominates. The recombination rate thus can also be

an written as R = -, where An is the density of photoexcited electrons. Under steady state, Tn recombination rate equals generation rate G , therefore An = GT. In highly photoconductive systems, An » no, so photogenerated carrier concentration can be treated as effectively the carrier concentration in the system.

The last component is carrier transport. This involves both carriers in the materials and contact metals. The contact metals can be ohmic, blocking or injecting. Ohmic contacts can supply carriers freely from one side when carriers depart from the opposite side.

Blocking and injecting contacts creates a charged region in the material as the entering rate

15 and departure rate of carriers are different. Photogenerated carriers contribute to conductivity through drift and diffusion. Consider n-type materials where electrons dominate. Drift current is proportional to the electric field across the material through

Jdrift = nE where an = neyn . Here, e is the elemental charge, and pn is electron mobility. Diffusion current is proportional to the gradient of electron density through

= . The total contribution is Jdiff = eDVn, where Dn is the diffusion coefficient Dn e the sum of drift and diffusion currents.

Photoconductivity can be affected by several parameters, including carrier density, carrier lifetime and mobility. To see this, we can write photoconductivity as

Uph = O+AU

= (no +An)e(pO +Ap)

Combined with ao = noepo, we have

Au = epoAn + neAp (1.1)

= epOGrn + nept (1.2)

From Eqn. 1.1-1.2, we identify three quantities affecting photoconductivity.

1. Carrier density. When carrier lifetime and mobility is constant, photoconductivity

is linearly proportional to the photoexcited carrier density. This occurs when the

occupation of the imperfections involved is not affected by the capture process.

2. Carrier lifetime. When carrier lifetime depends on illumination intensity,

photoconductivity may not exhibit a linear relationship with generation rate. If

unoccupied imperfections behave like recombination centers, then as more

16 electrons are excited from these sites with stronger light intensity, capture

probability increases and carrier lifetime decreases, ocGY-1 with y< 1. If this

is the case, then Au oc Gy. Similary, capture probability could decrease with light

intensity, leading to y > 1.

3. Carrier mobility. Mobility affect the second term in Eqn. 1.1-1.2, and usually its

effect is small compared with the previous two. Mobility could be increased by

decreased scattering, reduced barrier height, and carrier transferred to a band with

higher mobility.

Traps are ubiquitous in photoconductive materials and are important in determining the characteristics of the photoconductivity. The most significant one is that they decrease the speed of response. Trapping and detrapping are time-dependent phenomenon involving filling and emptying traps. The characteristic decay response timwhentrapdensity is low is just the carrier lifetime. However, when trap density is high compared with photoexcited carrier concentration, decay response time is approximately proportional to trap density.

The second effect of traps is that they decrease carrier mobility as they trap carrier during transport, so that the effective drift velocity is reduced. Traps can also reduce photoconductivity, as a fraction of free electrons which would have been in the conduction band when no trap exists are now trapped and become immobile [3].

1.1.2 Recombination processes

Photoexcited free electrons and holes can recombine and emit energy. Recombination is an essential process which determines the decay dynamics and various characteristics of photoconductivity. There are three fundamental processes to relax the emitted energy due

17 to recombination: radiative recombination, nonradiative recombination and Auger recombination [3].

Radiative recombination involves the recombination of free electrons and holes with the emission of photons, which is commonly referred to as luminescence. This occurs more likely in direct band gap materials than in indirect band gap materials. There are three main ways of radiative recombination involving imperfections: (1) recombination of a free electron (hole) with a localized hole (electron); (2) recombination of localized electron and hole when they are close enough in space such that their wave functions interact with each other; (3) recombination of electrons and holes in the excited state and ground state of an imperfection with incompletely filled atomic levels.

Nonradiative recombination is a process releasing phonons instead of photons to dissipate the excess energy. For radiative recombination, 1 eV of energy is typically released within one photon. However, for phonon emission, 1 eV of energy requires 20 to 30 phonons. It is not very likely that such many phonons can be emitted simultaneously, and they are usually released in a sequential way. When the recombination centers exhibit attractive

Coulomb force to electrons, the capture probability increases. Electrons can be captured by some excited states of the defects and sequentially relax down to the ground states. Phonons are emitted during this process, only one or a few in a step. When the defect repels electrons, capture probability could be much smaller. This usually requires extra thermal activation or tunneling to overcome the potential barrier surrounding the imperfection.

The third fundamental recombination process is the Auger process, where the released energy is used to raise the energy of other carriers, most likely a free majority carrier. After the secondary excitation, the carriers may relax by the emission of phonons.

18 The existence of two types of imperfections could give rise to many interesting phenomena.

An imperfection with a large capture coefficient is called a recombination center, and one with a small capture coefficient is called a sensitizing center. Sensitizing centers can increase the photosensitivity of the photoconductors. If defects in a semiconductor have small capture cross section for electrons but large capture cross section for holes, then photoexcited holes will be quickly captures, while the lifetime of photoexcited electrons will be increased [4]. Another phenomenon caused by sensitizing centers is thermal quenching of photoconductivity, where photoconductivity decreases with increasing temperature [5,6]. Holes captured by sensitization centers are no longer stable under elevated temperatures. They are released and more likely to be captured by the electron traps, removing the sensitization effects. A related effect is optical quenching of photoconductivity, where photons are used as energy sources to release holes from sensitization centers. With a secondary illumination of shorter wavelength, photoexcitation may reduce the photo-quenching effect. But as the secondary wavelength increases into infra-red region, optical quenching effect could be quite significant [7-9].

1.1.3 Recombination models

Recombination in direct band gap semiconductors without defect levels is relatively simple.

Free holes and electrons at the same position in k-space and real space find each other and release energy. The recombination rate is proportional to the concentration of electrons and holes, such that the recombination rate R = fnp, where 8 is the recombination coefficient.

This is called band-to-band recombination. The situation becomes more complex with the existence of defect levels, and the most common defect-assisted recombination model the

19 Shockley-Read-Hall (SRH) model [10,11]. A schematic of the SRH process is displayed in Fig. 1.2.

Et E,

Fig. 1.2. Energy band diagram showing the Shockley-Read-Hall trap-assisted recombination process. A recombination center with energy level Et below conductionband is shown in the band gap.

Defect-assisted recombinationoccurs when a hole falls into a trap withinthe band gap. The trap cannot accept another hole, but a free electron can be subsequently captured by the hole-occupied trap, completing the recombination process. It could also be that a hole is captured by an electron-occupiedtrap. Unlike band-to-band recombination, trap-assisted recombination can absorb the differences in momentum between the carriers. This mechanism is dominant in indirectband gap materials, such as silicon,as well as materials with a high density of traps. The net recombination rate for this process can be derived to be

USRH = R - G = n E E NtvthS p + n + 2ni cosh 'k t

20 Where ni is the intrinsic carrier concentration, Et is the trap energy level, Ei is intrinsic

Fermi level, Nt is trap density, Vth is thermal velocity of carriers, and S is carrier capture

cross section of the traps [12]. It is possible to include a number of different recombination

centers within a material, and solve for the recombination rate based on the SRH approach using a computer program.

Direct recombination usually exhibits capture cross section many orders of magnitude

smaller than trap-assisted recombination, and is only important at high carrier concentrations. The capture cross section of direct recombination is on the same order of magnitude as that of a recombination center with coulomb repulsion, and much smaller than that of one with coulomb attraction or a neutral center [2].

1.2 Cadmium sulfide as a photoconductor

Cadmium sulfide (CdS) is a widely studied material in investigating the interaction between light and semiconductors. There are two crystal structures exist for CdS: zinc blende (cubic) and wurtzite (hexagonal). The hexagonal phase is more stable under higher temperatures, and cubic phase can convert to the hexagonal phase under elevated temperatures [13]. The properties of the two phases are similar, and both are direct semiconductors with a band gap of 2.42 eV [14]. Two most studied effects in CdS are photoconductivity and photovoltaics (PV), with both single crystals and thin films deposited by various methods. These applications involve CdS being used as the sole material or used together with other semiconductors to form junctions. We introduce previous studies on this material and its applications in optoelectronic in the following sections.

21 1.2.1 Thin film deposition techniques

There are several techniques that could be used to deposit CdS thin films, and can be categorized into physical deposition and chemical deposition.

Physical deposition does not involve any chemical reactions, and typically uses CdS as the source material. Vacuum evaporation are widely used for CdS deposition where CdS is heated and vaporized and subsequently deposited onto a suitable substrate [15,16].

Another physical deposition technique is sputtering. Sputtering is the use of inert gas ions to bombard a material target such that particles of the material are ejected from the surface.

Sputtered CdS films are continuous and dense, and sputtered CdS/CdTe solar cell can achieve efficiency above 14% [17-20]. Screen printing can also be used to produce sintered films. It involves mixing CdS power with some organic binder to form a paste, followed by subsequent sintering [21-23].

Chemical deposition techniques are probably more prevalent in applications. Spray pyrolysis is a process where precursors are sprayed onto a hot substrate in liquid droplet to form desired films. For CdS, the precursors are typically CdCl2for Cd and thiourea (TU) for S [24,25]. Another technique called chemical vapor deposition (CVD) involves transporting organic precursors in an inert carrier gas, and let the precursors react on the substrate to form a thin film. Tuning deposition parameters such as deposition position, temperature and carrier gas flux can result in different nanostructures and morphologies [26,27]. The third commonly used chemical method is electrodeposition.

This method exploits the electroreduction of a Cd2+ and NaS2O3 at the substrate in an aqueous solution. Perhaps the most frequently used growth technique for PV applications in lab-scale studies is chemical bath deposition (CBD). Similar to the CVD process, CBD

22 involves the reaction of Cd and S precursors in a solution instead of a vapor phase. It is

suitable for deposition of highly insoluble compound at relatively low temperatures. The

advantages of CBD include large and uniform production of thin films, good

reproducibility, as well as no wastage on Cd sources [28]. There are two phases of

deposition in a chemical bath, the ion-by-ion mechanism and cluster-by-cluster

mechanism [29]. In the ion-by-ion phase, nuclei form on the substrate, and precursor ions

precipitate and grow on the nuclei. The films grown in this period is uniform and adheres

well onto the substrate. In the cluster-by-cluster mechanism, colloids agglomerate to form

large particles on the film. There is always competition between these two mechanisms,

and the dominant one is determined by thermodynamics and kinetics of the nucleation

process, as well as the catalytic behavior of the substrate [30].

The most typical chemical bath for CdS deposition contains Cd ions, TU and ammonium

hydroxide [29,31,32]. The exact deposition mechanism is still controversial, but a simple

one can provide insight into understanding the process. The mechanism involves the

following steps [33]:

2 1. Cd ++ 4NH3 <-* Cd(NH 3)42+

2. SC(NH2) 2 + OH- *- CH2N2 + H20+SH-

2 3. SH-+ OH- <- S -+ H 2 0

4. S2 -+ Cd(NH3)42++-* CdS + 4NH3

Here, TU is the S source, and ammonium hydroxide serves as a complexing agent to reduce the reaction rate. Without the complexing agent, the precipitation will be instantaneous and no thin films will be obtained. Cd2+ ions are slowly released from the complex compound

23 and S- ions are slowly released through the hydrolysis of TU. Maintaining a low concentration of free ions in the chemical bath, a CdS thin film can grow onto the substrate.

Other proposed mechanisms are more complex and involve many complicated chemical equilibria and intermediate products. For example, studies have shown that Cd(OH) 2 could be an important intermediate products during the growth of CdS [34]. However, in this thesis we assume the simpler mechanism, which is sufficient to gain understanding of the mechanism of photoconductivity in CdS.

1.2.2 Optoelectronic applications

CdS as a photoconductor has been widely studied. The photoconductivity characteristics introduced in the previous section, including imperfection sensitization, thermal and optical quenching of photoconductivity have all been observed in CdS [2,7,9,35,36]. Other than these properties, CdS also exhibit other unusual effects, such as decrease in photoconductivity with time, dependence of thermally stimulated conductivity on illumination history, and persistent photoconductivity (PPC), where photoconductivity lasts for a long time after photoexcitation is terminated [13]. Early studies attributed some of these light-induced effects to photochemical changes in CdS, which means the creation of certain defects as a result of photoexcitation [37-39].

Due to its excellent photosensitivity, CdS has been commercially used as a photoresistor for light sensors and flame detectors. However, a widespread application is the buffer layer in Cu(In,Ga)Se (CIGS) and CdTe thin film PV. PV is a photoelectric effect where light- induced electrons and holes are separated and transported to the opposite electrodes before recombination occurs to convert solar energy into electricity. Many thin film PV

24 technologies incorporate a large band gap n-type buffuer layer in contact with a smaller

band gap p-type absorber material, such as CIGS/CdS and CdTe/CdS [40-42]. The buffer

layer helps selectively conduct one carrier type over the other. One common way to achieve

this is to create a depletion region at the junction in order to separate the carriers. It also

has a large enough band gap such that light is admitted through the layer with little

absorption. At the same time, buffer layers should add minimal electrical resistance to the

whole stack. Therefore, the band gap of the buffer layer should be large while the thickness

should be small. CdS serves as one of the best buffer layer materials due to its excellent

properties to meet these requirements as well as various deposition techniques to optimize

its characteristics in a solar cell. Although there are more and more investigations on buffer

materials to replace Cd due to its environmental risk, CdS remains the oldest and most well

studied buffer layer. Although CBD is less commonly used than physical vapor deposition

(PVD) in the PV industry, commercial CIGS solar cells use CdS as the buffer layer deposited via CBD can also achieve high efficiency [43,44]. The device performance might be superior to the counterparts deposited by PVD partly due to improvements on the surface chemistry of the absorber and protection on the sensitive interface [45].

1.3 Persistent photo-effects in semiconductors

1.3.1 Observation of persistent photoconductivity

Persistent photoconductivity (PPC), the phenomenon where photoconductivity response of the semiconductor can persist for a long time after illumination is turned off, is associated with crystalline defects. It has been observed in many semiconductors, including Si, III-

Vs, oxides and chalcogenides [46-54]. This important property has implications in a number of optoelectronic applications, including solar cell [55,56], rewritable

25 optoelectronic devices [57], and photodetectors [49]. Among these materials, CdS has been widely recognized as an excellent photoconductor with large photoresponse. Nair synthesized CdS films using triethanolamine (TEA) as the complexing agent and thiourea as the sulfur source. The reported photosensitivity, characterized by light-to-dark-current ratio, reaches 106 under AM2 illumination for freshly prepared samples and 107 after these samples have been stored for several hours at 50 °C [58]. The reason for this increase remains undetermined, but is suggested to be due to chemisorbed oxygen at the surface and grain boundaries serving as additional recombination centers, which is consistent with

Bube's findings [59]. The same group also achieved films with about 109 photosensitivity under AM1 light and a decay time of 13 h/decade using the same synthesis method above, but slightly varying the ratio between cadmium and sulfur source [60]. Despite the amount of the work on PPC effect in CdS, it has never been systematically studied to confirm the source of PPC or the real underlying mechanism.

1.3.2 Theory of persistent photoconductivity

The mechanism of PPC in different materials could be very diverse, depending on the material and defect properties. The basic picture is that photogenerated minority carriers are trapped at crystal defects, so that transport is mainly due to majority carriers with prolonged lifetime. The recombination centers, usually having small capture cross sections for majority carriers and large cross section for minority carriers, slowly capture majority carriers so that conductivity decays with time. Revealing the true mechanism of PPC in a material is difficult, and even for the same material, PPC dynamics may be dissimilar due to different material processing history. Existing methods usually involves DFT

26 calculations of defect properties and modeling PPC decay dynamics with various defects to fit experimental data.

One of the most studied persistent photoconductors is AlGaAs. In earlier studies, Lang proposed a large-lattice-relaxation model which successfully explains the large Stokes shift and small, thermally activated, electron capture cross section that characterize PPC in

AlGaAs [61]. An analytical expression was derived by Queisser showing that PPC decay in n-type GaAs and AlGaAs is logarithmic and achieved good agreement with their experimental results [62]. Light-induced persistent metastable states are also observed in a-Si. In a-Si the metastability arises in part from photo-breaking of weak Si-Si bonds, creating dangling bonds that are deep traps [63-65]. In a-Si there is also evidence that hydrogen diffusion stabilizes light-induced metastable lattice configurations, enhancing persistent photoconductivity [66,671. PPC effect is also observed in SrTiO3. Sub-bandgap excitation and slow decay are attributed to photoexcitaion of electrons from Ti vacancies into the conduction band with very small recapture rate. Other materials showing significant photoconductivity include TiO2. The origin of large and persistent photoconductivity in these materials is inconclusive. TiO2 is known to feature large electron-phonon interactions, and minority holes may be trapped as small polarons [68,69].

Grain boundaries or oxygen-impurity defect pairs may also be responsible [70,71]. The salient point is that, in all of these materials, photoconductivity can be large and persistent due to photo-induced lattice distortion leading to metastable atomic configurations.

II-VI semiconductors feature a great amount of theoretical and experimental studies on

PPC. Early studies found different mechanisms leading to PPC in II-VI semiconductors.

Iseler et al. found that some impurity donor levels in CdTe are associated with the

27 conduction band minima above the lowest minimum at the F point, and there is a potential barrier to transfer electrons between the non-f levels and the F conduction band minimum, leading to a slow rate of electron transfer [72]. Another study observed that the storage of conductivity in CdS is accompanied by local deformation around the defect centers [73].

Modem theoretical proposals by Zhang, Wei, Lany and Zunger indicate that photo-induced lattice distortions at anion vacancies in II-VI and chalcopyrites are responsible for large and persistent photoconductivity [74,75]. Defect-localized states (DLS) can be categorized into two classes: a-type and p-type. a-type behavior refers to the situation where DLS is within the bandgap, and p-type behavior is shown when DLS is in resonance with the conduction-band. In ZnO, oxygen vacancy can be of different charge states. The neutral state, V , would result in an inward relaxation of nearest-neighbor Zn atoms, exhibiting a- type behavior. However, the charged state, V , leads to an outward relaxation of Zn neighbors. The vacancies become resonant in the conduction band and create a perturbed- host state (PHS) below the DLS, showing p-type behavior. Photoexcitation of V ground state to V2+ causes the DLS to move upward into the conduction band and creates photoelectrons shallowly bound in the PHS. Since the a-type behavior is non-conducting and -type is conducting, this transition results in an electron photoconductivity. However, relaxing from p-type to a-type requires thermal activation of electrons in the PHS to overcome a barrier between the PHS and the empty DLS in the conduction band, leading to a slower decay and the persistent effect. This is a theoretical result, but has never been experimentally verified. Due to the fact that CdS is categorized as aII-VI semiconductor and the similarities between CdS and ZnO, we want to provide experimental support for the theoretical results proposed above.

28 1.4 Motivation and main findings for this study

A lot of studies have been devoted to PPC in CdS and various other photoconductivity related phenomena, including thermal and optical quenching of PPC [7,8,58,73]. However, the connections between synthesis conditions, composition and morphology with PPC is still not clear. PPC is known to be associated with crystal defects, but the species of defects leading to the effect in CdS is also controversial. The negative-U effect theory due to

ZWLZ is a plausible one, but lacks experimental support [74,75]. Therefore, we would like to systematically study the defect species related to PPC in CdS and experimentally verify the ZWLZ theory. We want to model the dynamics of carrier processes involved in PPC and reveal any necessary conditions required to observe this phenomenon.

Understanding the physics related to the PPC processes can boost its application potential.

As observed, CdS is very sensitive to light, with huge resistance change over many orders of magnitude. One intuitive application of this characteristic is photodetection. However, it can be of great interest if this photoconductivity mechanism is applied to all-electrical resistive switches. Resistive switching devices has great potential in neuromorphic computing, memory technologies, and reconfigurable devices [76-78]. There are multiple mechanisms which can lead to the design of resistive switching devices, and most commonly are based on the formation of conductive filaments or phase change [77,79].

Light controlled resistive switches have also been demonstrated, but they are not suitable as all-electrically controlled devices in most applications [80,81]. Combined with our knowledge in the defect physics associated with photoconductivity in CdS, we can design an electrical resistive switching device by controlling the defects responsible for

29 photoconductivity. This would be a brand new route for designing resistive switches for various applications, without any mass transport of materials or collective phenomena.

Another application of PPC is in solar cells. CdS has been widely used as a window/buffer layer in CIGS or CdTe thin film photovoltaics. However, its properties as an excellent buffer layer is not correlated with its huge photosensitivity. In fact, this could be a point which has been neglected by people for years. Carrier selective contact (CSC) layers exist in all heterojunction solar cells. It is responsible for collecting charge carriers of one type, repelling the opposite type, and suppressing non-radiative recombination at the interface.

Although there are existing design rules for CSC layers, the role of photoconductivity has rarely been recognized because most CSC layers are depleted during operation [40,82].

However, if CSC materials are highly photosensitive, as it turns out to be true for many

CSC materials, they can change the band structure of the heterojunction under illumination.

This effect can lead to better carrier selection and reduced recombination, and therefore better solar cell performance.

In this thesis, we first systematically study the effect of synthesis conditions on CdS leading to a testable hypothesis that sulfur vacancies are the source of PPC in CdS. Use this understanding, we can tune photoconductivity of CdS films over nine orders of magnitude and vary the photoconductivity decay time constant from seconds to 104 seconds. We perform detailed analysis and modelling to test the negative-U behavior proposed by

ZWLZ, and identify relevant carrier processes during lattice relaxation. We also perform experiments to identify the defect transition energies involved in CdS and propose an energy level diagram for dark and illuminated cases. We subsequently suggest a screening method to identify other materials with long-lived, non-equilibrium, photoexcited states

30 based on the results of ground-state calculations of atomic rearrangements following defect redox reactions. This work has been published in the journal Physical Review Materials

Understanding the mechanism of PPC in CdS allows us to design electrical resistive switching devices. Photoconductivity is CdS is essentially the switching between deep donors (neutral sulfur vacancies) and shallow donors (doubly charged sulfur vacancies).

We design a switching device by introducingMoO3as a hole injection layer to trigger the donor level switching effect. We hypothesize thatMoO3 extract electrons from the deep donor levels in CdS due to its high electron affinity and nearly type-III band alignment with CdS, and thereby switches CdS into a "photoconductive" state in a thin layer near the interface. We test the hypothesis by providing experimental evidence that conductive filaments are not formed duringswitching events, and perform device modeling to confirm and reveal the key conditions required for switching effect. The defect-level switching device provides a new mechanism to design resistance switching elements.

To test the effect of photoconductivity of CSC layers in thin film solar cells, we conduct numerical modeling to estimate the effect of CSC photoconductivity on the performance of CdTe and CIGS solar cells. We model conventional device architectures and use conventional material parameters for all but the n-type CSC layers. We simulate CSC large and PPC by switching a population of defect levels between deep and shallow donor states, in accordance with our understanding of the PPC mechanism in CdS. We find that CSC photoconductivity can substantially affect solar cell performance, and greatly enhance power conversion efficiency by over 4% in both CIGS and CdTe solar cells. The primary cause is the influence of CSC shallow donor concentration on the junction depletion region.

31 These results suggest that better solar cells can be engineered by optimizing CSC photoconductivity. The generality of the effect implies that similar approaches may be successful across multiple thin-film and wafer-based solar cell platforms. We propose a non-contact method to measure photoconductivity of the CSC layers in a thin film stack, and experimental optimization for solar cell performance based on CSC layer photoconductivity.

