The Effects of Chlorine and Selenium in Cadmium Telluride Solar

The Effects of Chlorine and Selenium in

Cadmium Telluride Solar Cells

Thomas Fiducia

Abstract

Solar photovoltaics (PV) holds great promise to change the way that electricity is produced and used globally. As it stands, electricity is generated mainly by large coal and gas-fired power stations, which are expensive to build and rely on a fuel supply that becomes more expensive over time. By contrast, the costs of solar PV are falling rapidly, and solar is already producing electricity at lower levelised costs than coal and gas power stations. Moreover, it can do so with a very low environmental impact, and since it is a ‘distributed’ power source that does not require a fuel supply, also improves access to electricity and the overall security of supply.

However, if these benefits are to be realised, deployment of solar PV needs to continue to scale significantly. Solar PV currently supplies only ~3% of worldwide electricity demand, and demand for electricity is set to nearly double by 2050 as a result of the electrification of heating and transport and rising living standards. In order to continue its rapid growth and help to meet a significant portion of future electricity demand, solar module efficiencies need to continue to rise and production costs need to decrease further.

Fast-deposited thin-film PV technologies like cadmium telluride (CdTe) offer a promising route to achieve the necessary price decreases and industry scale-up because they are intrinsically less expensive to produce than the incumbent silicon PV modules, which require careful crystal growth and individual processing of each wafer, cell and module. The downside of fast thin-film deposition however is that the devices invariably have polycrystalline absorber layers with small crystal ‘grains’, and high defect densities. This not only limits power conversion efficiency compared to single crystal counterparts like silicon, but also makes the devices much more microstructurally and compositionally complicated, and hence more difficult to characterise and control. In particular, device-level characterisation techniques that were developed for homogeneous single crystal

1 absorber layers are not sufficient to resolve the complexities of thin-film cells and high-resolution characterisation techniques have been under-used, slowing device development.

Here we use high-resolution correlative characterisation techniques to investigate the effects of two elements that are vital to producing high efficiency cadmium telluride solar cells – chlorine and selenium. Using 3-dimensional NanoSIMS compositional mapping we find that following the essential cadmium chloride heat treatment, chlorine is not just present in grain boundaries – where it is known to have a passivation effect – but permeates every region of the CdTe absorber layer. It is found segregated at the front interface between the CdTe and the buffer layer, at incoherent twin boundaries that span grain interiors, and at dopant concentrations in grain interiors. In selenium- alloyed CdTe devices we use high resolution NanoSIMS and SEM-based cathodoluminescence, on the same area of the absorber, to reveal that selenium alloying lowers non-radiative recombination levels in CdSeTe grain interiors, helping to explain the record performance of selenium-graded devices. We then use TEM-based cathodoluminescence to show that selenium also has a passivation effect on grain boundaries in CdSeTe, which is the first time that high-resolution TEM-CL mapping has been achieved on a solar cell.

Together, these results help to explain how cadmium telluride devices have achieved efficiencies of over 22%, despite their fast absorber layer deposition and small grain sizes. The results suggest new routes for further efficiency improvement of CdTe solar cells, including by increasing selenium concentrations at grain boundaries and in the bulk material at the back of the absorber layer. This can reduce costs further for what is currently the lowest cost of all solar and fossil fuel electricity generation technologies, and hence help to spread the cost, security, and environmental benefits of solar photovoltaics. It is also intended that the work will encourage more high resolution, correlative characterisation of thin-film PV technologies in general.

Publications List

- T. A. M. Fiducia, B. G. Mendis, K. Li, C. R. M. Grovenor, A. H. Munshi, K. Barth, W. S. Sampath, L. D. Wright, A. Abbas, J. W. Bowers and J. M. Walls, ‘Understanding the role of selenium in defect passivation for highly efficient selenium-alloyed cadmium telluride solar cells’, Nature Energy, volume 4, pages 504–511 (2019).

- T. A. M. Fiducia, A. Howkins, A. Abbas, B. Mendis, A. H. Munshi, K. Barth, W. S. Sampath, and J. M. Walls, ‘Passivation of Grain Boundaries in Selenium-alloyed CdTe Solar Cells’, manuscript in preparation.

- T. A. M. Fiducia, K. Li, A. H. Munshi, K. Barth, W. S. Sampath, C. R. M. Grovenor, and J. M. Walls, ‘3D Imaging of Selenium and Chlorine Distributions in Highly Efficient Selenium-Graded Cadmium Telluride Solar Cells’, IEEE Journal of Photovoltaics, vol. 10, no. 2, pp. 685–689, Mar. 2020.

- M. J. Watts, T. A. M. Fiducia, B. Sanyal, R. Smith, J. M. Walls, and P. Goddard, “Enhancement of photovoltaic efficiency in CdSexTe1− x (where 0 ⩽ x ⩽ 1): insights from density functional theory,” J. Phys. Condens. Matter, vol. 32, no. 12, p. 125702, Mar. 2020.

- T. A. M. Fiducia, K. Li, A. H. Munshi, K. Barth, W. S. Sampath, C. R. M. Grovenor, J. M. Walls, ‘3D Distributions of Chlorine and Sulphur Impurities in a Thin-Film Cadmium Telluride Solar Cell’, MRS Advances, vol. 3, no. 56, pp. 3287–3292, May 2018.

- T. A. M. Fiducia, A. H. Munshi, K. Barth, D. Proprentner, G. West, W. S. Sampath, and J. M. Walls, ‘Defect Tolerance in as-deposited Selenium-alloyed Cadmium Telluride Solar Cells’, IEEE 2018 World Conference on Photovoltaic Energy Conversion.

- T. A. M. Fiducia, K. Li, A. H. Munshi, K. Barth, W. S. Sampath, C. R. M. Grovenor, and J. M. Walls. ‘Large Area 3D Elemental Mapping of a MgZnO/CdTe Solar Cell with Correlative EBSD Measurements’, IEEE 2018 World Conference on Photovoltaic Energy Conversion.

- T. A. M. Fiducia, A. Abbas, K. Barth, W. S. Sampath, J. M. Walls, ‘Intragranular Defects in As- Deposited and Cadmium Chloride-Treated Polycrystalline Cadmium Telluride Solar Cells’, 43rd IEEE Photovoltaic Specialist Conference (PVSC), 2016

Acknowledgements

There are several people that I would like to thank for their help and support during the PhD. First, I would like to thank my supervisor, Professor Mike Walls, who gave me the opportunity to pursue the PhD. Mike is perhaps the most supportive (but not everbearing) supervisor you could ask for. I have enjoyed the research process hugely and that is largely down to him and his enthusiasm and positive attitude. I would also like to thank our many collaborators and co-authors, who have been instrumental in making the work for this thesis possible. At Colorado State University, many people including Amit Munshi, Tushar Shimpi, Kurt Barth, Sampath, and Jim Sites have provided general support as well as their world class CdTe samples. They were also excellent hosts during my three months in Fort Collins. At Oxford, Chris Grovenor, Kexue Li and Junliang Liu performed the vital NanoSIMS measurements. I would particularly like to thank Kexue (now at Manchester) for his long hours gathering data at the Cameca. Budhika Mendis at Durham performed the SEM-based cathodoluminescence for chapter 5 and helped to make sense of the resulting data. He also put in a lot of time helping with publication of papers generally, which I am extremely grateful for. Ashley Howkins at Brunel performed the TEM-CL measurements for Chapter 7. At Loughborough there is a great network of people within the offices and labs at CREST that help each other out and I am particularly thankful to them. Jake bowers has been an excellent second supervisor and is always ready to help with good advice and PV knowledge. Likewise, Patrick Isherwood is a fountain of useful information and I have appreciated our solar-based chats. Ali Abbas is a fantastic microscopist and I am grateful for all his help over the years, and for our regular talks and meetings with Mike. At the LMCC, Ryan MacLachlan, Zhaoxia Zhao, and many others have always been on hand to help. My fellow CdTe PhD students and desk mates Rachael Greenhalgh and Christos Potamialis have provided much amusement and support throughout the project. Staff and students on the CDT-PV (Centre for Doctoral Training in New and Sustainable Photovoltaics) were helpful and entertaining throughout the PhD, particularly during the rotational training program in first year. Lewis Wright was there from the beginning: at the CDT-PV training, in the office, and as my housemate throughout most of the last 4 years. Lewis was instrumental for the python-based plotting in Chapter 5, and has greatly enhanced my time so far at Loughborough. Likewise, Rhys Comissiong has provided a lot of helpful advice and insight as a PV ‘outsider’. Finally, I want to thank my parents; for putting up with me during lockdown, for taking extra interest in my work, and for supporting me all the way up to this point. I hope that this can be a small repayment for everything that you have done.

Table of Contents

Abstract 1 Publications List...... 3 Acknowledgements ...... 4 Chapter 1 - Introduction ...... 9 1.1 Why solar photovoltaics ...... 9 1.1.1 The rise of solar PV ...... 11 1.1.2 Limitations of solar PV ...... 16 1.2 Why thin-film photovoltaics ...... 18 1.2.1 Silicon PV: production, efficiency, and costs ...... 18 1.2.2 Thin-film PV: production, efficiency, and costs ...... 22 1.3 CdTe photovoltaics ...... 26 1.3.1 Thesis aims ...... 28 Chapter 2 - Approach to cell characterisation ...... 29 2.1 The simplicity of silicon devices ...... 29 2.1.1 Silicon cell microstructure ...... 30 2.1.2 Silicon cell chemistry ...... 31 2.2 The complexity of thin-film devices ...... 32 2.2.1 Complex microstructure ...... 32 2.2.2 Complex chemistry ...... 34 2.3 CdTe cell designs...... 38 2.4 Characterisation methods ...... 40 2.4.1 NanoSIMS ...... 40 2.4.2 Cathodoluminescence ...... 45 Chapter 3 - 3D Distributions of Chlorine and Sulphur Impurities in a Thin-Film Cadmium Telluride Solar Cell ...... 53 3.1 Introduction ...... 53 3.2 Experimental ...... 54 3.3 Results ...... 55 3.4 Discussion ...... 57 3.5 Conclusions ...... 60

Chapter 4 - Large Area 3D Elemental Mapping of a MgZnO/CdTe Solar Cell with Correlative EBSD Measurements ...... 62 4.1 Introduction ...... 62 4.2 Experimental ...... 63 4.3 Results ...... 64 4.4 Discussion ...... 68 4.5 Conclusions ...... 70 Chapter 5 - Understanding the role of selenium in defect passivation for highly efficient selenium-alloyed cadmium telluride solar cells ...... 72 5.1 Introduction ...... 72 5.2 Selenium diffusion into the CdTe layer ...... 73 5.3 Selenium-induced defect passivation ...... 76 5.4 Selenium related sub-bandgap states ...... 78 5.5 Impact of selenium concentration on diffusion lengths ...... 80 5.6 Selenium-induced bandgap gradients ...... 85 5.7 Discussion ...... 77 5.8 Methods ...... 79 5.9 Supplementary Figures ...... 84 Chapter 6 - 3D Imaging of Selenium and Chlorine Distributions in Highly Efficient Selenium-Graded Cadmium Telluride Solar Cells ...... 88 6.1 Introduction ...... 88 6.2 Experimental ...... 89 6.3 Results and Discussion ...... 90 6.4 Conclusions ...... 98 Chapter 7 - Passivation of Grain Boundaries in Selenium-alloyed CdTe Solar Cells Revealed by TEM-based Cathodoluminescence ...... 99 7.1 Abstract ...... 99 7.2 Introduction ...... 100 7.3 Results ...... 103 7.4 Discussion ...... 112 7.5 Methods ...... 114 7.6 Supplementary information ...... 117 Chapter 8 - Conclusions ...... 123 8.1 Chlorine and the cadmium chloride treatment ...... 123 8.2 The effects of selenium alloying in CdTe ...... 126

8.3 Future work ...... 130 8.3.1 Chlorine in CdTe ...... 130 8.3.2 Selenium in CdTe ...... 132 8.4 The future of solar photovoltaics ...... 135 8.5 References ...... 138

Abbreviations and acronyms

BG Bandgap

CL Cathodoluminescence

CSS Close space sublimation

CST CdSe(x)Te(1-x)

DFT Density functional theory

EBSD Electron back-scatter diffraction

EDX Energy dispersive X-ray spectroscopy

GB Grain boundary

LCOE Levelised cost of electricity

PL Photoluminescence

PV Photovoltaic

SEM Scanning electron microscope

SIMS Secondary ion mass spectroscopy

STEM Scanning transmission electron microscopy

TCO Transparent conducting oxide

TRPL Time-resolved photoluminescence

Voc Open circuit voltage

Chapter 1 - Introduction

1.1 Why solar photovoltaics

Electricity is essential for almost every area of 21st century life, including health, work, education, and entertainment.

In most developed nations, the provision of electricity when it is needed is taken for granted. This is a result of decades of development of electricity generation, transmission, and distribution infrastructure. However, the electricity system that has been built to provide this flip-of-a-switch service is far from perfect.

Firstly, the electricity that is provided comes at a significant financial cost. For households, this includes electricity bills that reduce disposable income. For businesses, this includes the cost of running lights, equipment, computers, and factory machinery – to name a few examples – which significantly add to business expenses and hence the prices that they charge for most goods and services, increasing inflation. In addition to financial costs, there are significant environmental costs associated with electricity generation, since it is currently produced mostly by coal and gas-fired power stations that release pollutants such as sulphur dioxide, nitrogen oxide, and carbon dioxide into the atmosphere. The current system is also overly reliant on large centralised power stations that require a constant supply of fuel, exposing millions of users to electricity supply shocks and price fluctuations [1].

While the electricity system is not perfect in developed nations, it is non-existent in parts of many developing nations. An estimated 0.86 billion people worldwide do not have access to electricity, mainly in remote regions where extension of the grid infrastructure is too expensive. This includes, for instance 46 million people in Pakistan, 78 million people in

Nigeria, and 74 million people in India [2].

Increased proliferation of solar photovoltaics has the potential to significantly improve this situation for individuals in both developed and developing nations. This is for a number of reasons.

In terms of costs, solar PV is already a leader amongst electricity generation technologies.

Average levelised costs of electricity for solar PV in the US are now lower for any other technology except large-scale hydro, and costs continue to decrease rapidly worldwide [3],

[4]. The environmental impact of solar is also smaller than for traditional fossil fuel-based generation technologies. Lifetime greenhouse gas emissions for solar PV are 5% those of coal, and no pollutants are released during PV plant operation. In addition, whereas coal and gas-fired power plants are only effective at large scale, solar can be economic even at small system sizes because it is modular (i.e. built up of lots of smaller units – the solar cells and modules). This modularity, along with the fact that solar does not require a fuel supply, means that solar can act as an effective ‘distributed’ renewable energy source. This has benefits in terms of improving overall security of electricity supply, and for providing electricity access for those in remote regions that do not currently have it.

Solar PV therefore has the potential to improve access to electricity, improve its security of supply, reduce its costs, and reduce its environmental impacts, and hence improve global

10 living standards. However, despite its rapid growth so far, solar PV only generates approximately 3% of global electricity supply. Clearly, in order for solar to make a significant impact on the electricity generation system, and to provide the benefits described above, solar deployments and module production levels need to continue to scale significantly. The lower the costs of producing solar electricity, the higher the incentive there is for solar farm development and investment in module production capacity. It is therefore essential for further scale up of the industry that the levelised costs of solar electricity continue to decrease. (Put another way, lower solar costs are not just important for reducing electricity prices from future and planned solar farms, they also increase the number of farms that will be built, spreading the other solar benefits described above). Reducing solar costs is therefore the goal of most photovoltaics research and development, although it is not often explicitly mentioned. In the next section, we introduce solar PV in general and discuss the development of the industry so far.

1.1.1 The rise of solar PV

Solar photovoltaic devices convert sunlight directly into electricity. During the conversion process, electrons that are trapped within chemical bonds in a material absorb a photon and are excited to a higher energy state where they are free to move. A built-in field in the device, either across the absorber layer or at the contacts, is needed to ensure that electrons enter the external circuit through one contact and holes, which are essentially an absence of electrons, through the other contact.

The first practical solar cell was invented in the 1950’s in Bell Laboratories. Since then, solar cells based on various materials have been developed, and their conversion efficiencies have steadily increased. This is shown by the well-known chart in Fig 1.1, which plots the progression of the best research cell efficiencies for the different technologies since 1975 [5]. The highest efficiencies that have been achieved are for multi-junction devices (purple curves in the figure), which have recently achieved the all-time solar PV efficiency record of 39.2% for a standard, non-concentrated AM1.5 spectrum [6]. Multi- junction cells are used in high-value applications such as on satellites and are currently too expensive for terrestrial flat-plate power generation (for instance, the record NREL device had six junctions and 140 different semiconductor layers, making it difficult to manufacture at low cost). Other notable technologies include silicon-based cells (blue), thin-film cadmium telluride and copper indium gallium (di)selenide devices (green), and thin-film perovskites (red, with yellow fill).

Silicon solar cells have a record efficiency of 26.7% for single-crystal, research-scale devices and 22.3% for multi-crystalline devices, and silicon holds a ~90% share of the solar module market [7]. Along with silicon, thin-film cadmium telluride (CdTe) and copper indium gallium (di)selenide (CIGS) are the only technologies that have been commercialised for utility-scale power generation, and have reached efficiencies of 21.0% and 23.4% respectively (cell area >1cm2) [7]. Perovskite PV devices have been developed only relatively recently, but in that time have rapidly reached efficiencies of 21.6% (un- stabilised, small-area perovskite efficiencies of 25.2% have recently been reported, but details are as yet unpublished) [7].

Fig 1.1 Plot of the best research cell efficiencies from 1975 to 2020. Source: the national renewable energy laboratory (NREL) [5].

Despite the impressive early performance improvements shown in the figure, it was not until the 1990’s that solar PV was first used for commercial, utility-scale electricity generation [8]. At this point, module performance was high enough, and production costs low enough, to enable commercial power production – albeit with the help of significant government subsidies.

This started a positive feedback loop, whereby increased module production volumes and manufacturing efficiencies lowered per-unit module prices, raising profits and incentivising expansion of production capacity, which further increased volumes and lowered prices [9].

This process rapidly reduced module costs for both silicon and thin-film CdTe technologies on a dollar per watt basis. The price decrease, along with higher module efficiencies and crucial improvements at the PV system level (such as cheaper inverters, lower module

13 degradation rates, lower financing costs, and larger system sizes) has led to rapid decreases in the total costs of electricity from solar photovoltaics at the residential, commercial, and utility scale. The most recent decade of this cost decrease for utility-scale

PV is shown in Fig 1.2, which plots the estimated levelised costs of electricity (LCOE) from new-build solar farms in the US each year since 2009 [3]. In that 10-year time period, the cost of solar electricity has decreased by nearly 90%.

Fig 1.2 Plot of the levelised costs of electricity (LCOE) from new-build solar farms in the US each year since 2009 [3]. LCOE is a standard way to calculate the total costs of electricity over the lifetime of a power generation project.

For comparison, in Fig 1.3 the solar LCOE curve from above is plotted alongside those of the other main electricity generation technologies. The figure shows that levelised costs of electricity for new-build solar in the US are now on average 60% lower than those of new- build coal, and 74% lower than new-build nuclear costs. Natural gas, until recently the

14 price leader, is now 40% more expensive than solar PV, despite lower gas prices over the last decade because of US shale production. Solar electricity costs are now on a par with onshore wind, which itself decreased costs 70% over the decade.

400

350

300

250

200 Mean LCOE ($/MWh) LCOE Mean 150

100

0 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

Nuclear Coal Gas (CCGT) Wind Solar PV

Fig 1.3 Plot of the LCOE of different electricity generation technologies over the last decade in the US. Replotted and adapted from [3].

While the data shown in Fig 1.2 and Fig 1.3 is for the US, the same trends are present globally. Analysis by Bloomberg shows that electricity costs from utility-scale solar or wind projects are now lower than from any fossil fuel-based technologies for regions covering over 2/3rds of the world’s population. This is illustrated in Fig 1.4 below, which shows the technology with the lowest subsidy-free LCOE in different countries. The idea is that as solar prices continue to decrease, the technology will become the lowest-cost option in

15 more and more areas of the world, dragging global electricity prices lower as it does so and spreading the benefits of PV to more people and wider areas.

Fig 1.4 Graphic showing the technology with the lowest subsidy-free LCOE ($/MWh) in different parts of the world. Source Bloomberg [4].