32 Chapter 2. Mechanism of Large and Persistent

Photoconductivity in CdS

2.1 Introduction

Large and persistent photoconductivity (LPPC) is the phenomenon where photoconductive

response of a semiconductor is enormous and can persist for many hours after illumination

is turned off. LPPC in CdS has long been reported and is relevant to the operation of thin-

film solar cells that use CdS as an n-type layer [8,40,73]. Here, we show that chemical bath

deposition (CBD) can be used to tune the photoconductivity of CdS films over nine orders

of magnitude, and vary the photoconductivity decay time from seconds to 104 seconds. We

vary the activities of Cd2and S2 ions in the chemical bath and demonstrate that LPPC

results from sulfur deficiency in CdS. We provide experimental support for the theoretical

result that LPPC is caused by so-called negative-U sulfur vacancies, at which charge-lattice

coupling results in an effective attractive force between positively-charged holes. Due to this lattice distortion, recombination becomes thermally-activated, slowing the return to

equilibrium. This is conceptually related to the "DX-center" lattice distortion responsible for dopant de-activation in AlGaAs, and to long photo-excitation lifetimes observed in

systems with polaron excited states such as organic semiconductors and halide perovskites [84-88]. The idea that atoms rearrange to create physical separation between photo-excited charge carriers, thus slowing the decay to equilibrium, is also reminiscent of biological light absorbers such as photosystem II and dye molecules used for solar energy conversion [89,90].

33 Photoconductivity in large band gap II-VI materials is due to majority-carrier transport while the photo-generated minority carriers are trapped at crystal defects. Large and persistent photo-effects are associated with atomic lattice distortions around defects in response to changes in electron occupation, for instance at DX centers in AlGaAs, and at wrong-coordinated sites in amorphous Se [91,92]. Zhang, Wei, Lany, and Zunger (ZWLZ) published theoretical proposals that LPPC in II-VI and chalcopyrite semiconductors is due to lattice relaxation at anion vacancies [74,75]. Anion vacancies (VAn) in equilibrium in n-type semiconductors tend to be neutrally-charged, deep donors (e.g. Vo in ZnO). In CdS at equilibrium, the cations surrounding Vs distort inwards, as they are attracted by the two un-bound electrons and no longer repelled by a sulfur ion. Upon photo-excitation, anion vacancies can release two electrons and become doubly-charged ( V"*) . Without subsequent lattice relaxation, the positively-charged, deep donors would quickly capture electrons from the conduction band. However, Vi' sites no longer contain electrons to screen cation-cation interactions: the cations repel each other and distort outwards. We denote this transition to a non-equilibrium, metastable configuration as V' -> (Vn)* + ph, where ph indicates interactions with lattice vibrations. The resulting, distorted lattice presents an activation barrier to re-capture of photoexcited electrons by (V' )*. The distorted lattice may even raise the (V")**-> (V'n)* and (V')* -> (V' )* charge transition levels above the conduction band edge (Ec), showing p-type behavior and further reducing the rate of electron re-capture. This may be the case for ZnO, for which theory finds that the (V' )* -> (V')*transition level is resonant with the conduction band [75].

In Fig. 2.1 we visualize the distortion around Vs in the Vs' and (V*')* charge states, as calculated by Nishidate et al. [93]. These predictions using density functional theory are

34 consistent with the above description of contraction and expansion of Vs-Cd4tetrahedra in response to photo-excitation.

(a) 00 (b) O O 0 0

00 00

OS eCd CQVs

Fig. 2.1. Illustration of lattice distortions around Vs in the Vs' ground state (a) and (Vs')* non- equilibrium state (b). The bond lengths are as calculated by Nishidate et al. [93]. The Vs site is indicated by a grey sphere. The orthographic projection along a Cd-S bond allows easy comparison of the distortion of the Cd4 tetrahedron in the undistorted lattice (bottom-right of each panel) and around the vacancy.

2.2 Experimental Section

2.2.1 Synthesis of CdS thin films

CdS thin films are deposited onto various substrates in a chemical bath. In a typical synthesis procedure, 75 mL DI water is added into a 250 mL beaker, and heated in a water bath to the desired temperature (70 C if not mentioned otherwise). Predetermined amount of cadmium source (cadmium nitrate if not mentioned otherwise) and thiourea (TU) are dissolved in 11 mL deionized water. Substrates (soda lime glass slides if not mentioned otherwise) are sonicated in 1% detergent (Liquinox) and 2% acid neutralizer (Citranox) solutions successively for 5 minutes at 60°C and then fully rinsed with DI water and isopropanol, and dried in air. The cleaned substrates are placed vertically in the chemical bath with the assist of plastic forceps. The prepared Cd source, along with 14 mL 28% ammonium hydroxide, is added into the bath. After 2 minutes, TU is added as well. The

35 reaction is allowed to carry on with stirring (300 rpm if not mentioned otherwise) for some time, and the substrates are taken out of the solution. The deposited films are rinsed with

DI water, and sonicated in isopropanol for 10 minutes, and finally dried in air. The dried film is then removed from one side of the substrate by wiping with a cloth soaked in 10%

HCl.

2.2.2 Structural and morphology characterizations

X-ray diffraction (XRD) patterns are collected on a Rigaku Smartlab diffractometer with

Cu Ka (k = 1.5405A) radiation. Phase and grain size analysis are performed in PANalytical

X'Pert High Score Plus.

Grain size analysis is conducted using LaB6 powders from NIST as a calibration standard, scanned under the same conditions as the CdS thin films, to determine the instrumental profile. Scherrer's method is used to calculate the grain size:

KA fCos 0 where r is the mean size of the crystallites, K is the Scherrer constant and is set to be 0.94 for spherical particles, A is the wavelength of X-ray, # is the broadening at full width at half maximum (FWHM), and 6 is the Bragg angle. The average of grain sizes calculated from all the available peaks is used.

Thickness and roughness of the CdS thin films are acquired from a Dektak XT profilometer.

Thin films are etched with 10% HCl to create a sharp step for the measurement of thickness.

36 2.2.3 Optical measurements

Transmission and reflection data are measured on a PerkinElmer LAMBDA 1050

UV/Vis/NIR spectrophotometer with an integrating sphere module. The wavelength is scanned from 850 nm to 300 nm with a data interval of 2 nm. Absorption coefficients are obtained using the approximation formula

a= - (In (2.1) d (1 - R)2 where a is the absorption coefficient, d is the thickness of the film, T and R are transmittance and reflectance at a specific wavelength, respectively.

The band gap Eg is related to the absorption coefficient as [94]:

n K(hv - E) a =hv where K is a constant, h is the Planck constant, v is the frequency of the photon, Eg is the band gap, n equals to 1 for direct band gap semiconductors such as CdS. The band gap is determined by plotting (ahv) 2 versus hv, and then extrapolating the fitted straight line to get the intercept with the hv axis. This method is only accurate for photon energies close to Eg, where the effective mass approximation is valid. For CdS, this corresponds to approximately a < 105 cm-1 [95-97]. We therefore only analyze data in this range. This has the effect of limiting band gap analysis to samples thicker than 50 nm according to Eqn.

2.1.

37 2.2.4 Compositional measurements

X-ray fluorescence (XRF) measurements are conducted on a Bruker Tracer III-SD portable

XRF unit. The standard for quantification analysis is prepared by mixing commercially purchased, stoichiometric CdS powder with silicon powder in various weight percentages, and then adhered to glass slides using petroleum jelly (Vaseline) to make thin films of comparable thickness to those deposited by CBD.

Wavelength dispersive spectroscopy (WDS) was performed on a JEOL JXA-8200

Superprobe, and results were analyzed using the GMRFILM software. The standards for

Cd and S are elemental Cd and PbS, respectively.

It is challenging to prepare standards that exactly match the thickness of the films deposited by CBD, and as a result the quantification could suffer from systematic error due to matrix effects. The presence of the silicon signal may also systematically affect quantification.

Therefore, WDS is used to calibrate XRF results. We picked two samples with high and low photosensitivity to perform WDS measurement, and then linearly scaled the XRF results according to these two samples.

2.2.5 Electrical measurements

Coplanar silver-print electrodes are applied to the surface of the thin films with a separation of 24 mm. Resistance data of the films are collected using a Keithley 6517B electrometer with a bias voltage of 100 V. Samples are stored in the dark until the resistance stabilizes, then illuminated for 3 hours until reaching steady state. The illumination sources used for this study are an AM1.5 solar simulator, a cold white LED (Thorlabs Solis 1A), and a 455 nm LED (Thorlabs M455L3). The light is then turned off and the decay of

38 photoconductivity is measured for an extended period. Sample temperature is controlled using a thermal chuck throughout the measurement. The temperature is set to be 25C if not stated otherwise. The process is automated by a LabView program.

CdS thin films are highly resistive, necessitating the use of an electrometer and two-point resistance measurements. In order to assess the impact of non-ohmic contacts on our results, we measured current-voltage (I-V) sweeps for a representative sample over a wide range

(108 orders of magnitude) of resistance. The data (Fig. 2.2) show that the data are linear for current greater than 1 pA, corresponding to resistances up to 100 TO near the measuring voltage of 100 V. The data shown in the text are all below this resistance value. While we certainly expect that the contacts are non-ohmic, and that significant charge may be injected during measurements, this result shows that our results are representative of the photoconductivity of the material without influence from the contacts.

5 10 104

103

102

10~ • Reference s =1

101

10-2

10 100 Voltage (V) Fig. 2.2. I-V curves of the electrometer measured with different resistances on a log-log scale. The curves show good linearity except for very high resistances.

Hall carrier mobility was obtained from the rotating parallel dipole line (PDL) Hall measurement [98]. A 5x5 mm square was cut out from CdS thin film, and indium contacts were pressed onto the corners of the sample. The sample was then wire-bonded to the pins

39 on a sample stage, mounted on a micro manipulator and positioned at the center between a diametric magnet pair. The magnet pair rotates at 3 rpm during the measurement, creating oscillations in B-field. A parameter analyzer was embedded to perform all the tasks, including sourcing current/voltage, measuring voltages, reading magnetic field sensor, and switching through different van der Pauw states. Fourier transform and spectral density analysis were used to ensure that the Hall oscillation is real. The in-phase Hall signal can then be separated from the out-of-phase parasitic components by integrating the transversal resistance over a lock-in time constant:

t RH (t) = RXY(t') sin(wre t') dt'

RHis the in-phase Hall resistance, Rxy is the transversal resistance, T is the lock-in time constant, and Lrefis the angular rotating frequency of the magnet pair. The final Hall resistance was extracted from the stabilized value of the in-phase Hall signal. The carrier mobility can be calculated by the following equation:

RHt

t the thickness of the thin film, B is the magnetic field, and p is the resistivity obtained from resistivity measurement.

2.3 Results

2.3.1 CdS structure and morphology

We make CdS thin films on glass substrates by chemical bath deposition (CBD) [34,99].

The bath contains a cadmium source, a sulfur source such as thiourea (SC(NH2)2), and a

40 complexing agent. One proposed reaction mechanism is as follows, as stated previously [33]:

2+ 1. Cd2++ 4NH 3 +-+ Cd(NH 3)4 (2.2)

2. SC(NH2)2 + OH- +-+ CH2N2 + H20 + SH-

2 3. SH-+ OH- *-- S - + H2 0

4. S2- + Cd(NH 3)42+ *-* CdS + 4NH3

Chemical equilibria in the bath affect the composition of the growing CdS films. Therefore,

CBD offers a convenient way to control the composition of the resulting films by adjusting the bath chemistry.

Our film growth is consistent with an initial ion-by-ion process followed by a cluster-by- cluster process. Growth characteristic of CdS thin films follows an S-shaped curve (Fig.

2.3). The ion-by-ion process, corresponding to the heterogeneous nucleation on the substrate, is relatively slow. Once nucleation finishes, there is an acceleration since the energy required for growth is less than that for nucleation. As the concentration of reactants decreases, the deposition slows down. It is noted that temperature not only has an impact on the growth rate, but also on the final thickness of the films. Films tend to be thinner as temperature rises. This is because higher temperature favors homogeneous nucleation, leading to the formation of large particles in the solution as opposed to deposition onto the substrates.

41 120 *- 60 C 110 - 70C 1e80 C 100 i 90

C 70 U E60

301 0 5 10 15 20 25 30 35 40 45 Time (min) Fig. 2.3. Typical S-shaped growth characteristics for CdS thin films deposited at different temperatures. Samples: 60 °C: CdS_160709_1-4, 70 °C: CdS_160709_5-8,80 °C: CdS_160709_9- 12.

We measure film composition using X-ray fluorescence (XRF) data, which we calibrate using wavelength diffraction spectroscopy (WDS) measurements on select samples. We selected one sample each with high (107) and no photoresponse for high-accuracy WDS measurements, and then we linearly scaled the XRF results according to this two-point calibration. X-ray diffraction (XRD) reveals that the films are polycrystalline with the dominant phase being cubic sphalerite, as shown in Fig. 2.4(a). The broad peak at 20-35° is due to the glass substrates. The peaks show a slight shift of about 0.5° higher than the reference pattern, probably indicating compressive strain. Grain sizes obtained using

Scherrer's method range from 40 to 50 nm, and increase with deposition time.

Transmission electron microscopy (TEM) (Fig. 2.4(b)) shows that the thin film is well- crystalized with no evidence for amorphous regions.

42 (a) -42 nm ---- 77 nm 96 nm 103 nm Sphalerte CdS -Wurtzite CdS {220} c (31 1}

25 35 45 55 20 (degree) Fig. 2.4. Structural characterization of CdS thin films. (a) XRD spectra measured in grazing incidence geometry for CdS films (CdS_160709_5-8) grown at 70 °C with different thickness for a fixed bath chemistry (0.0015 M cadmium nitrate, 0.075 M thiourea and 1.87 M ammonia). Purple lines and labeled peaks correspond to the cubic sphalerite phase (ICDD #96-900-0109), and green lines to the hexagonal wurtzite phase (ICDD #04-004-8895). (b) TEM micrograph of a representative sample (CdS_1607042) shows that it is well-crystallized.

2.3.2 Basic optical and electrical measurements

We used optical reflection and transmission measurements to estimate the band gap (Eg) using Tauc's method. For films thicker than approximately 50 nm we find Eg =2.45 ±0.08 eV. For thinner films, Eg tends to increase and the data is difficult to analyze accurately due to multiple film-substrate reflections, as shown in Fig. 2.5(a) and (b). We measured

Hall mobility pH = 1.57 ± 0.27 cm2 V-1 s- for a typical sample using a rotating parallel dipole line Hall system with the film under white-light illumination [98]. In the analysis that follows, we assume that the drift mobility is equal to pH and is independent of carrier concentration.

43 El

(a) 3.2 (b) • 42 nm y' 2. * 77 nm > ,3.0- • 96 nm !) * 103 nm 22.8 j (9 > 02.6 ol

2.4

2.2 , 0 20 40 60 80 100 120 140 160 180 2.4 2.6 2.8 3.0 Thickness (nm) hv (eV)

Fig. 2.5. Band gap determined from optical measurements and its relationship with film thickness. (a) Band gap vs. sample thickness. (b) Representative Tauc plots used to determine band gap for films of different thickness (CdS_160709_5-8). The figure shows that the band gap ofCdS thin films decreases and approaches the nominal value of 2.42 eV as thickness increases. However, the films are hard to analyze when they become thinner due to multiple film-substrate reflections.

2.3.3 The relationship between photoconductivity and synthesis parameters

We measure photoconductivity in two-point configuration using silver paint contacts on bare CdS films and an electrometer (Keithley 6517B). We estimate carrier concentration

(n) from conductivity (a) using the relation n = ---. For the data reported here, we held ePH samples in the dark until steady-state conductivity was reached, and then switched on illumination for three hours. The measurement probe station is surrounded by a dark enclosure. The illumination sources used for this study are an AMl.5 solar simulator, a cold white LED (Thorlabs Solis IA), and a 455 nm LED (Thorlabs M455L3). We used a thermal chuck to control sample temperature.

The photoconductivity measurements were set to the following routine. CdS thin films were kept in the dark until they reach steady state conductivity, which is shown as the first-

44 -I

hour data in the conductivity curves. The films were then illuminated for 3 hours, and finally were allowed to relax in the dark. Typical conductivity data during the measurement are shown in Fig. 2.6(a). The photoconductivity decay rate was determined by fitting a single exponential function to the decay data (Fig. 2.6(b)). Note that this rate is a simple estimator of a complex process. Here, we study the relationship between photoconductivity and deposition time. The photoconductive response is strongest at some middle stage of the deposition, as shown in Fig. 2.6(c). This phenomenon can be readily related to the changing composition of the chemical bath during deposition, indicating that the chemical equilibria have significant impact on the composition of the resulting films, which in turn affect photoconductivity. We investigate the effect of bath chemistry on photoconductivity more systematically in the next.

0 (a) - mm (160709_5) (b)10 S08 min (160709.5) 101 -- mm (160709.6) 10-1 e•12 min (160709_-6) - 16 min (1607097) ••16 min (160709_7) 10 101 - 20min (160709_S) - e • 20 min (16070908) 10.3

10 10-4

105 C:10-1

t:106 b 106

10-1 io 7

101 101

10 0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 Time (h) Time (h) (C) Photosensitivity -- Decay rate -5 106 -4

0L 2

5 1'0 15 20 Deposition Time (min)

45 Fig. 2.6. Relationship between PPC and deposition time. (a) Photoconductivity time series data for CdS thin films deposited at 70 °C bath for different time.(b) Fitting the decay data to an exponential function. (c) Photosensitivity and decay rate plotted versus deposition time of the films in (a).

To understand and control photoconductivity in CdS thin films, we study the effect of changing the bath chemistry. The bath composition was varied around a baseline chemistry, i.e. 0.0015 M cadmium source, 0.075 M thiourea and 1.87 M ammonia. In Fig. 2.7(a)-(c) we show the effect of varying the concentration of cadmium, thiourea, and ammonia on the phoconductivity magnitude and decay. For this chemical study we characterize photoconductivity by two metrics: photosensitivity and decay rate. Photosensitivity is the ratio of conductivity under illumination to that in the dark, and the decay rate is obtained by fitting a single exponential function to the decay curve. The full decay dynamics are more complex than can be captured by a single time constant (c.f Fig. 2.12(a)), but a simplification suffices here. For high cadmium concentration, films exhibit LPPC. The trend is reversed with thiourea (TU) concentration: for high TU concentration, films are not photosensitive (and therefore the decay rate trends towards zero). Ammonia has a similar effect as TU: higher concentrations of ammonia weaken photoconductivity. The role of ammonia in CBD is twofold: to complex with free Cd 2 ions, and to accelerate the hydrolysis of TU, as illustrated by the following reactions:

2 2+ Cd ++ 4NH3 <-> Cd(NH 3)4 (2.3)

SC(NH2)2 + OH~ +-+ CH2N 2 + H20 + SH~

2 As a result, a higher concentration of NH3 reduces the concentration of free Cd + and increases the concentration of sulfur ions. Therefore, the effect of NH 3 on photoconductivity is consistent with the previous two cases: LPPC is enhanced by cadmium-rich, sulfur-poor bath chemistry.

46 -I

(a) 2Photosensitivity2 0 (b) - -20 107 -e-- Decay Rate (10. 6.1; -166g -1: -0-Photosensitiv' 10° @ Decay rate * ~ ~ .12 -' 4) '103. 8 8

01 0 13 101 14 2 ~0110 0 -0 10-3 10-2 102 10-1 2 c[Cd + (mol/L) c[TU] (mol/L)

IA - 70 (c) -0-Photosensiivty (d) 107 -0-Decay rate 10 %

.- 1051 4 2 .I 100 ,, •pa

2 8 10 0 M0 10 1011 .. 0~~ 10 1 2 3 4 100101O 1O* 101010 c[NH,]J(mol/L) Photosensitivity Fig. 2.7. Variation of photoconductivity sensitivity and decay rate with CBD bath chemistry using AM1.5 illumination. (a) Varying cadmium source concentration. Sample names from left to right: CdS_1606112, CdS_1603032, CdS_1606114, CdS_160611_6. (b) Varying thiourea (TU) concentration. Sample names from left to right: CdS_160611_8, CdS_160611_10, CdS_160303_2, CdS_160611_12. (c) Varying ammonia concentration. Sample names from left to right: CdS_1606132, CdS_1603032, CdS_160611_16, CdS_160611_18. These and other studies point to the ratio of cadmium-to-sulfur activity in the bath as the determining factor for photoconductivity. Note that a decay rate approaching zero can result when the film is not photosensitive, e.g. for high [TU]. (d) Decay rate and sensitivity are negatively correlated for photosensitive films.

Other variables that affect photoconductivity include the concentration of ammonium

(NH4*) ions, the choice of cadmium salt, the bath stirring speed, and the bath temperature.

NH4' ions serve as apH buffer together with NH3. However, the results in Fig. 2.8(a) and

(b) indicate that the addition of ammonium suppresses photoconductivity. The reason is that, although NH4* could slow down the hydrolysis of TU, it also ionizes to generate more

NH3 to cheldte Cd2 ions. Thus the net effect is similar to reducing the concentration of cadmium source. In Fig. 2.8(c) and (d) we show that using cadmium nitrate as the Cd source produces the largest photoconductive effect. The thin film deposited by sulfate has

47 similar photoconductivity as that by nitrate, but exhibits much higher dark conductivity.

Acetate gives less photosensitive films due to the basic nature of acetate, which accelerate the hydrolysis of TU. Stirring speed can affect the chemical equilibria near the surface of the substrate. It is noted in Fig. 2.8(e) and (f) that theCdS film deposited without stirring exhibit appreciably faster decay than the other samples synthesized with stirring. Without stirring, the deposition is limited by diffusion. Free cadmium ions are at very low concentration due to the chelation of ammonia and are quickly consumed near the surface of the substrate. In other words, the solution near the substrate is likely to be cadmium deficient. In comparison, the bath with stirring does not have this problem since the transport of chemicals is much faster. Temperature can considerably alter the chemical equilibria of the bath, and its effects on photoconductivity are shown in Fig. 2.8(g) and (h).

However, the influence of temperature is too complex, as many parameters of the bath would change at the same time. Thus, it is difficult to make detailed analysis as in previous cases.

48 6 -50 (a) lo - 0 M(180613_2) (b) 10 - Photosensitivity io 0.234 M (160505_8) Decay rate - 0.468 M (1604292) 105. -40 1 10 104- 0 -30 0

- - A0 103- C -20 10-7 102. -C 0 -10 0 101. - -0 0 10-9 1001 -0 0

0 2 4 6 8 10 i: 10 i -1 0 Time (h) 0.0 0.1 0.2 0.3 0.4 0.5 c[NHJ (mol/L)

rw I (C) 102r Sulfate (160215_8) 10 Photosensitivity -Nitrate (160303_2) 10.3 Decay rate -- Acetate (260504_2) I V -50

10- ~10° -40'0

10.5 104-, -30 b 10'

101 103_ 20 c

10-8g 102. -10

10-9 2 4 6 8 1o 12 101. LO Time (h) Sulfate Nitrate Acetate Anion Species 10'1 (e) - 0rpm (16061122) - Photosensitivity -45 104 - 6rpm (160303_2) -- Decay rate -40 - 300rpm (160613.4) - 600 rpm (160611.20) -35 0 V 104 -30

C 10°- -25

10.8 -20 U

1048 -15 10-9 -10 I' .5 0 2 4 6 8 10 12 100. Time (h) 0 100 200 300 400 500 600 Stirring Speed (rpm) b 10-2 7 -140 - 60 c(1606124) (h)10 Photosensitivity - 70 C (160303_2) ---Decay rate -120 10.4 ,.-----D- oC (160612_1o) 100 |-100-5 10- ~~ 10-' -800 10-6 -60 104--1e 10 10-8 .c i - -40 >% 0. -201 '0-20 I In 2 4 8 a 10 n If Time (h) 6 80 Temperature (°C)

49 Fig. 2.8. Relationship between PPC and various deposition parameters. Photoconductivity time series data (a) and metrics (b) for CdS films deposited with different concentration of ammonium. Photoconductivity time series data (c) and metrics (d) for CdS films deposited with different anion species. Photoconductivity time series data (e) and metrics (f) for CdS films deposited with different stirring speed. Photoconductivity time series data (g) and metrics (h) for CdS films deposited at different temperature.

Since sulfur-deficient chemical baths produce CdS films with pronounced photoconductivity, we hypothesize that highly-photosensitive films are sulfur-deficient.

Our film composition measurements by XRF and WDS support this hypothesis. We show in Fig. 2.9 that photosensitivity is strongly correlated with sulfur deficiency. This trend is also confirmed by Rutherford backscattering spectroscopy (RBS) measurements with the help of Prof. Nate Newman's group at the Arizona State University. We performed RBS on the same two samples used for WDS calibration. The Cd/S ratios are 1.25 ±0.04 for the photosensitive sample and 0.91 0.04 for the one without photoresponse, compared with

1.27 ±0.03 and 0.96 ±0.03 measured by WDS. The Cd/S ratios in Fig. 2.9 span a range from 0.96 to 1.31, which would correspond to enormous point defect concentrations (up to

5x1021 cm). We note that the error bars in Fig. 2.9 represent the counting statistics of the

XRF and WDS data, and do not represent systematic errors that would affect all data in the same way. The transition between strongly- and weakly-photosensitive samples takes place over a narrow range of less than ±2%. This suggests that intrinsic point defects are primarily responsible for LPPC, and that the concentration of relevant point defects saturates for larger deviations in overall composition. Inspired by previous studies connecting anion vacancies to LPPC in other materials [75,100,101], we propose that sulfur vacancies are responsible for large and persistent photoconductivity in CdS.

50 109-'

5 S 10

.' 0 101 - - -

11 *

0.9 1.0 1.1 1.2 1.3 1.4 Cd/S Fig 2.9. Relationship between photosensitivity and Cd/S ratio in CdS films. Error bars represent 68% confidence intervals from the counting statistics of the XRF measurements.

Knowing the chemical cause of LPPC allows us to engineer it. To obtain extremely photosensitive CdS thin films with very slow decay, we tried different combinations of cadmium and sulfur sources, while restricting the ratio of S:Cd to be no more than 2. The results in Fig. 2.10(a) validate this method. To further enhance the photoresponse, we used the same recipe for the sample with the strongest photoresponse in Fig. 2.10(a), but changed deposition temperature. The photoresponse increases again when the sample is deposited at 60 °C. By systematically varying bath chemistry and temperature, we can make films with sensitivity of up to 10' and decay rate of 0.1 decade/h.

51 --- 60 C160709_4) - 1 -70 'C (160704_2) -- 80 C (160709_12)

104 i103

10-4 10 10'S 8 b 10-6 Ce - o.0seN.CTw - Ls1o01 ca- Om c-sK onU-Leme1W>M)A 10-7 NU- Cd-0.0045KCd1U-1*.2S(140?004) - Cd041:A KCtTh- .. 00305 0 ) Cd 0.0005 KCOU-1IA.0(00NW) -Cd G0.00)0 N.W MI)..4) Cd-. 060K Cd2(J-101042) 4 0.000C10--1001002 Time (h) Fig.2.10. Fine-tuning photocondcutivity by varying the parameters ofthe chemical bath to achieve the strongest PPC effect. (a) Tuning cadmium and sulfur source concentration. (b) Tuning deposition temperature using the recipe for the sample in (a) with the largest photoresponse.

2.3.4 Modeling photoconductivity

2.3.4.1 The standard model

Boer and Vogel described majority-carrier photoconductivity (n-type discussed here)using a generic model that considered two defects with charge transition energy in the band gap: a recombination level and an electron trap level [102]. Electrons can be optically excited from the recombination level to the conduction band at a rate of F, and re-captured with rate coefficient R. The trap level has emission and capture rate coefficients P and Rt, respectively. The rate equations are:

dn~ S = Rt(Nt - nt)n - Pnt (2.5) dt where n is the density of conduction band electrons, nt the density of trapped electrons, and Nt the total trap density. The model assumes that all conduction band electrons are generated from the recombination level, and that the concentration of holes in the valence

52 band is negligible. The recombination level is typically a deep donor, and is sometimes called the sensitizing level [2]. Under illumination, electrons are effectively transferred from deep donors to shallow traps. When illumination is turned off, relaxation to equilibrium is often limited by the trap escape rate. In the following discussion, we will refer to this as the standard model for photoconductivity.

To apply the standard model to CdS, we hypothesize that the deep donors are sulfur vacancies. The traps could include several species, possibly including ionized sulfur vacancies. There will be band-to-band generation, and we assume that the capture of free holes by recombination levels is effectively instantaneous. In the following we discuss our experimental results in light of the standard model, suggesting modifications as needed to describe the data.

2.3.4.2 Spectral dependence of photoconductivity

In Fig. 2.11 we show the dependence on photoconductivity on excitation wavelength. For this experiment the sample was equilibrated in the dark at room temperature, and then monochromatic incident light was varied from long-to-short wavelength (k). The light was modulated using a mechanical chopper at 31 Hz, a constant voltage was applied across the sample, the current response was amplified and de-modulated, and the resulting response curve was normalized by the wavelength-dependence of the incident light power. The data show a flat response and are noise-limited above 950 nm (below 1.3 eV), a rising trend for k < 950 nm, and a steep rise near Eg. The broad photoconductive response for 600 < k <

950 nm (1.3 - 2 eV) indicates the energy inside the band gap of deep donors ("sensitizing centers") that contribute to photoconductivity.

53 104

E 102 *0. C101

10o

N10-2 E 0 10.3 Z 10______3_ __ 450 600 750 900 1050 Wavelength (nm)

Fig. 2.11 Photoconductivity excitation spectra for a representative sample with large photoresponse(CdS_160613_4).

2.3.4.3 Modeling photoconductivity decay data

In Fig. 2.12(a) we show a typical photoconductivity timeseries measured at room temperature. The photoconductivity decay after illumination is turned off has three distinct regimes (1) an initial, rapid decay, (2) an intermediate, nearly single-exponential decay, (3) a long-time decay, slower than exponential. The short-time decay is well-modeled by a stretched-exponential:

Opc(t) = opc,oexp -(t)--) (2.6)

opc(t) and aC, are the conductivity due to excess carriers at times t and time t = 0, respectively. opC is defined as o(t) - os, whereos is the conductivity in equilibrium in the dark. We approximate os as the steady-state conductivity measured prior to illumination, after a long time held in the dark.Tstr and f are fitting parameters corresponding to a characteristic decay time and the width of the decay time distribution, respectively. In Fig.

2.12(b) we show fits of Eqn. 2.6 to decay data at short times for different measurement

54 temperatures. The stretched exponential function is the Laplace transform of function

P(s,fi):

exp - )=f° P(s,f) exp )-sds (2.7)

P(s, 8) describes a distribution of single-electron exponential decay processes with decay times r = . The distribution of f at different temperatures is 0.43 ±0.01. This narrow S distribution means that the activation energy distribution is unchanged within the measured temperature range. For a thermally-activated process, r is related to an activation energy

Ea through the Arrhenius equation r = ro exp (- . Our temperature-dependent data

(Fig. 2.12(c)) show that photoconductivity decay is thermally-activated with zo = 15 ±12

Ps.

The intermediate-time decay is well-modelled by a single exponential. This can be shown explicitly by fitting the data to a stretched exponential (Eqn. 2.6-2.7): for this fit f=0.9960

±0.0003 and the distribution of activation energy is sharply peaked.

In Fig. 2.12(d) we plot the distribution of activation energy photoconductivity decay. The distribution for short-time decay does not change with temperature over the measured range of 25 - 105 °C, which is reasonable because relatively small changes should not affect the distribution of defect levels or the phonon spectra. The intermediate-time decay process has an activation energy of Ea = 0.55 eV, which falls at the high end of the distribution of activation energy for the short-time decay. This is therefore the slowest single-electron process and remains active after the faster processes with lower activation energies have completed.

55 Light Intensity (mW cm-2) 31 (a) 10-3 10-2 10~' 100 101 102 (b) 10-2 0., 10161 _10-3 0 o104E 1015 0 210. . . 1.15..2530.. •• • 25 °C 1031014 1- 10-6 CL * • 45 °C ID - 10 0 10.s C 2. *:'• 85 °C 0 10-8 3- 1010 •105 °C 10~9 -41 1 Time (h) 8 In(t / sec)

(C) 7 S(d) - 25°C -45 0 C 6- 650C c 5. 85OC - 105 °C 4.- &w 4 - Intermediate time decay 3. Short-time decay 2- Arrhenius plot to fitto TO = (15 112) ps 1 0. 2.6 2.8 3.0 3.2 3.4 0.0 0.2 0.4 0.6 1 /T (10-3 K-1) Ea (eV)

Fig. 2.12. Analysis of photoconductivity decay for a representative sample with large photoresponse (CdS_160613_4). (a) Bottom axis, blue curve: photoconductivity time-series data measured using a white LED, which was turned on at time t = 1 h and off at time t = 4 h. Top axis, black curve: power law dependence of steady-state photoconductivity on light intensity, measured using a 455 nm LED. (b) Stretched exponential fits (blue lines) to short-time decay data measured at different temperatures. (c) Arrhenius plot; red points are data, blue line is Arrhenius fit. (d) Probability distribution (p(E)) of activation energy for short- and intermediate-time decay processes.

We make a small discursion here to apply the stretched exponential decay model to the data shown in Fig. 2.7(a)-(c). The best-fit parameters for each time-series data and dark conductivity are shown in the following tables, Tables 2.1-2.3. It is observed that in most cases, # values lie between 0.6 and 0.8, meaning that the single exponential model is not the best one to fit these processes. However, the decay time constant derived from the

56 stretched exponential fits are in the same trend with those derived from the single exponential model. Thus, the data shown in Fig. 2.7 are still valid.

Table 2.1. Data table for samples with varying cadmium concentration Dark

2 Sample name c[Cd +j / M conductivity/ Tstr / S

f-I cm-1

CdS_160611_2 0.00075 3.21x10-9 1.0 298.7

CdS_160303_2 0.0015 7.01x10-7 0.6535 30.9

CdS_160611_4 0.0045 8.19x10-9 0.6520 197.4

CdS_160611_6 0.009 5.70x10-9 0.8071 602.5

Table 2.2. Data table for samples with varying thiourea concentration Dark

Sample name c[TU] / M conductivity/ Tstr / S

1..1 cm-'

CdS_160611_8 0.0075 2.45x10-9 0.7806 1118.9

CdS_160611_10 0.0375 3.05x10-9 0.7240 520.4

CdS_160303_2 0.075 7.01x10-7 0.6535 30.9

CdS_160611_12 0.15 9.60x10-10 Undefined Undefined

Table 2.3. Data table for samples with varying ammonia concentration

57 Dark

Sample name c[NH3]/M conductivity/ Tstr / S

A-I cm-1

CdS_160613_2 0.94 2.18x10-9 0.5715 40.2

CdS_160303_2 1.87 7.01x10-7 0.6535 30.9

CdS_160611_16 2.81 1.14x10-9 0.3855 29.8

CdS_160611_18 3.74 6.87x10-WO Undefined Undefined

The photoconductivity decay dynamics can be understood in the context of the standard model. The rapid decay at short-times is due to trapping and recombination of electrons in the conduction band. The relative magnitude of recombination and trapping rate depends on temperature, which will be proved in the appendix of this chapter. The single- exponential decay at intermediate times implies nt » n: most of the non-equilibrium electrons are trapped. In this case ~ 0 and Eqn. 2.4 becomes = -Rrfntn, which dt dt produces single-exponential decay. This approximation relies on recombination being slow, so that nt changes slowly. These approximations are supported by numerical simulation of the standard model, which will also be demonstrated in the appendix. At long times the system approaches equilibrium, we have detailed balance between the trap level and conduction band. Eqn. 2.5 gives nt = n considering nt « Nt. Then Eqn. 2.4 p becomes = -Rr(tNt + 1) n2 and the decay rate is non-linear. The validity of this dt / ecay approximation is also shown in the appendix.

58 2.3.4.4 Power-law model of steady-state photoconductivity

In Figs. 2.12(a) and 2.13 we show the dependence of steady-state photoconductivity on light intensity. The data can be divided into two regimes: under low illumination, it follows a power law upc oc Ib with exponent b between 0.3 and 0.5 that depends on temperature, while under higher illumination b is between 0.6 and 0.7.

10-2. • 25 °C • 45 °C

10-3 A 65C Slope= 1 10-4 Slope =2/3 0

510-5 Slope =1/3

16Slope =1/2 10~6.-,.. --. --- ,-.--,-.-,.-.. 10-3 10-2 10~1 100 101 102 Light Intensity (mW cm-2 ) Fig. 2.13. The power-law relationship opC °Clb between the steady-state photoconductivity (pc) and light intensity (I) measured at different temperatures for a representative sample with large photoresponse (CdS_160613_4). The data at 25 °C (black curve) is the same as plotted in Fig. 2.12(a). At low intensity b is between 0.3 and 0.5. At higher illumination b is between 0.6 and 0.7. Data was measured using a 455 nm LED.

We can model the generation rate as F = g), where 4) is photon flux and g is a constant.

Adding this term to the standard model and setting time derivatives to zero yields the steady-state condition g4= R,(n + nt)n. For high trap occupancy nt Nt » n, the standard model predicts a power law n oc D. For the case of low trap occupancy n « Nt« , setting Eqn. 6 to zero produces nt = n. In this case g( = R, R + 1) n 2, which leads to n oc ). Therefore, the standard model with linear recombination can produce an

59 apparent power law with exponent 0.5 < b < 1, approaching 1 under high illumination. In order to model our results showing 1/3 < b < 2/3, we consider the effect of quadratic recombination. If we replace the linear recombination term by R'(n + nt)n2 , the power law relationships becomes n oc 4when nt ~ Nt, and n oc

2.3.4.5 Temperature-dependence of steady state photoconductivity

In Fig. 2.14(a) we plot the temperature-dependence of the steady-state conductivity in the dark and under illumination. At equilibrium in the dark, the conductivity increases with increasing temperature. This is as-expected for a non-degenerate and large band gap semiconductor. In contrast, the photoconductivity decreases with increasing temperature.

In the context of the standard model, this shows that recombination is thermally-activated.

Thermally-activated recombination may have several causes. By comparing energy scales, we can rule out thermal release of holes trapped at sensitizing levels. The spectral photoconductivity data in Fig. 2.11 show that the sensitizing levels are in a band approximately 1.3 - 2 eV below the conduction band edge, or 0.4 - 1.1 eV above the valence band edge. The activation energy determined from the steady-state photoconductivity data in Fig. 2.14(a) is 0.144 ±0.024 eV. Therefore, thermal release of holes trapped at the same energy levels as the sensitizing levels cannot explain the observations.

60 2.3.4.6 Temperature-dependence of photoconductivity transients

In Fig. 2.14(b) we show photoconductivity transient data measured at different temperatures. The transient behavior under illumination can be described by two time scales: a fast rise with time constant ti, and a slow decay with time constant T2.

Photoconductivity decay with the light off is characterized by a third time scale, T3. The time constants Ti and T3 appear to be correlated: both get smaller with increasing temperature. The correlation between11 andT3 is shown in Fig. 2.14(c). The best-fit slope is 0.94 0.15 and the p-value is 0.004 for a one-tailed t-test, which means that the slope is greater than 0 at 95% confidence level. That the slope is nearly unity means that ti andT3 evolve similarly with temperature. This suggests that the same thermally-activated process is responsible for these two transients.

The combination a fast rise (with time constantti) and a slow decay (with time constant

T2)leads to a transient maximum (an "overshoot") in photoconductivity under illumination.

This maximum is visible in the data in Fig. 2.14(b) measured between 378 and 338 K; at lower temperatures, the maximum moves outside our time window. We find that Rr > Rt is a necessary condition for the rate equations to allow a transient maximum (dn/dt = 0) under illumination. The proof will be shown in the appendix. The observation that the maximum occurs at shorter times with increasing temperature means that recombination is thermally-activated, and grows faster than Rt as temperature increases. A similar conclusion can be drawn when considering quadratic recombination, also shown in the appendix.

61 In Fig. 2.14(d) we show the results of numerical simulation of the standard model with thermally-activated recombination, i.e. Rr = Rroexp(-). Forsimplicity,inthis simulation we assume that trapping rate is temperature-independent, because our data do not clearly indicate the temperature dependence of the trapping rate (adding thermally- activated trapping leads to the same qualitative conclusions). The simulations reproduce the conductivity overshoot and its temperature dependence, and the positive correlation between time constantst 1and13. From the data and simulations in Fig. 2.14 we conclude that thermally-activated recombination is responsible for the temperature dependence of

LPPC in CdS.

(a) b) 10-2 _10-3 E E 1T T 10-41 10- C:10-5 r2 -298 10-6 K

10-7 - 318 K 0 10 10-5_ - 338 K 358 K - -8 Dak 1 0 Photo 378 K 01-6 300 330 360 390 1 2 3 4 5 6 Temperature (K) Time (h) (c) ( d) 7- Light On Light Off

0 6-

S Ca) 5 Increasing 4- S C b temperature 0) 3. 0

2 2 3 4 5 6 In(c decay (s)) Time (a.u.) Time (a.u.)

62 Fig. 2.14. Evidence for thermally-activated recombination in the temperature-dependence of photoconductivity for a representative sample with large photoresponse (CdS_160613_4). (a) Temperature dependence of steady-state conductivity in the dark and under illumination using the white LED. (b) Transient photoconductivity measured at different temperatures. The light is switched on at I hr and off at 4 h. The transient under illumination shows a fast rise with time constant ri and a slow decay with time constant T2. The decay with the illumination off has characteristic time constant T3. (c) Positive correlation between time constants Ti and T3. (d) Numerical simulations of the standard model, showing how transient behavior changes with temperature.

2.4 Discussion

2.4.1 Modeling large and persistent photoconductivity in CdS

The standard model of photoconductivity is simple but possess high explanatory power. In

Sec. 2.3.4 we analyzed our results on CdS in terms of the standard model, and suggested modifications needed to describe the data. Both linear and quadratic recombination processes are needed to describe the observed power law dependence of steady-state photoconductivity on illumination. The temperature dependence of steady-state photoconductivity and of the transient maximum observed under illumination indicate that recombination of conduction electrons with trapped holes is thermally-activated.

In Sec. 2.3.3 we showed that photosensitivity is strongly correlated with sulfur deficiency

(Fig. 2.9). We also showed that the photosensitivity and photoconductivity decay rate are strongly correlated (Fig. 2.7(d)). Therefore, the same defects that are responsible for the large photoconductive response are also responsible for the slow decay. This is supported by our simulations showing that recombination is the rate-limiting process during photoconductive decay. This is unlike the usual interpretation of the standard model in which the size of the steady-state photoresponse depends on the sensitizing centers, while the slow decay depends on the trap levels.

63 These observations support the hypothesis that sulfur vacancies featuring hole-hole correlation with negative-U behavior are responsible for large photoconductivity and its slow decay in CdS. Both the large photoresponse and its slow decay originate from thermally-activated recombination between conduction electrons and trapped holes.

The resulting, modified standard model to describe LPPC in CdS is consistent with the

ZWLZ model developed to explain LPPC in ZnO (Sec. 2.1). The sensitizing levels are neutral sulfur vacancies, VS'. The model features thermally-activated linear and quadratic recombination processes, i.e. (Vs")* - Ys, (Vs')* -+ Vs, and (Vs")* -+ Ys. Each of these processes involves a transition between metastable and equilibrium lattice distortions around Vs. The kinetic pathway for quadratic recombination could be (1) Thermally- assisted, two-electron excitation into the vacancy level, followed by spontaneous lattice relaxation: (Vs")* + 2e- -+ (Vsx)* - Vs + ph; or, (2) Thermally-assisted lattice distortion, followed by two-electron capture: (Vs")* + ph -+ V;', Vj' + 2e- - VsX .

Pathways for linear recombination involving a metastable state (Vs)* can be proposed in a similar way. The chemical identity of the traps is unknown. The population of traps may include singly-ionized sulfur vacancies with or without a lattice distortion, Vs or (Vs)* in our notation.

2.4.2 Energy level diagram

In Fig. 2.15(a) we show a generic energy level diagram for the standard model of photoconductivity. Energy level diagrams show equilibrium charge transition levels and do not clearly represent kinetic effects such as thermally-activated recombination and metastable lattice distortions. Nevertheless, we can illustrate features of the energy level diagram for photoconductive CdS based on our data.

64 The Fermi level (EF) at thermal equilibrium in the dark can be calculated using the formula no = Nc exp (-Ec-EF), where no is the carrier density in the dark, and Nc is the conduction band effective density of states, Nc = 2.2 x 1018 cm 3 [103]. Our Hall measurements give no = (7.0 + 1.0) x 10 9cm-3 at room temperature (298 K), from which we calculate Ec - EF = 0.507 0.003 eV. Similarly, our Hall data give npc,o =

(1.0 ± 0.1) x 10 16cm-3 and Ec - EF = 0.140 0.003 eV under 149 mW/cm 2 white

LED illumination. The photoconductivity excitation spectrum gives Ec - ER > 1.3 eV, where ER is the charge transition level of deep donors in equilibrium in the dark.

In Fig. 2.15(b) we show the energy level diagram in the dark. EF lies in between the deep donors (including Vsx) and the shallow traps. In Fig. 2.15(c) we show one possibility for the energy level diagram under illumination with sulfur vacancies in their ionized, metastable configuration (Vs)*. Here we have drawn the transition level (Vs")* -+ (Vs')* resonant with the conduction band, as for ZnO, but this is speculative. The conclusion that the dominant recombination process is thermally-activated does not depend on the location of the charge transition levels relative to the band edges.

65 (a) E_ EF

(b) (c)

eV Ec 0 EF ------. 14 F.45 eV J1.3eV EFp Ey-..-. Ey. ~-~ ______EIVI

•Electron o Hole

Figure 2.15: Energy transition level diagrams. (a) Standard model of photoconductivity including trap and recombination levels. (b) Energy level diagram for photoconductive CdS at equilibrium in the dark. (c) One possibility for energy level diagram forCdS under illumination.