1.1.2 Limitations of solar PV

While electricity generation from solar PV has many advantages, there are some limitations to the technology that should be noted. The main disadvantage of solar PV compared to traditional turbine-based electricity generation sources is that it is intermittent. Clearly, no useful power can be generated by PV modules at night, so they are inactive for at least half of the time, decreasing PV capacity factors. While this is an issue for the balancing of

16 electricity demand and supply at high PV penetration levels, below about 20% PV penetration these fluctuations can be balanced by the grid. Short-term (<4 hours) electricity storage options like lithium-ion batteries can also help in this regard, and these have recently experienced similar cost reductions to solar modules [10]. Another important consideration for power generation technologies is their power density, since it determines the amount of land that is needed to supply a certain amount of electricity. The annually averaged power densities for solar PV range from about 10-40 W/m2. While this is a factor of 10x lower than for fossil-fuel-based power plants, it is higher than for all other renewable technologies [8].

Ultimately, in any one location the best option for provision of power depends on numerous factors including local geography, climate, population density, power requirements, and engineering expertise, etc. Inevitably, the technologies that are adopted will therefore vary depending on where you are in the world, as illustrated by the map in

Fig 1.4.

The increased spread of solar PV therefore requires further rises in module efficiency and decreases in module production costs, to incentivise deployment in areas with less favourable local conditions such as a poor solar resource. In the next section, we will describe why thin-film PV modules such as CdTe are a promising option to achieve the required cost decreases.

1.2 Why thin-film photovoltaics

There are two main approaches to producing solar panels for utility-scale solar photovoltaics.

The first is a wafer-based approach where large, slow-grown crystals are sliced into thin wafers, which are each made into a cell and then connected to one another form a module.

This is how silicon PV modules are produced. While the production method is expensive, it results in high efficiency modules.

The second approach is based on fast deposition of thin, highly absorbing semiconductor layers over large areas (>1m2) of a supporting substrate. This is a faster, lower cost production method, but generally results in lower efficiency modules.

In this section we will introduce the two different approaches, their advantages and disadvantages, and how they affect module costs and levelised costs of electricity.

1.2.1 Silicon PV: production, efficiency, and costs

The most common production process for high efficiency silicon modules is summarised in

Fig 1.5 below. First, a large single crystal silicon ingot is drawn from melted polysilicon using a seed crystal. The ingot is then sawn into thousands of wafers that are typically

~100–200 µm thick with an area ~100cm2. These are each turned into a cell by several processes including surface texturing, doping, surface passivation, and contacting. Finally,

18 the cells are electrically interconnected in series and parallel and laminated and framed to form a module.

Fig 1.5 Schematic of the production process for Czochralski-grown single crystal silicon modules. Schematic adapted from [11].

Because the ingot is cooled slowly at close to thermal equilibrium, the wafers have a very low crystal defect density. They have few or no grain boundaries (even grains in multi- crystalline ingots are typically of the order of 1 cm3 volume). By distorting atomic bonding within the lattice, crystal defects such as grain boundaries introduce electronic states within the bandgap of the material that can trap electrons and holes, causing them to recombine non-radiatively and not be extracted at the contacts, harming device efficiency.

The low defect density in silicon devices therefore enables cells to be produced that are close to their thermodynamic efficiency limits (the thermodynamic limit is ~30% for a single junction solar cell with 1.1 eV bandgap, compared to the silicon record of 26.7%).

While this approach enables defect-free crystals and high efficiency devices to be made, the crystal growth process is slow, energy intensive, and expensive. In addition, the three key processes of: 1) crystal growth, 2) cell production, and 3) cell interconnection and module assembly, are separate processes that take place in separate production lines or factories.

Production of silicon modules therefore typically requires four or more different factories and takes over three days from start to finish.

Despite the involved production process, silicon module costs now average less than 23 cents/watt [12]. This is the result of a process of silicon module efficiency increases and rapid expansion of production capacity over the last few decades, principally in China, which have brought increased economies of scale and lower module costs. Fig 1.6 shows an analysis by Fraunhofer ISE of the change in silicon solar module prices with increased cumulative production between 2006 and 2015 [13]. It shows that with each doubling of cumulative module production, module costs have decreased by an average of 28.2%

(called the learning rate). This drop in the price of silicon modules has been the main driver of the dramatic reduction in overall costs for solar farms shown in Fig 1.2 and Fig 1.3.

Fig 1.6 The price vs experience ‘learning’ curve for silicon PV module production between 2006 and 2015. Source: Fraunhofer ISE [13].

1.2.2 Thin-film PV: production, efficiency, and costs

The relatively high production cost of wafer-based, silicon solar cells have driven efforts to develop modules that can be produced at significantly lower cost. One way to do this is through thin-film photovoltaics.

In thin-film PV, thin semiconductor films and contact layers are deposited directly onto the supporting substrate of the module, which is typically glass and has an area >1m2. The module area is then divided into cells by a process such as laser scribing. A scribed device structure for CdTe is shown in the schematic in Fig 1.7. Making the modules in this way means that the functional semiconductor material can be deposited over large areas rapidly, enabling fast module production. Importantly, it also means that the three key processes of semiconductor crystal growth, cell production, and module production can all be performed in a single, continuous production line, lowering costs. For instance CdTe module production, from glass to finished module, takes less than 3.5 hours and only requires one factory, compared to 3 days and multiple factories for silicon.

Fig 1.7 Schematic of the superstrate CdTe device structure and series interconnection [14]. The transparent conducting oxide acts as the front contact for the delineated cells, and the CdTe as the absorber layer.

While the thin-film production process is faster and cheaper than that of silicon modules, it is more difficult to achieve such high device efficiencies. This is because fast-deposited thin

22 semiconductor films generally have higher crystal defect densities than slow-grown bulk crystals. The films are polycrystalline, with crystal grain sizes on the micron scale. Thin- film devices therefore have higher levels of defect-mediated non-radiative recombination of carriers, which makes it more difficult for carriers generated in the absorber to be extracted at the contacts of the cell. While one benefit of thin-film absorber layers is that carriers do not have to diffuse so far to reach the contacts, meaning thin-film cells can generally tolerate shorter carrier diffusion lengths than silicon devices, this does not compensate for the higher defect density in thin-film devices (since the absorbing layer is only a few microns thick, direct-bandgap semiconductors must be used in thin-film PV so that the incident light can be absorbed in this short distance). The key issue in thin-film device development is therefore to try and minimise both the number and impact of crystal defects present in the semiconductor stack.

Despite having lower efficiency modules, thin-film technology has been able to compete with silicon because of its lower $/W production costs. Fig 1.8 below shows the learning curve for the leading thin-film PV technology, cadmium telluride, plotted alongside the silicon learning curve from Fig 1.6. The chart shows that in 2006, thin-film module prices were slightly lower than those of silicon. However, at this point the annual production capacity of CdTe modules was only ~0.2 GW, whereas capacity for silicon was a factor of ten higher at 2 GW. CdTe module prices, which are a reasonable proxy for module costs, were therefore lower despite not enjoying the same economies of scale as silicon. Since then, this trend has continued, and so the CdTe curve is shifted down from the silicon curve

(learning rates are similar for the two technologies, at 28.2% for Si and 25.6% for CdTe).

Using the plot, it is possible to compare the module prices of the two technologies when

23 both are at equivalent levels of production. For instance, at 10 GW production capacity,

CdTe module prices were ~0.7-0.8 Euro/W, whereas silicon prices were four times higher at ~3 Euro/W. This helps to illustrate the inherent cost advantages of thin-film photovoltaics compared to wafer-based technologies like silicon (this is backed up by analysis by Lazard that shows that in the US, levelised cost of electricity from thin-film

CdTe solar farms is lower than from silicon farms [3]).

Fig 1.8 The price vs experience learning curve for silicon (blue) and CdTe (green) PV module production between 2006 and 2015. Source: Fraunhofer ISE [13].

Thin-film PV, and CdTe in particular, is therefore a promising candidate to achieve the cost reductions and scale-up that is required in the solar industry.

In the following section, we briefly describe the recent improvements in record CdTe efficiency through the addition of chlorine and selenium to the absorber layer, and state the two main aims for the thesis.

1.3 CdTe photovoltaics

Despite containing a high density of crystal defects, thin-film cadmium telluride devices have achieved efficiencies of over 22% (21.0% for a device that is over 1cm2 [15]). This is nearly as high as the record for multi-crystalline silicon at 22.3% [7], [16], even though absorber grain sizes in CdTe are a factor of ~1,000 times smaller than silicon grains by diameter (part of this tolerance to higher defect densities is due to the thinness of the CdTe absorber, which means that carriers have a shorter distance to travel before being extracted at the contacts and so can accommodate shorter diffusion lengths). In addition to matching multicrystalline silicon, the 22.1% record for CdTe is significantly higher than the record for multi-crystalline GaAs cells at 18.2%, even though single crystal GaAs cells hold the single junction solar PV efficiency record at 29.1% [7], [17]. Investigating the reasons why CdTe has managed to reach these performance levels despite its fast growth and small grain structure is a central theme of the thesis.

In 2011, the record cadmium telluride solar cell efficiency was 16.7%. In a period of just 6 years after that, a series of device improvements from GE and First Solar Inc raised efficiency to the current record of 22.1% (see Fig 1.9). First Solar attributed early improvements, from 18% to 19.5%, to optimisation of the cadmium chloride (CdCl2) heat treatment process [18]. During the treatment, the CdTe absorber layer is exposed to CdCl2 or MgCl2 (as a solid, liquid or gas) while the substrate is heated to typically ~420°C. While the treatment has been used for decades to improve CdTe performance, it is not well understood how it improves efficiency [19].

Fig 1.9 Graph showing the increase in the record CdTe research cell efficiency since 1990, with shading showing what the performance improvements were attributed to [18]. FSLR refers to First Solar Inc.

The most recent CdTe efficiency improvements, from 19.5% to 22.1%, were attributed to the addition of selenium to the front of the absorber layer, creating a CdSexTe1-x alloy

(where x is typically ~0.2). Selenium was initially added to the CdTe to lower its bandgap and increase absorption in the long wavelength part of the visible spectrum. However, as well as increasing device current densities, cell voltages were maintained or improved – despite the lower band gap. As such, while the addition of selenium to the CdTe absorber is critical for the production of the highest efficiency modern CdTe-based devices, it is not well understood how the alloying improves cell voltage and efficiency.

The optimisation and addition of chlorine and selenium to CdTe devices has been instrumental in their recent increase in efficiency from under 17% to over 22%, however the mechanisms behind the performance improvements for both elements are not well

27 understood. Furthering knowledge of the effects of chlorine and selenium in CdTe devices therefore provides the two main aims of the thesis.

1.3.1 Thesis aims

Aim 1:

To improve understanding of the ways in which the cadmium chloride heat

treatment improves the efficiency of CdTe solar cells.

For instance, it is known that chlorine segregates to CdTe grain boundaries and

partially passivates them following cadmium chloride treatment, but what other

parts of the cell are affected by CdCl2 treatment? What is the effect of the treatment

on grain interior defects and the front interface, and what are the effects of the

treatment on sulphur interdiffusion in CdS/CdTe solar cells?

Aim 2:

To determine the sources of selenium-related efficiency improvements in CdSeTe

solar cells.

For instance, to what extent are improvements down to a change in the 1D band

structure through the device, an improvement in the front interface, and/or

reduction in carrier recombination in grain interiors and grain boundaries?

Chapter 2 - Approach to cell characterisation

The best approach to characterisation and performance improvement of a PV cell depends on the type of cell that is being characterised. Different characterisation approaches are therefore needed for silicon and thin-film cells. Because of their slow crystal growth, silicon cells have a relatively simple device structure and have generally well-known, standardised wafer materials properties, and so there is less of a need for advanced materials characterisation. Thin-film devices on the other hand have a much more complicated microstructural and chemical makeup, because of their rapid deposition and high defect densities. This makes it more difficult to predict the microstructure and composition of each cell that is fabricated, and therefore how this will affect electronic performance. For thin-film photovoltaics there therefore needs to be a much greater emphasis on materials characterisation to find out why a device is not performing as well as hoped.

2.1 The simplicity of silicon devices

Fig 2.1 shows a schematic of a basic monocrystalline silicon solar cell. The thickness of the absorber layer is about 100 to 200 microns. Most of this thickness comprises of p-type, boron-doped silicon. A p-n junction is formed by n-type doping the material at the front of the device with phosphorous. The front of the device can be textured by using an anisotropic chemical etch to form pyramids that help reduce reflection of light form the

29 front surface and increase the optical path length in the silicon. Modern devices also have some form of surface passivation at the front and back of the device.

Fig 2.1 Schematic of the different layers in a basic silicon solar cell [20].

2.1.1 Silicon cell microstructure

In single crystal silicon devices there are no grain boundaries, so there is little variation in materials properties laterally across the cell. While grain boundaries are present in multi- crystalline silicon, the densities are low because grain sizes are ~1cm, which is 100x larger than the absorber thickness. The main extended defects that are present in single crystal silicon devices are dislocations. These are linear defects at the ends of missing, extra, or shifted planes of atoms in the lattice. However, dislocation densities are generally relatively low and manufactures offer wafers with a range of different dislocation densities. Wafers are grown at close to thermal equilibrium, so point defect densities are lower than in rapidly cooled semiconductor material.

2.1.2 Silicon cell chemistry

Silicon is an elemental semiconductor, so the only intrinsic point defects that can form are vacancies, where an atom is missing from the lattice, or interstitials, where there is an extra atom in the lattice. This simplifies the cell’s chemical make-up significantly. Intentional impurities in silicon devices include boron for p-type doping and nitrogen for n-type doping. Doping is well controlled whether the doping is achieved during wafer production, or extrinsically during cell production. Unintentional impurity densities are low in solar- grade silicon, which is at least 99.9999% pure silicon by weight.

The result of the simplicity of silicon devices is that it is easy to predict the microstructural and chemical make-up of the active semiconductor layers following cell fabrication.

Differences from one cell to another will be small, and generally intentional and predictable. The simple, homogeneous, and predictable nature of the devices means that bulk cell characterisation will suffice and there is not often need for advanced microstructural or chemical characterisation following cell fabrication, and little need for frequent high-resolution studies.

2.2 The complexity of thin-film devices

2.2.1 Complex microstructure

Fig 2.2 shows a schematic of the microstructure of a typical thin-film device stack. The main features are labelled in the schematic, and described below (while CdTe is used as an example, the microstructure of most thin-film devcies is similar).

Fig 2.2 Schematic of the typical microstructural features present in a thin-film CdTe solar cell.

Grain boundaries. Because thin-film PV devices are fast-deposited, grain sizes in each layer are not much greater than the film thickness (so grains are generally < 10 µm). This means they have a high density of grain boundaries, adding significantly to the

32 microstructural complexity of the absorber layer. Disruption of the normal atomic bonding at the boundaries (dangling bonds, wrong bonds etc) introduces electron and hole traps that act as centres for non-radiative recombination of carriers and a significant source of power loss from the device. The exact atomic configuration at the boundary is dependent on a number of factors such as the relative misorientation angles of the two grains and the angle of the plane of the grain boundary, and so there are thousands of different types of grain boundary even in just one material system and one polycrystalline film.

Twins and stacking faults: Most semiconductor materials for thin-film solar cells have the adamantine crystal structure and have a low twinning formation energy. Stacking faults and twins can therefore form particularly easily, especially in fast-grown films [21].

Dislocations: dislocations are linear defects that form at the end of planes of extra, missing or shifted atoms in a crystal. They are present when stacking faults terminate mid-grain.

The low stacking fault energy of thin-film PV materials therefore increases the likelihood of dislocation formation.

Front and back interface: as shown in the schematic, the absorber layer of thin-film solar cells is typically deposited onto thin semiconductor layers that have very small grains, tens of nm in diameter. Interfaces in polycrystalline PV devices are therefore almost always non-epitaxial (i.e. like a grain boundary), because each grain in the absorber is in contact with tens or hundreds of buffer layer grains, and so can’t be epitaxially related to all of them, adding to the microstructural complexity of the devices.

The high density of closely spaced defects present in thin-film devices makes diagnosis of the causes of efficiency loss problematic. For instance if efficiency is limited by low carrier

33 lifetimes in the absorber layer, which is frequently the case with thin-film PV devices, then it is difficult to know to whether it is due to poor quality bulk, linear defects in the grain interior, grain boundary recombination or a poor front/back interface.

Bulk characterisation techniques such as X-ray diffraction XRD can give an indication of the overall structural quality of the absorber layer but cannot show what kind of defects are present, and so cannot inform how device deposition and processing might be changed to eliminate the harmful defects. Any high-resolution microstructural characterisation technique that can inform of the type, location and density of defects is therefore an important tool in identifying how efficiency might be improved in thin-film devices.

2.2.2 Complex chemistry

Characterisation techniques such as electron backscatter diffraction and especially transmission electron microscopy are excellent at determining the type, density and location of extended defects in thin-film PV devices. However, even if it is known what defects are present in a device, it is still difficult to be sure which of them are having a significant effect on performance. Predictions of the electronic effects of defects can be made by reviewing the literature on prior, more controlled experiments, or based on atomic scale modelling using techniques such a density functional theory (DFT). However, the electronic effects of defects are strongly influenced by their local chemical environment, which changes from one film to another and may even vary across the length of a single defect. Therefore, the likely electronic effects of defects that are present within a film are not predictable without knowledge of elemental distributions within the film. In

34 addition, the basic electronic properties of semiconductors are determined by their chemical make-up. Taking a purely microstructural view of a thin-film stack is therefore in many cases not sufficient to ascertain why one device performs better than another. This is particularly important given the inherently complex and unpredictable chemistry of thin- film PV devices. In Fig 2.3 below we show a schematic of the chemical complexity of a CdTe cells, alongside our schematic from Fig 2.2. The typical compositional features are shown in the figure and described in the text below.

Microstructure Composition

Fig 2.3 Labelled schematics of typical CdTe microstructure (left) and composition (right). Information on the example CdTe impurities is taken from [22].

Intrinsic defects: the types of point defect that can form in elemental semiconductors such as silicon are limited because only one element is typically present in the crystal. Most thin-

35 film PV technologies are based on semiconductors comprising more than one element, which immediately increases the chemical complexity of the devices. For instance, since cadmium telluride is a compound semiconductor with two elements making up the crystal, antisite point defects can form where a cadmium atom is on a tellurium site, or visa versa, and larger off-stoichiometric regions can form where the Cd:Te ratio isn’t exactly 1:1. This complexity is increased in ternary semiconductor materials like copper indium sulphide, and even more so in quaternary semiconductors made up of four elements, like CIGS.

Intentional impurities: in addition to elements that make up the crystal structure of the material, impurities are often added to semiconductor layers to alter the material’s properties. This could be through doping, such as copper or group V doping in CdTe. In addition to doping, treatments are often needed to try and ‘passivate’ the high density of defects present in the fast-deposited absorber layer, such as with the cadmium chloride treatment in CdTe. Introduction of these elements can have unintended consequences on device performance and adds more compositional complexity to the thin-film device stack.

Unintentional impurities: as well as intentional impurities it is inevitable that other, unintentional impurities will enter the various semiconductor layers during fabrication of thin-film devices [22]. To start with, source materials typically have one part per million impurities. Impurities can also be introduced during film deposition, either from the deposition system (in vacuum deposited films) or from the ambient, particularly when films are deposited in atmosphere such as perovskites. Finally, materials such as CuCl and

CdCl2 that are used for doping or passivation treatments are themselves not totally pure.

‘Self’ impurities: Another source of impurities into any particular layer of the film is through diffusion of elements from other layers in the device stack. Because of the thinness of films in PV devices, the large number of different layers in the semiconductor stack (and therefore numerous different materials in the system as a whole), and the high deposition and treatment temperatures, there is significant interdiffusion of elements between different layers in thin-film PV device stacks. This adds further mixing to the complicated chemical ‘soup’ present in thin-film PV devices.

This is indicative of a general theme within thin-film photovoltaics research. Because there are a large number of different defects in the thin-film stack, and because they are so closely spaced, the chances of the defects interacting with one another is increased.

Microstructural defects can influence the composition of the film by acting as conduits for diffusion of elements. Film composition itself can affect microstructure, since it determines the formation energy of different kinds of defect and material phases. Microstructural defects also interact with one another: dislocations determine the growth and movement of stacking faults, and point defects strain the lattice and so can be attracted to or repelled by one another, and extended defects. The result of this interconnectedness is that thin-film semiconductor stacks are a complicated, inter-related system of impurities and crystal defects. Any perturbation of the system, say through a change in device processing, can therefore cause a cascade of different effects, resulting in so many changes to the device microstructure and composition that it is impossible to tell which of the changes altered cell performance. An example of this is the cadmium chlorine treatment in CdTe.

The interrelatedness also means that is it difficult to change just one feature or part of the film stack without inadvertently changing several other parts. This lack of precise control limits the usefulness of experiments where researchers try and alter just one element of the cell and see how it affects performance, and so slows the deposition-characterisation- improvement cycle. In this case, the best way of understanding what has happened during device processing is to characterise the myriad of microstructural and compositional changes in as much detail as possible, and to match those to local changes in material electronic properties using high-resolution correlative characterisation of device electronics. As such, for the thesis we have tried to do this wherever possible.