2.5 Conclusion

The proposed mechanism for large and persistent photoconductivity is atomic rearrangement following a redox reaction at an optically-active defect. In both ZnO and

CdS, the transition VA + hv -> (VZ)* + ph + 2e- is accompanied by a large movement of the surrounding cations (Fig. 2.1). With the defect in the neutral state (VA), the cations are attracted to two un-bound electrons and move inwards relative to their crystal lattice positions, into the space that would otherwise be occupied by a large anion. After the transition to the doubly-charged state (V;), the cation-cation interaction is no longer screened and the cations flee the defect site, moving outwards relative to their lattice positions. This atomic motion physically isolates the trapped minority carriers, providing a kinetic barrier to recombination.

66 The size of this effect in ZnO and CdS can be understood by comparing the fractional change in ion-ion distance due to redox transitions at all intrinsic defects in these materials.

A comparison of density functional theory (DFT) ground-state calculations, performed by different research groups for ZnO and CdS, finds a striking quantitative agreement: the fractional change in cation-cation distance due to the anion vacancy double oxidation reaction is 29% and 33% in ZnO and CdS, respectively [75,93]. In Fig. 2.16 we compare the structure changes due to redox reactions at intrinsic defects in ZnO as calculated by

Lany and Zunger, to those same in CdS as calculated by Nishidate et al. [75,93]. These are much larger than the atomic rearrangements due to redox reactions at other intrinsic defects.

This suggests that additional materials with long-lived, non-equilibrium, photo-excited states may be identified by analyzing ground-state calculations of atomic motion following defect redox reactions. The idea that atoms rearrange to create physical separation between photo-excited charge carriers, thus slowing the decay to equilibrium, is reminiscent of

"DX-center" dopant de-activation in AlGaAs, to long photo-excitation lifetimes observed in systems with polaron excited states including organic semiconductors and halide perovskites, and to biological light absorbers such as photosystem II and dye molecules used for solar energy conversion [84-90].

67 El

Percent change in atom-atom spacing due to redox transitions at intrinsic point defects 35 r

M ZnO, Lany and Zunger (2005) 30- = CdS, Nishidate (2006)

25-

20-

15-

10-

01 M V0 Cds Cd (I) Cd1(II) Vs SI Si() Scd VCd

Fig. 2.16. Comparison of the structure changes due to redox reactions at intrinsic defects in ZnO as calculated by Lany and Zunger, and in CdS as calculated by Nishidate et al. [75,93]. Structure changes are represented by percentages, calculated as % = 100 X 2 X (dma - dmin)/(dma + dmin), where dm. and dmi are the maximum and minimum atom-atom separations realized over all possible charge states. The particular point defects and atomic distances are as follows (Vo is for ZnO, all others are for CdS). Vo: Zn-Zn distance at an oxygen vacancy, change from +2 to 0. Cds: substitutional-Cd distance at a Cd-on-S substitutional defect, change from +2 to 0. Cd(I): interstitial-Cd distance at an octahedrally-coordinated Cd interstitial defect, change from +2 to 0. Cd(II): interstitial-S distance at an octahedrally-coordinated Cd interstitial defect, change from+2 to 0. Vs = vacancy-Cd distance at a sulfur vacancy, change from +2 to 0. Si(I): interstitial-Cd distance at a tetrahedrally-coordinated S interstitial defect, change from 0 to -2. S(II): interstitial-S distance at a tetrahedrally-coordinated S interstitial defect, change from 0 to -2. Scd = substitutional-S distance at a S-on-Cd substitutional defect, change from 0 to -2. Vcd: vacancy-S distance at a Cd vacancy, change from -I to -2.

Persistent photoresponse often results from spatial non-uniformity, such as potential barriers due to resistive secondary phases or high doping levels. Spatial inhomogeneity is invoked to explain apparent recombination cross section (acs) many orders of magnitude smaller than 102 cm2 [62,104,105].For the samples studied here, the large range of decay rates measured and the systematic dependence of such on the CBD bath chemistry suggest that inhomogeneous photoconductivity is not dominant. We can estimate acs as (inT)- , where v is the electron thermal velocity, n is the instantaneous non-equilibrium carrier

68 concentration, andt is an instantaneous photoconductivity decay rate. Taking v = 107 cm/s and using our measured (n, t) data we estimate that acs falls in the range 10-27 - 10-21 cm-

2 for the samples studied here. The ZWLZ model can be considered an extreme version of inhomogeneous photoconductivity, in which the potential barriers are centered around individual point defects. It is possible that individual point defects and defect clusters both contribute to LPPC in CdS, thereby linking the ZWLZ model and spatial inhomogeneity models of persistent photo-effects. Persistent photo-effects in CdS have also been ascribed to extended defects including grain boundaries and screw dislocations [106,107]. It would be worthwhile to further understand the interplay between off-stoichiometry, point defects, and extended defects in CdS.

In summary, we demonstrate that the photoconductive response of CdS thin films can be widely tuned by varying the bath chemistry during CBD. We show that large photoresponse and slow decay are correlated with sulfur deficiency. Both linear and quadratic electron-hole recombination are relevant, and the rate-limiting process during photoconductivity decay is thermally-activated recombination. These observations provides experimental validation for the model that anion vacancies in II-VI semiconductors are negative-U defects, exhibiting strong hole-hole correlation due to lattice distortions that respond to changes in the vacancy charge state. Directions for future research include complementary spectroscopic experimental studies and in-depth theoretical modeling of the kinetic pathways for linear and quadratic recombination.

69 2.6 Appendix

2.6.1 Conditions for transient maximum photoconductivity

We start from the rate equations:

dn -= F + Pnt - Rt(Nt - nt)n - Rr(n+ nt)n dt

dn= Rt(Nt - t)n - Pnt dt

Setting = 0 to model a conductivity maximum or minimum, we have

Rrfn2 + RtNtn - F t-= l to(nl) Ren + P - Rrn

Suppose there exists a pair (n, nt) away from equilibrium that satisfies the above equation.

We want to determine under what conditions this corresponds to a maximum in n. First, we need to determine the sign of at nt = nto. Then, we add a small perturbation to nt dt

(positive or negative depending on the sign of ) and see how changes. dt dt

Since we can choose Rt and P freely, both casesdnt > 0 or < 0 could be dt lnt=nto dt nt=nto true.

Case 1: We first consider the case ofdnt > 0. In this case, the condition for a d nt=nto maximum in n isd = P + Rtn - Rrn < 0 because - = 0 at this point in time. It dnt dt follows that Rt - Rr< - P < 0. n

70 Case 2: We next consider the case of d < 0. We will show that in order to have

dt < 0 for someta, Rt - Rr < 0 must hold. Immediately after switching illumination on (F > 0) we know that > 0. Also, when there is a small increase in n dt there will also be an increase in nt, .- > 0. Therefore, for a short time after illumination dt is turned on, both n and nt are increasing. Suppose at some later time ta we have

dntl < 0. Then it follows that there exists time tm < ta at which nt is at a local dt'ttadft _ dlI 0.ft maximum: tmsatisfies It = 0 and > 0. Inspecting the formula for -s-, we at It=tm at ttm tm we have n < 0. It follows that there exists time tn < tm at which dn =0 dt=tm dt=tn

(n(tn) is a temporary maximum) and ant > 0. This is just the situation that we

discussed in Case 1, which can only be true when Rt - Rr < - - < 0. n

We conclude that Rr > Rt is a necessary condition to observe a transient photoconductivity maximum under illumination.

The above analysis considered the case of linear recombination. For quadratic recombination, the recombination term is replaced by R'(n + nt)n 2 , and the necessary condition for a transient photoconductivity maximum becomes R'n > Rt.

A similar argument as made in Sec.2.3.4 can be used to show that quadratic recombination also must be thermally activated to explain the observed transient maximum photoconductivity. Here, Rn plays the same role as Rr does for the case of linear

71 recombination. In particular, since for the case of linear recombination we concluded that

R, must be thermally-activated (increase with temperature), then for the case of quadratic recombination R'n must increase with temperature. We have observed that under illumination n decreases with temperature. Therefore, the quadratic recombination rate coefficient R' must be thermally-activated and increase with temperature.

2.6.2 Analyzing numerical simulations of the standard model

In Figs. 2.14(d), 2.17, and 2.18 we show the results of numerical simulations of the standard model with thermally-activated recombination. In Fig. 2.17 we plot the rates of trap electron captrue, trap electron emission, and recombination during photoconductivity decay for the case of low and high temperature, where temperature ranges from 298 K to

378 K and the activation energy for recombination process is taken to be 0.5 eV. The initial fast decay is be due mainly to trapping and/or recombination, with recombination becoming more important at higher temperature. During decay at intermediate times, recombination is the rate limiting step and is responsible for the slow decay of photoconductivity. When recombination is very much slower than trapping and emission, then trapping and emission are nearly in detailed balance.

(a) -_-cobintion -Recombination - -Trapping - -Trapping Emssion Emission

T (T (

Time (a.u.) Time (a.u.)

72 Fig. 2.17. Rates of trapping, emission, and recombination at (a) low temperature, and (b) high temperature.

Fig. 2.18 shows that the ratio between nt and n approaches a constant at long times. In this limit, the conductivity decay deviates from single exponential and becomes nonlinear. In

dn themaintextweshowthatinthislimittheconductivitydecayisquadratic=dt

R,.r + 1) n2 . This quadratic decay at long times is consistent with the data, c.f Fig.

2.12(a).

Time (a.u.)

Fig. 2.18. Ratio of concentrations of trapped and free electrons during photoconductivity decay. The ratio approaches a constant at long times when the decay curve deviates from single exponential.

73 Chapter 3. Two-Terminal Resistive Switches Based on Defect-

Level Switching in CdS

3.1 Introduction

3.1.1 Two-terminal electronic devices

Two terminal resistive devices are widely used in all kinds of electronics. Most of the devices obey Ohm's law, where the relationship between current and voltage is linear.

Semiconductor materials also exhibit nonlinear characteristics, such as diode behavior when forming junctions. One important nonlinear phenomenon is hysteretic behavior, which is essential for many applications, such as circuit protection (e.g. fuses), oscillators

(e.g. Gunn diode), selectors (e.g. Ovonic switch), and resistive switches (e.g. memristors) [108-112]. Particularly, there is tremendous interest in resistive switching devices for applications in storage-class memory, resistive computing, compute-in- memory, and neuromorphic computing [113-119,76]. This nonlinear and hysteretic phenomenon has been observed in various material systems, and different mechanisms contribute to each case. One most common mechanism in reports relies on mass transport of atoms by drift and diffusion [120-123]. This usually involves growth of conductive filaments within a dielectric matrix material under voltage bias. Another mechanism with sustained interest is based on phase transitions, such as metal-insulator transitions and structural transformations [124-128]. Switching between phases with different conductivity under electrical stress leads to the desired nonlinear behavior. We introduce a novel mechanism for highly nonlinear, hysteretic two-terminal devices that does not rely on mass transport or phase transitions. This mechanism is called defect level switching

74 (DLS), as it is based on the giant change in conductivity of materials and heterojunctions due to discrete changes in point defect atomic configurations. We expect this to be a generalizable mechanism, which may be realized in many different materials and device configurations.

3.1.2 Point defects in semiconductors

Point defects are common in semiconductors. At most simple level, point defects in semiconductors are characterized by their transition levels, the thermodynamic potential at which they become ionized. Defect levels are essential for semiconductor device engineering. They guide doping strategies and control recombination activities. In some cases, defects can feature multiple, widely-separated transition levels when defect ionization is accompanied by a substantial defect lattice relaxation. This usually occurs when the atomic configurations around a defect are bistable, or due to electron-electron correlation. Point defects are the main cause of photoconductivity in a variety of semiconductors. Lattice distortions around DX-centers in AlGaAs can lead to large and persistent photoeffects [61]. Theoretical studies have also shown that lattice relaxations around anion vacancies in ZnO and chalcopyrite semiconductors is the cause of persistent photoconductivity [75]. Recently, we demonstrated that CdS exhibits similar phenomenon, where lattice reconfigurations around sulfur vacancies could result in large and persistent photoconductivity [83]. The hole-hole correlation stabilizes doubly ionized sulfur vacancies and creates a metastable conductive state. Such defect level bistability is challenging to measure in semiconductor device design, and has not been exploited as a functional basis for devices.

75 3.1.3 Defect-level switching device concept

The giant change in conductivity due to point defect transition level bistability is an attractive potential mechanism for resistive switching. These bistable defects can exist in two energy levels, a deep donor configuration and a shallow one. Conductivity change between drain and source in conventional MOSFET devices is limited by the Boltzmann term 60 mV/decade in the subthreshold regime. This number is larger when there are defect levels in the bulk or interface. However, the conductivity change due to switching defect levels between deep and shallow configurations can in principle be larger when the density of bistable defects is much higher than background doping. To date, DLS has been observed and catalogued in photoconductors, with photons responsible for directly ionizing the bistable defects, or for generating the ionization potential to do so. We establish here the concept for an all-electrical DLS device to re-create the minority carrier injection and defect ionization as occurs in photoconductivity, but using electrical device design and without light.

Here, we propose and demonstrate a two-terminal DLS device using CdS as an active layer andMoO3as a hole injection layer. CdS is a well-known photoconductor. It exhibits large and persistent photoconductivity, which is correlated with sulfur vacancies switching between deep and shallow donor configurations upon ionization. Photoresponsivity can be tuned through processing conditions by controlling the vacancy concentration; thin films with higher sulfur deficiency demonstrate higher photosensitivity and longer decay time constant.MO03 is known to possess high electron affinity and wide band gap, and is commonly used as a hole injection layer in organic solar cells [129-131]. We design an all-electrical device usingMoO3 to inject holes into CdS, thereby ionizing the sulfur

76 vacancies and switching the defects in different configurations. I-V measurements confirm that the fabricated device show resistive switching and results from mechanism studies are consistent with the DLS hypothesis.

3.2 Experimental methods

3.2.1 Device fabrication

The device is fabricated on a silicon wafer with approximately 100 nm oxide layer. For the normal device, 4 nm Cr and 100 nm Ag are first deposited on the substrate using thermal evaporation. The CdS layer is then deposited through chemical bath deposition. Typically,

306 mg Cd(N0 3)2 and 76 mg thiourea are first dissolved in 10 mL DI water, respectively.

76 mL DI water is heated to 70 , and the substrate is immersed into the bath vertically.

Then precursors are added into the bath in the order of Cd(N0 3)2 solution, 14 mL ammonium hydroxide, and thiourea solution. The whole chemical bath is under constant stirring at 300 rpm, and the deposition last for 8 min. The sample is then rinsed with DI water and sonicated in isopropanol for 5 min. 20 nmMoO3 layer is then deposited onto the sample using thermal evaporation, followed by 4 nm Cr and 50 nm Au contact.

Device patterns are made with photolithography. The device is first spin-coated with SPR

700 photoresist, and then illuminated under a photomask for 15 s with the help of a Karl

Suss MA-6 mask aligner. The sample is then developed in MF CD-26 developer for 1 min.

Au top contact is first removed by ion milling, and the sample is further etched in dilute

(NH4)OH and dilute HCl to removeMoO3 and CdS, respectively. Finally, the sample is sonicated in acetone to remove photoresist.

77 3.2.2 Device characterizations

Electrical measurements are carried out in a custom-made probe station. I-V data are measured using a Keithley 2400 source meter in a two-point configuration. The voltage sweep rate is about 0.2 V/s. Sample temperature is controlled using a thermal chuck. It is set to 25 °C if not stated otherwise. Time series data are also collected using the Keithley

2400 with a LabView program. The data collecting rate is about 1 data point per second.

Pulsed voltage measurement using a Keithley 2450 source meter along with a Tektronix

AFG3052C arbitrary function generator. The random function generator is used to supply voltage pulses. This function generator has a rise and fall time of 7 ns, but is small compared to the pulse widths used in our study, which is 50 ms. The pulse width is chosen such that the resistance of the device changes within reasonable number of pulses. The

Keithley 2450 source meter is used to determine the resistive state of the device with a small test voltage of 0.05 V. We set resistance thresholds for the ON and OFF states, and voltage pulses reverse polarity when the device resistance moves across the thresholds.

3.2.3 Simulation methods

Device simulation is performed in SCAPS software [132]. The CdS layer contains a deep donor level and a metastable defect transition energy level within the conduction band. The distribution between the bistable defect species is calculated at an initial working voltage.

During voltage sweeps, the bistable defect states function as normal defect levels, respectively. The conversion between the deep level and shallow level can only be set as single levels. Another recombination level with a Gaussian distribution in the middle of the bandgap is also included. The metastable distribution for ON state is calculated by an

78 initial voltage of -0.4 V and for OFF state, 0.4V. Parameters used in the simulation are shown in Table 3.1.

Table 3.1. Material parameters for simulation CdS MoO3 Thickness (nm) 60 20 Eg (eV) 2.42 3.1 X (eV) 4.30 6.70 er 9.0 4.60 Nc (cm-3 ) 2.24x1018 2.0 x1019 Nv (cm-3 ) 1.80x1019 4.0x1019 ve (cm s') 2.6x107 1.0x107

vh (cm s-1) 1.3x107 1.0x107

Me (cm2 V-1 s-1) 1.57 100 Ph (cm2 V-1 s-1) 1.0 100 ND (cm-3 ) 1.0x 10 2.8x1017 Defect levels Type +/0 (Gaussian) AEc (eV) 1.4

Characteristic 0.2 energy (eV)

Nt (cm-') I x 1019

Metastable transition levels Nt (cm-3 ) 2x1018

Ec - ETR(eV) 0.9 Defect I type +/0

AEc 1 (eV) 1.1 Defect 2 type 2+/+

AEc 1 (eV) -0.2

79 3.3 Results

3.3.1 Device concept

Fig. 3.1 shows the device schematic and simulated band diagrams. Simulations predicts that the device will be in low-resistance state as fabricated, and will switch to high- resistance state under forward bias. This prediction indicates that injecting electrons into n-type CdS from the electrode will switch the device to high-resistance state because the ionized point defects get filled. This counterintuitive behavior is akin to the Gunn diode which also relies on bistable electronic state. In that case, there are two valleys in the conduction band with different electron masses and mobilities, and electrons can transfer between them under proper electric field, resulting in a negative differential resistance region. Although our device simulation provides some insights into how the device works, there is substantial challenge for traditional simulation packages due to bistable and history dependent defect transition levels. In addition, due to the large band offset, tunnel current might also play an important role.

(a) 2 - Mo~aCdS

0 ------M EcF

ED -- - -2 EV -- 3 3 20 0 20 40 60 Distance (nm)

80 Fig. 3.1. Two-terminal resistive switching device design concept. (a) Device structure and measurement schematics. (b) Simulated band structure at zero bias. The ground state of deep donor levels and the shallow metastable state are shown as EDand Es separately.

3.3.2 Current-voltage characteristics

Experimental results are as-expected for DLS hypothesis and device design. Fig. 3.2 shows the I-V and R-V curves of the device for the first 10 sweeping cycles. It demonstrates bipolar resistive switching behavior, which RESET at positive bias and SET at negative bias. The device is in ON state as fabricated, and exhibits no electro-forming process. The

RESET voltage for the first cycle is at about 0.4 V, and for the subsequent cycles are concentrated at about 0.55 V. The SET voltages are quite small, typically below 0.1V in absolute value. The asymmetry in SET/RESET voltages could be due to the closer proximity of the equilibrium state under no bias to the ON state, so that a small negative bias is enough to provide the driving force to switch ON. This is different from the conductive filaments based switching devices where the filaments are usually quite stable and non-volatile under zero bias. DLS devices can relax back to the equilibrium state slowly due to the charge state transitions. The device in the ON state shows linear I-V behavior like a resistor, but in the OFF state the device shows rectifying behavior like a diode. The on/off ratio is about 10 at the instance of switching, but can reach more than

100 at the largest due to the diode behavior.

81 El

111 (b) - Cycle 2

103. - ---Cycle 34 Cycle 4 10-2- 1 Cycle 5 1 2 --- Cycle -Cycle 6 Cycle 2 2 - Cycle 7 C -3 - Cycle3 E 1023- Cycle 8 10 - Cycle 4 -Cycle 9 - Cycle5 . -Cycle 10 S - Cycle 6 4) 10-4. -- Cycle 7 it 3 - Cycle 8 101 - Cycle 9 10 - Cycle 4 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Voltage (V) Voltage (V)

Fig. 3.2. Measured (a) I-V and (b) R-V characteristics ofthe resistive switching device (190516_2). The device is in its ON state as fabricated, and shows bipolar switching behavior. The size of the device is lxi mm 2 .

3.3.3 Switching mechanism

We study the mechanism of the switching effect using two categories of tests: by direct material processing and substitution, and more advanced characterization techniques. The most straightforward tests are achieved by modifying material selection and processing methods. Fig. 3.3(a) shows the I-V characteristics when either MoO3 or CdS is omitted in the device. When either layer is not included, the device shows pure resistive behavior and no switching effect can be observed. The Ag/MoO3/Au device shows smallest resistance, which demonstrates that MoO3 is highly doped and conductive. Four point probe measurements on bare MoO3 films give a resistivity of 2.49x10-3 cm, on the same order of graphite. This observation confirms the importance of the CdS/MoO3 junction, which is essential in creating the rectifying behavior in the OFF state.

82 (a)10 (b) 101

10 2 10-2.

C 10 3 ,CD 10-

4 10 - 10 - Ag/CdS/MoO3/Au AGICdS/MoO,/Au Ag/CdS/Au Ag/CdS/ITOIAu 5 1,* - -Ag/Mo3/Au Ag/NPC CdS/MoO/AL 10. 10-5 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 -1.0 -0.5 0.0 0.5 1.0 1.5 Voltage (V) Voltage (V) (C) 10I-1

10-1

4-a C 0) 10

0 7 10- - A/CdS/Mo^/Au Au/CdS/MoO,/Ag - Cr/CdS/MoO./Au -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 Voltage (V)

Fig.3.3. Tests to rule out conductive filaments based switching mechanisms.(a) I-V characteristics for devices without the CdS/MoO3 junction are purely resistive and show no switching behavior. (Samples: 190516_2, 190222_3,1902136, in the same order as in the legend) (b) Measurements showing no sulfur vacancy filaments is not likely for the switching device. (Samples: 190516_2, 190315_1, 190716__4) (c) Measurements showing no Ag filaments is not responsible for the switching behavior. (Samples: 1905162, 1906191, 1905161) Size of the devices measured are 1xl mm2 if not stated otherwise.

Next, we show that sulfur vacancies are essential for resistive switching, but sulfur filaments are not involved in the hysteretic process. Fig. 3.3(b) shows the I-V curves for a device with non-photoconductive CdS layer. Previously, we have demonstrated that photosensitivity of CdS is strongly correlated with sulfur deficiency in the film, and we can tune that by simply varying the concentration of precursors during film growth [83]. 83 We increased the sulfur precursor concentration and deposited a non-photoconductive CdS layer in the device. The resulting device shows only diode like behavior without any switching even when voltage bias is beyond ±2V. This confirms that sulfur vacancies are the key to observe resistive switching phenomenon. When these vacancies are eliminated in the active material, the device act as just an ordinary diode. However, as essential as the sulfur vacancies are for switching events, their movement does not seem to be critically involved. We replacedMoO3with ITO, another conductive oxide with a much lower work function (4.7 eV), but the I-V behavior shown in Fig. 3(b) is purely rectifying without turning conductive or resistive as in the case of a typical working device. If the mechanism of switching ON were due to growth of sulfur vacancy filaments, then the contacting oxide layer should not have prevented the filaments from forming. The reason why ITO/CdS junctions cannot exhibit resistive switching behavior is that the low work function of ITO cannot interact with the deep defect levels in CdS, and therefore cannot inject holes into the sulfur vacancies. The resulting device behaves just like a normal heterojunction.

Silver filaments are another common mechanism utilized in designing resistive switching devices. Silver is known to be mobile and easily diffuse into the dielectric matrix materials.

However, results in Fig.3.3(c) show that silver filaments are also not the cause of switching behavior. One piece of evidence is that we switched the Ag and Au contact but the resulted device still shows the same bipolar resistive behavior. In this case, switching direction is inconsistent with Ag filament growth. We also replaced Ag with Cr which is much less mobile than Ag. Same switching effect is observed, although the SET/RESET voltages are slightly higher due to the greater background resistance introduced by Cr. Ag is important

84 -I

for high conductivity and high on/off ratio due to its low resistance compared with other metal contacts, but not essential for the DLS switching mechanism.

One property of localized filaments based resistive switching device is no apparent areal dependence of ON resistance [133]. We measured the I-V curves of three devices of different area, as shown in Fig.3.4(a). In Fig.3.4(b), we see that both ON and OFF current scale with device area when device is smaller than 1 mm2 , which is inconsistent with filament based mechanism. The background resistance is about 3.6 K due to the silver substrate and internal resistance of the measurement setup, which is large compared to that of the device when cell area is larger than 1 mm2 (less than 1 K). In this case, the reduction in ON resistance is mostly immersed in the background, but still observable. However, the

OFF state resistance still apparently scales with device area for the larger devices because the series resistance does not matter as much in the OFF state. These results again indicate that conductive filaments are not likely responsible for the ON state.

(a) (b)10'

-- OFF 10-2 10.2

0(. Slp -1

-5 - 0.3 x 0.3 mm -l-- 1 x1 mm 10~ d d 02 x 2 mm 10 Series resistance limited -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 10-1 10° Voltage (V) Device area (mm2 )

Fig. 3.4. Size-dependence of equilibrium-ON device test results. (a) I-V characteristics of resistive switching devices with different sizes. (b) Device resistances extracted at 0.3 V for both ON and OFF states. (Sample name: 190516_2)

85 We also performed more advanced characterizations on the switching media to rule out alternative switching mechanisms. Since DLS is based on charge state transitions, the equilibrium state at zero bias would be different from those at SET/RESET voltages. This means that the device may be volatile and exhibit transient behaviors when resting at low bias after a SET/RESET event. We measured such transients at different temperatures, and results are shown in Fig. 3.5(a) and (c). After RESET at 0.8V, the bias is lowered to 0.1

V, and resistance jumps due to the diode like characteristic in the OFF state. However, resistance drops to the equilibrium value before RESET after some time, and the decay time constant shortens with increasing temperature. This demonstrates that the volatile behavior is possibly thermally activated. Using the duration for which the high resistance persists as the time constant, we can make the Arrhenius plot and obtain an activation energy of 0.57 ±0.02 eV, as shown in Fig. 3.5(b). This value is very close to the activation energy of 0.55 0.02 eV that we got previously for the slow recombination events associated with the deep donor levels in CdS [83]. Therefore, it seems that the resistance decay in the switching device is analogous to conductivity rise under photoexcitation, and both are related to the charge state transitions with sulfur vacancy levels. The volatile nature of the device provides further evidence to confirm the DLS hypothesis. Fig. 3.5(c) shows the transients after the device is being SET, and we can observe a comeback in resistance after the SET bias is removed. However, because the resistance in ON state is quite close to that under low bias, the transient is difficult to analyze due to the small signal- to-noise ratio.

86 -J

(a) 5.0 0.1 V 0.8V 0.1V - -- 25C (b) 104 - 30C - 35C 4.5- -- 40C -45C I- i0 -50C 0I 4.0-

C 3.5-

102 3.0- E = 0.57±0.024 eV j 2.5- S 100 200 3010 0.0031 0.0032 0.0033 0.0034 Time (s) 1/T (K)

E I En 30 102 - 25C -- 30C 35C 6 - 40C 45C - 50C 60090 . . 260SO 80 - 300 120 0 60 180 2620 .0 329) 340 0 li Time (s) Raman nrrcm)4

Fig. 3.5. Testing the DLS hypothesis by advanced characterization. (a) Dynamics of resistance after RESET at different temperatures. (b) Arrhenius plot shows resistance decay after RESET is thermally activated with an activation energy of 0.57 + 0.02 eV. (c) Dynamics of resistance after SET at different temperatures. The restoration of resistance is small compared to noise. Plots (a)- (c) are measured on device 190607_2. (d) Raman mapping of 30x30 pm2 device (1905163) area showing ILO peak position differences before and after switching.

In Fig.3.5(d) we show Raman spectra measured across 30x30 pm2 area of an equilibrium-

ON device in both the ON and OFF states, with 1 pm2 spatial resolution set by the excitation spot size. To make possible Raman spectromicroscopy, this device was fabricated with a semi-transparent top contact. The excitation laser wavelength is 532 nm, which is just below the band gap of CdS. This optical excitation may perturb the ON and

OFF states that everywhere else in this report are controlled and measured all-electrically

87 and in the dark, but we are still able to draw clear conclusions. We find that the 1LO Raman peak blue-shifts uniformly by 2.29:1:0.74 cm-' upon switching from ON to OFF, with no evidence of localized features (e.g. filaments) at the spatial resolution of the experiment.

DLS at sulfur vacancies involves a substantial lattice strain, which may be expected to affect the phonon spectra. The blue-shift in the 1LO peak position that we observe upon switching from ON to OFF - i.e. electron injection into CdS -- is consistent with a previous study of the effects of charge transfer on CdS Raman spectra, which study includes results for CdS/MoO3 interfaces [134]. The Raman spectromicroscopy data therefore further support the hypothesis that resistive switching results from ionization state transitions at sulfur vacancies, akin to photoconductivity but engineered in an all-electrical device and without photons.

We collected capacitance-voltage data with the help of Chris Thompson from the IEC at the University of Delaware. Capacitance measurements show that the device has very large capacitance, on the order of 10-6 F/cm2, but the measured values are negative due to small resistance of the device. The LCR meter has an impedance of 2 k, which is large compared to that of the device itself. The first moment of charge shifts to larger values after switching to the OFF state, but the shape of the charge density remains the same. This could be an indication of the change in charge states of the defect levels, but the results are inconclusive.

Device simulation is conducted in SCAPS software, and the distribution of bistable defect states are initiated before performing an I-V sweep. Simulated band structures show that the ON state introduces more band bending in CdS near the junction, and the edge of ionized deep donors is pushed deeper into the material, as shown in Fig. 3.6(a). The extra

88 electrons released by the deep donor levels contribute to conductivity, leading to the

difference in I-V simulations for the two different initializations. Fig. 3.6(b) shows that the

current density for the simulated ON state is indeed higher than that of the OFF state.

However, conventional simulation packages like SCAPS lacks the ability to simulate history dependent defect level transitions. It is also a challenge considering the large band offset between CdS andMoO3, which probably makes tunneling current important.

Because of these challenges, the simulated I-V curves behave as a diode in both ON and

OFF state instead of as a resistor in the ON state, and the on/off ratio is not as large as observed in experiments. We also simulate the as-fabricated state by applying zero bias in determining the metastable distribution of deep and shallow donors. Simulated J-V curve for the as-fabricated device is very close to the ON state J-V. This result is consistent with our experimental finding that the as-fabricated device is in the ON state.

(a) (b)1

c 1.0- - - OFF E ON 2 0.8- - As fabricated E V 0.6 10-9

0.4- C.-10 C0.2-+- 2 - OFF 010 0 0.0 - ON 0 1012.1 0 06 20 4-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 Distance (nm) Voltage (V)

Fig. 3.6. Simulation results on the switching mechanism. (a) Distribution of ground state deep donor configuration as a function of position. The value of 0 means deep donors are fully ionized and convert into shallow donor configuration. (b) Simulated J-V curves for ON, OFF and as- fabricated states. The as-fabricated state is very close to the ON state.

89 Although there are challenges in simulating the resistive switching effect using conventional simulation packages, we can still achieve some informative results that can explain out experimental findings. We find that the I-V characteristics are strongly affected by electron affinity of theMoO3 layer.MoO3is found to exhibit a large electron affinity of

6.7 eV, which is the reason why it acts as a hole injection layer in organic solar cells [135].

However, study has also found that the electron affinity of thermally evaporatedMoO3, when exposed to air for a few minutes, reduces to 5.5 eV [131]. Adjusting the electron affinityof MoO3 in the device simulation to the reduced value results in large ON/OFF current ratio, as shown in Fig. 3.7(a). The ON and OFF current densities are higher in this case because the junction barrier is significantly reduced compared to the previous simulation. The increased ON/OFF ratio is partly due to a shorter ionized width of deep donors in the OFF state and a slightly larger difference in ionized width between the OFF and ON states. Similar to the previous case, the as-fabricated state still shows J-V characteristics closer to the ON state. However, if we further reduce the electron affinity of the hole injection layer to 4.7 eV, which is the value for ITO, there is no apparent difference between the ON and OFF state current, as shown in Fig. 3.7(b) [136]. This is because the electron affinity of ITO is so low that the band bending in CdS is not enough to allow interactions between the deep donor levels with bands in ITO. This result is consistent with experimental observations (c.f Fig. 3.3(b)).

90 (a)10-2 (b) 107 - OFF 41 - OFF E ON E 10°- ON 1- 4 As fabricated 105 E E

0 10-84- 104

102 10 __0 101

10o-2 U 10° '0'1 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 Voltage (V) Voltage (V)

Fig. 3.7. Simulation results on the switching mechanism with lower electron affinity of the hole injection layer. (a) Simulated J-V curves for ON, OFF and as-fabricated states using 5.5 eV as the electron affinity of MoO3. (b) Simulated J-V curves for ON and OFF states using 4.7 eV (for ITO) as the electron affinity of the hole injection layer.

3.3.4 Pulsed voltage measurements

Pulsed voltage measurements are conductive with a random function generator to supply voltage pulses. Fig. 3.8(a) shows the results of the measurement, where the device shows incremental SET and a rapid RESET. This observation is partly consistent with our observation in the I-V curves, where the RESET process is instantaneous at some voltage threshold, while the SET process occurs during voltage sweep within a range. Fig. 3.8(b) shows the pulsed measurement results involving more switching events. It seems that successful switching rate decreases with pulse number, and the device is easier to stuck in the OFF state as pulses continue.

91 (a) - - .- . .

Z10-

10-4 % M WJ;

0.6 - 0.6 0.4 0.4 0.2

57 0.0

-0.2 •••••••-- _- 0.0 -0.4 -0.2 0 20 40 60 80 100 250 500 750 1000 1250 1500 1750 2000 2250 Pulse Number Pulse Number

Fig. 3.8. Pulsed voltage measurements of the resistive switching device. Sample name: 190516_2. (a) Pulsed measurements involving the first few switching event. (b) Pulsed measurements with more switching cycles. The device tested in this measurement has an area of 0.1 mm 2. The pulsing voltages are shown in the figure below the current measurements. Two thresholds are set to identify whether the device is in ON or OFF state, and the pulse polarity reverses when the device changes state.

3.3.5 Device endurance

Device endurance test was done by repetitively cycling the device under voltage sweeps.

Fig. 3.9(a) shows the I-V curves of the resistive switching device for 100 cycles. It can be seen from the figure that both SET and RESET voltages are stable for the initial-20 cycles and then start to drift to higher absolute values and ON/OFF ratio decreases as cycling goes on. Fig. 3.9(b) plots the SET and RESET voltages against cycle number to more clearly show the drifting. The device stuck in the ON state in the end because the RESET voltage drifts beyond 1 V. This phenomenon might be related to the drift and diffusion of sulfur vacancies under electrical bias. Although sulfur vacancies do not directly affect switching by forming conductive filaments, it is possible for them to move under bias to gradually form a sulfur vacancy rich area near the MoO3/CdS interface. When this happens, the

92 ON/OFF ratio decreases and switching voltage increases because it is more difficult

(requires higher voltage) to create a difference in the width of ionized donors.

(a) (b) 1.00

0.50 ( 10-0.25 10-4 1 0.00 10-s yca4 -0.25- 10-6 ccieo -0.50 1 -80 -0.75

-1.00______-1.00-0.75-0.50.-0.250.00 0.25 0.50 0.75 1.00 - 0 20 40 60 80 100 Voltage (V) Cycle No.

Fig. 3.9. Endurance test for device. Sample name: 190516_2. (a) I-V curves of the resistive switching device for 100 cycles. (b) Drift in SET and RESET voltages with cycle number. Device area is 1mm 2 and cycling rate is 0.2 V/s.

3.4 Discussion

Two-terminal resistive switches described to-date primarily fall into two categories: those based on mass transfer, i.e. ion diffusion and drift, and those based on collective phase transitions, i.e. martensitic transformations, metal-insulator transitions. DLS is distinct because it relies on independent ionization events at point defects. Compared with the switching devices based on the unpredictable growth of conductive filaments, the non- diffusive mechanism means greater predictive design and reliability. Also, the non- collective response implies a generalizable mechanism, which is not limited to transitions that naturally occur in technologically-relevant operation windows.

The underlying mechanism of bistable defect transition levels is known in multiple material systems. Lattice relaxations around anion vacancies have been shown to be able to create metastable defect transition levels in n-type II-VI semiconductors and p-type Cu-III-Se2

93 chalcopyrites [75]. DX-centers in AlGaAs possess similar bistable transition levels leading to large conductivity difference [61]. The universality of bistable defect levels suggests that DLS is generalizable, and could be applied to various photoconductive materials to design resistive switches. The potential application of DLS mechanism adds an extra dimension to point defect engineering, a mainstay of semiconductor physics and device engineering. Materials selection combined with advanced TCAD could enable engineering normally-on and normally-off devices, as well as devices with varying degrees of response and decay times.

3.5 Conclusion

In summary, we designed an all-electric resistive switching device exploiting the mechanism of photoconductivity in CdS. The device introduces aMoO3 layer to inject holes into CdS to trigger sulfur vacancies to switch between deep and shallow donor configurations. The device is in low resistance state as fabricated, and shows repeatable resistance switching behavior with no electro-forming process. The DLS mechanism is confirmed through material processing and substitution, as well as through advanced characterizations. Results show that the resistive switching behavior is not due to growth of Ag or sulfur vacancy filaments, and is analogous to the mechanism of photoconductivity in CdS. Numerical simulations also predict similar switching behaviors as observed in experiments. The DLS mechanism suggests a new way to control conductivity in electronic devices. We expect this mechanism to be also applicable to other material platforms with multi-stable defect levels to design new resistive switching devices.

94 Chapter 4. Effect of Contact Layer Photoconductivity on Solar

Cell Performance

4.1 Introduction

4.1.1 The role of carrier-selective contacts in solar cells

The majority of photovoltaic (PV) solar cells produced worldwide this year are based on crystalline Si (c-Si) homojunction technologies [137]. However, there are many more and different PV technologies based on heterojunctions than on homojunctions. All thin-film solar cells (CIGS, CdTe, halide perovskite, etc.) are heterojunction technologies, as are advanced c-Si technologies that are expected to enter production in the coming decade [138,139]. Design rules for optimizing heterojunction PV performance are therefore broadly important. All heterojunction solar cells employ at least one carrier- selective contact (CSC) which is responsible for collecting charge carriers of one type, repelling the opposite type, and suppressing non-radiative recombination at the interface.

Design rules for optimizing CSC materials selection have been long-established [40,140].

In this work we elaborate on a detail of heterojunction solar cell design: CSC layer photoconductivity.

4.1.2 Photoconductivity in carrier-selective contact layer materials

Large and persistent photoconductivity is observed in many CSC materials, including II-

VI semiconductors and amorphous Si (a-Si) [60,70,141,142]. The large photoconductive response in these materials is due to defects that trap minority carriers and enhance the lifetime of photoexcited majority carriers. Large and persistent photoconductivity is

95 typically associated with lattice distortions and/or atomic diffusion following charge-state transitions.

Theoretical and experimental studies indicate that in II-VI semiconductors, the photo- induced lattice distortions responsible for large and persistent photoconductivity occur at anion vacancies [74,75,83]. For instance, in CdS at equilibrium in the dark, sulfur vacancies are neutral deep donors, and the cations surrounding a vacancy distort inwards, attracted by the unbound electrons at the vacancy. When an anion vacancy is ionized by photoexcitation, the surrounding cations are no longer screened from each other and are repelled outwards. This distortion is a metastable state and presents an activation barrier to recapturing the photoexcited electrons; the result is a prolonged excited majority carrier lifetime.

Light-induced persistent metastable states are also observed in a-Si, which is used for CSCs in c-Si heterojunction PV (i.e. heterojunction with intrinsic thin layer, or HIT cells). In a-

Si the metastability arises in part from photo-breaking of weak Si-Si bonds, creating dangling bonds that are deep traps [63-65]. In a-Si there is also evidence that hydrogen diffusion stabilizes light-induced metastable lattice configurations, enhancing persistent photoconductivity; similar mechanisms have been reported for (Zn,Mg)O and for

SrTiO3 [52,66,67,143].

Other materials showing significant photoconductivity and used for CSCs or passivation in c-Si PV include TiO2and SiNx. The origin of large and persistent photoconductivity in these materials is inconclusive. TiO 2 is known to feature large electron-phonon interactions, and minority holes may be trapped as small polarons [68,69]. Grain boundaries or oxygen- impurity defect pairs may also be responsible [70,71]. Persistent photoeffects in SiNx are

96 also observed but have been less-studied than the other materials mentioned here. One

possibility is that holes are trapped at dangling bonds, while electrons conduct via phonon-

assisted hopping [144,145]. The salient point is that, in all of these materials,

photoconductivity can be large and persistent due to photo-induced lattice distortion

leading to metastable atomic configurations.

4.1.3 Contribution of CSC photoconductivity to light soaking-effects in

Light soaking refers to a collection of disparate effects by which the performance of solar

cells changes with the duration of light exposure. Light soaking effects can be reversible

or irreversible, beneficial or deleterious. When reversible, light soaking is often ascribed

to metastable configurations and slow equilibration of deep levels in the absorber [146-

149]. In CIGS and CdTe devices it is commonly observed that reversible light soaking

improves device performance, with a stabilization time from minutes to

hours [143,146,147,150-153]. Our results here build on previous studies that focused

attention on metastable effects in CSC layers and the effect of light-induced electronic

interface passivation [55,143,150,151].

We show that, for the case of photoconductive n-type CdS used as a CSC, the efficiency

of CIGS or CdTe devices could be improved by over 4% (absolute) by optimizing CdS photoconductivity within an achievable range for devices with interface defects. We then

survey the published literature on photoconductivity for a number of common CSC materials, most of which are highly-photoconductive. This survey suggests that persistent photoconductivity in CSC layers may be responsible for device improvement by light-

97 soaking in many PV technologies. Optimizing CSC layer photoconductivity through

materials processing may be a general route to improving PV performance.

4.2 Methods

4.2.1 Solar cell modeling

We use SCAPS to numerically simulate the effect of CSC layer photoconductivity on the

performance of thin-film solar cells [132]. SCAPS uses finite-element methods to solve

the Poisson equation and carrier continuity equations to calculate steady-state band

diagram, recombination profile, as well as carrier transport in one dimension. To keep the

models simple, we include one neutral defect level in each of the layers. Interface defects

are ubiquitous in real-world devices, but here we consider models with and without

interface defects to better understand the effect of CSC layer photoconductivity. We do

not input additional series or shunt resistance in our simulations, to maintain focus on

effects of CSC layer photoconductivity on the p-n junction. We use the AM 1.5 spectrum

for illumination, and we fix the temperature at 300 K. Our baseline parameters for

modeling CIGS and CdTe solar cells are summarized in Table 4.1. We introduce

photoconductivity in the CSC layer by tuning the shallow donor density, in accordance

with our understanding of large and persistent photoconductivity in CdS and

ZnO [74,75,83].

For CIGS solar cells, the structure is back contact/p-CIGS/n-CdS/n-ZnO/top contact. Here, p-CIGS is the absorber, n-CdS is the electron CSC, and n-ZnO is the transparent conductive

oxide (TCO). Some studies consider an ordered vacancy compound (OVC) layer between

CIGS and CdS to reduce recombination at the CIGS/CdS interface [154,155]. For

98 simplicity and clearer interpretation of our results, we do not explicitly consider this here.

The effect of an OVC is to reduce the interface defect density, and therefore the effect of

CSC photoconductivity on device performance in the presence of an OVC can be qualitatively assessed by comparing our results with and without interface defects. For

CdTe solar cells, the structure is back contact/p-CdTe/n-CdS/n-SnOx/top contact. Here,

SnOx is the TCO. For both CIGS and CdTe devices we assume 95% optical transmission through top contact and 80% reflection from back contact.

4.2.2 Surveying CSC layer materials

Heterojunction PV technologies (both thin-film and c-Si) employ a range of different CSC materials. The following materials are commonly-used as CSCs and exhibit substantial

photoconductivity: CdS, ZnS, and (Zn,Mg)O for CIGS and CdTe PV; TiO 2 for a-Si, c-Si, dye-sensitized, and halide perovskite PV; ZnO for organic PV; CuO for halide perovskite

PV. The simulation results below focus on CdS used in CIGS and CdTe PV. Then, our survey of previously-reported photoconductivity in this larger list of CSC materials suggests that the opportunities for improving performance that our simulations suggest for

CIGS and CdTe may be present more generally in heterojunction PV technologies.

99 Table 4.1. Baseline parameters for simulating CIGS and CdTe solar cells

CIGS CdS ZnO CdTe SnOx Layer Properties

Thickness 2000 50 50 4000 500 (nm) Eg (eV) 1.2 2.4 3.3 1.5 3.6 X (eV) 4.5 4.3 4.45 4.0 4.4 E'r 13.6 9.0 9.0 9.4 9.0 Nc (cm-3) 2.2x1018 2.2x1018 2.2x1018 8x1017 2.2x1018

N , (CM-3) 1.8x 1019 1.8x 1019 1.8x 1019 1.8x 1019 1.8x 1019 Ve (cm s-') 3.9x107 2.6x107 1x107 1 x107 1x107

Vh (CM s-') 1.4x 107 1.3 x107 1 X107 1 X 107 1 X107

2 Ie(cm V- 100 1.57 100 320 100 s- ) _ (cmV 25 1 25 40 25

ND (cM3 ) x 1017 1x1018 1x 1017

NA (cm-3) 1 x1016 5x1014

Bulk Defect Properties

AEc (eV) 0.6 1.2 1.65 0.75 1.8 Nt (cm-3 ) 1 x101 5x1016 1x1016 2x 1014 1 x1015

Ue (cm2) 5x10-15 1x10-16 1x10-15 5x10-13 1 x10-12

2 Uh (Cm ) 1x10-13 1 x10- 5x10-13 1x10-3 1x10-

Interface Defects CIGS/CdS CdTe/CdS

Position middle of interface gap middle of interface gap Nt (cm-2) 102 1012

Ue (cm2) 10-" 10-"

100 4.2.3. Simulation for non-contact measurements of photoconductivity

To suggest a plausible non-contact method to measure photoconductivity of the contact layer in a solar cell stack, we investigate the reflectivity change due to photoconductivity of the contact layer. Reflectivity simulations are performed using the RefFIT software [156]. We use the Drude-Lorentz model to simulate the dielectric function of CdS.