2.3 CdTe cell designs

Fig 2.4 shows schematic drawings of three different CdTe device designs. Part (a) on the left shows an example of the traditional CdS/CdTe cell stack that has been a staple of CdTe research for decades. This absorber/buffer combination held the world efficiency record at

16.7% until 2011 [23] (a CdS/CdTe device is characterised in chapter 3). To make a

CdS/CdTe device, a TCO, typically fluorine-doped tin oxide, is first deposited on a glass substrate. This is followed by cadmium sulphide, which acts as the n-type partner for the p- type CdTe absorber. Finally, a back contact is applied to the device (the contact shown is specific to CSU-made devices). A problem with this design is that the CdS ‘window’ layer has a band gap of 2.4 eV and so parasitically absorbs photons with higher energy than this, leading to a lower device current density. As a result, there has been significant research focus on replacing the CdS layer with a higher band gap buffer layer. An example of this is

38 magnesium-doped zinc oxide (MgZnO, Bg 3.7 eV), which is shown in part (b) of the figure.

MgZnO has been successfully implemented in CdTe devices and has led to higher device currents and efficiencies than CdS-based devices [24]. A MgZnO-based device is characterised in chapter 4. Part (c) shows a schematic of a device with a selenium-alloyed absorber layer, giving the bilayer CdSeTe/CdTe absorber structure that was described in the section 1.3, and that has resulted in high efficiencies. Bilayer CdSeTe/CdTe devices are characterised in chapters 5, 6, and 7.

Fig 2.4 Schematics of three different CdTe device architectures (a) CdS/CdTe, (b) MgZnO/CdTe, (c) MgZnO/CdSeTe/CdTe. Schematics are by Amit Munshi at Colorado State University [25].

2.4 Characterisation methods

In the thesis NanoSIMS and cathodoluminescence are used to map the chemical composition and electronic properties, respectively, of CdTe devices at high resolution. In chapter 5, the two techniques are combined on the same area of material, which is the first time that this has been done. The two techniques are introduced below.

2.4.1 NanoSIMS

During a secondary ion mass spectrometry (SIMS) measurement a beam of ions is accelerated towards the surface of a material that is in a vacuum (the ions typically have an energy of tens of keV). The accelerated beam is called the primary ion beam. When the ions impinge on the material, they transfer their momentum to atoms at the material surface, knocking them free into the vacuum in a process called sputtering. Depending on the ion beam energy, this happens to a depth of between a few nanometres and a few microns from the sample surface [26]. Most of the material that is released from this process is charge neutral. However, a small proportion of it is charged (between 10-5 and 10-2) [27].

These are what is referred to as secondary ions. Since they are charged, they can be extracted by an electric field and accelerated towards a mass spectrometer where they can be identified. In the mass spectrometer, which in the NanoSIMS is a magnetic sector mass spectrometer, species of a different charge-to-mass ratio are separated by a magnetic field that is perpendicular to the initial extraction direction of the secondary ions.

The sputtering process and the release of secondary ions is shown in Fig 2.5 Schematic of the sputtering process at the surface of a material in a NanoSIMS, showing the primary ion beam, atoms in the sample surface, sputtered material and secondary ions. The schematic

40 is from the Dynamic SIMS ‘Essential Knowledge Briefing’, published by John Wiley & Sons

[26].Fig 2.5. During the process, some primary ions are implanted into the material below the sputtering crater. The primary ion beam can also knock forward atoms from near the surface and bury them deeper into the material below. This needs to be considered for interpretation of the SIMS depth profiles.

Fig 2.5 Schematic of the sputtering process at the surface of a material in a NanoSIMS, showing the primary ion beam, atoms in the sample surface, sputtered material and secondary ions. The schematic is from the Dynamic SIMS ‘Essential Knowledge Briefing’, published by John Wiley & Sons [26].

In traditional SIMS measurements, the primary ion beam has very high current but a large spot size (~400 µm2). This gives high sensitivity but a low lateral resolution, and so it is typically used for 1D, depth-profile measurements. In NanoSIMS the beam current is much smaller, typically a few nA, but it is focussed onto a much smaller area giving a higher resolution (down to ~50 nm, where >68% of the beam flux is within this diameter). The

NanoSIMS can therefore be scanned across the sample and a high resolution 2D compositional map can be built up of the sample. During the raster, a thin layer of material is sputtered from the surface of the sample. Repeating this process on the same area enables a 3D map to be built up of the sputtered volume.

Fig 2.6 below shows a schematic of a Cameca NanoSIMS 50L (the machine used for the thesis work was a Cameca NanoSIMS 50 which, while otherwise similar, has parallel detection of only five ionic species rather than seven). In the NanoSIMS system the primary ion beam can be either Cs+ or O-. The Cs+ beam is used for the detection of electronegative elements, and the oxygen beam for electropositive elements.

Fig 2.6 Schematic diagram of the Cameca NanoSIMS instrument used for the thesis work (diagram shows the Cameca 50L, with parallel detection of seven ionic species, whereas the instrument used was the Cameca 50 which has parallel detection of 5 ionic species). The schematic is from the Cameca NanoSIMS instrumentation booklet. The caesium beam was used for the analysis in the thesis because most of the elements of interest for the work are electronegative (Cl, Se, Te, O, etc.). For the measurement, the Cs+ beam is accelerated over a net potential difference of 16 keV between the source and the sample surface. The beam current ranges from 0.1 pA to over 5nA [27]. The beam is focussed and deflected towards the sample surface by electrostatic lenses. At typical beam conditions the beam spot size is 100 nm (1pA). The smallest achievable beam spot size is less than 50 nm (0.3 pA). The beam can raster over a maximum area of 200 µm x 200 µm.

In practice, during imaging the raster areas are much smaller than this because the beam begins to defocus at the edges of the scan and resolution is lost.

In conventional SIMS measurements, the primary ion beam is incident at an angle to the sample surface, and the secondary ion beam is extracted normal to the surface (see Fig 2.7).

This is to make space for the two sets of lenses. In a NanoSIMS, both the primary ion beam and secondary ion beam are normal to the sample surface, since they are focussed and accelerated by the same set of lenses. Having a primary ion beam that is normal to the surface like this reduces shadowing effects from any surface topography. In addition, the co-axial configuration means that the ion optical elements can be much closer to the sample surface than in conventional SIMS (400 µm vs ~1 cm). This reduces aberrations, enables the beam spot size to be reduced, and increases the secondary ion yield [27].

Fig 2.7 Schematic diagram showing the arrangements of the primary and secondary ion beams in conventional SIMS and NanoSIMS. The schematic is from the Cameca NanoSIMS instrumentation booklet.

Much of the work that has been done characterising elemental distributions in cadmium telluride PV has been with time-of-flight SIMS (ToF SIMS). In 2016, Mao et al [28] investigated the 2D distribution of chlorine, sulphur, oxygen and copper in a cadmium chloride treated CdS/CdTe solar cell by ToF SIMS. They found that chlorine, sulphur and oxygen all segregated to grain boundaries in the CdTe absorber, while no segregation of copper was detected. In another paper in 2014, Mao et al [29] performed EBSD measurements on the same region mapped with ToF SIMS. This enabled the relationship between grain boundary misorientation angle and chlorine concentration at the boundary to be plotted. Chlorine concentrations were seen to be almost double at high misorientation angles (45 – 50 degrees) compared to lower angles (10 – 20 degrees). The

3D distribution of chlorine and sulphur in a substrate CdTe device has been measured by

Kranz et al, also by ToF SIMS [30]. Harvey et al performed 3D ToF SIMS in the absorber layer of a superstrate CdTe device, measuring the 3D distribution of chlorine only [31].

High resolution dynamic SIMS (NanoSIMS) has not been used before to characterise CdTe solar cells, despite its capability for higher sensitivity than ToF SIMS. In addition, high resolution elemental mapping like described above has not been combined with high resolution electronic characterisation using cathodoluminescence.

2.4.2 Cathodoluminescence

The term Luminescence refers to any light that is emitted from a material in excess of thermal radiation (i.e. due to it being a black body at a certain temperature) [32].

Luminescence can be emitted when electrons, which have been excited to a higher energy

45 level, relax to a lower energy level and recombine with a hole. The electrons can be excited by different sources of energy including, for instance: light (photoluminescence, PL), chemical reactions (chemiluminescence), or an electric field (electroluminescence). If electrons in the material are excited by external electrons that are incident on it then the process is called cathodoluminescence (CL). In a typical CL system, an incident electron beam excites electron-hole pairs where it enters the material, and light from resulting radiative recombination is collected and fed to a detector. The number and wavelengths of photons that are emitted from the material depend on its band structure, and therefore the emitted light provides information about the material electronic properties local to the beam. If the beam is then moved across the surface of the material, then a map can be built up of the variations in electronic properties over the area covered. This can then be correlated to maps of the material microstructure and composition. The principal advantage of CL over other luminescence techniques such as PL is that it has a higher resolution.

After an electron has been accelerated onto the sample surface, it undergoes a series of elastic and inelastic scattering events. As such, each incident electron can generate thousands of electron-hole pairs. This is in contrast to photoluminescence, where one photon generally generates a maximum of one electron. The number of carriers generated per incident electron depends linearly on the electron beam energy, and is inversely proportional to the bandgap of the material [32]. Fig 2.8 shows monte carlo simulations of electron trajectories in gallium arsenide for different beam energies. If enough trajectories are simulated, they can give an impression of the size of the ‘interaction volume’ between the electron beam and the material. The figure shows the increase in the size of the

46 interaction volume with increased beam energy. It also indicates that a large proportion of the beam energy is dissipated near the point where the electron beam enters the material.

Generally, the higher the atomic number of the material the smaller the interaction volume.

CL resolution is determined mainly by the size of the electron beam-sample interaction volume. Since the carrier density decreases rapidly outside the interaction volume, even in long diffusion length materials, resolution is mainly determined by the size of the interaction volume rather than the diffusion length [32]. Following carrier generation, electron-hole pairs can recombine either radiatively or non-radiatively. Non-radiative recombination is typically via trap states that are present near the middle of the band gap.

Fig 2.8 Plots of 100 Monte Carlo modelled electron beam trajectories in GaAs at incident beam energies of 10, 20, and 30 keV [32].

Fig 2.9 shows a schematic of a standard CL setup based in the SEM. The electron beam passes through a hole in a parabolic mirror and hits the sample. The resulting luminescence is then funnelled by the mirror and can either be directed straight to the PMT detector, for panchromatic imaging of all wavelengths, or via a monochromator to the

PMT/CCD. The monochromator splits the spectrum into its component wavelengths and enables hyperspectral imaging, where a full spectrum is collected at each pixel in the image

(this is used in chapter 5).

Fig 2.9 Schematic of a typical SEM-based cathodoluminescence system including electron beam, parabolic mirror, monochromator and PMT detector, from [33].

A body of work from Moseley et al [34]–[39], has used SEM-based cathodoluminescence imaging to characterise the electronic properties of CdTe films. Use of a low voltage beam in the SEM limits the interaction volume of the beam with the sample to less than around

200 nm [39], with most of the carriers generated in a volume smaller than this. This enables luminescence properties of grain boundaries to be distinguished from those of grain interiors. Results show that grain boundaries are significant recombination centres in untreated CdTe films [39]. Following treatment, grain boundary recombination is reduced. However, GB luminescence is still below that of grain interiors, indicating that even following CdCl2 passivation, grain boundaries remain detrimental to cell performance.

Measurements where CL intensity maps were compared to EBSD maps in exactly the same

49 region revealed that grain boundary CL intensity increases with increasing boundary misorientation angle [40].

Stechmann et al [41] performed high resolution CL measurements in combination with 3D

EBSD tomography. This meant a 3D representation of the grain boundary network deep into the material could be produced. This is beneficial because in ordinary circumstances, the beam’s interaction volume extends up to a few hundred nanometres below the surface, and the luminescence it causes is therefore influenced by microstructural features that are not exactly represented by the 2D grain structure at the surface. Mendis et al have also performed high resolution Cl on CdTe films [42]. Measurements again showed lower luminescence efficiency at grain boundaries.

Cathodoluminescence is most frequently used in the SEM. Recently, Gatan has developed a commercial holder that enables TEM-based CL imaging [43]. Fig 2.10 shows a schematic of the holder. Mirrors are positioned above and below the sample to collect CL emission and feed the light to optic fibres, and there are holes so that the electron beam can pass through the sample as normal.

Fig 2.10 Schematic of the tip of the Gatan TEM-CL holder, from the Gatan ‘Vulcan’ website [43].

One advantage of implementing CL in the TEM is that it makes higher resolution imaging possible. For instance, Fig 2.11 shows a comparison of monte carlo simulations of the spread of two different beams in a 50 nm thick GaN sample [44]. In part (a) of the figure the beam energy is 5keV, and it can be seen that there is significant spread of the beam in the material, just as seen in the monte carlo simulations in Fig 2.8. This is representative of

SEM beam energies. In part (b) the beam energy is 60 keV, which is more representative of the beam energies used in STEM imaging. It can be seen that the beam energy is deposited in only a very narrow region through the sample. While this enables a smaller interaction volume and hence higher resolution imaging, there is lower signal than in SEM-CL because of the smaller energy loss of the beam, and because of the close proximity of the two free surfaces of the TEM foil. TEM-CL is used in chapter 7 to assess carrier recombination at grain boundaries in selenium-alloyed CdTe solar cells.

Fig 2.11 Monte Carlo simulations of the spread of electron beams in 50 nm of gallium nitride at 5 keV (a) and 60 keV, from [44]. Lines show the percentage of the beam energy which is deposited within that region in the sample.

Chapter 3 - 3D Distributions of Chlorine and Sulphur Impurities in a Thin-Film Cadmium Telluride Solar Cell

This chapter is modified from a paper published in MRS Advances in 2017

3.1 Introduction

Chlorine and sulphur are known to have significant effects on the electronic properties of

CdTe. Chlorine passivates grain boundaries, affects doping, and can alter the band- structure of CdTe [35] [45] [46]. Sulphur alloying alters the material bandgap - affecting light absorption and carrier transport - and affects material microstructure by reducing the lattice constant [47]. It is therefore important that the location and distributions of the two elements throughout the device are known. However, their behaviour during the ubiquitous cadmium chloride (CdCl2) heat treatment is complex. For instance, chlorine segregates to grain boundaries in the CdTe, and sulphur is known to interdiffuse into the

CdTe layer from the CdS buffer layer [48][47]. Secondary Ion Mass Spectrometry (SIMS), specifically high resolution dynamic SIMS (NanoSIMS), is an ideal characterisation technique for monitoring these complex behaviours because of its high sensitivity, high spatial resolution, and its ability to produce 3D data (detection limits are µg g-1 or lower for most elements, and resolution is sub 100 nm [27]) . In this work, high resolution 3D dynamic SIMS measurements are performed on a CdS/CdTe solar cell to obtain a complete

53 analysis of the locations of chlorine and sulphur in the device. This can then be used to inform processing changes to improve cell performance.

3.2 Experimental

A plate of cadmium telluride devices was fabricated in an all-in-one vacuum process at

Colorado State University [49]. In the process, ~120 nm of cadmium sulphide (CdS) was sublimated onto an NSG Pilkington TEC12D glass substrate. This was followed by ~2 µm of

CdTe, resulting in the (glass/F:SnO2/SnO2/CdS/CdTe) device structure shown in Fig 3.1.

The stack was then exposed to previously optimised CdCl2 activation and CuCl doping treatments. The CdCl2 treatment was performed for 6 minutes and was carried out in an atmosphere of nitrogen with 2% oxygen. The stack was contacted with nickel-carbon paint and then sand-blasted to create 9 separate devices. These were electrically tested, and a

12.01% efficient cell selected for further characterisation. High resolution SIMS analysis was performed using a NanoSIMS 50 (CAMECA, France) with a 16 keV Cs+ primary beam and simultaneous collection of 5 secondary ions. The elements mapped were Cl, S, Te, F, and Cu (F and Cu are not included in this analysis). The 0.5-1 pA Cs+ beam was focused to a nominal spot size about 60 nm in diameter and stepped over the sample in a 256 × 256 pixel raster to generate secondary ions. The dwell time was 5000 µs/pixel, and the raster size was 5 × 5 µm. Measurement of the volume shown took 7 hours. To provide a flat surface for the SIMS measurements, a region of the CdTe back surface was first polished with a focussed ion beam (surface shown schematically in Fig 3.1).

Fig 3.1 Schematic of the CdS/CdTe device structure and the NanoSIMS measurement process. Layer thicknesses are to scale apart from the glass and back contact layers (Glass 3 mm, SnO2:F 400 nm, CdS 120 nm, CdTe ~2 microns). The raster area is 5 x 5 microns, and the data cube is 256 x 256 pixels on the top (the pixel side length is therefore 19.5 nm).

3.3 Results

Data on the chlorine and sulphur distributions is displayed in Fig 3.2. Part (a) on the left of the figure shows plan-view images of the Cl and S signal distributions at different sputter depths through the CdTe layer. Parts (b) (i and ii) in the centre of the figure show cross- sectional images, and part (c) on the right is a 3D rendering of the two signals in plan-view.

In all the figures, chlorine is shown in red and sulphur in yellow.

It is clear from the figure that the chlorine signal emanates mainly from grain boundary regions. There is however some signal from the grain interiors (GIs). This is clear in in the close-up in part (b) (iii) of the figure (this image has adjusted intensity ranges). A signal hot spot can also be seen in the interior of the grain. These are common in the data, and the larger 3D rendering in Fig 3.3 shows that in three dimensions they form rod- or ribbon- shaped features that generally span grains (see green arrows in the figure).

Fig 3.2 (a) Plan-view images of the chlorine (red) and sulphur (yellow) signal distributions. Left (i, iii) from near the back of the CdTe layer; right (ii, iv) from near the front interface with the CdS layer. (b) (i, ii) Cross- sectional images of the chlorine and sulphur distributions respectively. (b) (iii) Close-up of the large grain on the right-hand side of image (a) (ii), with intensity ranges altered to highlight the chlorine signal from the grain interior and the signal hot spot within the grain. (c) 3D rendering of chlorine and sulphur signal distributions. Chlorine appears as a red surface, and sulphur as a yellow ‘mist’. The blue lines and arrows indicate the location of the cross sections in the planar images and 3D reconstruction respectively. Scale bars are 1 µm.

The chlorine cross-section in Fig 3.2 shows that there is a thick horizontal layer of high chlorine signal deeper in the cell. The sulphur cross section shows that this coincides with a horizontal band of sulphur signal.

It is clear from all the images in Fig 3.2 that sulphur signal is present in the CdTe layer, and that it is of highest intensity in the grain boundary regions. This is revealed particularly clearly by the plan-view images, which show that towards the back of the cell, the sulphur signal closely mirrors the chlorine signal from the same plane (images 2.2a [i] and 2.2a

[iii]). Deeper in the cell and closer to the CdS layer, the sulphur signal is higher generally and has encroached further into the grain interiors (image 2.2a [iv]). When the signals are

56 viewed from the side in the cross-sections, this encroachment into the GI results in distinctive U-shaped intensity profiles within grains. It should be noted that this widening of the sulphur signal deeper in the cell does not occur with chlorine, and so is not a smearing or spreading effect from the beam as it sputters through the depth of the cell. The

Tellurium signal is homogeneous throughout the CdTe layer.

Fig 3.3 A 3D rendering of chlorine and sulphur distributions in the absorber of the CdS/CdTe cell (different region of the cell from Fig 3.2). Chlorine is in red, and Sulphur in yellow. Automatic drift correction is used in the reconstruction. The bounding box is 5 x 5 x 0.2 microns. Rod-shaped features in the grain interiors are indicated by the green arrows.

3.4 Discussion

The results show that following CdCl2 treatment chlorine permeates every region of the

CdTe absorber layer, and even moves into the CdS. This includes grain boundaries, grain interiors, rod-shaped features in the grain interiors, the front interface, and the CdS layer.

In each of these regions, it is likely to have significant effects on local electronic properties of the material, and therefore in each case influence the overall device performance.

At grain boundaries, chlorine is present in significantly higher concentrations than the surrounding material. This is consistent with observations from previous work on CdTe

[48]. Cathodoluminescence measurements have shown that the CdCl2 treatment has a passivation effect on the grain boundaries, but that they remain recombination centres relative to the grain interiors [35]. This passivation will clearly have a positive impact on grain boundary recombination velocity and hence device performance.

The presence of chlorine within grain interiors (GI) is expected but has not been directly measured before. Here, the signal intensity for chlorine in the interiors is two orders of magnitude above the nominal background level of the instrument [27]. Chlorine in the grain interiors will affect doping and also potentially act to reduce harmful non-radiative recombination in the grain bulk by forming complexes with, and passivating point defects

[45].