The model simulates the response of a set of harmonic oscillators and is given by

2 2

e()=e + ) 2 - 2 where e is the "high-frequency dielectric constant", which approximates the response of oscillators at very high frequencies. wp, ooi and yj are the plasma frequency, eigen- frequency and scattering rate. For the Drude response, toi = 0. In particular, we are interested in the Drude response of CdS because (1) it contributes mainly to reflectivity at low frequencies, which reduces overlapping with other oscillators, (2) the Drude response depends on carrier concentration. The plasma frequency can be calculated from (, =

e2m . Here, n is the electron density, e is unit charge, Er is relative permittivity of CdS,

EO is the permittivity of free space, and m* is the effective mass of electrons in CdS. The scattering rate y can be calculated from y = -- , where p is the electron mobility. Normal- Am incidence reflectivity R(o) can be calculated given dielectric functionE(O), by R(W)=

The electron density in the calculation of plasma frequency is varied from 1010 -

1020 cm, according to our knowledge of photocarrier concentration in CdS obtained in

Chapter 2.

101 For CIGS, we use two Tauc-Lorentz oscillators to model the dielectric function [157]. The

Tauc-Lorentz oscillator is given by the following equations:

1 Awoy(w - ()g) 2U(- Wg) 2 2 E2= - + u -)

f°w'E2(w')d' €1(2 (2= co +12 2 r 0 W -

Here, wo is the peak transition frequency, w is frequency corresponding to optical band gap, u(co) is the step function. A, &,,wo, y are all fitting parameters. The model is used to fit extinction coefficient data derived from an absorption model:

)( Eg) hv a (hv) =ao + plo hv/ Eg 1

Ic=- 4w

The parameters used in this model are ao = 1.095 x 10s cm-' , and f 0 = 1.0 x

10-12 cm-1. Table 4.2 summarizes the parameters fitted for dielectric function models of

CdS and CIGS.

Table 4.2. Parameters for the simulation of dielectric function of CIGS CIGS op/cm~ 1 y/cm~ 1 Drude 38.02 3352.4 1 CIGS w0 /cm-' wg/cm-1 A/cm- y/cm-1 Tauc Lorentz 1 48529.19788 8550.046334 514955.2651 198135.0647 Tauc Lorentz 2 8360.233689 9380.133737 3338011.305 602.283043

102 Reflectance of CIGS/CdS multilayer is then simulated using the method described by

Harbecke [158]. This approach accounts for multiple reflections between layers, and is also able to take into account the coherent and incoherent propagations within individual layers.

A thickness of 30 nm is used for CdS layer, and 2 pm for CIGS layer.

4.3 Results

4.3.1 The effect of CSC photoconductivity on CIGS and CdTe solar cell performance

We first study the effect of CSC layer photoconductivity on the performance of CIGS and

CdTe solar cells without interface defects. In previous work, we showed that the photosensitivity of CdS thin films made by chemical bath deposition (CBD) can be increased to PS = 109 by changing synthesis parameters; PS =alight/Udarkwhere odarkand alight are the steady-state film conductivity in the dark and under AM.5 illumination, respectively [83]. For our films, this corresponds to a carrier concentration between 1010-

1019 cm-3. We simulate this effect by varying the shallow donor density (SDD) ofCdS within this range, and we present the results in Figures 4.1-4.2. For CIGS solar cells, the

J-V curves shift towards higher voltage and current as we increase the CdS SDD (Fig.

4.1(a)), and the efficiency improves from 15.46% to 18.30%. The efficiencyjumps notably when the CdS SDD increases from 1016 to 1017 cm-3 , and it achieves a maximum for CdS

SDD = 1018 cm-3 . The efficiency gains mirror the fill factor. Voc is little-affected by CdS

SDD, whereas Jsc is decreases somewhat with increasingCdS photoconductivity.

The trends in cell performance with CdS photoconductivity can be understood by considering the device band diagram, shown in Fig. 4.1(c). Higher SDD in CdS creates a

103 more one-sided junction and increases the band bending within the CIGS absorber. This pushes the depletion region deeper into CIGS, which is desirable for heterojunction solar cell performance because it facilitates charge separation without non-radiative recombination; this is a form of electronic (as opposed to chemical) interface passivation [140]. Fig. 4.1(d) shows that increasing CdS SDD substantially reduces the interface recombination rate by moving the position of peak recombination into the CIGS layer. Although this does increase the recombination rate within the CIGS layer, the overall effect is a performance improvement.

The effect of CSC layer photoconductivity on the p-n junction is the dominant effect on device performance, and the contribution of the CSC layer to the series resistance is a secondary effect. Taking for instance material properties at flat-band conditions, the series resistance contribution of our modeled CdS layers is below 0.1 Q cm2 for ND > 10 cm-3 .

Therefore the trends reported here - and in particular the performance improvement for

ND > 10' cm-3 - are not due to a simple resistance in series with the p-n junction. Our results are consistent with previous reports of electronic interface passivation by CSC layer photoconductivity in CIGS, CZTS, and CdTe solar cells [55,143,150,151].

104 (a) 10- 0. 2 33-

0- 101 cm4 3 >0.68 -10- 10 cm- E 0-66 3,8111 3 1 s 16 17 18 19 31.0 10"1010610710o18119 1017 cm- 10 01 10 1 0 10 10 -20- 19 E 101 cm-3 -13 iOW9cm-3 80 18 -30- 2751a 17 - 70 C" 16 , -404- - . . . . . - . I . . . 65 aid 11I 1smilli 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0. ' 60 14 1010101h1010'710''01 10'°101510110171018101 Voltage (V) 3 N. (cm ) N 0 (cm-) (c) (d ) 5. CIGS CdS °cm-a 1022 --- 10 E - -101 *cm-a 5-...... E~n 0. 0 Q) 0. 1021 E F a ~~r -0. 0- -.... .2E0 -1. -1. 5 ------100°cm-a ---. E -2. 0. -- 1011*CM- . • -0.2 -0.1 0.0 -0.2 -0.1 0.0 x (pm) x (gm)

Fig. 4.1. Simulation results on the effect of photoconductivity in CdS layer on CIGS solar cell performance under AMI.5 illumination. (a) Simulated J-V curves for different photocarrier concentrations in CdS. (b) Changes in Voe, J., FF and r with photocarrier concentration in CdS. (c-d) Effect on CdS SDD on device band structure and recombination rate for the cases SDD = 1010 and 101 cm'. x is the thickness through the device, the p-n junction is at x = 0. (c) Higher photocarrier concentration in CdS results in more band bending in CIGS and creates a deeper depletion region. (d) Photoconductivity in CdS reduces recombination rate at thejunction interface; we show simulation results for the maximum power point.

We find similar results for CdTe solar cells. CdTe typically has a carrier density on the order of 1014 cm-3, lower than CIGS. Therefore, photoconductivity in CdS can push the depletion region even deeper into the absorber than for CIGS. In Fig. 4.2(a) we show J-V curves simulated for varying CdS SDD. As for CIGS solar cells, the FF increases with CdS

105 U!

SDD. The device efficiency improves by 1.81% (absolute) as the CdS SDD increases from

1010 to 10'9 cm-3 . The recombination rate near the maximum power point and at the interface is actually enhanced for the case of highly-photoconductive CdS, but the total non-radiative recombination through the cell is suppressed, leading to better performance.

(a) (b) O • 0-90 24.5

0------1 - - 24.0 3 0.86 EE E -5- -- 105 cm- se23.5UU.- N 0.84 U I 10 1cm- -10- 3 1010101016101710 10 7cm~3 10 2 1011011011017101810 -15- 3 80 17 10 1cm 1m- 75 16 10o" -20- cm-a -70 15 -25 65 14 ).0 0.2 0.4 0.6 0.8 1.0 60 13 i10o1016¶01710181019 Voltage (V) N (cm4 4 0 ) N 0 (cm ) (c) (u) 1022 CdTe .S I. 'co 1.5 - 100 cm-3 1.0 E >0.5 EFn 1021.: 0 0.0 Z) -0.5 w 8~ 1020 12

.....1 10 C -3

-2.01 1019 -0.4 -0.3 -0.2 -0.1 0.0 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 x (Pm) x (pm)

Fig. 4.2. Simulation results on the effect of photoconductivity in CdS layer on CdTe solar cell performance. (a) Simulated J-V curves for different photocarrier concentrations in CdS. (b) Changes in Ve, J., FF and r with photocarrier concentration in CdS. (c-d) Effect on CdS SDD on device band structure and recombination rate for the cases SDD = 1010 and 101 cm-3. x is the thickness through the device, the p-n junction is at x = 0. (c) Higher photocarrier concentration in CdS results in more band bending in CIGS and creates a deeper depletion region. (d) Photoconductivity in CdS reduces recombination rate in the bulk of CdTe, but effect is reversed near junction interface.

106

- We next consider cases including interface defects between the absorber and the n-type

CdS. For this study we use single-level defects at the middle of the interface gap. We show in Fig. 4.3(a-b) that CdS photoconductivity can substantially improve Vo and the efficiency of both CIGS and CdTe solar cells. The efficiency increases by up to 4.10% and

4.38% absolute for CIGS and CdTe, respectively, as the CdS SDD varies between 1010-

1019 cm-3 .

Interface defects at the p-n junction in heterojunction solar cells reduce Voc by accelerating nonradiative recombination at the interface. This effect can be ameliorated by moving the ambipolar region of the device away from the interface through higher doping of one side, resulting in a one-sided junction [140]. This well-accepted solar cell design guideline well- explains the results that we show here. For CdS CDD higher than 1017 cm-3 for CIGS (10'9 cm 3 for CdTe), increasing interface defect density no longer has an effect on Voc and cell performance; the junction is fully one-sided, and the interface is effectively passivated.

107 (a) 10- ()5-

0------0------101°0cm 3 -- ---10' cm-3 4 -5- E -10- 1016 cm E 4 - 1017 cm -10- -1015 Cr-a -11 --i0" cm4 4 I E -20- E -- 10" cm - 1018 cm-a -15-

101 cm-a - -30- 1019 cm -20 -

-1n -25 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0 0.2 0.4 0.6 0.8 Voltage (V) Voltage (V) (C) (d) 3 --- 100° cm-a -- 101f cm 8S1022 -- 1018 cm-3 1022 101 8m-a

0S 1021 E 8o~ ~1020. La0'. E zzzzz~ 8 0 Ix 101 -0.2 -0.1 0.0 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 x (Pm) x (gm)

Fig. 4.3. Simulation results for CIGS and CdTe solar cells with absorber/buffer interface defects at a concentration N = 1012 cm-2 (see Table I). J-V curves for (a) CIGS and (b) CdTe solar cells. Total recombination rate as a function of thickness through the device for (c) CIGS and (d) CdTe solar cells. x is the thickness through the device, the p-n junction is at x=0.

4.3.2 Photoconductive response of common n-type CSC materials

In Table 4.3 we summarize the experimentally-measured photoconductivity of common

CSC materials: ZnS [159-163], ZnO [141,164-168], (Zn,Mg)O [143,169,170],

CuO [171-176], TiO2 [71,177-181], and a-Si [105,182-186]. This is not a comprehensive list of all CSC materials commonly used in solar cells, because for some such materials including spiro-OMeTAD, CdO, Zn(O,S), and MoO3 the photoconductivity

108 Table 4.3. Summary of photoconductivity ranges for common CSC materials. Results summarized for different illumination conditions: A = AM1.5, B = Hg lamp, C = UV lamp, D = Xe lamp, E = UV LED, F = UV laser, H = Halogen lamp, I = Nd-YAG laser, J =Red LED, K = White LED. The * label indicates filtered or monochromatic UV. Carrier Highest Equilibrium Material (c 2 V- -) photosensitiv carrier densitunder Illumination (cm 2~) ity density (cm-) illmnaio(cm-3)

ZnS [159 102 105 10'-1014 1013-1018 B, D*

ZnO [141,16 102 104 10-1 104-1018 B, B*, D*, E 4-168,188]

(Zn,Mg)O [1 1 104 10"-01 17-0V A, B, C 43,169,170]

CuO [171- 1 <101 10"4-10"5 14-10 A, F, H 176]

TiO 2 [71,177 10-2 108 108-1012 101O-1018 B, D, E -181,189] a- Si [105,182- 10-1 106 1010-1012 1012-1017 F, H, I, J 186]

CdS [83] 1 10101-02 1010-1019 A, K

109 -I

E 1019 E 1 17 o1510

C 010 -3 0 10 E109 Dark 7 M Light

ZnS ZnO ZMO CuO TiO2 a-Si CdS

Fig. 4.4. Estimated steady-state carrier concentrations in the dark and under illumination for various and common CSC materials; assumptions and illumination conditions are summarized in Table 4.2. ZMO = (Zn,Mg)O. of thin films has not been widely-studied. In most cases papers report conductivity values, and we assume reasonable values for the mobility to determine carrier concentration; we list these assumed mobility values in Table II [83,176,185,187-189]. In Fig. 4.4 we visualize the range of majority carrier concentration achievable between dark and AM1.5 illumination. We see that for all n-type materials the electron concentration can rise to 1017 cm-3 under illumination, and that ZnS, ZnO and TiO2 can all achieve 1018 cm 3 or higher.

TiO2 shows the highest photosensitivity (c.f Sec. 4.3.1). Referring to our simulations of

CIGS and CdTe devices that showed substantial improvement for n-type CSC majority carrier concentration above 1016 cm 3, this survey suggests that similarly substantial effects may be found with other PV technologies and other CSC materials.

110 4.4 Discussion

4.4.1 Opportunities for solar cell engineering: CSC photodoping and material processing

Basic research and development in PV often focuses on improving absorber materials. For

CIGS, this work includes band gap engineering by Ga composition grading and alloying with Ag [41,190-193]. The beneficial effect of increasing the majority carrier concentration in CSC layers has long been recognized, but chemical doping studies on CSC layers are typically intended to improve properties such as interface stability and band offsets but not to increase the carrier concentration in the CSC itself [42,194-196]. This is sensible because many candidate elements for extrinsic shallow donors in II-VI n-type

CSCs may also be compensating donors in p-type CIGS and CdTe, and therefore should be avoided. Chemical doping may also introduce additional, unwanted deep level defects.

We suggest that this conundrum can be avoided by focusing on photodoping (i.e. large and persistent photoconductivity), which can greatly increase majority carrier concentration apparently without drawbacks.

The large photoresponse of the CSC materials surveyed in Fig. 4 is caused by point defects, such as anion vacancies in II-VI materials and dangling bonds in a-

Si [74,75,83,55,142,66,67]. In our previous experimental work on CdS, we found that simple changes to the bath chemistry during CBD can result in substantial changes in the sulfur vacancy concentration and the photoconductivity [83]. The role ofCdS point defects in the performance improvement of CIGS solar cells during light soaking has been numerically studied, and a similar approach to optimizing PV performance by controlling

CdS point defects during CBD has been demonstrated experimentally for CZTS solar 111 cells [55,151]. Simple changes to processing parameters during deposition of other CSC materials and by other methods can also strongly-affect the concentration of defects responsible for photoconductivity. For instance, the photoconductivity of ZnO is expected to vary with oxygen vacancy concentration, for amorphous indium-gallium-zinc-oxide

(IGZO) it can be tuned by metals composition, and for a-Si the photoconductivity (albeit not necessarily the photosensitivity) can be tuned by changing the density of dangling bonds [74,75,197,198]. This emphasizes the importance of CSC processing conditions on solar cell performance, and the potential device improvements that could be achieved by relatively simple changes to processing parameters.

4.4.2 Measurement challenges

The photoconductivity of bare films deposited on insulating substrates is relatively easy to measure. However, it is much less straightforward to measure the photoconductivity of a thin over- or under-layer in a device structure. Due to the sensitive dependence of photoconductivity on processing conditions, it is important to characterize the photoconductivity of CSC layers in real devices. For a substrate-style CIGS solar cells, the n-type CSC (typically CdS) is a layer approximately 30 nm thick deposited on top of the absorber, which is typically several microns thick and rough. Direct-current electrical measurements of the CSC layer conductivity are not possible, and indirect measurements

(e.g. from device series resistance) are open to interpretation. We suggest that this challenge could be resolved by non-contact microwave reflectivity measurements, such as microwave photoconductivity in the frequency range of 1-10 GHz. Such measurements are currently under-development in our research group.

112 Our simulation results suggest that the experimental signature of contact layer conductivity is enhanced at microwave frequencies. For conductivity values realistic for photo-sensitive

CdS under AMI.5 illumination, the experimental signature is large below -30 GHz. Fig

4.5(a) shows the simulated reflectivity curves for CIGS/CdS stacks with different carrier concentration in CdS, varying from 1010-1019 cm- 1. A dashed vertical line is plotted at 8.3

GHz where the difference between reflectivity is quite significant. This suggest that the stack reflectivity in microwave range could be a good indicator for differentiating CSC layer photoconductivity. We propose to measure this effect using cavity microwave spectroscopy with chopped UV illumination and lock-in discrimination. In Fig. 4.5(b), we propose a design of a cavity for the microwave reflectivity measurements, and simulate the microwave field profile at 8.3 GHz. Simulation results suggest that the resonance frequency changes monotonically with SDD in CSC layer, and the difference could be readily detected by a lock-in measurement.

(a) 1.o 11m 4 (b) -- 17 cm~3 101 CM-3 4 0.8- - 101 cm -- 10'scma

0.6

0.4

0.2 10- 10-1 100 10' 102 103 Wavenumber (cm-)

Fig. 4.5. Microwave photoconductivity experiment design. (a) Calculated reflectivity of CIGS semi-fabricate with varying contact layer SDD (cm-). (b) Simulated microwave field profile at 8.3 GHz (indicated by vertical dashed line in (a)) for cavity designed for the proposed work. (Inset) Drawing of sample in designed cavity, including feedthroughs for microwave antenna and sample illumination.

113 4.5 Conclusion

We present numerical simulations predicting that the performance of CdTe and CIGS solar cells could be improved by over 4% (absolute) by optimizing CdS photoconductivity within an achievable range. For simulated devices without interface defects, the effect of

CSC layer photoconductivity is most pronounced in the FF. When interface defects are included, we find that substantial improvement is possible in both FF and Voc. The effect of CSC layer photoconductivity is to push the depletion region farther into the absorber, in agreement with longstanding guidelines for optimizing heterojunction solar cells by electronic interface passivation [140].

We survey a range of CSC materials and find that many are highly photoconductive.

Therefore, the lessons presented here for CIGS and CdTe devices may apply to a broad range of heterojunction PV technologies. CSC photoconductivity can be thought of as photo-doping to enhance CSC shallow donor/acceptor concentration while avoiding the pitfalls of chemical doping. Photoconductivity is highly sensitive to material processing, and therefore it may be possible to optimize CSC photoconductivity for improved device performance with relatively simple changes to existing processing methods. We emphasize that testing this hypothesis experimentally in real solar cells is a metrology challenge, which may be addressed by measurements of microwave photoconductivity.

114 Chapter 5. Summary and Outlook

5.1 Summary

In summary, we systematically study the cause of LPPC inCdS and provide experimental

support for the negative-U model. We then find two new applications for LPPC, to design

novel resistive switching devices, and to optimizing contact layer photoconductivity in thin

film solar cells.

LPPC is due to the trapping of photo-generated minority carriers at crystal defects. We

demonstrate that the photoconductive response of CdS thin films can be widely tuned by

varying the bath chemistry during CBD. We show that large photoresponse and slow decay

are correlated with sulfur deficiency. Both linear and quadratic electron-hole

recombination are relevant, and the rate-limiting process during photoconductivity decay

is thermally-activated recombination. These observations provides experimental validation

for the model that anion vacancies in II-VI semiconductors are negative-U defects, exhibiting strong hole-hole correlation due to lattice distortions that respond to changes in the vacancy charge state.

We then apply our understanding of the mechanism of LPPC to design novel electrical devices. The DLS mechanism suggests a new way to control conductivity in electronic devices by switching defects between deep and shallow configurations. We fabricate and test vertical thin-film devices consisting of Ag/CdS/MoO3/Au layers. MoO3 is a hole- injection layer, due to its large electron affinity and nearly type-III alignment with CdS.

We hypothesize that M003 extracts electrons from the deep donor levels in CdS (possibly via the valence band), and thereby switches CdS into a "photoconductive" state in a thin

115 layer near the interface. The device is in low resistance state as fabricated, and shows repeatable resistive switching under electrical bias with no electro-forming process. We conduct mechanism studies by material substitution as well as more advanced characterizations. Results suggest that the observed behavior is due to donor-level switching, rather than mass transport (e.g. filament formation). We further support the hypothesized defect-level switching mechanism with numerical device simulations. We expect this mechanism to be also applicable to other photoconductive materials with bistable defect levels, and could be used to design other resistive switching devices.

LPPC also finds its application in heterojunction thin film solar cells. LPPC is pronounced in n-type CSC materials in thin film solar cells; such materials include CdS, ZnO,

(Zn,Mg)O, Cd(,S), and Zn(,S). We use numerical modeling to estimate the effect of

CSC photoconductivity on the performance of CdTe and CIGS solar cells. We simulate

CSC large and persistent photoconductivity by switching a population of defect levels between deep and shallow donor states, in accordance with current theoretical understanding and using published experimental results to constrain the parameters. We find that CSC photoconductivity can substantially affect solar cell performance.

Particularly, the power conversion efficiency of both CIGS and CdTe solar cells can be improved by over 4% (absolute) by optimizing the photoconductivity of CdS layer. Higher shallow donor density increases the band bending of the absorber near the interface and reduces carrier recombination, which is the primary cause of the improvements. The generality of this effect in various CSC materials suggests that optimizing CSC photoconductivity may be an effective method in solar cell engineering across multiple thin-film and wafer-based platforms.

116 Some notes on the sample names are mentioned here. The naming scheme starts with the finish date followed by a sequence number. Details on the samples can be found in the lab notebook. Measurement data (e.g. I-V data) files also contain the corresponding sample names.

5.2 Outlook

We propose several tasks which could further the studies presented in this thesis. To better understand the mechanism of persistent photoconductivity in CdS, we could benefit from complementary spectroscopic experimental studies and in-depth theoretical modeling of the kinetic pathways for linear and quadratic recombination. Another project aimed at better understanding carrier dynamics in semiconductors in general is to build algorithms to automatically select processes involved in experimental data. This requires using the sparse selection of parameters using Bayesian methods, and is currently under development.

DLS is a novel mechanism in designing resistive switching devices, and could be applied to various semiconductors exhibiting LPPC, offering wide range of responses by materials selection. Future directions include comprehensive project to realize defect-level switches in other compound semiconductor materials, including normally-on (e.g. for selectors) and normally-off (e.g. for protection circuits) devices. A better modeling technique to handle the metastable configurations of defect levels is also required to further understand the switching mechanism.

Optimizing contact layer LPPC could be effective in engineering thin film solar cells. To experimentally realize this route, we first need a method to measure photoconductivity of the contact layer in a stack of low-resistance materials. Non-contact reflectivity

117 measurement is one possible solution to this issue, and is still under development by our group. A complementary measurement to microwave reflectivity is to use op-XPS to measure the change in Fermi level in contact layers upon illumination. Op-XPS complements microwave measurements because, while it is a less-direct measurement of photoconductivity, it can be used to confirm the microwave results, and yields band edge position data that can be useful in device simulations. Further projects requires developing deposition techniques to synthesize highly photoconductive CSC layers onto absorber materials across different PV platforms, and experimentally verify their effect.

118 References

[1] Effect of Light on Selenium During the Passage of An Electric Current *, Nature 7, 303(1873). [2] R. H. Bube, PhotoconductivityofSolids (Wiley, New York, 1960). [3] R. H. Bube, in Wiley Encycl. Electr. Electron. Eng., edited by J. G. Webster (John Wiley & Sons, Inc., Hoboken, NJ, USA, 1999). [4] R. H. Bube, in Solid State Phys. (Elsevier, 1960), pp. 223-260. [5] M. Q. Tran, On Thermal Quenching of the Photoconductivity in Hydrogenated Amorphous Silicon, Philos. Mag. B 72, 35 (1995). [6] R. H. Bube and E. L. Lind, Photoconductivity of Gallium Selenide Crystals, Phys. Rev. 115,1159 (1959). [7] S. 0. Hemila, Optical Quenching of Photoconductivity in CdS and ZnS Crystals, J. Appl. Phys. 38, 5258 (1967). [8] R. H. Bube, Infrared Quenching and a Unified Description of Photoconductivity Phenomena in Cadmium Sulfide and Selenide, Phys. Rev. 99, 1105 (1955). [9] S. Tutihasi, Quenching of Photoconductivity in Cadmium Sulfide, JOSA 46, 443 (1956). [10] W. Shockley and W. T. Read, Statistics of the Recombinations of Holes and Electrons, Phys. Rev. 87, 835 (1952). [11] R.N. Hall, Electron-Hole Recombination in Germanium, Phys. Rev.87,387 (1952). [12] B. Van Zeghbroeck, PrinciplesofSemiconductor Devices (Bart Van Zeghbroeck, 2011). [13] R. H. Bube, in Encycl. Mater. Sci. Technol. (Elsevier, 2001), pp. 873-879. [14] S. M. Sze and K. K. Ng, Physics ofSemiconductor Devices (John Wiley & Sons, Inc., Hoboken, NJ, USA, 2006). [15] H. Chavez, M. Jordan, J. C. McCLURE, G. Lush, and V. P. Singh, Physical and Electrical Characterization of CdS Films Deposited by Vacuum Evaporation, Solution Growth and Spray Pyrolysis, J. Mater. Sci. Mater. Electron. 8, 151 (1997). [16] S. Ray, R. Banerjee, and A. K. Barua, Properties of Vacuum-Evaporated CdS Thin Films, Jpn. J. Appl. Phys. 19,1889 (1980). [17] B.-S. Moon, J.-H. Lee, and H. Jung, Comparative Studies of the Properties of CdS Films Deposited on Different Substrates by R.F. Sputtering, Thin Solid Films 511-512, 299(2006). [18] K. El Assali, M. Boustani, A. Khiara, T. Bekkay, A. Outzourhit, E. L. Ameziane, J. C. Bernede, and J. Pouzet, Some Structural and Optical Properties of CdS Thin Films Prepared by RF Sputtering, Phys. Status Solidi A 178, 701 (2000). [19] A. Gupta and A.D. Compaan, All-Sputtered 14% CdS/CdTe Thin-Film Solar Cell with ZnO:AI Transparent Conducting Oxide, Appl. Phys. Lett. 85, 684 (2004). [20] A. D. Compaan, A. Gupta, S. Lee, S. Wang, and J. Drayton, High Efficiency, Magnetron Sputtered CdS/CdTe Solar Cells, Sol. Energy 77, 815 (2004). [21] I. Clemminck, M. Burgelman, M. Casteleyn, and B. Depuydt, SCREEN PRINTED AND SINTERED CdTe-CdS SOLAR CELLS, Int. J. Sol. Energy 12, 67 (1992). [22] N. Nakayama, H. Matsumoto, A. Nakano, S. Ikegami, H. Uda, and T. Yamashita, Screen Printed Thin Film CdS/CdTe Solar Cell, Jpn. J. Appl. Phys. 19, 703 (1980).

119 [23] S. Ikegami, CdS/CdTe Solar Cells by the Screen-Printing-Sintering Technique: Fabrication, Photovoltaic Properties and Applications, Sol. Cells 23, 89 (1988). [24] Y. Y. Ma, Properties of CdS Films Prepared by Spray Pyrolysis, J. Electrochem. Soc. 124, 1430 (1977). [25] Y. Y. Ma, A. L. Fahrenbruch, and R. H. Bube, Photovoltaic Properties of n -CdS/ p -CdTe Heterojunctions Prepared by Spray Pyrolysis, Appl. Phys. Lett. 30,423 (1977). [26] J. P. Ge and Y. D. Li, Selective Atmospheric Pressure Chemical Vapor Deposition Route to CdS Arrays, Nanowires, and Nanocombs, Adv. Funct. Mater. 14, 157 (2004). [27] X.-P. Shen, A.-H. Yuan, F. Wang, J.-M. Hong, and Z. Xu, Fabrication of Well- Aligned CdS Nanotubes by CVD-Template Method, Solid State Commun. 133, 19 (2005). [28] P. K. Nair and M. T. S. Nair, Prospects of Chemically Deposited CdS Thin Films in Solar Cell Applications, Sol. Cells 22,103 (1987). [29] I. Kaur, D. K. Pandya, and K. L. Chopra, Growth Kinetics and Polymorphism of Chemically Deposited CdS Films, J. Electrochem. Soc. 127, 943 (1980). [30] M. Froment and D. Lincot, Phase Formation Processes in Solution at the Atomic Level: Metal Chalcogenide Semiconductors, Electrochimica Acta 40,1293 (1995). [31] M. Nagao and S. Watanabe, Chemically Deposited Thick CdS Films and Their Properties, Jpn. J. Appl. Phys. 7, 684 (1968). [32] W. J. Danaher, L. E. Lyons, and G. C. Morris, Some Properties of Thin Films of Chemically Deposited Cadmium Sulphide, Sol. Energy Mater. 12, 137 (1985). [33] N. R. Pavaskar, C. A. Menezes, and A. P. B. Sinha, Photoconductive CdS Films by a Chemical Bath Deposition Process, J. Electrochem. Soc. 124, 743 (1977). [34] R. Ortega-Borges and D. Lincot, Mechanism of Chemical Bath Deposition of Cadmium Sulfide Thin Films in the Ammonia-Thiourea System In Situ Kinetic Study and Modelization, J. Electrochem. Soc. 140, 3464 (1993). [35] F. Cardon and R. H. Bube, Theory of Superlinear Photoconductivity in CdS and Related Materials, J. Appl. Phys. 35, 3344 (1964). [36] W. Palz and S. Ravelet, The Influence of Chemisorption on Photoconductivity in CdS Crystals, Solid-State Electron. 13, 1241 (1970). [37] T. A. T. Cowell and J. Woods, Photochemical Effects in CdS Crystals, J. Phys. Appl. Phys. 2,1053 (1969). [38] H. B. Im, H. E. Matthews, and R. H. Bube, Evidence for Photochemical Changes in Traps in CdS Crystals, J. Appl. Phys. 41, 2581 (1970). [39] J. Woods and K. H. Nicholas, Photochemical Effects in Cadmium Sulphide Crystals, Br. J. Appl. Phys. 15, 1361 (1964). [40] R. Scheer and H.W. Schock, Chalcogenide Photovoltaics:Physics, Technologies, and Thin Film Devices (Wiley-VCH, Weinheim, Germany, 2011). [41] I. Repins, M. A. Contreras, B. Egaas, C. DeHart, J. Scharf, C. L. Perkins, B. To, and R. Noufi, 19-9%-Efficient ZnO/CdS/CuInGaSe 2 Solar Cell with 81-2% Fill Factor, Prog. Photovolt. Res. Appl. 16, 235 (2008). [42] N. Romeo, A. Bosio, and A. Romeo, An Innovative Process Suitable to Produce High-Efficiency CdTe/CdS Thin-Film Modules, Sol. Energy Mater. Sol. Cells 94, 2 (2010).

120 [43] D. Herrmann, P. Kratzert, S. Weeke, M. Zimmer, J. Djordjevic-Reiss, R. Hunger, P. Lindberg, E. Wallin, 0. Lundberg, and L. Stolt, in 2014 IEEE 40th Photovolt. Spec. Conf PVSC (IEEE, Denver, CO, USA, 2014), pp. 2775-2777. [44] M. Powalla, S. Paetel, D. Hariskos, R. Wuerz, F. Kessler, P. Lechner, W. Wischmann, and T. M. Friedlmeier, Advances in Cost-Efficient Thin-Film Photovoltaics Based on Cu(In,Ga)Se 2, Engineering 3, 445 (2017). [45] H.-W. Schock and R. Noufi, CIGS-Based Solar Cells for the next Millennium, Prog. Photovolt. Res. Appl. 8, 151 (2000). [46] E. Bhattacharya and A. H. Mahan, Microstructure and the Light-Induced Metastability in Hydrogenated Amorphous Silicon, Appl. Phys. Lett. 52, 1587 (1988). [47] J. R. Haynes and J. A. Hornbeck, Trapping of Minority Carriers in Silicon. II. n - Type Silicon, Phys. Rev. 100, 606 (1955). [48] D. E. Theodorou and H. J. Queisser, Illumination-Dose Dependence of Persistent Photoconductivity Ofn-GaAs Epitaxial Layers, Appl. Phys. 23, 121 (1980). [49] 0. Katz, G. Bahir, and J. Salzman, Persistent Photocurrent and Surface Trapping in GaN Schottky Ultraviolet Detectors, Appl. Phys. Lett. 84, 4092 (2004). [50] G. Yusa and H. Sakaki, Trapping of Photogenerated Carriers by InAs Quantum Dots and Persistent Photoconductivity in Novel GaAs/n-AlGaAs Field-Effect Transistor Structures, Appl. Phys. Lett. 70, 345 (1997). [51] A. Bhatnagar, Y. H. Kim, D. Hesse, and M. Alexe, Persistent Photoconductivity in Strained Epitaxial BiFeO3 Thin Films, Nano Lett. 14, 5224 (2014). [52] M. C. Tarun, F. A. Selim, and M. D. McCluskey, Persistent Photoconductivity in Strontium Titanate, Phys. Rev. Lett. 111, 187403 (2013). [53] P. Liu, G. She, Z. Liao, Y. Wang, Z. Wang, W. Shi, X. Zhang, S.-T. Lee, and D. Chen, Observation of Persistent Photoconductance in Single ZnO Nanotube, Appl. Phys. Lett. 94, 063120 (2009). [54] P. K. Nair, M. T. S. Nair, V. M. Garcia, 0. L. Arenas, A. C. Pefia Y., I. T. Ayala, 0. Gomezdaza, A. Sanchez, J. Campos, H. Hu, R. Suirez, and M. E. Rincn, Semiconductor Thin Films by Chemical Bath Deposition for Solar Energy Related Applications, Sol. Energy Mater. Sol. Cells 52, 313 (1998). [55] M. Neuschitzer, Y. Sanchez, S. L6pez-Marino, H. Xie, A. Fairbrother, M. Placidi, S. Haass, V. Izquierdo-Roca, A. Perez-Rodriguez, and E. Saucedo, Optimization of CdS Buffer Layer for High-Performance Cu2 ZnSnSe 4 Solar Cells and the Effects of Light Soaking: Elimination of Crossover and Red Kink: CdS and Effects of Light Soaking: Elimination of Crossover and Red Kink, Prog. Photovolt. Res. Appl. 23, 1660 (2015). [56] D. Hariskos, R. Menner, P. Jackson, S. Paetel, W. Witte, W. Wischmann, M. Powalla, L. Birkert, T. Kolb, M. Oertel, B. Dimmler, and B. Fuchs, New Reaction Kinetics for a High-Rate Chemical Bath Deposition of the Zn(S,O) Buffer Layer for Cu(In,Ga)Se2-Based Solar Cells: New Reaction Kinetics for a High-Rate CBD Zn(S,O) Buffer, Prog. Photovolt. Res. Appl. 20, 534 (2012). [57] K. Roy, M. Padmanabhan, S. Goswami, T. P. Sai, G. Ramalingam, S. Raghavan, and A. Ghosh, Graphene-MoS2 Hybrid Structures for Multifunctional Photoresponsive Memory Devices, Nat. Nanotechnol. 8, 826 (2013). [58] P. K. Nair, M. T. S. Nair, and J. Campos, Photocurrent Response in Chemically Deposited CdS Thin Films, Sol. Energy Mater. 15, 441 (1987).

121 [59] C. Wu and R. H. Bube, Thermoelectric and Photothermoelectric Effects in Semiconductors: Cadmium Sulfide Films, J. Appl. Phys. 45, 648 (1974). [60] P. K. Nair, J. Campos, L. E. Sansores, and others, A Critical Discussion of the Very High Photoconductivity in Chemically Deposited Cadmium Sulfide Thin Films: Implications for Solar Cell Technology, Sol. Cells 22, 211 (1987). [61] D. Lang and R. Logan, Large-Lattice-Relaxation Model for Persistent Photoconductivity in Compound Semiconductors, Phys. Rev. Lett. 39, 635 (1977). [62] H. J. Queisser and D. E. Theodorou, Decay Kinetics of Persistent Photoconductivity in Semiconductors, Phys. Rev. B 33, 4027 (1986). [63] J. I. Pankove and J. E. Berkeyheiser, Light-induced Radiative Recombination Centers in Hydrogenated Amorphous Silicon, Appl. Phys. Lett. 37, 705 (1980). [64] H. Dersch, J. Stuke, and J. Beichler, Light-induced Dangling Bonds in Hydrogenated Amorphous Silicon, Appl. Phys. Lett. 38, 456 (1981). [65] M. Stutzmann, W. B. Jackson, and C. C. Tsai, Light-Induced Metastable Defects in Hydrogenated Amorphous Silicon: A Systematic Study, Phys. Rev. B 32, 23 (1985). [66] S. Nonomura, N. Yoshida, T. Gotoh, T. Sakamoto, M. Kondo, A. Matsuda, and S. Nitta, The Light-Induced Metastable Lattice Expansion in Hydrogenated Amorphous Silicon, J. Non-Cryst. Solids 266-269, 474 (2000). [67] W. B. Jackson and J. Kakalios, Evidence for Hydrogen Motion in Annealing of Light-Induced Metastable Defects in Hydrogenated Amorphous Silicon, Phys. Rev. B 37, 1020 (1988). [68] H. Tang, F. Levy, H. Berger, and P. E. Schmid, Urbach Tail of Anatase TiO 2, Phys. Rev. B 52, 7771 (1995). [69] C. Itoh and A. Wada, Relaxation of Photogenerated Carriers in the Anatase Form of Crystalline Titanium Dioxide, Phys. Status Solidi C 2, 629 (2005). [70] T. Sekiya, Y. Takeda, H. Takeda, S. Ohya, and T. Kodaira, Persistent Trapping of Photogenerated Carriers in Colorless Anatase TiO2Single Crystals, J. Phys. Soc. Jpn. 81, 124701 (2012). [71] D. Comedi, S. P. Heluani, M. Villafuerte, R. D. Arce, and R. R. Koropecki, Power- Law Photoconductivity Time Decay in Nanocrystalline TiO2Thin Films, J. Phys. Condens. Matter 19, 486205 (2007). [72] G. W. Iseler, J. A. Kafalas, A. J. Strauss, H. F. MacMillan, and R. H. Bube, Non-F Donor Levels and Kinetics of Electron Transfer in n-Type CdTe, Solid State Commun. 10, 619 (1972). [73] H. C. Wright, R. J. Downey, and J. R. Canning, Conductivity Storage in CdS, J. Phys. Appl. Phys. 1, 1593 (1968). [74] S. B. Zhang, S.-H. Wei, and A. Zunger, Intrinsic n -Type versus p -Type Doping Asymmetry and the Defect Physics of ZnO, Phys. Rev. B 63, (2001). [75] S. Lany and A. Zunger, Anion Vacancies as a Source of Persistent Photoconductivity in II-VI and Chalcopyrite Semiconductors, Phys. Rev. B 72, 035215 (2005). [76] S. H. Jo, T. Chang, I. Ebong, B. B. Bhadviya, P. Mazumder, and W. Lu, Nanoscale Memristor Device as Synapse in Neuromorphic Systems, Nano Lett. 10, 1297 (2010). [77] P. A. N. Feng, C. Chao, Z. Wang, Y. Yang, Y. Jing, and Z. Fei, Nonvolatile Resistive Switching Memories-Characteristics, Mechanisms and Challenges, Prog. Nat. Sci. Mater. Int. 20, 1 (2010).

122 [78] S. H. Jo, T. Kumar, S. Narayanan, W. D. Lu, and H. Nazarian, in Electron Devices Meet. IEDM2014 IEEE Int. (IEEE, 2014), pp. 6-7. [79] D. B. Strukov and H. Kohlstedt, Resistive Switching Phenomena in Thin Films: Materials, Devices, and Applications, MRS Bull. 37, 108 (2012). [80] M. Ungureanu, R. Zazpe, F. Golmar, P. Stoliar, R. Llopis, F. Casanova, and L. E. Hueso, A Light-Controlled Resistive Switching Memory, Adv. Mater. 24,2496 (2012). [81] M. Lee, W. Lee, S. Choi, J.-W. Jo, J. Kim, S. K. Park, and Y.-H. Kim, Brain- Inspired Photonic Neuromorphic Devices Using Photodynamic Amorphous Oxide Semiconductors and Their Persistent Photoconductivity, Adv. Mater. 1700951 (2017). [82] A. Niemegeers, M. Burgelman, and A. De Vos, On the CdS/CuInSe 2 Conduction Band Discontinuity, Appl. Phys. Lett. 67,843 (1995). [83] H. Yin, A. Akey, and R. Jaramillo, Large and Persistent Photoconductivity Due to Hole-Hole Correlation in CdS, Phys. Rev. Mater. 2, (2018). [84] P. M. Mooney, Deep Donor Levels (DX Centers) in III-V Semiconductors, J. Appl. Phys. 67, RI (1990). [85] K. Meng, Q. Ding, S. Wang, Y. He, Y. Li, and Q. Gong, Spatial Conformation and Charge Recombination Properties of Polythiophene Deriatives with Thienylene-Vinylene Side Chains Investigated by Static and Femtosecond Spectroscopy, J. Phys. Chem. B 114, 2602 (2010). [86] R. D. Pensack, K. M. Banyas, and J. B. Asbury, Charge Trapping in Organic Photovoltaic Materials Examined with Time-Resolved Vibrational Spectroscopy, J. Phys. Chem. C 114, 5344 (2010). [87] F. Deschler, E. Da Como, T. Limmer, R. Tautz, T. Godde, M. Bayer, E. von Hauff, S. Yilmaz, S. Allard, U. Scherf, and J. Feldmann, Reduced Charge Transfer Exciton Recombination in Organic Semiconductor Heterojunctions by Molecular Doping, Phys. Rev. Lett. 107, (2011). [88] X.-Y. Zhu and V. Podzorov, Charge Carriers in Hybrid Organic-Inorganic Lead Halide Perovskites Might Be Protected as Large Polarons, J. Phys. Chem. Lett. 6, 4758 (2015). [89] C. Kupitz, S. Basu, I. Grotjohann, R. Fromme, N. A. Zatsepin, K. N. Rendek, M. S. Hunter, R. L. Shoeman, T. A. White, D. Wang, D. James, J.-H. Yang, D. E. Cobb, B. Reeder, R. G. Sierra, H. Liu, A. Barty, A. L. Aquila, D. Deponte, R. A. Kirian, S. Bari, J. J. Bergkamp, K. R. Beyerlein, M. J. Bogan, C. Caleman, T.-C. Chao, C. E. Conrad, K. M. Davis, H. Fleckenstein, L. Galli, S. P. Hau-Riege, S. Kassemeyer, H. Laksmono, M. Liang, L. Lomb, S. Marchesini, A. V. Martin, M. Messerschmidt, D. Milathianaki, K. Nass, A. Ros, S. Roy-Chowdhury, K. Schmidt, M. Seibert, J. Steinbrener, F. Stellato, L. Yan, C. Yoon, T. A. Moore, A. L. Moore, Y. Pushkar, G. J. Williams, S. Boutet, R. B. Doak, U. Weierstall, M. Frank, H. N. Chapman, J. C. H. Spence, and P. Fromme, Serial Time-Resolved Crystallography of Photosystem II Using a Femtosecond X-Ray Laser, Nature 513, 261 (2014). [90] R. P. Sabatini, M. F. Mark, D. J. Mark, M. W. Kryman, J. E. Hill, W. W. Brennessel, M. R. Detty, R. Eisenberg, and D. W. McCamant, A Comparative Study of the Photophysics of Phenyl, Thienyl, and Chalcogen Substituted Rhodamine Dyes, Photochem. Photobiol. Sci. 15, 1417 (2016). [91] D. V. Lang and R. A. Logan, Large-Lattice-Relaxation Model for Persistent Photoconductivity in Compound Semiconductors, Phys. Rev. Lett. 39, 635 (1977).

123 [92] X. Zhang and D. A. Drabold, Direct Molecular Dynamic Simulation of Light- Induced Structural Change in Amorphous Selenium, Phys. Rev. Lett. 83, 5042 (1999). [93] K. Nishidate, T. Sato, Y. Matsukura, M. Baba, M. Hasegawa, and T. Sasaki, Density-Functional Electronic Structure Calculations for Native Defects and Cu Impurities in CdS, Phys. Rev. B 74, 035210 (2006). [94] H. Khallaf, I. 0. Oladeji, G. Chai, and L. Chow, Characterization of CdS Thin Films Grown by Chemical Bath Deposition Using Four Different Cadmium Sources, Thin Solid Films 516, 7306 (2008). [95] W.-D. Park, Optical Constants and Dispersion Parameters of CdS Thin Film Prepared by Chemical Bath Deposition, Trans. Electr. Electron. Mater. 13, 196 (2012). [96] A. J. Aristizibal and M. A. Mikan, Optical Properties of CDS Films by Analysis of Spectral Transmittance, IOSR J. Appl. Phys. 08, 24 (2016). [97] 0. A. Niasse, M. A. Tankari, F. Dia, N. Mbengue, A. Diao, M. Niane, M. Diagne, B. Ba, and G. Levebvre, Optimization of Electrics Parameters CdS/CdTe Thin Film Solar Cell Using Dielectric Model, World J. Condens. Matter Phys. 06, 75 (2016). [98] 0. Gunawan, K. F. Tai, and Y. Virgus, in (IEEE, 2016), pp. 2002-2005. [99] D. Lincot and R. 0. Borges, Chemical Bath Deposition of Cadmium Sulfide Thin Films. In Situ Growth and Structural Studies by Combined Quartz Crystal Microbalance and Electrochemical Impedance Techniques, J. Electrochem. Soc. 139, 1880 (1992). [100] D. V. Lang, R. A. Logan, and M. Jaros, Trapping Characteristics and a Donor- Complex (DX) Model for the Persistent-Photoconductivity Trapping Center in Te- Doped A1xGaixAs, Phys. Rev. B 19, 1015 (1979). [101] T. Endo, A. Hoffmann, J. Santamaria, and I. K. Schuller, Enhancement of Persistent Photoconductivity by UV Excitation in GdBa2Cu30.3,Phys. Rev. B 54, R3750 (1996). [102] K. W. B6er and H. Vogel, Nichtstationare Vorginge in Photoleitern. I Langzeitiges Anklingen der Photoleitung in CdS-Einkristallen als Methode zur Strtermanalyse, Ann. Phys. 452, 10 (1955). [103] S. Khosroabadi and S. H. Keshmiri, Design of a High Efficiency Ultrathin CdS/CdTe Solar Cell Using Back Surface Field and Backside Distributed Bragg Reflector, Opt. Express 22, A921 (2014). [104] M. K. Sheinkman and A. Ya. Shik, Long-Term Relaxation and Residual Conductivity of Semiconductors, Sov. Phys. - Semicond. 10, 128 (1976). [105] J. Kakalios and H. Fritzsche, Persistent Photoconductivity in Doping-Modulated Amorphous Semiconductors, Phys. Rev. Lett. 53, 1602 (1984). [106] D. Azulay, 0. Millo, S. Silbert, I. Balberg, and N. Naghavi, Where Does Photocurrent Flow in Polycrystalline CdS?, Appl. Phys. Lett. 86, 212102 (2005). [107] A.A. Istratov and 0. F. Vyvenko, DX -like Center Generated by Uniaxial Strains of Screw Dislocations in CdS, J. Appl. Phys. 80, 4400 (1996). [108] A. Wright and P. G. Newbery, Electric Fuses, 3rd ed (Institution of Electrical Engineers, London, UK, 2004). [109] P. J. Bulman, G. S. Hobson, and B. C. Taylor, Transferred Electron Devices (Academic Press, London, New York, 1972). [110] W. Czubatyj and S. J. Hudgens, Invited Paper: Thin-FilmOvonic Threshold Switch: Its Operation and Application in Modem Integrated Circuits, Electron. Mater. Lett. 8, 157(2012).