Another interesting observation from the data is the presence of rod-shaped features of high chlorine signal in the grain interiors. These are likely to be thin regions of higher order twins that contain segregated chlorine. Extended defects like these would have a sufficiently open atomic structure to allow chlorine segregation to occur at them. This open structure would also mean that they are likely to be recombination centres, despite the passivation effect of chlorine. The defects could therefore be harming device performance.

In addition, the presence of these defects in treated devices suggests that they are also present in untreated devices, where they would not be passivated. These could contribute

58 to the poor performance of untreated devices, especially since the higher density of stacking faults in untreated material indicates a higher density of higher order twins is present.

Given the necessarily open atomic structure at the front interface of the absorber layer it is not surprising that chlorine would diffuse there from the grain boundaries. Nevertheless, previous detection of chlorine at the interface has been limited to a few TEM studies [48] and an atom probe tomography (APT) investigation [50]. At the interface, chlorine is again likely to have a significant impact on device performance. This could be through: 1) passivation of mid-level states in the bandgap; and 2) through a local change in the conduction and valence band profiles across the p-n junction [50]. The local and device- level effects of chlorine at the front interface have not been widely considered, despite its potentially significant impact on device performance.

The results show that after the cadmium chloride heat treatment sulphur has moved into the CdTe from the CdS layer. The diffusion is primarily up grain boundaries, but some sulphur has entered the grain bulk as well, forming fringes around the grain boundaries.

This U-shaped diffusion profile is indicative of a type B diffusion regime in the Harrison classification system i.e. mixed grain boundary and lattice diffusion [51], and has been observed in other CdTe alloys following cadmium chlorine treatment [52][53]. There is some diffusion directly across the CdS/CdTe interface, some up CdTe grain boundaries, and some out-diffusion from the grain boundaries.

Once in the CdTe, the sulphur has several effects on the material. Firstly, sulphur alloying reduces the band gap of CdTe through band gap bowing [54]. Its U-shaped interdiffusion

59 would therefore affect band structure profiles across grain boundaries in the CdTe and through the depth of grains. This will influence light absorption and carrier transport in the device. Secondly, it has been suggested that sulphur alloying in the CdTe might reduce the lattice constant in the CdTe, and hence act to reduce the ~10% lattice mismatch between the CdTe and the CdS [55]. In addition to these factors, the effects of sulphur on recombination activity at grain boundaries are unknown. The behaviour of sulphur in CdTe is useful to study as an example system of an alloyed group 6 element in the CdTe lattice, since this occurs in the new selenium-graded CdTe cells [56].

The work suggests that as well as factors like grain size and grain boundary passivation, device processors should consider how their chosen deposition and processing conditions affect three other important factors: 1) concentrations of chlorine at the front interface; 2) the density and passivation of extended defects in the grain interiors, like higher order twin boundaries; and 3) the shape and extent of sulphur interdiffusion into the CdTe (this will affect the band gap grading profile through the depth of the device, and also laterally across grain boundaries). In addition, makers of devices with selenium graded absorbers should consider that behaviour of selenium in the CdTe layer may be similar to the mixed grain boundary and lattice diffusion seen here with sulphur.

3.5 Conclusions

High resolution 3D dynamic SIMS measurements have been performed on a thin-film

CdS/CdTe solar cell. The measurements provide a detailed analysis of the complex distributions of chlorine and sulphur in the device following the cadmium chloride

60 activation treatment. Following treatment chlorine is found to permeate every region of the absorber layer, including grain boundaries, grain interiors, extended defects within the grains, the front interface, and the CdS layer. In each of these regions, chlorine will affect the local electronic properties of the material, and therefore efficiency to varying extents.

In future work each of these effects can be quantified and put into a 2D/3D device model in order to establish the most important mechanisms for the remarkable performance improvement following the chloride heat treatment. Sulphur is found to have a U-shaped diffusion profile, which is indicative of a mixed grain boundary and lattice diffusion regime, and can provide a model for the behaviour of other group 6 elements such as Selenium in

CdTe.

Contributions:

See ‘publications list’ (page 3) for a full list of authors of the associated manuscript.

TF: conception, direction of measurements, sample preparation using focused ion beam, data analysis & image processing, writing of manuscript/chapter.

Cells were fabricated by Amit Munshi at Colorado State University. NanoSIMS was performed by Kexue Li at Oxford University.

Chapter 4 - Large Area 3D Elemental Mapping of a MgZnO/CdTe Solar Cell with Correlative EBSD Measurements

This chapter is modified from a paper published in the proceedings of the 2018 IEEE 7th

World Conference on Photovoltaic Energy Conversion

4.1 Introduction

For CdTe to approach its single junction efficiency limit of ~30%, cell voltages need to improve [57], [58]. While device simulations have shown that this can be achieved by simultaneously improving carrier lifetimes and carrier concentrations, non-radiative carrier recombination at crystal defects in the material currently limits lifetimes and hence cell voltages [59].

The ubiquitous cadmium chloride (CdCl2) heat treatment is known to help in this regard, as chlorine that enters the device passivates grain boundaries to some extent [35]. However, the treatment – which takes devices from a maximum of a few percent efficiency, up the typical values seen of over 10% – causes a myriad of other microstructural and chemical changes in the device. It is therefore difficult to discern which are the most important in improving efficiency. The changes include a reduction in the density of stacking faults and twins in the grain interiors, inter-diffusion of sulphur from the cadmium sulphide (CdS)

62 layer into the CdTe (in CdS/CdTe devices), and doping/segregation of chlorine at the grain interiors and front interface respectively [60]–[62]. In this work we map the 3D distributions of chlorine, tellurium, oxygen, magnesium and zinc in a MgZnO/CdTe device using high resolution dynamic SIMS. This enables us to further investigate the role of these elements in improving efficiencies following the cadmium chloride treatment. The devices incorporate an MZO window layer instead of the traditional CdS since it reduces parasitic absorption and increases efficiency. The study is large area, passes through the depth of the cell, and involves correlative microstructural measurements to aid data interpretation.

4.2 Experimental

A high efficiency MgZnO/CdTe device was fabricated by Close Space Sublimation (CSS)

[49]. To make the cell, a 100 nm Mg0.23Zn0.77O (MZO) buffer layer was first deposited onto a

TEC 10 TCO-coated glass substrate by magnetron sputtering. This was followed by ~4 µm of CdTe deposited by CSS, resulting in the device structure shown in Fig 4.1. The stack was then exposed to a cadmium chloride activation treatment lasting six minutes and a copper chloride doping treatment. Finally, the cell was contacted with a layer of nickel-carbon paint. The cell was measured to have an efficiency of 15.0% under a standard AM 1.5 spectrum. To prepare the cell for further characterisation a region of the CdTe back surface was polished with a gallium Focused Ion Beam (FIB) to remove surface roughness (see schematic in Fig 4.1). Following the Ga ion polish, an Electron Back-Scatter Diffraction

(EBSD) measurement was performed on a 20 x 20 µm region on the surface of the cell. This provides information on crystal orientations and therefore grain boundary types in the analysed area. Correlative high-resolution SIMS measurements were then taken on the

63 same area with a Cameca 50 NanoSIMS. During the measurements a 0.5-1 pA Cs+ ion beam with a nominal diameter of 60 nm was rastered over the surface and sputtered secondary ions were analysed with a double-focused magnetic sector mass spectrometer.

Fig 4.1 Schematic of the device structure and the polished back surface on which the SIMS and EBSD was performed. Layer thicknesses are to scale apart from the glass and back contact (Glass 3 mm, SnO2 400 nm, MgZnO 100 nm, CdTe 4 µm, back contact 25µm).

Masses analysed were 35Cl-, 130Te-, 24Mg16O-, 64Zn16O-, and 16O-, and so a high-resolution map of each of these elements is formed following the raster (MgO and ZnO are used instead of

Mg and Zn as the oxides are negatively charged and therefore detectable by the NanoSIMS using a Cs+ primary ion source). On repeating the process a 3D data cube is built up of elemental distributions in the analysed volume.

4.3 Results

Fig 4.2(a) shows a chlorine intensity map of a region on the back surface of the CdTe layer.

Apart from the chlorine signal network at the grain boundaries there are numerous signal hot spots in the grain interiors. Part (c) of the figure shows that the hot spots do not correspond to fluctuations in the Te signal. However, the EBSD imaging performed on the

64 same region prior to the SIMS measurement reveals that the spots often occur along Σ3

(111) twin boundaries – lamellar twins. This can be seen by comparison of

Fig 4.2 (a) Chlorine intensity map taken at the back of the CdTe. The yellow annotation shows where the line profile in Fig 4.4(c) was taken. (b) EBSD band contrast map of the same region in (a). A selection of Σ3 (111) twin boundaries are highlighted in red. Numbers 1 – 4 show the equivalent grains in (a). (c) Te map of the same region shown in (a) and (b). The yellow annotation shows where the line profile in Fig 4.4(c) was taken.

the chlorine map to the EBSD map in part (b) of the figure, which has had a selection of lamellar twins highlighted. For example, hot spots in grains 1, 2, 3 and 4 clearly lie on twins in those grains (it is also worth noting that there is often more than one chlorine hot spot along a single twin boundary in a grain). Spots occur at a density of around one per 2.9 µm2, and careful inspection of the data reveals that around one third of these can be directly

65 associated with a lamellar twin boundary (it may be that more are associated, but higher resolution EBSD would be needed to show it clearly). Tomographic SIMS shows that in three dimensions the hot spots form rod-shaped features of chlorine signal that generally span grains. This can be seen in the 3D renderings of the chlorine signal in Fig 4.3 (a – d).

The close-up in Fig 4.3(c) shows two rod-shaped grain interior (GI) chlorine features that lie in the same plane, and part (d) shows two separate instances of a pair of linear chlorine features lying in the same plane.

Fig 4.3 (a) 3D rendering of the chlorine signal in the CdTe absorber layer (different region from map in Fig 4.2). The bounding area is 8.3 µm x 7.4 µm. (b-d) Re-oriented and close-up views of linear chlorine features in the grain interiors.

Another feature that is apparent from Fig 4.2 is that there are faint lines of lower signal in the tellurium map. These can be mapped directly onto the grain boundary network of

66 chlorine signal seen in Fig 4.2(a). In addition, the profile in Fig 4.4(c) below shows a dip in the tellurium signal at the point where the chlorine signal rises at the boundary. In this case the drop in the tellurium counts is between 5-10%, and this is typical for the boundaries seen in the Te map.

Fig 4.4 (a) Schematic of a kink in a Σ3 (111) lamellar twin boundary forming a Σ3 (112) double-positioning twin boundary (vertical dotted lines and horizontal red dotted line respectively). The <111> direction points vertically up the page. (b) Schematic of kinks in Σ3 (111) twins within the setting of a grain. (c) Line profile of the chlorine and tellurium counts in the regions shown in Fig 4.2(a) and Fig 4.2(c).

The 3D nature of the SIMS analysis means that the behaviour of elements through the depth of the cell can be tracked. Fig 4.5 shows reconstructed cross-sectional images of the chlorine, oxygen, magnesium and zinc signals. A layer of chlorine signal is seen at the front interface of the cell, between the CdTe and the MZO. It can be seen that there is a thin region of oxygen signal extending around 500-600 nm into the CdTe layer. However, no magnesium or zinc signal is present in the CdTe.

Fig 4.5 Cross sectional images of the 35Cl-, 16O-, 24Mg16O-, and 64Zn16O- signals across the front interface of the MZO/CdTe device. In general brighter means higher signal. In the oxygen map red corresponds to high signal intensity and blue low intensity.

4.4 Discussion

The 3D and correlative nature of the measurements, performed over a large surface area, means that they provide fresh insights into the microstructure and chemistry of MZO/CdTe solar cells. For instance, hot spots in the chlorine signal are frequently seen to lie along Σ3

(111) lamellar twin boundaries, which suggests that they are due to chlorine segregation at small kinks in the (111) twins. These kinks would form thin ribbon-shaped features connecting misaligned twins, as shown schematically in Fig 4.4(b). The linear nature of these kinks is consistent with the chlorine signal seen in the 3D NanoSIMS. In addition, the kinks would tend to span the grain, which is again consistent with the SIMS data.

Kinks in lamellar twin boundaries form incoherent twins, a common example of which is the Σ3 (112) ‘double positioning’ (DP) twin boundary, shown schematically in Fig 4.4(a).

Twins form between two crystals whose orientations are closely related. In Σ3 twin

68 boundaries they are related by a 180° rotation around the <111> direction (see grains 1 and 2 in the schematic in Fig 4.4a). In Σ3 twins 1/3rd of atomic sites are identical in both crystals; hence boundaries between two crystals like this are CSL Σ3 twin boundaries in the

CSL convention. When the boundary between the two crystals lies on the (111) plane in both the crystals – such as in the vertical dotted boundaries in Fig 4.4(a) – it is called a Σ3

(111) lamellar twin boundary. At these boundaries there is little disruption of the atomic structure across the interface and tetrahedral bonding is maintained. As such Σ3 (111) twin boundaries are thought to be electrically benign [63]. However, when the boundary lies on another crystal plane between the two CSL Σ3 grains – such as the (112) plane shown in the figure by the horizontal red dotted line - there must be non-tetrahedral bonding at the boundary and it is potentially electrically active. Modelling performed on supercells containing a proposed atomic structure for the Σ3 (112) DP twin has suggested that the defect introduces mid-gap states into the material [64]. It is therefore likely that unpassivated Double Positioning twin boundaries are non-radiative recombination centres.

However, as the SIMS measurements have shown, their open atomic structure means that chlorine is able to enter the defects during the treatment.

This would be expected to passivate the defects to some extent, as it does with random grain boundaries (cathodoluminescence measurements have shown that chlorine improves the electrical properties of general grain boundaries but that they remain non-radiative recombination centres [35], [65]). Combined with the high density of the defects in the material and their position in the centre of grains, this means that incoherent twins are probably influencing CdTe device performance. In addition, in untreated devices, the twin and stacking fault density is much higher than in treated material, meaning that the density

69 of twin and stacking fault kinks and terminations will also be higher. The defects in untreated devices will also not be passivated. The reduction in density and passivation of incoherent and higher order twins during the CdCl2 treatment is therefore another reason for efficiency improvement following the treatment.

The NanoSIMS measurements also show that there is a build-up of chlorine at the front interface of the device, where it would be expected to passivate mid-level defects as it does for grain boundaries. In contrast to cells with a CdS buffer layer, chlorine does not appear to enter the MZO in significant concentrations [62]. Importantly, no magnesium or zinc is seen to move into the CdTe, but a 500-600 nm layer of oxygen does move into the absorber.

The slight dip in tellurium signal at the grain boundaries suggests that chlorine is to some extent substitutional with the tellurium, rather than purely interstitial. This is in line with previous work employing STEM EDX and Electron Energy Loss Spectroscopy mapping

[66].

4.5 Conclusions

High resolution 3D SIMS measurements and correlative EBSD has been performed on an

MZO/CdTe solar cell. A large density of chlorine hot spots (forming rods in three dimensions) are found to be present in grain interiors in the cell. Their location along the line of Σ3 (111) lamellar twin boundaries suggests that the spots are due to chlorine segregation at kinks in the twins, i.e. incoherent twin boundaries like the Σ3 (112).

Passivation and reduction in density of these defects is therefore likely to be a reason for device efficiency improvements following CdCl2 heat treatment – alongside other known effects like passivation of general grain boundaries. In addition, the remaining incoherent

70 twins will still harm device efficiency, and so further reductions in defect density would be a worthwhile goal in CdTe device processing. Finally, neither magnesium nor zinc is found to diffuse into the CdTe layer from the MZO, however a small amount of oxygen is detected in the CdTe.

Contributions:

See ‘publications list’ (page 3) for a full list of authors of the associated manuscript.

TF: conception, direction of measurements, sample preparation using focused ion beam, EBSD measurement, data analysis & image processing, and writing of manuscript/chapter.

Cells were fabricated by Amit Munshi at Colorado State University. NanoSIMS was performed by Kexue Li at Oxford University.

Chapter 5 - Understanding the role of selenium in defect passivation for highly efficient selenium- alloyed cadmium telluride solar cells

Modified from a paper published in Nature Energy

5.1 Introduction

In the last 7 years the efficiency of cadmium telluride (CdTe) solar cells has improved from

16.7% to 22.1% [15], [67]. This has enabled the cost of CdTe photovoltaic electricity to decrease to the point where it is lower than for silicon photovoltaics, and lower than for conventional fossil fuel sources in many regions of the world [68], [69]. The most recent efficiency improvements, from 19.5% to 22.1%, have been achieved by the addition of selenium to the front of the CdTe absorber layer [18]. This creates a CdSe(x)Te(1-x) (CdSeTe) alloy that lowers the material bandgap, increases absorption in the long wavelength part of the spectrum, and increases device short-circuit current density [70]–[72]. However, in addition to improved current generation, selenium alloying maintains or improves open circuit voltages – despite the lower bandgap [56]. Recent studies have shown that this is associated with improved minority carrier lifetimes in the absorber, but it is not known why the lifetimes increase [56], [73], [74].

One reason that has been proposed for the higher carrier lifetimes is that the selenium improves band alignments between the absorber and buffer layers at the front interface of the device, reducing interfacial recombination [9]. Another is that the wider-bandgap, non- alloyed CdTe at the back of the device acts as an electron reflector and reduces back surface recombination [75]. These explanations both relate to the effects of selenium at the device level, i.e. its effects on the one-dimensional band structure through the depth of the cell.

However, the effects of selenium on the basic optoelectronic properties of the absorber material have not been investigated.

Here we use high resolution cathodoluminescence (CL) microscopy to map nanoscale variations in electronic properties in a high efficiency selenium-graded CdTe device, including variations in luminescence efficiency, diffusion length, and effective bandgap. We then compare these with secondary ion mass spectrometry (SIMS) elemental maps of the selenium distribution from the same area of the film, so that the electronic property variations can be directly related to nanoscale changes in the selenium concentration.

The results reveal that the presence of selenium causes clear and dramatic improvements in the local luminescence efficiency of the absorber material, indicating that it passivates deep-level defects in CdTe. This provides an explanation for the superior voltage and performance of selenium-alloyed CdTe devices and may enable further gains in efficiency beyond the current record level of 22.1%.

5.2 Selenium diffusion into the CdTe layer

In order to perform the investigation, a cell stack incorporating a ~1.5 µm layer of CdSeTe at the front of the absorber layer was fabricated as shown schematically in Fig 5.1(a) and

73 described in the Methods section. Following a cadmium chloride activation treatment, which is used universally to produce high efficiency CdTe-based devices [60], [62], [76],

[77], the cell was measured to have an efficiency of 16.8%. This value is high amongst CdTe photovoltaics research laboratories, and devices made using this exact deposition system and method have achieved certified conversion efficiencies of up to 18.3% [78] (only exceeded by the 22.1% champion cell fabricated by First Solar Inc [15]). After the efficiency measurement, a shallow 7° bevel was milled through the device stack using a Focussed Ion

Beam (FIB). This provided direct access to the CdSeTe layer and presented an extended, smooth cross-section for the CL and elemental mapping. Cathodoluminescence measurements were performed on the bevel surface on an area as outlined on the schematic in Fig 5.1(a). Correlative SIMS elemental mapping at high resolution was then performed on exactly the same region as the CL.

A NanoSIMS map of the selenium concentration on the bevel surface in the CdCl2-treated cell is shown in Fig 5.1(b). At the bottom of the bevel, near the front interface of the absorber layer, selenium concentrations are in the range 8-10.5 at% (cyan/white). This is similar to the as-deposited device. However, in contrast to the as-deposited device, which has a sharp interface between the CdSeTe and CdTe layers (see [79], and Supplementary

Figure 1), the CdCl2-treated cell shows a gradual decrease in the selenium concentration with distance away from the front interface. This means that selenium signal is detected towards the back of the cell, in the CdTe layer (for clarity now referred to as the

‘interdiffused CdTe’ region).

Fig 5.1 (a) Schematic of the CdSeTe/CdTe device structure and 7° bevel, with an outline of the SIMS and CL measurement area (dotted black line). (b) SIMS elemental map of the selenium distribution on the bevelled surface of the CdCl2 treated device, with the ‘skeletonised’ chlorine signal overlaid in white to help delineate grain boundaries (see Methods section for calibration of the selenium concentration). Darker blue corresponds to lower selenium concentrations and brighter cyan/white corresponds to higher concentrations, towards 10 at% (see colour bar). (c) Cathodoluminescence (CL) map of the panchromatic CL intensity, taken on the same area as the selenium map. Black and dark grey corresponds to lower CL signal intensity and brighter white to higher signal intensity (see colour bar). (d) High magnification selenium map of a region at the top of the bevel, shown by the red annotations in (b) (the image has no skeletonised chlorine overlay, so that selenium signal at the grain boundaries can be seen). The scale bar is 5 µm. (e) Corresponding high magnification CL map to the SIMS map in (d). The dashed lines in (d) and (e) highlight similarity in positioning of the selenium and CL signals. (f) Scatter plot of selenium concentration versus CL counts for equivalent regions (i.e. pixels) in (b) and (c), with grain boundary regions and voids omitted from the analysis (voids can be seen as bright white spots in the panchromatic CL in (c)).

In addition to this grading through the absorber depth, there is also some non-uniformity in the selenium concentration around the grain structure of the absorber. In the interdiffused CdTe region at the top of the bevel, close to the back contact, there is an enhanced selenium signal at the grain boundaries and around the fringes of the grains [this can be seen clearly in the high magnification image in Fig 5.1(d)]. However, at the bottom of the bevel, in the CdSeTe region, the selenium concentration is lower around the grain boundaries than in the grain interiors.

This distribution of selenium around grain boundaries indicates that during the high- temperature cadmium chloride activation treatment the grain boundaries provide channels for fast diffusion of selenium from the CdSeTe layer into the CdTe above, which then slowly out-diffuses from the grain boundaries into the grain interiors. This is a mixed grain boundary and lattice diffusion regime of type B in the Harrison classification system, and is also observed with sulphur interdiffusion in conventional CdS/CdTe devices [51], [62].

5.3 Selenium-induced defect passivation

To assess the electronic effects of selenium alloying in CdTe we have performed correlative cathodoluminescence imaging on the bevel, on the same area as the selenium map (see Fig

5.1(c)). The map shows that at the top of the bevel, in the interdiffused CdTe region, the luminescence intensity is relatively low. However at the bottom of the bevel, in the CdSeTe region, the luminescence intensity is much brighter, with counts 10 to 20 times higher than at the top (150,000 - 300,000 counts versus ~ 15,000). This steep increase in CL signal through the depth of the absorber layer mirrors the increase in selenium concentration.

In addition to this general trend down the bevel, there are variations in CL signal around the grain structure of the absorber. Firstly, it can be seen that luminescence is lower at grain boundaries than in grain interiors. This is in line with previous CL measurements on non-alloyed CdTe, and shows that grain boundaries act as non-radiative recombination centres [35], [39]–[41]. Secondly, brighter cathodoluminescence signals can be seen around the fringes of grains in the interdiffused CdTe region. This closely matches the distribution of selenium observed around grains in the SIMS maps (see the higher magnification images in Fig 5.1(d) and Fig 5.1(e)).

To quantify this selenium concentration vs luminescence relationship we have exactly aligned and repixellated the SIMS and CL images so that the intensity values in each can be compared pixel-for-pixel. This alignment is relatively straightforward because of the obvious features that delineate grain boundaries in both maps (see Supplementary Figure

2b, where the SIMS and CL images have been superimposed on top of one another). The results of this comparison are shown in the scatter plot in Fig 5.1(f). The plot shows that in regions containing < 2% selenium the CL intensity is generally less than 30,000 counts, and in regions containing > 9% selenium the CL intensity is up to 300,000 counts. Since high luminescence efficiency is indicative of low levels of defect-mediated non-radiative recombination in a semiconductor, this clear positive correlation suggests that selenium passivates a non-radiative recombination centre in the alloyed material. Moreover, we have observed this effect in multiple SEM systems and on multiple samples, including on untreated CdSeTe/CdTe devices (see Supplementary Figure 3). Together these results suggest that selenium alloying creates a passivation effect in CdTe that explains the higher open circuit voltage and improved performance of selenium-graded CdTe solar cells.

5.4 Selenium related sub-bandgap states

The cathodoluminescence image shown in Fig 5.1 is of the panchromatic CL signal. This means that the intensity value in each pixel is the total number of photons collected by the

CL system over all wavelengths (i.e 700 – 999 nm) when the electron beam is in that region of the sample. However, important information is contained in the specific wavelengths of light that are emitted from the material, which is not displayed in the panchromatic image since it is a simple photon count. As such, in this section we build on the general picture provided by the panchromatic image by analysing the underlying spectra of wavelengths that make it up. This is possible because the CL data we collected on the bevel is hyperspectral, meaning that a full spectrum is collected at each pixel.

Fig 5.2(a) shows a waterfall plot of a representative sample of CL spectra from the panchromatic image of the bevel. To make the plot, a sample of 73 of the spectra in the hyperspectral map was selected at random for display. The curves from these pixels were then sorted by the concentration of selenium present in the corresponding pixel in the

SIMS map, with the curves with the highest selenium content at the back of the plot and the lowest at the front.

The data shows a steep increase in CL intensity as the selenium concentration in the corresponding pixel increases. In addition, it shows a shift of the emission peak towards lower energies with increasing selenium content. This shift is expected, since it is known that increasing the selenium concentration initially decreases the bandgap of CdSeTe alloys due to a bandgap ‘bowing’ effect [71], [80] (this effect is explored in more detail in the

“Selenium-induced bandgap gradients” section, where we map the effective bandgap of

CdSeTe material over the bevel measurement area).

Fig 5.2 Hyperspectral CL imaging shows that selenium is associated with a sub-bandgap emission peak. (a) A waterfall plot of a random sample of CL spectra taken from the bevel measurement area. Curves are sorted by the concentration of selenium present in the corresponding pixel in the SIMS map. The curves highlighted in blue and red are deconvoluted in Supplementary Figures 4a and 4b respectively. (b) A waterfall plot of the low-selenium CL spectra in (a), representing a selenium concentration range from 0 to 0.42%. The curve highlighted in red is the same as that highlighted in (a) and is deconvoluted in Supplementary Figure 4b. (c) Energy band diagram showing a selection of possible electron and hole recombination channels in the material: (i) defect-mediated non-radiative recombination, (ii) radiative band- to-band recombination with emission at 1.46 eV, (iii) radiative band-to-band recombination with emission at 1.36 eV, (iv)-(vi) radiative recombination with emission at 1.36 eV, either from or to selenium-related states in the band gap.

In the curves with a low selenium concentration (0-0.42 at%) plotted in Fig 5.2(b) we observe two distinct emission peaks: one at ~1.46 eV, which we attribute to band-to-band recombination, and a sub-bandgap peak at ~1.36 eV. The intensity of the sub-gap peak

79 increases with selenium content, indicating that selenium is the cause of the emission and creates states within the bandgap when present in the CdTe in low concentrations. Possible transitions from or to such selenium-related states are depicted in Fig 5.2(c). While the intensity of the sub-gap peak increases with selenium content, it is noted that that this does not affect the intensity and position of the band-to-band recombination peak. This is more clearly shown in the deconvoluted CL spectra in Supplementary Figure 4a and 4b, where the highlighted curves in the waterfall plots are analysed). It is also noted that the emission tail of the band-to-band recombination peak can make a contribution to the intensity of the

1.36 eV peak (see Supplementary Figure 4).

5.5 Impact of selenium concentration on diffusion lengths

The luminescence data presented here shows that higher levels of selenium in the CdTe material lead to an increased CL signal intensity. This suggests that selenium passivates a defect in bulk CdTe, decreasing non-radiative recombination and increasing cell performance. To confirm this, we can use CL to estimate local diffusion lengths in the material. This is done by analysing how the CL intensity varies in proximity to grain boundaries. An example of this analysis is shown in Fig 5.3(a). Here we take a line profile of the selenium and CL signals across a grain boundary in the interdiffused CdTe region of the cell, as shown in the inset. The profile shows that the selenium levels in grain (i) are higher than grain (ii), at 1 at% compared to ~0.03 at% respectively, resulting in a step in selenium concentration across the grain boundary. While there is some variation in the selenium

80 concentration in grain (i) the profile is uniform on the length scale of the CL analytical volume, which is ~1 µm.

Along with the change in selenium concentrations, the figure shows that there is clear asymmetry in the intensity of the CL profiles either side of the boundary. In the grain with higher selenium concentration, the total (panchromatic) CL signal plateaus at a distance > 1

µm from the grain boundary (referred to from now on as the ‘plateau distance’, and shown by red shading in the Fig 5.3(a). This compares to ~ 0.7 µm in the low-selenium grain. This is indicative of a greater diffusion length in the high-selenium grain because here, even when the electron beam is up to a micron away from the grain boundary, generated carriers experience the effects of the boundary (i.e. they are able to travel the ~ 1 µm distance to the boundary and recombine there, quenching the CL signal compared to its maximum level at the plateau). Calculated diffusion lengths from the profiles back this up, with the diffusion length in the ~1 at% selenium-containing grain being 0.25 µm, compared to 0.14 µm in the low-selenium grain (it should be noted that calculations of diffusion lengths are approximate since the model assumes that the grain boundary is a free surface where carriers generated in one grain cannot diffuse past the grain boundary into the adjacent grain [81], [82]. In addition, states on the bevelled surface may artificially decrease minority carrier diffusion lengths).

Fig 5.3 (a) Profile plot of the selenium concentration (blue line, plotted on the left y-axis) and CL intensities (black lines, plotted on the right y-axis) across a grain boundary in the interdiffused CdTe region at the top of the bevel. The exact profile area is shown by the yellow line in the inset (red in the inset is the chlorine SIMS signal, blue is the selenium SIMS signal, and (i) and (ii) indicate the grains with high- and low- selenium concentrations respectively). The solid, dashed, and dot-dashed black lines show the profile of the total, 1.43 – 1.49 eV, and 1.34 – 1.40 eV CL signals respectively (see legend). Red shading indicates the ‘plateau distance’ for the total CL signal on either side of the grain boundary i.e. the distance from the grain boundary, at 0 µm, to where the CL signal plateaus. (b) Plot of the average CL spectrum for the area analysed in panels (c - e), with shading showing the spectral windows used in the analysis. (c) Map of total panchromatic CL signal in a region at the top of the bevel, with the yellow line showing the region where the profile in (a) was taken. (d) Map of CL counts in the 1.43 – 1.49 eV spectral window shown by the shading in (b). (e) Map of CL counts in the 1.34 – 1.40 eV spectral window shown in (b). (f) Profile plot of the selenium concentration (blue line, plotted on the left y-axis), chlorine SIMS counts (dashed black line, plotted on the right y-axis), and total CL intensity (solid black line, plotted on the right y-axis) across the high-selenium CdSeTe region shown in the inset (red in the inset is the SIMS chlorine signal and cyan is the SIMS selenium signal). Chlorine signal intensity is included in the plot to show the position of the grain boundaries. Note that the profile is from a region with a relatively uniform selenium concentration so as not to influence the shape of the CL profile.

This kind of analysis was performed in two other regions with a similar step in selenium concentration across grain boundaries, with the results shown in table 1. The data shows that diffusion lengths in the grains with higher selenium concentration are consistently longer than in the low-selenium grains (note that grain boundaries were only chosen for this analysis if they had a step-change in selenium concentration across the boundary, as well as no other grain boundaries nearby that could influence the CL profile).

To gain further insight into why the diffusion lengths differ, we can assess the relative contributions to total CL signal variations of the band-to-band peak at ~ 1.46 eV and the selenium-related sub-gap peak at ~ 1.36 eV (Fig 5.3a). The 1.43 – 1.49 eV line in the figure plots the variation in CL counts within just that spectral window (shown by the right-hand grey region in Fig 5.3b). This captures intensity variations in just the band-to-band peak, and shows that variations in band-to-band transitions are symmetric either side of the grain boundary, despite the differences in selenium content (the plateau distance is ~0.7

µm on both sides of the GB). However, variations in the CL intensity in the 1.34 to 1.40 eV

83 window, where the 1.36eV selenium-related peak falls, are not symmetric across the boundary. The plateau distance for this energy range in the high-selenium grain is > 1 µm, compared to ~ 0.65 µm in the low- selenium grain. The 1-µm long plateau distance in the sub-gap peak is therefore the reason for the similarly long plateau distance observed in the total panchromatic CL curve. This demonstrates that the long lifetime of selenium-related transitions enables the longer diffusion lengths in regions with higher selenium content. In addition, the symmetry of the band-edge luminescence profile (1.43 – 1.49 eV) shows that any field-effects due to the stepped selenium profile across the grain boundary have not influenced the CL profiles, since the effect of an electric field is to weaken CL emission at all wavelengths. Therefore, any asymmetry in the field due to the selenium step would have caused asymmetry in the 1.43 – 1.49 eV profile.

Mapping variations in CL counts in these different energy windows shows that the trend for longer plateau distances in the 1.34 – 1.40 eV energy window vs the 1.43 – 1.49 eV window is present across large regions of the bevel measurement area (see Fig 5.3c – Fig

5.3e). For instance, sharp image contrast is seen in the band-to-band transitions map in Fig

5.3(d), indicating steep V-shaped drops in CL signal across grain boundaries and therefore short plateau distances. However, when the sub-gap energy window is mapped (Fig 5.3e), the contrast in the image is low. This means that signal variations across grain boundaries in the image have a shallower V-shape, indicating larger plateau distances, as demonstrated by the 1.34 – 1.40 eV line profile in Fig 5.3(a).

The diffusion length analysis has only been performed on the upper part of the bevel where the concentration of selenium is low compared to the CdSeTe region at the bottom of the

84 bevel. This is because in the CdSeTe region there are no obvious plateaus in the CL signal within the grain interiors, making it difficult to perform either a rough ‘plateau distance’ analysis as done in Fig 5.3(a) or to calculate diffusion lengths. However, this in itself is an indication of long diffusion lengths in the region, since the signal does not plateau even in large grains, such as the 2-micron diameter grain shown in Fig 5.3(f). These results provide further evidence that selenium passivates bulk defects in CdSeTe alloys, even at very low alloying fractions.

5.6 Selenium-induced bandgap gradients

The initial purpose of alloying selenium into the front of CdTe solar cell absorbers was to decrease the material bandgap and therefore increase absorption in the low-energy part of the visible spectrum. However, very little research has been published quantifying the effects of selenium on the bandgap of CdTe [71]. In this section we map the effective bandgap of CdSeTe material at high resolution over the bevel measurement area by tracking the photon energy of the dominant emission in each pixel in the CL. Fig 5.4(a) shows a SIMS map of the selenium distribution on the bevel surface, as seen in Fig 5.1(b).

Alongside this in Fig 5.4(b) is a map of the peak CL photon emission energy (i.e. effective bandgap), taken on the same area as the SIMS. The map shows that at the top of the bevel, towards the back contact of the device, the effective bandgap of the absorber is ~ 1.46 eV

(yellow). This is typical for non-alloyed CdTe [71]. Further down the bevel, there is a steady decrease in the effective bandgap down to ~ 1.36 eV in the CdSeTe region (purple).

Fig 5.4 (a) SIMS map of the selenium concentration on the bevel surface, as seen in Fig 5.1 but without the skeletonised chlorine signal overlay. Darker blue corresponds to lower selenium concentrations and brighter cyan/white corresponds to higher concentrations, towards 10 at% (see colour bar). (b) Map of the peak emission energy of the CL (i.e. the effective band gap of the material) on the same area as (a), with a skeletonised chlorine signal overlay and with pixel dimensions matched to the SIMS map. This enables pixel- for-pixel comparison of the selenium concentration and the effective band gap at each point on the bevel surface, as shown in the scatter plot in (c). Yellow/orange corresponds to a higher effective bandgap of ~1.47 eV and purple corresponds to a lower effective bandgap of ~1.37 eV (see colour bar). (c) (Blue) scatter plot of the effective bandgap vs selenium concentration at each point/pixel on the bevel, and (green) plot of the band gap vs selenium concentration reported in [71]. (d) Scatter plot of the effective bandgap vs panchromatic CL signal at each point on the bevel. Regions containing voids were omitted from the analysis in producing the scatter plot (voids can be seen as white spots mid-way down the bevel in the CL map in Fig 5.1c).

This mirrors the grading of the selenium concentration and will form a built-in field through the depth of the device.

In Fig 5.4(c) we have plotted a scatter graph of selenium concentration against effective bandgap for each pixel in the two images, in the same way as performed previously for the total CL intensity scatter plot (Fig 5.1f). The plot has a classic bowing shape with an initial, steeper decrease in bandgap followed by a slight levelling off towards 10 at% selenium.

This is consistent with the data points plotted in green (replotted from [71]), which show bandgap vs composition for untreated CdSeTe films deposited on glass using the same deposition system as used in this study. For these films, composition was obtained using energy-dispersive x-ray spectroscopy (EDX) measurements and the bandgap was extracted from absorption edge Tauc plots. Although the two curves are well-matched, the curve produced with the CL data is shifted down by ~ 0.03 eV. This may be due to the influence of excitonic peaks in the emission spectra, which would lower the energy of the luminescence peak compared to the true bandgap [83]. In addition, the CL is measured in a CdCl2 treated film incorporated into a working device, as opposed to untreated CdSeTe films on a glass substrate, so intentional absorber impurities such as chlorine may slightly alter the material bandgap.

As well as a grading through the depth of the absorber the map in Fig 5.4(b) shows variations in the effective bandgap around grain boundaries in the material. At the top of the bevel, in the interdiffused CdTe region, there is a decrease in the effective bandgap around grain boundaries and grain fringes (darker orange) vs the grain interiors (lighter

yellow). This matches the positioning of selenium seen in the SIMS maps and corresponds to a decrease in bandgap of ~0.01 eV in these regions. However, at the bottom of the bevel, in the CdSeTe region, there is an increase in effective bandgap around grain boundaries

(lighter purple). This tracks the decrease in the selenium content seen at grain boundaries in the SIMS maps in this region.

Fig 5.4(d) shows a scatter plot of effective bandgap versus total CL signal at each point on the bevel. The plot shows an initially shallow, and then steep rise in material luminescence efficiency with decreasing bandgap. This means that, for instance, decreasing the bandgap by from 1.46 eV to 1.42 eV gives a much smaller increase in defect passivation than an equal step from 1.42 eV to 1.38 eV. This information will be useful to device designers and and fabricators when deciding on, or modelling the optimum selenium grading profile.

5.7 Discussion

Remarkable progress has been made in cadmium telluride photovoltaics despite the lack of a full, fundamental understanding of absorber layer material properties. This is especially true of the record-breaking selenium-graded devices. In this work, by means of correlative cathodoluminescence and SIMS measurements, we have shown that selenium alloying enables high luminescence efficiency in alloyed CdTe, suggesting that it passivates defects in both CdCl2-treated and untreated bulk CdTe. Further evidence for this passivation effect is provided by an analysis of CL signal variations across grain boundaries, which shows longer carrier diffusion lengths in alloyed regions with higher selenium content.

Hyperspectral CL imaging shows that selenium not only shifts the band-to-band emission

77 peak from ~1.46 eV to ~1.36 eV, creating a bandgap gradient and built-in field across the absorber, but it also creates radiative sub-gap states at ~1.36 eV, which is the same energy as the lowest point on the bandgap bowing curve. The high-resolution SIMS measurements show that during the CdCl2 heat treatment process selenium diffuses from the CdSeTe into the CdTe, primarily along grain boundaries but with some out-diffusion from grain boundaries into grain interiors. This causes an excess of selenium at grain boundaries in the CdTe layer that decreases the material bandgap in those regions. In addition, it causes a selenium deficiency at grain boundaries in the CdSeTe layer, resulting in higher band gaps in these regions relative to the grain interiors. This will have unknown effects on carrier transport around grain boundaries in selenium-graded devices.

These results show that in non-alloyed CdTe there are deep-level defects in the bulk material that act as recombination centres and limit device efficiency, even following the

CdCl2 treatment. Potential candidates for these defects are cadmium vacancies (VCd) and tellurium-on-cadmium antisites (TeCd), which Density Functional Theory (DFT) modelling has shown act as harmful recombination centres [84], [85]. We suggest that these defects can be passivated or made less likely to form when selenium is present, with an increasing passivation effect for concentrations of selenium up to 10 – 11 at%. A slight levelling out of the passivation effect towards 10-11% suggests this may be close to the optimum alloying concentration in terms of absorber material quality.

This defect passivation explains the remarkable performance of CdSeTe devices. However, current high efficiency devices only have selenium in significant concentrations at the very front of the device, leaving much of the back of the absorber layer unpassivated. This

78 means that extending the selenium profile further towards the back of the device, whilst maintaining a concentration gradient and built-in field, will passivate more defects at the back of the absorber and might improve efficiencies further. In addition, we note that the energy of the sub-gap selenium-related emission (~ 1.36 eV) is similar to the lowest achievable effective bandgap energy in the selenium-rich layer, which also shows the highest luminescence intensity. This indicates that all these effects can be related to a common underlying mechanism.