124 [111] T. Prodromakis, C. Toumazou, and L. Chua, Two Centuries of Memristors, Nat. Mater. 11, 478 (2012). [112] R. Kozma, R. E. Pino, and G. E. Pazienza, editors , Advances in Neuromorphic Memristor Science and Applications (Springer Netherlands, Dordrecht, 2012). [113] J.-J. Huang, Yi-Ming Tseng, Wun-Cheng Luo, Chung-Wei Hsu, and T.-H. Hou, in 2011 Int. Electron Devices Meet. (IEEE, Washington, DC, USA, 2011), pp. 31.7.1- 31.7.4. [114] K.-H. Kim, S. Gaba, D. Wheeler, J. M. Cruz-Albrecht, T. Hussain, N. Srinivasa, and W. Lu, A Functional Hybrid Memristor Crossbar-Array/CMOS System for Data Storage and Neuromorphic Applications, Nano Lett. 12, 389 (2012). [115] S. H. Jo and H. Nazarian, Next Generation Non-Volatile Memory and Its Impact on System Performance, (2013). [116] J. J. Yang, D. B. Strukov, and D. R. Stewart, Memristive Devices for Computing, Nat. Nanotechnol. 8, 13 (2012). [117] B. Chen, F. Cai, J. Zhou, W. Ma, P. Sheridan, and W. D. Lu, in Electron Devices Meet. IEDM2015 IEEE Int. (IEEE, 2015), pp. 17-5. [118] Z. Wang, S. Joshi, S. E. Savel'ev, H. Jiang, R. Midya, P. Lin, M. Hu, N. Ge, J. P. Strachan, Z. Li, Q. Wu, M. Barnell, G.-L. Li, H. L. Xin, R. S. Williams, Q. Xia, and J. J. Yang, Memristors with Diffusive Dynamics as Synaptic Emulators for Neuromorphic Computing, Nat. Mater. 16, 101 (2016). [119] S. Yu, B. Gao, Z. Fang, H. Yu, J. Kang, and H.-S. P. Wong, Stochastic Learning in Oxide Binary Synaptic Device for Neuromorphic Computing, Front. Neurosci. 7, (2013). [120] Y. C. Yang, F. Pan, Q. Liu, M. Liu, and F. Zeng, Fully Room-Temperature- Fabricated Nonvolatile Resistive Memory for Ultrafast and High-Density Memory Application, Nano Lett. 9, 1636 (2009). [121] K. Szot, W. Speier, G. Bihlmayer, and R. Waser, Switching the Electrical Resistance of Individual Dislocations in Single-Crystalline SrTiO3, Nat. Mater. 5, 312 (2006). [122] N. Xu, L. Liu, X. Sun, X. Liu, D. Han, Y. Wang, R. Han, J. Kang, and B. Yu, Characteristics and Mechanism of Conduction/Set Process in TiNZnOPt Resistance Switching Random-Access Memories, Appl. Phys. Lett. 92, 232112 (2008). [123] S. H. Jo and W. Lu, CMOS Compatible Nanoscale Nonvolatile Resistance Switching Memory, Nano Lett. 8, 392 (2008). [124] H. S. P. Wong, S. Raoux, S. Kim, J. Liang, J. P. Reifenberg, B. Rajendran, M. Asheghi, and K. E. Goodson, Phase Change Memory, Proc. IEEE 98, 2201 (2010). [125] D. Kuzum, R. G. D. Jeyasingh, B. Lee, and H.-S. P. Wong, Nanoelectronic Programmable Synapses Based on Phase Change Materials for Brain-Inspired Computing, Nano Lett. 12, 2179 (2012). [126] M. Wuttig and N. Yamada, Phase-Change Materials for Rewriteable Data Storage, Nat. Mater. 6, 824 (2007). [127] S. Raoux, W. Welnic, and D. lelmini, Phase Change Materials and Their Application to Nonvolatile Memories, Chem. Rev. 110, 240 (2010). [128] F. Xiong, A. D. Liao, D. Estrada, and E. Pop, Low-Power Switching of Phase- Change Materials with Carbon Nanotube Electrodes, Science 332, 568 (2011).

125 [129] M. Kr6ger, S. Hamwi, J. Meyer, T. Riedl, W. Kowalsky, and A. Kahn, Role of the Deep-Lying Electronic States of MoO3 in the Enhancement of Hole-Injection in Organic Thin Films, Appl. Phys. Lett. 95, 123301 (2009). [130] C. Girotto, E. Voroshazi, D. Cheyns, P. Heremans, and B. P. Rand, Solution- ProcessedMoO3Thin Films As a Hole-Injection Layer for Organic Solar Cells, ACS Appl. Mater. Interfaces 3, 3244 (2011). [131] J. Meyer, R. Khalandovsky, P. Grrn, and A. Kahn, MoO3 Films Spin-Coated from a Nanoparticle Suspension for Efficient Hole-Injection in Organic Electronics, Adv. Mater. 23, 70 (2011). [132] M. Burgelman, P. Nollet, and S. Degrave, Modelling Polycrystalline Semiconductor Solar Cells, Thin Solid Films 361-362, 527 (2000). [133] S. Zhang, S. Long, W. Guan, Q. Liu, Q. Wang, and M. Liu, Resistive Switching Characteristics of MnO , -Based ReRAM, J. Phys. Appl. Phys. 42, 055112 (2009). [134] M. Zhang, Z. Shao, T. Jiang, X. Wu, B. Zhang, X. Zhang, F. Xia, and J. Jie, Tuning Electrical and Raman Scattering Properties of Cadmium Sulfide Nanoribbons via Surface Charge Transfer Doping, J. Phys. Chem. C 123, 15794 (2019). [135] J. Meyer, Electronic Structure of Molybdenum-Oxide Films and Associated Charge Injection Mechanisms in Organic Devices, J. Photonics Energy 1, 011109 (2011). [136] G. G. Pethuraja, R. E. Welser, A. K. Sood, C. Lee, N.J. Alexander, H. Efstathiadis, P. Haldar, and J. L. Harvey, Current-Voltage Characteristics of ITO/p-Si and ITO/n-Si Contact Interfaces, Adv. Mater. Phys. Chem. 02, 59 (2012). [137] D. J. Feldman and R. M. Margolis, Q1/Q2 2019 Solar Industry Update (2019). [138] International Energy Agency, Technology Roadmap Solar Photovoltaic Energy - 2014 Edition (OECD/IEA, Paris, France, 2014). [139] M. Woodhouse, R. Jones-Albertus, D. Feldman, R. Fu, K. Horowitz, D. Chung, D. Jordan, and S. Kurtz, On the Path to SunShot - The Role of Advancements in Solar PhotovoltaicEfficiency, Reliability, and Costs (2016). [140] A. Niemegeers, M. Burgelman, and A. De Vos, On the CdS/CuInSe2Conduction Band Discontinuity, Appl. Phys. Lett. 67, 843 (1995). [141] J. Reemts and A. Kittel, Persistent Photoconductivity in Highly Porous ZnO Films, J. Appl. Phys. 101, 013709 (2007). [142] S.-H. Choi, G.-L. Park, C. Lee, and J. Jang, Persistent Photoconductivity in Hydrogenated Amorphous Silicon, Solid State Commun. 59, 177 (1986). [143] S. Kim, C.-S. Lee, S. Kim, R. B. V. Chalapathy, E. A. Al-Ammar, and B. T. Ahn, Understanding the Light Soaking Effect of ZnMgO Buffer in CIGS Solar Cells, Phys Chem Chem Phys 17, 19222 (2015). [144] Y. Nakayama, P. Stradins, and H. Fritzsche, Metastable Centers and Photoconduction in A-SiNx:H, J. Non-Cryst. Solids 164-166, 1061 (1993). [145] H. Fritzsche and Y. Nakayama, Light-Induced Metastable Changes in Amorphous Silicon Nitride, Philos. Mag. B 69, 359 (1994). [146] S. Merdes, R. Siez-Araoz, A. Ennaoui, J. Klaer, M. Ch. Lux-Steiner, and R. Klenk, Recombination Mechanisms in Highly Efficient Thin Film Zn(S,O)/Cu(In,Ga)S2 Based Solar Cells, Appl. Phys. Lett. 95,213502 (2009). [147] A. Ferguson, P. Dippo, D. Kuciauskas, R. Farshchi, J. Bailey, G. Zapalac, and D. Poplavskyy, in 2018 IEEE 7th World Conf Photovolt. Energy Convers. WCPEC Jt.

126 Conf 45th IEEE PVSC 28th PVSEC 34th EU PVSEC (IEEE, Waikoloa Village, HI, 2018), pp. 3918-3922. [148] V. Nadenau, U. Rau, A. Jasenek, and H. W. Schock, Electronic Properties of CuGaSe2-Based Heterojunction Solar Cells. Part I. Transport Analysis, J. Appl. Phys. 87, 584 (2000). [149] A. Jasenek, U. Rau, V. Nadenau, and H. W. Schock, Electronic Properties of CuGaSe2-Based Heterojunction Solar Cells. Part II. Defect Spectroscopy, J. Appl. Phys. 87, 594 (2000). [150] M. G. Panthani, J. M. Kurley, R. W. Crisp, T. C. Dietz, T. Ezzyat, J. M. Luther, and D. V. Talapin, High Efficiency Solution Processed Sintered CdTe Nanocrystal Solar Cells: The Role of Interfaces, Nano Lett. 14, 670 (2014). [151] I. L. Eisgruber, J. E. Granata, J. R. Sites, J. Hou, and J. Kessler, Blue-Photon Modification of Nonstandard Diode Barrier in CuInSe2 Solar Cells, Sol. Energy Mater. Sol. Cells 53, 367 (1998). [152] T. Kobayashi, T. Kumazawa, Z. Jehl Li Kao, and T. Nakada, Cu(In,Ga)Se2 Thin Film Solar Cells with a Combined ALD-Zn(O,S) Buffer and MOCVD-ZnO:B Window Layers, Sol. Energy Mater. Sol. Cells 119, 129 (2013). [153] Y.-J. Hsiao, C.-H. Lu, and T.-H. Fang, To Enhance Performance of Light Soaking Process on ZnS/ CuIn 1 - x Ga x Se 2 Solar Cell, Int. J. Photoenergy 2013, 1 (2013). [154] N. Devi, A. Aziz, and S. Datta, in (Bikaner, India, 2018), p. 100031. [155] S. Ouedraogo, F. Zougmord, and J. M. Ndjaka, Numerical Analysis of Copper- Indium-Gallium-Diselenide-Based Solar Cells by SCAPS-1D, Int. J. Photoenergy 2013,1 (2013). [156] A. B. Kuzmenko, Kramers-Kronig Constrained Variational Analysis of Optical Spectra, Rev. Sci. Instrum. 76, 083108 (2005). [157] G. E. Jellison and F. A. Modine, Parameterization of the Optical Functions of Amorphous Materials in the Interband Region, Appl. Phys. Lett. 69, 371 (1996). [158] B. Harbecke, Coherent and Incoherent Reflection and Transmission of Multilayer Structures, Appl. Phys. B Photophysics Laser Chem. 39, 165 (1986). [159] X. J. Zheng, Y. Q. Chen, T. Zhang, B. Yang, C. B. Jiang, B. Yuan, and Z. Zhu, Photoconductive Semiconductor Switch Based on ZnS Nanobelts Film, Sens. Actuators B Chem. 147, 442 (2010). [160] R. K. Srivastava, N. Pandey, and S. K. Mishra, Effect of Cu Concentration on the Photoconductivity Properties of ZnS Nanoparticles Synthesized by Co-Precipitation Method, Mater. Sci. Semicond. Process. 16, 1659 (2013). [161] A. Bera and D. Basak, Photoluminescence and Photoconductivity of ZnS-Coated ZnO Nanowires, ACS Appl. Mater. Interfaces 2, 408 (2010). [162] X. Fang, Y. Bando, M. Liao, U. K. Gautam, C. Zhi, B. Dierre, B. Liu, T. Zhai, T. Sekiguchi, Y. Koide, and D. Golberg, Single-Crystalline ZnS Nanobelts as Ultraviolet- Light Sensors, Adv. Mater. 21, 2034 (2009). [163] X. J. Zheng, Y. Q. Chen, T. Zhang, C. B. Jiang, B. Yang, B. Yuan, S. X. Mao, and W. Li, A Photoconductive Semiconductor Switch Based on an Individual ZnS Nanobelt, Scr. Mater. 62, 520 (2010). [164] S. K. Mishra, R. K. Srivastava, S. G. Prakash, R. S. Yadav, and A. C. Panday, Photoluminescence and Photoconductive Characteristics of Hydrothermally Synthesized ZnO Nanoparticles, Opto-Electron. Rev. 18, (2010).

127 [165] X. G. Zheng, Q. Sh. Li,J. P. Zhao, D. Chen, B. Zhao, Y. J. Yang, and L. Ch. Zhang, Photoconductive Ultraviolet Detectors Based on ZnO Films, Appl. Surf. Sci. 253, 2264 (2006). [166] S. Mridha and D. Basak, ZnO/Polyaniline Based Inorganic/Organic Hybrid Structure: Electrical and Photoconductivity Properties, Appl. Phys. Lett. 92, 142111 (2008). [167] J. Bao, I. Shalish, Z. Su, R. Gurwitz, F. Capasso, X. Wang, and Z. Ren, Photoinduced Oxygen Release and Persistent Photoconductivity in ZnO Nanowires, Nanoscale Res. Lett. 6, 404 (2011). [168] Y. Takahashi, M. Kanamori, A. Kondoh, H. Minoura, and Y. Ohya, Photoconductivity of Ultrathin Zinc Oxide Films, Jpn. J. Appl. Phys. 33, 6611 (1994). [169] C.-Y. Tsay, S.-T. Chen, and M.-T. Fan, Solution-Processed Mg-Substituted ZnO Thin Films for Metal-Semiconductor-Metal Visible-Blind Photodetectors, Coatings 9, 277(2019). [170] J. Pettersson, C. Platzer-Bj6rkman, A. Hultqvist, U. Zimmermann, and M. Edoff, Measurements of Photo-Induced Changes in the Conduction Properties of ALD-Zn -

x Mg x O Thin Films, Phys. Scr. T141, 014010 (2010). [171] Y. H. Ko, G. Nagaraju, S. H. Lee, and J. S. Yu, Facile Preparation and Optoelectronic Properties of CuO Nanowires for Violet Light Sensing, Mater. Lett. 117, 217 (2014). [172] R. Jayakrishnan, A. S. Kurian, V. G. Nair, and M. R. Joseph, Effect of Vacuum Annealing on the Photoconductivity of CuO Thin Films Grown Using Sequential Ionic Layer Adsorption Reaction, Mater. Chem. Phys. 180, 149 (2016). [173] K. H. Al-Mayalee, N. Saadi, E. Badradeen, F. Watanabe, and T. Karabacak, Optical and Photoconductive Response of CuO Nanostructures Grown by a Simple Hot-Water Treatment Method, J. Phys. Chem. C 122, 23312 (2018). [174] D. A. Kumar, F. P. Xavier, and J. M. Shyla, Investigation on the Variation of Conductivity and Photoconductivity of CuO Thin Films as a Function of Layers of Coating, 10 (2012). [175] H.-J. Song, M.-H. Seo, K.-W. Choi, M.-S. Jo, J.-Y. Yoo, and J.-B. Yoon, High- Performance Copper Oxide Visible-Light Photodetector via Grain-Structure Model, Sci. Rep. 9, (2019). [176] N. Wongcharoen and T. Gaewdang, Effect of Oxidation Temperature on the Properties of CuO Thin Films Prepared by Thermal Oxidation of Sputtered Cu Thin Films, Adv. Mater. Res. 1098,6 (2015). [177] N. Serpone, D. Lawless, J. Disdier, and J.-M. Herrmann, Spectroscopic, Photoconductivity, and Photocatalytic Studies of TiO2 Colloids: Naked and with the Lattice Doped with Cr3+, Fe3+, and V5+ Cations, Langmuir 10, 643 (1994). [178] A. Brajsa, K. Szaniawska, R. J. Barczynski, L. Murawski, B. Koscielska, A. Vomvas, and K. Pomoni, The Photoconductivity of Sol-Gel Derived TiO2 Films, Opt. Mater. 26, 151 (2004). [179] K. Pomoni, A. Vomvas, and Chr. Trapalis, Transient Photoconductivity of Nanocrystalline TiO2 Sol-Gel Thin Films, Thin Solid Films 479, 160 (2005). [180] S. Zhang, Y. F. Zhu, and D. E. Brodie, Photoconducting TiO2 Prepared by Spray Pyrolysis Using TiCl4, Thin Solid Films 213,265 (1992).

128 [181] G. Liu, N. Hoivik, X. Wang, S. Lu, K. Wang, and H. Jakobsen, Photoconductive, Free-Standing Crystallized TiO2 Nanotube Membranes, Electrochimica Acta 93, 80 (2013). [182] R. A. Street, Transient Conductivity and Photoconductivity in a -Si:H, Phys. Rev. B 32, 3910 (1985). [183] G. Moddel, D.A. Anderson, and W. Paul, Derivation of the Low-Energy Optical- Absorption Spectra of a -Si: H from Photoconductivity, Phys. Rev. B 22, 1918 (1980). [184] S.C. Agarwal and S. Guha, Persistent Photoconductivity in a- Si: H / a- Si N X: H Layered Structures, Phys. Rev. B 31, 5547 (1985). [185] C. Main, R. Russell, J. Berkin, and J. M. Marshall, Transient Photoconductivity in N-Type a-Si: H, Philos. Mag. Lett. 55,189 (1987). [186] D. L. Staebler and C. R. Wronski, Reversible Conductivity Changes in Discharge- produced Amorphous Si, Appl. Phys. Lett. 31, 292 (1977). [187] M. Aven and C. A. Mead, ELECTRICAL TRANSPORT AND CONTACT PROPERTIES OF LOW RESISTIVITY n -TYPE ZINC SULFIDE CRYSTALS, Appl. Phys. Lett. 7, 8 (1965). [188] A. Janotti and C. G. Van de Walle, Fundamentals of Zinc Oxide as a Semiconductor, Rep. Prog. Phys. 72, 126501 (2009). [189] P. Tiwana, P. Docampo, M. B. Johnston, H. J. Snaith, and L. M. Herz, Electron Mobility and Injection Dynamics in Mesoporous ZnO, SnO2 , and TiO2Films Used in Dye-Sensitized Solar Cells, ACS Nano 5, 5158 (2011). [190] T. M. Friedlmeier, P. Jackson, A. Bauer, D. Hariskos, 0. Kiowski, R. Wuerz, and M. Powalla, Improved Photocurrent in Cu(In,Ga)Se2Solar Cells: From 20.8% to 21.7% Efficiency with CdS Buffer and 21.0% Cd-Free, IEEE J. Photovolt. 5, 1487 (2015). [191] 0. Lundberg, M. Edoff, and L. Stolt, The Effect of Ga-Grading inCIGS Thin Film Solar Cells, Thin Solid Films 480-481, 520 (2005). [192] J. A. M. AbuShama, S. Johnston, T. Moriarty, G. Teeter, K. Ramanathan, and R. Noufi, Properties of ZnO/CdS/CuInSe2 Solar Cells with Improved Performance, Prog. Photovolt. Res. Appl. 12, 39 (2004). [193] J. H. Boyle, B. E. McCandless, W. N. Shafarman, and R. W. Birkmire, Structural and Optical Properties of(AgCu)(InGa)Se2Polycrystalline Thin Film Alloys, J. Appl. Phys. 115, 223504 (2014). [194] 0. Vigil-Galin, F. A. Pulgarin, F. Cruz-Gandarilla, M. Courel, G. Villarreal-Ruiz, Y. Sinchez, D. Jimenez-Olarte, and E. Saucedo, Optimization of CBD-CdS Physical Properties for Solar Cell Applications Considering a MIS Structure, Mater. Des. 99, 254(2016). [195] Y. Sanchez, M. Espindola-Rodriguez, H. Xie, S. Lopez-Marino, M. Neuschitzer, S. Giraldo, M. Dimitrievska, M. Placidi, V. Izquierdo-Roca, F. A. Pulgarin-Agudelo, 0. Vigil-Galin, and E. Saucedo, Ultra-Thin CdS for Highly Performing Chalcogenides Thin Film Based Solar Cells, Sol. Energy Mater. Sol. Cells 158,,138 (2016). [196] S. Alhammadi, H. Jung, S. Kwon, H. Park, J.-J. Shim, M. H. Cho, M. Lee, J. S. Kim, and W. K. Kim, Effect of Gallium Doping on CdS Thin Film Properties and Corresponding Cu(InGa)Se2/CdS:Ga Solar Cell Performance, Thin Solid Films 660, 207(2018). [197] M. Lee, W. Lee, S. Choi, J.-W. Jo, J. Kim, S. K. Park, and Y.-H. Kim, Brain- Inspired Photonic Neuromorphic Devices Using Photodynamic Amorphous Oxide

129 Semiconductors and Their Persistent Photoconductivity, Adv. Mater. 29, 1700951 (2017). [198] 0. Astakhov, R. Carius, F. Finger, Y. Petrusenko, V. Borysenko, and D. Barankov, Relationship between Defect Density and Charge Carrier Transport in Amorphous and Microcrystalline Silicon, Phys. Rev. B 79, (2009).

130