In summary, in this work we have correlated variations in the electronic properties of selenium-alloyed CdTe with local variations in the selenium concentration. We find a strong correlation between selenium concentration, high material luminescence efficiency, sub-gap transitions at 1.36 eV, and longer diffusion lengths, all at the sub-micron scale. This indicates that selenium passivates defects present in bulk CdTe, and provides an explanation for the remarkable performance of selenium-alloyed CdTe devices. In addition, the results provide crucial insights into the fundamental electronic behaviour of selenium alloyed CdTe, which could unlock further improvements in the photovoltaic performance of

CdSeTe solar cells.

5.8 Methods

Cell Fabrication and Electrical Testing. The two types of cells used in this study were deposited on TEC 10 glass substrates supplied by NSG Pilkington ltd. The substrates consist of 3 mm of soda lime glass coated with a ~400 nm layer of fluorine-doped tin oxide, which acts as the transparent conducting oxide (TCO). The other layers of the cells were then

79 fabricated as follows: 100 nm of MgZnO was deposited on the TCOs by magnetron sputtering, forming the buffer/window layer of the devices. This was followed by 1.5 – 2

µm of CdSeTe, deposited with the substrates held at ~ 420 °C and from a source containing

40% CdSe at 575 °C. A ~ 3.5 µm layer of CdTe was then deposited on top of the CdSeTe with the substrates held at 500 °C and the CdTe source at 555 °C. A cadmium chloride

(CdCl2) activation treatment was then performed on one of the substrates. This involved sublimation of a CdCl2 vapour onto the back surface of the substrate whilst it was maintained at 430 °C for 600 seconds, followed by a 180 °C cooling step with the substrate removed from the vapour. Both device stacks then received a copper doping treatment whereby copper chloride was deposited on the back surface of the CdTe for 110 seconds whilst the substrate was held at

~ 140 °C. This was followed by an anneal in vacuum at 220 °C for 220 seconds to drive copper into the device. Finally, a ~ 30 nm Te film was evaporated on to the back of the

CdTe to improve the back contact. At this stage the two substrates were split in half, with one half of each substrate undergoing contacting and performance testing, and the other half left bare for materials characterisation.

For cell contacting a layer of carbon and nickel paint in a polymer binder was sprayed on the back of the device stack, forming the back electrode of the device. This was then masked and sand- blasted to delineate 10 separate cells of area 0.55 cm2. The devices were then tested for electrical performance using current density vs voltage measurements using an AM1.5 spectrum. An ABET Technologies 10,500 solar simulator with uniform illumination accessory was used to illuminate the devices for measurements. The

80 lamp used for illumination is an ozone free DC xenon arc lamp that produces 1 Sun power output over a 35mm diameter field and meets ASTM, IEC and JIS Class A AM1.5G output requirements. Current density-voltage curves were generated based on electrical measurements performed using a Keithley 2420 SourceMeter controlled by a LabView program. Short-circuit current density was calibrated to CdTe cells measured by NREL.

Device areas were measured using a webcam that took an image of a backlit solar cell and counted the pixels below certain brightness. Both the light intensity and area were calibrated for each set of measurements. The cells were contacted by a fixture of spring- loaded gold pins that provided a 4-point connection and collect current from all around the front contact of the device. The setup accurately measures the J- V parameters and the agreement of these measurements with an externally certified photovoltaic device has been shown in previous work [18].

TEM. The specimen foil for scanning transmission electron microscopy (STEM) was prepared using an FEI focused ion beam (FIB) dual beam system using a standard in-situ lift out method [86]. STEM imaging was performed using a FEI Tecnai F20 S/TEM equipped with Gatan Bright and Dark field STEM detectors, Fischione High Angle Annular Dark Field

(HAADF) STEM detector and an Oxford Instruments X-Max 80mm2 windowless energy- dispersive spectrometer (EDX). STEM imaging was performed at 200 kV with a camera length of 100mm and condenser aperture size of 70µm.

Cathodoluminescence. To present an extended cross-section for cathodoluminescence

(CL) and SIMS characterisation, a bevel was milled through the CdCl2 treated device stack at an angle of 7 degrees to the horizontal using a 30 keV gallium focussed ion beam in an

FEI Nova 600 Nanolab Dual Beam scanning electron microscope(SEM). A platinum over- layer was deposited on the back surface of the CdTe to homogenize the FIB milling of the bevel. Hyperspectral CL characterisation was carried-out at room temperature in a Hitachi

SU-70 scanning electron microscope (SEM). During the measurement a 7 keV beam was rastered across an area on the bevel surface. Luminescence emitted from the sample surface was collected by a parabolic mirror and fed through a diffraction grating to a Gatan

MonoCL system for CL detection. This resulted in a 102 x 282 pixel image, with each pixel containing a full CL spectrum. At 7 keV, CASINO Monte Carlo simulations show that most of the carriers excited by the beam in CdTe (75%) are generated within a tear-drop shaped volume that extends 200 nm below the sample surface and has a diameter of ~ 200 nm

(note that results are for simulations performed at 7.5 keV) [21].

NanoSIMS. Following the CL, high resolution elemental mapping was performed on the same area of the bevel surface using a Cameca secondary ion mass spectrometer

(NanoSIMS 50) with a 16 keV Cs+primary beam. The diameter of the D1 aperture was set to 100 µm (D1-4). Entrance and Aperture slits are 50 x 220 µm (ES-1) and open, respectively. During the measurement, a 0.5 – 1 pA Cs+ primary beam with a nominal diameter of 60 nm was rastered over the measurement area and sputtered secondary ions analysed with a double-focused mass spectrometer. The raster size was 25 x 25 µm (512 x

512 pixels) and the dwell time was 500 µs per pixel. Masses analysed were 35Cl- and 80Se-, giving high resolution images of distributions of chlorine and selenium in the measured area. The scan was repeated 20 times giving 20 stacked images of the distributions of each element and sputtering a total depth of ~ 200 nm below the bevel surface. Images used in the figures are a sum of each stack of 20 images. The ‘auto-track’ feature in ImageJ was

82 used to correct a small amount of image drift before the images were summed. Summing the images ensures that the information depth of the SIMS, at ~ 200 nm, is similar to the CL information depth (100- 200 nm excluding carrier drift/diffusion). EDX measurements were taken on the bevel at to calibrate the selenium counts obtained in the NanoSIMS measurements (beam energy 20 keV). They showed that the average selenium concentrations in the CdSeTe region were 8.9%. This value could then be used to calibrate the average selenium counts over the same region/area given by the NanoSIMS.

Image processing. Some image manipulations were required to allow exact alignment of the pixels of the CL and SIMS images on the same area of the bevel. First, the SIMS maps were rotated to match the orientation of the CL map. The CL map was then scaled by

+13.5% in the Y-direction, i.e. height (this was necessary because of some drift in the raster of the electron beam in the CL measurements, creating a slight distortion that shortened the image). The two sets of images could now be superimposed on top of one another with exact matching of grain boundary features in each (grain boundaries are delineated clearly by the chlorine signal in the SIMS and darker valleys in the CL). At this point the images are matched in everything but the pixel size (the SIMS images are higher resolution). For the pixel-for-pixel comparisons given in the scatter graphs, the pixel size of the SIMS was increased to match the CL, giving images of 102 x 282 pixels that could be directly compared. For the maps used in the figures, the resolution of the SIMS image was maintained (389 x 1152 pixels). Image rotation, scaling, and repixellation were performed using ImageJ.

Contributions:

See ‘publications list’ (page 3) for a full list of authors of the associated manuscript.

TF: conception, direction of experiments, bevel preparation using focused ion beam, data analysis & image processing, and writing of manuscript/chapter.

Cells were fabricated by Amit Munshi at Colorado State University. NanoSIMS was performed by Kexue Li at Oxford University. CL was performed by Budhika Mendis at Durham University with TF present.

5.9 Supplementary Figures

0.0 at% Se

10.1 at% Se

a) b) c)

Supplementary Figure 1. As-deposited CdSeTe/CdTe bilayer. (a) TEM cross section of the untreated CdSeTe/CdTe cell. (b) Corresponding tellurium EDX map. (c) Corresponding selenium EDX map. White boxes show areas over which EDX spectra were acquired to provide the labelled selenium concentrations.

a ) b)

Supplementary Figure 2. Alignment of bevel SIMS and CL maps. (a) SIMS map of selenium and chlorine signals on the bevel area (selenium is blue, chlorine is orange/red, and regions where the signals combine are white). (b) Chlorine signal (green) from the

SIMS map combined with cathodoluminescence signal from the same area, showing good alignment of the grain boundaries. The width of the images is 10.2 µm.

5 µm a) b) c)

Supplementary Figure 3. Bright luminescence in as-deposited CdSeTe layer vs CdTe.

(a) panchromatic CL image on a bevel milled through the untreated CdSeTe/CdTe absorber, measured in a Hitachi SU-70 scanning electron microscope with a Gatan MonoCL system for CL detection. (b) Panchromatic CL image through an untreated CdSeTe/CdTe absorber, measured in a Zeiss SUPRA 55-V (field of view 28 x 30 µm). (c) Panchromatic CL image on a non-alloyed CdTe bevel.

Supplementary Figure 4. Deconvolution of CL spectra from different areas of the bevel. (a) Deconvolution of the CL curve highlighted in blue in the waterfall plot in Fig

5.2(a), corresponding to a region of the bevel containing 2.86 at% selenium. The plot shows a main peak in the spectrum (black) at 1.41 eV, which is significantly red-shifted from the pure CdTe peak emission energy at ~ 1.46 eV (see dashed line and arrow in the figure). The deconvolution shows that the sub-gap peak at ~1.35 eV (green) contributes only a small shoulder to the main peak and does not influence its energy or height. The red-shift is therefore caused by movement of the band-edge emission peak (magenta) with decreasing bandgap of the material. (b) Deconvolution of the CL curve highlighted in red in the waterfall plot in Fig 5.2(b), corresponding to 0.28 at% selenium. Again, it can be seen that the sub-gap peak at ~1.36 eV (green) has not influenced the energy of the band edge peak

(deconvoluted, magenta) at 1.45 eV in the spectrum. The dashed cyan curves show the sum of the deconvoluted spectra in each plot.

Chapter 6 - 3D Imaging of Selenium and Chlorine Distributions in Highly Efficient Selenium-Graded Cadmium Telluride Solar Cells

Modified from a paper published in the IEEE Journal of Photovoltaics.

6.1 Introduction

The addition of selenium to the absorber layer of cadmium telluride (CdTe) solar cells is thought to increase efficiency for two main reasons. Firstly, selenium alloying at the front of the absorber layer decreases the bandgap of the material, increases absorption in the long-wavelength part of the spectrum, and increases device short-circuit current density

[70]. Secondly, the higher levels of selenium at the front of the device creates a grading in electron affinity that helps sweep electrons and holes to the front and back contacts of the device respectively [75]. However, in addition to these two factors, recent work has shown that selenium passivates harmful defects in the bulk of the CdTe, contributing to the measured increases in carrier lifetime and higher than expected voltage and performance

(see chapter 5) [87]–[90]

Despite the importance of selenium in CdTe photovoltaics – and the new evidence of its passivation effects – there has been little research on how it is distributed in CdTe devices, and what has been done has relied on 2D cross-sections through devices [79], [87], [91]. In

88 this work, we map the precise location of selenium in 3-dimensions in a high-efficiency selenium graded cell for the first time, revealing significant inter-diffusion of selenium from

CdSeTe grain boundary regions up into grain boundaries in the CdTe. This raises new questions about the role of selenium-induced band bending and grain boundary passivation in the operation and performance of selenium-alloyed CdTe devices, and provides a starting point for the optimisation of selenium positioning in high efficiency graded devices.

6.2 Experimental

A high efficiency CdSeTe/CdTe bilayer device was fabricated by Close Space Sublimation

(CSS) at Colorado State University. To make the cell, a 100 nm thick Mg0.23Zn0.77O (MZO) buffer layer was first deposited onto a TEC 10 TCO-coated glass substrate. This was followed by ~1 µm CdSeTe (~10 at. % selenium) deposited by CSS, and then ~3 µm of

CdTe, resulting in the device structure shown in Fig 6.1. For the CST deposition, which lasted 110 seconds, the substrate temperature was 420°C and the source temperature was

575°C. For the CdTe deposition, which lasted 160 seconds, the substrate temperature was

360°C and the source temperature was 555°C. The stack was then exposed to cadmium chloride (CdCl2) vapour for 20 minutes with the substrate at 450°C, and for 3m 40s with the substrate at 383°C. Chamber ambient during the CdCl2 treatment was ultra-high purity

N2 at 40 mTorr. There was no intentional oxygen during the process. The cell was measured at 16.8% efficiency under a standard AM 1.5 spectrum. A region of the CdTe back surface was then polished with a gallium Focused Ion Beam (FIB) to remove surface roughness (see schematic in Fig 6.1).

Fig 6.1 Schematic of the device structure and polished back surface on which the SIMS analysis was performed. Layer thicknesses are to scale apart from the glass and back contact (Glass 3 mm, SnO2 400 nm, MgZnO 100 nm, CdSeTe 1 µm, CdTe 3 µm, back contact 25µm). Arrow shows the erosion direction of the sputtered crater during the SIMS measurement.

Following the FIB polish, high-resolution SIMS measurements were obtained from a ~10 x

10 µm area (512 x 512 pixels) on the surface with a NanoSIMS 50 (CAMECA, France). A 16 keV Cs+ beam with a current of 0.6-0.8 pA was scanned over the surface (dwell time is 500

µs/pixel) to generate negative secondary ions, which were analysed with a double-focused mass spectrometer. Masses analysed were 35Cl-, 80Se- 130Te-, 24Mg16O-, and 16O-, and so a high-resolution map of each of these species was formed from each complete scan. On repeating the process, a 3D data cube is built up of elemental distributions in the analysed volume. Data processing was performed with ImageJ with the OpenMIMS plugin (Harvard,

Cambridge, MA, USA), and 3D reconstruction was performed with AVIZO.

6.3 Results and Discussion

Fig 6.2(a) shows a plan-view image of the chlorine signal intensity over an ~8 x 8 µm area at the polished back surface of the absorber layer of the device. It can be seen that the

90 strongest chlorine signal emanates from grain boundary (GB) regions. This is in agreement with previous SIMS (ToF-SIMS) and TEM measurements, which have shown that during the

Fig 6.2 (a) NanoSIMS images of the chlorine (a & b), selenium (c & d), and combined chlorine and selenium (e and f) distributions in a CdSeTe/CdTe bilayer solar cell. The top row shows plan-view images of the distributions on the polished back surface of the cell. The bottom row shows cross-sectional images of the elemental distributions, formed by vertically re-slicing the data cube of the measurement volume. Scale bars are 2 µm and for the vertically resliced images, apply only in the x-direction.

cadmium chloride activation process chlorine segregates to grain boundaries in the absorber layer [28], [31], [60], [62]. Cathodoluminescence measurements have also shown that the presence of chlorine reduces non-radiative recombination at grain boundaries, and it is thought that grain boundary passivation is a primary reason for performance improvement following CdCl2 treatment [35]. Chlorine signal hot spots can also be seen in

91 the grain interior regions of Fig 6.2(a). These are where chlorine has segregated to thin ribbons of incoherent twin boundaries that form at kinks in Ʃ3 (111) twin boundaries [92].

In addition to plan-view images, the 3D data cube can be sliced vertically to present a cross- sectional view of the elemental distributions within the film. This is shown for the chlorine signal in Fig 6.2(b), revealing that as well as segregating at grain boundaries, chlorine is present at the front interface of the absorber with the MZO layer. This is likely to have a passivation effect on the front interface just as it does at grain boundaries, improving the performance of the cell.

Fig 6.2(c) shows a map of the selenium signal intensity over the same area at the back of the CdTe film as the chlorine map in Fig 6.2(a). It can be seen that the selenium signal is concentrated mainly in the grain boundary regions (selenium concentrations here are in the range 0.2 - 0.5 at%). This suggests that during the CdCl2 treatment process, grain boundaries provide channels for fast diffusion of selenium from the as-deposited CdSeTe

(CST) layer into the CdTe above. This behaviour can also be seen in the cross-sectional image in Fig 6.2(d), where the selenium signal rises in ‘plumes’ from the CdSeTe layer into the CdTe.

In Fig 6.2(e) the plan-view chlorine and selenium signals have been superimposed on top of one another. Here it can be seen that, as well as being concentrated at the grain boundaries, some selenium signal is present in the fringes of the grains. This indicates that selenium first diffuses up grain boundaries into the CdTe, and then begins to out-diffuse into the grain interiors. This is a U-shaped diffusion regime in the Harrison classification system [51] and has been observed with sulphur inter-diffusion in traditional CdS/CdTe

92 solar cells [62]. Contrary to selenium, higher chlorine signal is not seen in the grain fringes, indicating that the fringe selenium signal has not come from any smearing effect created by the moving ion beam during the measurement. The selenium signal is also seen to be present in clusters in grain interior regions in Fig 6.2(c), where two example clusters have been highlighted with arrows. Comparison with the chlorine signal shows that these tend to coincide with the chlorine hot spots, suggesting that selenium also preferentially diffuses up incoherent twin boundaries in the grain interiors (as well as along general grain boundaries). 3D renderings of the chlorine and selenium signals in the measurement volume are shown in Fig 6.3. These clearly demonstrate the way in which selenium is associated with grain boundaries in the inter-diffused CdTe. It can be seen that diffusion is not evenly distributed up all grain boundaries, with some showing less selenium present.

In some cases, the amount of selenium diffusion is not even on either side of the same boundary.

Fig 6.3 (a) 3D renderings of the chlorine (green) and selenium (purple) distributions in the absorber layer SIMS measurement volume, showing tilted and plan-view perspectives (a and b respectively). The x- and y- dimensions of the renderings are 8.2 µm and 7.7 µm respectively.

Fig 6.2(d) shows that there are lateral inhomogeneities in the selenium signal in the CST layer. The selenium signal is lowest below the ‘plumes’ of selenium signal that are present in the CdTe layer. This suggests that the excess selenium around grain boundaries in the

CdTe has diffused mainly from the grain boundary regions in the CST, leaving a lower selenium concentration at the grain boundaries relative to the grain interiors. While Fig

6.2(d) and Fig 6.2(f) suggest this, interpretation of reconstructed cross-sectional images, especially at greater sputter depths, is difficult. This is for two main reasons. Firstly, the incident ions can accentuate any small voids in the material because of a preferential sputtering rate at the edge of the voids. As sputtering progresses through the depth of the film, these voids can grow into larger pits which can cause signal to be artificially enhanced at the edges of the pits (for instance, see the line of enhanced signal in Fig 6.2(b), shown by the arrow). In addition, as the sputtered surface progresses through the depth of the device with each raster, the unevenness in the sputtered surface caused by the pits creates some distortion of the reconstructed image. This effect can be seen in Fig 6.2(b), where the front interface chlorine signal appears uneven despite being flat in the measured cell.

For these reasons, we have also directly imaged the CST layer of a graded bilayer device by first using a FIB to remove the CdTe material above it. Selenium and chlorine signal maps from the exposed region are shown in Fig 6.4(a). The selenium map clearly shows regions of higher and lower selenium signal, and comparison with the chlorine map from the same region shows that the lower selenium signal generally coincides with regions of high

94 chlorine signal. This confirms that the source of the excess selenium in the CdTe grain boundaries is diffusion from grain boundary regions in the CdSeTe.

Fig 6.4 a) Plan-view SIMS map of the selenium signal on an ion milled region of CdSeTe, with corresponding chlorine map below (scale bars are 2 µm). b) Line profile of the selenium (blue) and chlorine (red) counts from the region shown by the yellow line in a). c) Schematic of selenium-induced downward band bending across a grain boundary in the CdTe layer. d) Schematic of upward band banding at grain boundaries in the CST layer. e) Surface plot of variations in the estimated CBM energy across the milled CST region (the pure CdTe VBM energy is pinned to 0 eV, as per convention [93]).

Fig 6.4(b) shows line profiles of the chlorine and selenium signals across two grain boundaries in the area. Here, the GB selenium counts drop ~18% from the level in the grain interior, and this is typical of the region. While the high chlorine and low selenium signals generally coincide in the measurement area, they are not perfectly aligned. This is likely to be because of the recrystallisation of the CST layer that occurs during the cadmium

95 chloride heat treatment. Previous work has shown that, before the heat treatment, the

CdSeTe and CdTe are present as two distinct layers with no inter-diffusion of selenium, and that the CdSeTe layer is made up of small columnar grains [79]. Following the heat treatment there are no longer two distinct layers, and grains often run through the thickness of the absorber.

So far, we have discussed the distributions and diffusion of selenium in the absorber. In the following section we will discuss its likely device-level effects. Firstly, there are effects relating to defect passivation. Since it is known that selenium passivates defects in bulk

CdTe, it is likely that selenium also has a passivation effect on grain boundaries. Indeed, recent Density Functional Theory (DFT) calculations have suggested that together chlorine and selenium have a co-passivation effect on grain boundaries [94], [95]. It is therefore possible that the diffusion of selenium into CdTe grain boundaries reduces the active grain boundary defect density and therefore reduces recombination in the CdTe layer. However, the loss of selenium from grain boundaries in the CST may increase GB recombination in this layer versus a case where there is no inter-diffusion [89].

In addition to possible passivation effects, there are also effects related to the distribution of fields within the device. Experiments and modelling have shown that both the conduction band minimum and valence band maximum (CBM and VBM) energies decrease with higher selenium content [93], [96]. This means that electrons will tend to move towards regions of higher selenium content, whereas holes will move away from the selenium. Given the high selenium concentrations observed at the front of the cell, the most obvious effect of this is that electrons will be attracted towards the front contact of the cell,

96 and holes repelled to the back. However, the lateral variations in selenium content due to the grain boundary inter-diffusion means that there will also be lateral fields. For instance, the excess selenium at and around grain boundaries in the CdTe will cause downward GB band bending of both the CBM and VBM in these regions, and therefore tend to attract electrons towards the boundaries and repel holes. This is shown in the schematic diagram in Fig 6.4(d). Conversely, in the CST layer, the highest selenium concentration is in the centre of grains. This will tend to attract electrons towards the grain interiors, and push holes towards grain boundaries (see schematic in Fig 6.4(d)).

Fig 6.4(e) shows a surface plot of the variations in the estimated CBM energy over an area of the CdSeTe region of the bevel. The surface was plotted using the calculated CBM positions for each selenium alloying fraction, as seen in [93]. The plot demonstrates how the lateral selenium concentration variations in the CST will act to funnel electrons into the centre of grains. Although not plotted here, the spatial VBM variations have a similar shape and so will act the opposite way on holes, funneling them towards grain boundaries. It is not yet clear what the overall effects of these lateral fields are on device performance, and even whether they are likely to be beneficial or detrimental. Since the selenium concentrations are low in the CdTe grain boundaries it is hard to assess the significance of the selenium-related band-bending effect in this region. In addition, segregation of other impurities and dopants at the grain boundaries, such as chlorine, may also influence the positions of the CBM and VBM locally. Because of the combination of passivation, field, and possibly doping effects of selenium, these would need to be incorporated into a device-level model to be fully understood (given the lateral variations in selenium concentration shown here, the modelling would need to be 2D). This would enable the optimum selenium

97 distributions throughout a typical grain in the absorber to be predicted. To an extent, it should be possible to independently control the inter-diffusion of selenium using pre-CdCl2, purely thermal anneals.

6.4 Conclusions

High resolution 3D SIMS measurements have been performed on a high efficiency bilayer

CdSeTe/CdTe solar cell. It has been found that during the cadmium chloride heat treatment process, selenium inter-diffuses from the CdSeTe layer into the CdTe, primarily up grain boundaries and then out-diffusing into the fringes of grains. This results in an excess of selenium in and around grain boundaries in the CdTe material, and a deficit of selenium around grain boundaries in the CdSeTe. This has implications in terms of band bending at the grain boundaries and across grains, and likely for grain boundary passivation. These findings are important for the understanding and further improvement of high efficiency

CdSeTe devices.

Contributions:

See ‘publications list’ (page 3) for a full list of authors of the associated manuscript.

TF: conception, direction of measurements, sample preparation using focused ion beam, data analysis & image processing, and writing of manuscript/chapter.

Cells were fabricated by Amit Munshi at Colorado State University. NanoSIMS was performed by Kexue Li at Oxford University.

Chapter 7 - Passivation of Grain Boundaries in Selenium-alloyed CdTe Solar Cells Revealed by TEM-based Cathodoluminescence

7.1 Abstract

In the US, levelized costs of electricity from cadmium telluride (CdTe) photovoltaic modules are lower than for any other solar or fossil fuel technology. This is because the modules combine high power conversion efficiencies with fast, in-line deposition of their semiconductor layers. This combination is particularly difficult to achieve because fast- deposited semiconductor films inevitably have small grains and a high density of grain boundaries, which cause carrier recombination in the absorber and low device efficiencies.

CdTe solar cells have achieved efficiencies over 22%, despite having absorber layer grain sizes less than 10 µm. Recent research has shown that this is possible because of partial grain boundary passivation during the widely used cadmium chloride heat treatment, and passivation of grain interior defects by selenium. Here we use state-of-the-art TEM-based cathodoluminescence measurements to show that selenium also passivates grain boundaries in alloyed CdSeTe solar cells, helping to explain the superior performance of these devices and providing routes for further efficiency improvement and electricity cost reduction.

7.2 Introduction

The record efficiency of cadmium telluride (CdTe) solar cells is now 22.1% [18], [67]. This is higher than for the best multi-crystalline gallium arsenide cells at 18.4%, and near the record for multi-crystalline silicon cells at 22.3%, despite the fact that CdTe grains are more than 1,000 times smaller than silicon grains by diameter [15]–[17]. This apparent paradox can be explained by three recent discoveries in CdTe solar cell research, all of which relate to the introduction of either chlorine or selenium into the CdTe.

Firstly it was found that during the cadmium chloride (CdCl2) heat treatment, which is used universally to produce high efficiency CdTe cells, chlorine segregates to grain boundaries in the CdTe and partially passivates them [19], [29], [35], [60]. Barnard et al then showed that that in addition to grain boundaries, the treatment also increases carrier lifetimes in the interiors of CdTe grains, and at the front interface of the absorber [77], [86], [92]. Finally, in

2019 our group used SEM-based cathodoluminescence (SEM-CL) to show that selenium, which was initially alloyed to the front of the CdTe absorber layer to decrease its bandgap, also has a strong passivation effect on grain interiors in both treated and untreated CdSeTe

(see chapter 5) [87], [88]. This helped to explain the remarkable optoelectronic properties of polycrystalline CdSeTe, which can have higher carrier lifetimes than single crystal CdTe

[97]. However, while the SEM-CL data clearly showed the positive effects of selenium in the interiors of CdSeTe grains, it did not show whether it affects recombination at grain boundaries. This was also an issue in a more recent paper by Zheng et al that uses SEM-CL to analyse a selenium-alloyed device [89]. In both cases, the problem was due to the limited resolution of the SEM-based CL technique, which has an electron beam-sample interaction

100 volume of more than 250 nm at high resolution beam settings (7.5 eV beam energy, with

75% of carriers generated within this volume [39]). With limited spatial resolution low contrast defects, such as passivated grain boundaries, are harder to resolve, making it difficult to make any firm conclusions on the role of selenium on CdSeTe grain boundaries.

Here, we determine the electronic effects of selenium on CdSeTe grain boundaries using high resolution CL imaging in a scanning transmission electron microscope (STEM-CL)

[44]. Whereas high resolution STEM-CL imaging of a solar cell has previously not been achieved because of problems with low signal, we use cryogenic cooling of the TEM foil and xenon ion milling of the sample to boost the CL signal and overcome this issue [98]. Using

STEM-CL allows us to directly correlate the CL maps to TEM micrographs of the absorber layer microstructure and high-resolution STEM-EDX maps of elemental composition, which is a key benefit of the TEM-based technique. The results show that selenium has a strong passivation effect on grain boundaries in alloyed CdSeTe material, on top of what can be achieved with chlorine alone. This further explains how polycrystalline selenium-graded

CdTe devices can compete on efficiency with large-grained, slow-grown, and more expensive competitors like silicon.

To perform the investigation, two bilayer CdSeTe/CdTe solar cells were fabricated at

Colorado State University as described in the methods section. One of the samples was left as-deposited while the other received a cadmium chloride (CdCl2) heat treatment. Cross- sectional TEM foils were then ion milled and ‘lifted out’ from the samples. Crucially, in order to maximise the luminescence signal from the foils, the ion milling was performed using a xenon focused ion beam (FIB), as opposed to the traditional gallium FIB. Because of

101 the higher atomic mass of xenon there is less implantation of ions into the sample during milling. This reduces the number of harmful point defects that are introduced to the sample, reducing the number of non-radiative recombination channels available to carriers and hence increasing luminescence signal. In addition, because xenon is inert, the defects that are formed are less likely to be harmful than those created by gallium implantation. To confirm the superior electronic properties of xenon milled CdTe compared to gallium milled material, we milled two bevelled trenches adjacent to each other in a CdTe film, one with a xenon beam and one with a gallium beam, and compared the SEM-CL signal from the two bevels. The results are shown in supplementary information Fig 1, where it can be seen that the CL signal from the Ga milled bevel is significantly lower than the signal from the Xe milled bevel. Moreover, the level of CL signal from the xenon milled bevel is similar to that from the unprepared CdTe surface either side of the trench, showing that xenon ion milling introduces minimal defects compared to a normal, unprepared CdTe surface.

In addition to using a Xe milled lamella, another way to improve the luminescence signal from a sample is to cryogenically cool it down as this increases the efficiency of radiative recombination. As such, during the STEM-CL measurements, liquid nitrogen was used to cool the lamellae to around -170°C. We found cooling to be particularly important, and without it, the CL signal was very low even with a xenon ion milled sample.

102

7.3 Results

As-deposited device

A cross-sectional TEM micrograph of the as-deposited CST/CdTe bilayer device is shown in

Fig 7.1(a). Generally, small columnar grains are seen in the CST layer, and larger grains in the CdTe. However, there are some instances where the CdTe has grown epitaxially on the

CST material and formed continuous grains that span the two layers (see dashed lines in the figure). The distribution of selenium within the cross-section is shown in the STEM-EDX map in Fig 7.1(b). It shows that selenium (~10 at%) is contained within the CST layer, with no detectable diffusion into the CdTe during deposition [79]. This is mirrored by the tellurium EDX signal distribution in Fig 7.1(c), which shows higher tellurium signal in the

CdTe and lower signal in the CdSeTe as expected.

The cathodoluminescence signal distribution over the bilayer is shown in Fig 7.1(d). In the

CdTe region at the top of the absorber, the CL signal is low. This is expected because it is known that as-deposited CdTe material has a low luminescence efficiency, particularly compared to CdCl2 treated CdTe [77], [99], [100]. However, there is sufficient signal to distinguish dark contrast at the CdTe grain boundaries, which is due to increased non- radiative carrier recombination at grain boundary defects (dangling bonds, wrong bonds, etc) compared to the grain bulk. In the CST layer, the cathodoluminescence signal is significantly brighter than in the CdTe, despite the smaller CST grains (counts reach ~1.2 x106 in the CST, compared to a maximum of ~1.5 x105 in the CdTe). This is consistent with our recent SEM-based CL results which show that selenium alloying significantly increases

103

Fig 7.1 (a) Cross-sectional TEM micrograph of the as-deposited CST/CdTe device, with dashed lines showing where CdTe has grown epitaxially on CST grains. (b) EDX map of the selenium signal distribution over the cross-section, with brighter blue showing higher signal. (c) EDX map of the tellurium signal distribution over the cross section, with brighter orange showing higher signal. (d) STEM-based cathodoluminescence (CL) map of the luminescence signal over the cross-section, with the field of view shifted slightly from that in a (the brackets in (a) and (d) show the same grain). (e) Higher magnification image of the CL intensity variations over a region in the CST layer. The measure in the top right shows the average contrast width of the CdTe grain boundaries at this magnification. All scale bars are 1 µm except in (e).

luminescence efficiency within the grain bulk of both treated and untreated CdSeTe material [87], [88]. Values of grain boundary contrast in the CST and CdTe are similar at

~50%, however the width of the grain boundary contrast in the CST is only ~250-300 nm, compared to an average contrast width of ~500 nm in the CdTe (shown to scale on the

104 image for comparison). This narrower CST grain boundary contrast might suggest that there are lower levels of non-radiative recombination at untreated CST grain boundaries compared to untreated CdTe grain boundaries. However, the small grain sizes in the CST and the low signal-to-noise ratio in the CdTe make it difficult to draw firm conclusions on this.

As well as dark contrast between grains, there are also signal variations within grains in the CST layer. These are shown more clearly in the higher magnification image in Fig 7.1(e), where it can be seen that the signal variations are bands of brighter and darker contrast within the CST grains. Comparison with the TEM micrographs shows that these bands run parallel to the (111) twinning plane of the grains, indicating that the in-grain signal variations are caused by twinning of the CST. This could be due to increased carrier recombination at regions of highly faulted or hexagonal phase material, or to variations in the defect density at different crystal surfaces (111, 100, etc) which are exposed during the milling of the TEM lamella surfaces, and which are affected by twinning [88].

Cadmium chloride treated device

A TEM micrograph of the cadmium chloride treated CST/CdTe device is shown in Fig

7.2(a). In contrast to the untreated device, which has a bilayer structure, there are no distinct CST and CdTe layers in the treated absorber and grain sizes are generally larger.

This shows that there has been recrystallisation of the absorber layer during the cadmium chloride heat treatment. An EDX map of the selenium signal distribution in the cross- section is shown in Fig 7.2b, and in Fig 7.2c this map has been superimposed on top of the

105

Fig 7.2 (a) Cross-sectional bright field TEM micrograph of the cadmium chloride treated CST/CdTe device, with a field of view of 6.8 µm x 6.1 µm (same for all the panels in the figure). (b) EDX map of the selenium distribution in the cross section (brighter blue shows higher selenium signal intensity). (c) Map of the selenium signal distribution in (b), superimposed on top of the micrograph in (a). (d) STEM-based cathodoluminescence (CL) map of the panchromatic CL intensity over the cross-section, with arrows highlighting the thinner grain boundary contrast in the interdiffused region (dashed arrow) and CST (solid arrow) versus the CdTe. (e) EDX map of the chlorine signal intensity over the cross-section. The arrow shows an ‘edge on’ grain boundary in the CdTe.

TEM micrograph. Compared to the untreated device, the maps show a more gradual decrease in selenium signal from the front to the back of the film. This indicates that during the CdCl2 treatment, selenium diffuses from the CST layer into the CdTe [101]. In the interdiffused region between the top and bottom of the film, higher selenium signal is seen at grain boundaries compared to the adjacent bulk, indicating that grain boundaries

106 provide a pathway for preferential diffusion of selenium into the CdTe layer [79]. The EDX map for chlorine is shown in Fig 7.2e and indicates segregation along the grain boundaries, consistent with previous reports [29], [60], [79]. In Fig 7.2e there appears to be less chlorine segregation at CdTe grain boundaries. This is an experimental artefact caused by the large inclination of these grain boundaries within the TEM foil (see Fig 7.2a), resulting in less sensitivity of the measured EDX signal to chlorine segregation. CdTe grain boundaries that are close to being 'end-on' (e.g. the arrowed grain boundary in Fig 7.2e) show the expected chlorine signal enhancement. Furthermore, EDX measurements on different regions of the sample show that chlorine is present along grain boundaries throughout the film thickness (see Supplementary Figure 5).

A STEM-CL map acquired on the device cross-section is shown in Fig 7.2d. Grain interior CL signal in the CdTe is higher than in the untreated CdTe sample as expected, giving clearer grain boundary contrast. However, there is significant variation in the amount of contrast at grain boundaries depending on their depth through the absorber. In the CdTe region towards the top of the film there are thick, dark bands of grain boundary contrast. At the bottom of the film, in the CdSeTe material, there are only thin, faint lines of contrast (see the solid arrowed grain boundary in the figure for example). For instance, the width of GB contrast in three of the CdTe boundaries in the image are all between 500-600 nm, with a boundary contrast of between 63-78%. This means that the signal at the trough of the V-shaped CL profile is 63-78% lower than the averaged signal on both sides of the boundary (see the CL profile in Supplementary Information Fig 2 for example). This compares to the arrowed boundary in the CST, where the GB contrast is only 29%, with a width of 100 nm (see Fig 2 in the SI). This reduction in grain boundary contrast and width

107 shows that there is significantly lower carrier recombination in the CST grain boundaries compared to CdTe. Since the bilayer has been CdCl2 treated, this is ‘extra’ GB passivation – on top of what can be achieved purely with chlorine at the grain boundaries (see the chlorine EDX map in Fig 7.2e). In addition, the data shows that the level of grain boundary passivation is dependent on the amount of selenium at each specific boundary. For instance, a boundary with an intermediate concentration of selenium around it has been circled in the selenium map in Fig 7.2b. It can be seen in the CL map (see dashed arrow) that the contrast at this boundary is also intermediate, i.e lower than that at pure CdTe boundaries, but higher than at boundaries in the CST. The CL map shows that generally for the selenium concentration ranges present in this cell (i.e. 0 – 10 at%), the more selenium present at and around a boundary, the greater the passivation of the boundary. We have quantified the recombination velocity of the grain boundaries using the method described in [81], [102]. The recombination velocity is a measure of the carrier 'lifetime' at a grain boundary; the larger its value the stronger the recombination and therefore more harmful to device performance. Many of the grain boundaries deep within the CST layer show too little contrast to carry out a meaningful quantitative analysis. We have nevertheless been able to analyse grain boundaries in the regions with intermediate selenium concentration, where the contrast is slightly higher. The results indicate that grain boundary recombination in the intermediate CST layer is only 20-40% of that at CdTe grain boundaries (see Supplementary Information). The true value would be even smaller for

CST grain boundaries with high selenium concentration. It should be noted that grain boundary projected width does not have a large effect on grain boundary contrast in the

108

CdTe since the CL resolution is governed by the carrier diffusion length, which is ~100 nm even for TEM-CL (see tables 1 and 2 in the Supplementary Information).

One feature of the STEM-CL map in Fig 7.2d is that the bulk CST material does not show brighter CL signal than the CdTe bulk, as would be expected from our previous SEM-CL measurements. We believe that this is because of the proximity of the free surfaces of the

TEM lamella, which are separated by less than 150 nm. In this situation, any increase in the carrier diffusion length caused by selenium alloying only makes it more likely that the generated carriers will diffuse to the lamella surfaces and be quenched. This highlights one disadvantage of STEM-CL, which is the proximity of the ion milled lamella surfaces, and suggests that lamella surface passivation, perhaps with alumina, could be a good way to improve STEM-CL imaging further [103]. Despite this, our other STEM-CL measurements of bilayer films have shown the expected brighter luminescence in the CST. One of these is shown in Fig 3 in the Supplementary information. It can be seen that as well as having brighter CST compared to CdTe, there is the thin grain boundary contrast in the CST layer that we have seen in Fig 7.2. In addition, another example of a measurement showing brighter CST is shown in Fig 7.3a in the next section. It could be argued that the thin grain boundary contrast observed for CST is an artefact of electron beam injection, since the doping concentration is likely to be different between the CdTe and CST layers. The fact that these thicker specimens with brighter CL signal for the CST layer also show thin grain boundary contrast effectively rules out electron beam injection artefacts. The thicker specimens have higher injection levels due to the incident electron beam losing more of its energy and due to the diminished role of free surface recombination. Despite this there is

109 still a clear difference between CdTe and CST grain boundary contrast, indicating that it is a real effect.

Hyperspectral STEM-CL

A panchromatic map of the CL signal intensity over a treated bilayer cross-section, from the same device as that in Fig 7.2 but taken from a different area of the film, is shown in Fig

7.3a. Darker CL signal is seen at the top of the film in the CdTe, and brighter signal in the

CST, and there is a thick region of grain boundary contrast in the CdTe. A low-temperature hyperspectral CL map, where a full luminescence spectrum is collected in each step of the electron beam raster, was performed on this sample. Fig 7.3b shows a comparison of the average CL spectrum in the CdTe part of the sample (black curve) with the average spectrum in the CST part of the sample (blue curve). The CdTe spectrum shows a sharp excitonic peak at 780 nm and a broader peak at 870 nm, which we attribute to donor- acceptor-pair (DAP) emission. The CST spectrum has similar excitonic and DAP peaks, but their peak maxima are red-shifted to lower energies (830 nm and 965 nm respectively).

This is due to the band gap narrowing that occurs when CdTe is alloyed with selenium [71],

[87], [104]. The total CL signal in the CdTe region is small at 1.1 x106 counts, compared to

110

Fig 7.3 (a) STEM-based CL map of the panchromatic luminescence intensity over a cross-section of the treated CST/CdTe device, showing a brighter CST layer. (b) Comparison of the average CL spectrum from the CdTe layer of the cross section (black curve, with the region shown by the top brackets in (a)) versus the spectrum from the CST layer (blue curve, with the region shown by the bottom brackets in (a)). (c), normalised plot of the CdTe and CST spectra, with the DAP peaks centred on a common wavelength.

9.5 x106 counts over the same area in the CST. In order to more directly compare the shapes of the excitonic and DAP peaks for both materials, in Fig 7.3c we have superimposed the spectra such that both the DAP peaks are normalised and centred at a common wavelength. It can be seen from the figure that the wavelength difference between the DAP and excitonic peaks is larger in the CST than in the CdTe (135 nm vs 90 nm, or 0.204 eV vs

0.163 eV). Since the excitonic binding energy will be similar in materials with similar relative permittivity, this indicates that DAP emission in CST material is from deeper donor and acceptor states than in CdTe. This could either be because the deeper defects are not present in CdTe, or because they are present but not undergoing radiative recombination like they are in CST. In addition, the normalised DAP peak is broader in the CST spectrum

111 than in the CdTe, with a FWHM 75% larger (103 nm vs 59 nm). This again suggests that the addition of selenium to CdTe increases the density of donor and acceptor defects. Finally the CL spectra in Fig 7.3 show only a weak transition radiation signal [98] compared to luminescence generated by electron-hole pair recombination. In reference [98] transition radiation was found to dominate the TEM-CL signal in CdTe. The fact that this is not the case for our samples is due to the improved specimen preparation (i.e. less ion beam damage) from xenon FIB-milling (reference [98] on the other hand used conventional gallium ion beam milling). The suppression of transition radiation artefacts by xenon FIB is crucial for the correct interpretation of grain boundary contrast in this work.

7.4 Discussion

The results presented here show that in CdCl2 treated, selenium-graded CdTe cells there are significantly lower levels of non-radiative recombination at CdSeTe grain boundaries compared to CdTe boundaries. This suggests that selenium has a passivation effect on grain boundaries in CdSeTe material, in addition to what can already be achieved with chlorine passivation [35]. Given the high density and well-known detrimental effects of grain boundaries in thin-film polycrystalline solar cells this is a particularly important finding.

Alongside the grain interior passivation effect that has recently been discovered, the result provides an explanation for the superior carrier lifetimes and performance of selenium- alloyed CdTe. In addition, the results show that the more selenium that is present around the boundaries, the stronger the passivation effect at the boundary. This suggests that a selenium concentration above 10 at% at grain boundaries could have a stronger

112 passivation effect than is already achieved and could lead to higher efficiency devices. It also suggests that if more selenium can be incorporated at grain boundaries at the back of the device, in the nominally CdTe region, then the amount of non-radiative carrier recombination in the absorber can be reduced and efficiencies increased. This could be achieved by performing selenization treatments on the absorber to diffuse selenium into the grain boundaries.

In terms of the potential passivation mechanisms, it is not clear to what extent the reduced recombination is due to either: 1) the presence of selenium in the host material immediately either side of the grain boundaries, changing the electronic band structure that the boundary defects exist within; or 2) whether selenium interacts with the boundary defects themselves (i.e. selenium interacting directly with the wrong/dangling bonds); or a combination of the two. If there is no segregation of selenium at the CdSeTe grain boundaries (we could not detect any with TEM-EDX line scans) then it is worth noting that only ~1 in 10 of the atoms at the boundaries in our CdSeTe layer will be selenium. In this situation, selenium could only directly interact with and passivate ~1 in 10 wrong/dangling bonds at the grain boundary. Hence it seems unlikely that the mechanism is direct passivation of wrong/dangling bonds (option 2), unless there is some selenium segregation that is below the detection limits of TEM-EDX. This is in contrast to the situation with chlorine, where there is strong, clear segregation of chlorine to grain boundaries and therefore ample opportunity for segregated chlorine atoms to interact with wrong/dangling bonds at the boundary. Atomic resolution measurements using HRTEM or atom probe tomography could shed light on this, in addition to DFT modelling and further

STEM-CL [95], [105].

113

In summary, in this work we have successfully performed TEM-based cathodoluminescence imaging on a selenium-graded CdTe solar cell by using xenon ion milling and sample cooling to significantly increase the luminescence signal from the TEM foil. The results show that selenium reduces harmful non-radiative recombination at grain boundaries in alloyed CdSeTe material, which helps to explain the superior carrier lifetimes and performance of selenium graded CdTe solar cells. This could lead to further efficiency improvement of selenium-graded CdTe solar cells if concentrations at boundaries in the

CdTe part of the absorber can be increased. In addition, the results demonstrate that TEM-

CL has the potential to become a more standard technique for characterising solar cells, enabling a full package of microstructural, chemical and electronic characterisation at high resolution.

7.5 Methods

Solar cell fabrication

The two cells used in this study were deposited on TEC10 TCO-coated glass substrates supplied by NSG Pilkington. The substrates comprise 3mm soda lime glass with a 400 nm fluorine doped SnO2 TCO. Initially, a 100 nm MgZnO buffer layer was deposited on the TCO by magnetron sputtering. This was followed by ~1.5 microns of CdSeTe deposited using

Colorado State University’s ARDS close space sublimation system [49]. During CdSeTe deposition a graphite source containing 40% CdSe was held at 575 °C, while the substrate was held at 420 °C. A ~3 micron layer of CdTe was then deposited with the CdTe source material held at 555 °C and the substrate at 500 °C (the CdTe and CdSeTe source material

114 was supplied by 5N Plus). One of the cells then underwent a cadmium chloride activation process. During the process a CdCl2 vapour was sublimated on to the back surface of the

CdTe while the substrate was maintained at 430 °C for 600 seconds. It then went through a

110 s cooling step whilst held at 180 °C, removed from the vapour. Both devices then underwent a 110s copper doping treatment where copper chloride was deposited onto the back surface of the CdTe whilst held at 140 °C. The copper was then diffused into the device by a 220 °C, 220s anneal in vacuum. 30 nm of tellurium was then deposited onto the CdTe to form the back contact. The efficiency of the CdCl2 treated device was measured at 16.8%.

TEM

The TEM lamellae for each sample were prepared by xenon ion milling in a FEI Helios

Plasma-FIB using a standard in-situ lift out technique [86]. During final thinning the beam energy was 5 kV. STEM-CL was carried out in a JEOL 2100F FEG TEM at Brunel University.

For the CL measurements the lamellae were cryogenically cooled to minus ~170 °C using liquid nitrogen. The microscope is fitted with a Gatan Vulcan CL system that has two parabolic mirrors, one either side of the TEM foil. A photomultiplier tube was used for acquiring the panchromatic CL images, and a CCD camera for the spectrum images. The electron beam energy during the CL measurements was 80kV. Due to the positioning of the parabolic mirrors of the CL holder, combined CL-EDX measurements were not possible, hence STEM-EDX imaging was performed separately. TEM and STEM-EDX measurements were carried out in a JOEL 2000FX TEM fitted with an Oxford Instruments EDX detector.

115

Contributions:

See ‘publications list’ (page 3) for a full list of authors of the associated manuscript.

TF: conception, direction of experiment, data analysis, and writing of manuscript/chapter.

Cells were fabricated by Amit Munshi at Colorado State University. TEM EDX was performed By Ali Abbas at Loughborough University. STEM-CL was performed by Ashley Howkins at Brunel University. Recombination calculations were carried out by Budhika Mendis at Durham University.

116

7.6 Supplementary information

Supplementary Figure 1 SEM-based CL map of the luminescence intensity from Ga-milled and Xe-milled bevels in the treated CST/CdTe bilayer device (ion beam milling was performed at 30 kV for both trenches).

Supplementary Figure 2 (a), STEM-based cathodoluminescence (CL) map of the panchromatic CL intensity over the cadmium chloride treated CdSeTe/CdTe bilayer cross-section, as seen in Fig 7.2(d) in the main text. Boxes show the positions where the line profiles in (b) are taken. (b), Line profiles of the CL signal across grain boundaries CdTe (black) and CST (blue) layers. In each case the grain boundary position is arbitrarily set to zero microns.

Supplementary Figure 3 STEM-based CL map of the panchromatic luminescence intensity over a cross- section of the treated CST/CdTe device, showing brighter Cl intensity and thin grain boundary contrast in the CST layer.

118

It is possible to extract the recombination velocity of individual grain boundaries from CL images as has previously been demonstrated for SEM-CL [1]. The CL contrast I(x) at a distance x from the grain boundary is given by [81]:

(1)

where L is the minority carrier diffusion length and Sred is the reduced recombination velocity. Sred is related to the grain boundary recombination velocity (S) by Sred = S/L, where  is the carrier lifetime. By plotting log[I(x)] as a function of x the diffusion length L and reduced recombination velocity Sred can be extracted.

In this model it is assumed that free surface recombination can be ignored. While this can be approximately satisfied in SEM by increasing the energy of the incident beam, it is not possible to ignore surface recombination in TEM-CL. Despite this a recent TEM-CL study by

Yoon et al [102] has shown that Equation (1) can still be applied, provided the lifetime  is replaced by an effective lifetime eff that is lower than the bulk value. eff is determined by surface recombination and is a function of the TEM specimen thickness. The effective diffusion length is then Leff = (Deff), where D is the carrier diffusion coefficient.

Applying Equation (1) to our TEM-CL data would yield effective values for the diffusion length and reduced recombination velocity. Since these values depend on the specimen thickness they are not very useful on their own. However, it does enable us to compare different grain boundaries provided the data are all extracted from the same specimen.

119

Figure 4a shows the TEM-CL map (Fig 7.2d in the manuscript) with the grain boundaries analysed indicated in numerical order. Four CdTe and three CST grain boundaries were found to be suitable for quantitative analysis. Figure 4b shows an example CL intensity profile across a grain boundary and Figure 4b its linearisation according to Equation (1).

The linearisation plots for the grain boundaries all had a regression coefficient larger than

0.91. When selecting CST grain boundary profiles care was taken to ensure the selenium concentration was uniform over the region of interest; this was done by comparing with the EDX map for selenium (Fig 7.2, main text).

Supplementary Figure 4 (a) TEM-CL image of the treated bi-layer device with the grain boundaries analysed indicated in numerical order. (b) shows the CL intensity profile across one of the grain boundaries (grain boundary 1) and (c) is its linearisation according to Equation (1).

120

The extracted values for CdTe and CST grain boundaries are listed in Tables 1 and 2 respectively. The diffusion length for CST is smaller than CdTe. Since it is assumed that eff is the same throughout the specimen this variation can be attributed to different diffusion coefficients D between CdTe and CST regions. To account for this variation we calculate the product (SredL), which is equal to Seff and therefore proportional to the grain boundary recombination velocity. This enables us to compare CdTe and CST grain boundaries with one another. As can be seen from Table 2 recombination at CST boundaries are on average

~20-40% smaller than CdTe.

Supplementary Table 1 values for CdTe grain boundaries (numbers 1 to 4 in Supplementary Figure 4a).

Supplementary Table 2 values for CST grain boundaries (numbers 5 to 7 in Supplementary Figure 4a). The column SLCST/SLavg expresses the product (Sred x L) for the CST grain boundary as a percentage fraction of the average value for CdTe grain boundaries (from Table 1 SLavg = 0.04 µm).

121

Supplementary Figure 5 TEM-EDX map of the chlorine distribution in the CdCl2 treated cell, from a different region as that shown in Fig 7.2 in the main text.

122

Chapter 8 - Conclusions

As set out in the introductory chapter, the purpose of the thesis is to contribute to the spread and decrease in cost of solar power by improving the characterisation, understanding, and eventually the efficiency of thin-film cadmium telluride (CdTe) solar cells. Specifically, the aims were set to improve understanding of the effects of chlorine and selenium in CdTe devices. This is because the addition of the two elements has helped CdTe device efficiencies to surpass 22%, despite rapid, low-cost semiconductor deposition and small grain sizes. Because of the small-grained, complex nature of thin-film PV devices, untangling the myriad of effects of processing changes such as chlorine and selenium addition is nearly impossible using only device-level characterisation. As such, we used high-resolution and correlative characterisation of material microstructure, composition, and electronics where possible to help determine the effects of the two elements. Our findings for chlorine and selenium are summarised below. This is followed by a ‘future work’ section. Finally, there is a section on our assessment of the future of solar photovoltaics.

8.1 Chlorine and the cadmium chloride treatment

Chlorine is introduced into the absorber layer of CdTe devices during the cadmium chloride

(CdCl2) heat treatment, which is essential for the production of high efficiency CdTe devices. Recently, Abbas et al and Mao et al showed that during the treatment chlorine

123 accumulates at grain boundaries [28], [60]. In 2015, Moseley et al used SEM-CL to show that recombination at grain boundaries is reduced following the CdCl2 treatment, indicating that grain boundary passivation by chlorine is an important mechanism for efficiency improvement in CdTe solar cells [35]. In particular, this gave the first suggestion as to how polycrystalline CdTe could achieve high efficiencies and carrier lifetimes despite having small grains and therefore a high grain boundary density.

In this work we used high resolution and high sensitivity dynamic SIMS mapping

(NanoSIMS) to show that during the CdCl2 treatment chlorine permeates every part of the

CdTe absorber layer, not just grain boundaries. Chlorine is detected in the grain interiors at dopant concentrations, at incoherent twin boundaries that span grain interiors, and at the front interface of the device, i.e. between the CdTe and the CdS or MgZnO buffer layer. In each of these areas, chlorine is likely to have significant effects on the local electronic properties of the material and therefore, to varying extents, the electronic performance of the device as a whole.

Our measurements showed strong segregation of chlorine to grain boundaries as expected.

There were significant variations in chlorine concentration between different boundaries, sometimes varying by a factor of two. We also showed that the tellurium signal drops at grain boundaries in treated CdTe, suggesting that chlorine at the boundaries is somewhat substitutional in place of tellurium.

NanoSIMS measurements showed that there are hotspots of chlorine signal in the interior of CdTe grains. In three dimensions, these form rod-shaped features that span grain interiors. Comparison of the SIMS images with correlative EBSD maps acquired on the

124 same region of the films showed that the chlorine hotspots often lay on mid-grain kinks and terminations of (111) lamellar twin boundaries. This indicated that the hotspots were regions of chlorine segregation at incoherent twin boundaries. Unlike coherent twin boundaries, these have non-tetrahedral bonding and so, without chlorine present, are expected to act as non-radiative recombination centres. Given the positioning of the incoherent twin boundaries in the centres of grains, and the generally high twin density in untreated CdTe, these will act as a source of efficiency loss in as-deposited devices.

Segregation of chlorine is expected to partially passivate incoherent twins in the same way as it does at general grain boundaries that surround grains (indeed, DFT modelling on

(211) double positioning twins suggests that it will [65]). Our results therefore indicate that the passivation of incoherent twins in CdTe grain interiors is one reason for improved device performance following CdCl2 treatment, and that further reduction in the density of incoherent GI twins is a worthwhile goal for further CdTe device development.

Our SIMS measurements show that following the CdCl2 treatment there is segregation of chlorine at the front interface of the absorber with the MZO buffer layer. Because buffer layer grain sizes are much smaller than those for the CdTe absorber, this interface is non- epitaxial and will contain a high density of defects, in the same way as a grain boundary does. The detection of chlorine at this interface suggests that it plays an important role in passivating the front interface and improving device performance, however the electronic effects of chlorine at the front interface in CdTe solar cells has not yet been studied.

In addition to observing chlorine segregation at extended defects, chlorine signal was detected in grain interiors at levels twice as high as the detection limit of the instrument.

125

Our results suggest that the cadmium chloride anneal is so effective because during the treatment it is more likely to end up at areas of disrupted atomic bonding – including grain boundaries, incoherent twins, heterointerfaces, dislocations – which are also the regions that typically contain wrong or dangling bonds and therefore require passivation. In other words the regions where chlorine is most likely to segregate are also the regions where it is most ‘needed’. Details on suggested further work related to chlorine in CdTe are given in the ‘Future work’ section in 8.3.1.

8.2 The effects of selenium alloying in CdTe

Selenium was initially added to the front of the absorber layer of CdTe solar cells to lower the material band gap and increase device current densities. However, it was also found that cell voltages were maintained or improved, despite the lower band gap. For this work we investigated how the addition of selenium affects alloyed CdTe material properties.

We performed correlative SEM-based CL and NanoSIMS on the same bevelled area of a

CdCl2 treated CST/CdTe cell (with 10 at% selenium in CST). Cathodoluminescence signal was found to be factor of 10-20x higher in the CST than the CdTe layer, and brighter CL tracked the positioning of selenium within grains. Since brighter CL signifies lower levels of non-radiative recombination, this indicated that selenium has a passivation effect on defects in CdSeTe grain interiors and helped to explain the higher carrier lifetimes and efficiencies of selenium-graded devices. CL signal increased for concentrations up to ~10 at% selenium, and data on the effect of selenium on grain boundaries was inconclusive.

126

One factor that was not considered in the work was the extent to which material band gap variations have a natural effect on luminescence efficiency between CdTe and CdSeTe. The lower band gap in the CdSeTe means that electrons require a smaller amount of energy to be excited to the conduction band, increasing the number of carriers generated per incident beam electron (called the generation factor), and therefore also increasing the recombination rate and luminescence. The generation factor is inversely proportional to material band gap [32]. This means that for a given electron beam energy a decrease in band gap from 1.46 eV to 1.36 eV, as is the case between CdTe and CdSeTe, increases the generation factor by only ~7% and would therefore increase the maximum possible luminescence by a similar amount. This is therefore not a significant factor in the 10-20x increase in CL counts between CdTe and CST grain interiors.

There are a number of additional reasons why it is likely that the higher CL signal is mainly due to defect passivation rather than decreased band gap in the CdSeTe. The first is that longer effective diffusion lengths were measured in CdSeTe material than in CdTe (chapter

5). The second is that TRPL measurements have shown higher carrier lifetimes in CdSeTe compared to CdTe [56]. This is not only single-photon TRPL measurements, which are affected by front or back interface of the absorber and by grain boundaries, but 2-photon

TRPL measurements have also since shown higher carrier lifetimes within CST grain interiors, separating the higher lifetimes from any grain boundary or front interface effects

[89], [106]. This is a direct measurement of reduced defect-mediated recombination in

CdSeTe grain interiors. Finally, CL experiments on sulphur-alloyed CdTe, where sulphur is shown to reduce the material band gap of CdTe in a similar way to selenium alloying, do

127 not show brighter CL emission despite the lower the band gap (in fact, emission decreases slightly with the lower bandgap) [34].

Analysis of per-pixel CL emission spectra showed that, in regions with a low selenium concentration (< ~1%), a sub-bandgap emission peak at 1.36 eV was present in addition to the 1.46 eV band-to-band emission. With increased selenium content, the peak emission energy decreased as expected due to bandgap bowing.

Our NanoSIMS measurements showed that there is interdiffusion of selenium between the

CST and CdTe layers during the CdCl2 treatment. This results in an excess of selenium at

CdTe grain boundaries and a deficit at CST grain boundaries compared to the grain interiors. This alters the position of the conduction band minimum (CBM) and valence band maximum (VBM) at and around grain boundaries, creating lateral fields in the absorber layer. In the CdTe, selenium at and around the grain boundaries will cause downward band bending of both the CBM and VBM locally. This will tend to cause electrons to be attracted towards the boundaries and holes repelled. However, because of the low concentration of segregated selenium in these regions, and the potential impact of other segregated atoms, this effect is likely to be relatively minor. In the CST layer more selenium is present in the grain interiors than in the grain boundary regions, so there is upward band bending of the

CBM and VBM at the grain boundaries. This will have the opposite effect as above, causing electrons to be funnelled towards the grain interiors, and holes towards the grain boundaries. By our estimates, the lateral selenium concentration variations in CST are enough to cause approximately -0.02 eV of band banding from GBs to the centre of grains.

128

In addition to SEM-based CL measurements, we performed CL measurements in the TEM.

This has a number of advantages compared to SEM-CL. It enables CL imaging to be compared directly to high-resolution TEM micrographs of cell microstructure and elemental distributions. It also enables higher resolution CL imaging than SEM-CL, because of the smaller electron beam-sample interaction volume. However, there are also some disadvantages to using TEM-CL. These include that there is a smaller field of view, and lower signal compared to SEM-CL – a result of the smaller interaction volume and extra free surface of the TEM foil. In order to increase the TEM-CL signal, we prepared the TEM- foil using a Xe FIB and cryogenically cooled the sample using liquid nitrogen. This gave much higher signal than had been achieved previously and enabled good TEM-CL imaging on a solar cell for the first time. TEM-CL imaging was successfully performed on an as- deposited and a treated CST/CdTe bilayer cell. This was complemented with high resolution bright-field imaging of the microstructure of the cross-section, and TEM-EDX imaging of the elemental distributions, including chlorine and selenium. In the treated device, there were clear differences in grain boundary CL contrast between the CdTe and

CST layers. Grain boundaries in the CdTe layer at the top of the absorber had think, dark regions of grain boundary contrast, indicating high levels of non-radiative recombination.

Grain boundaries in the CST layer however showed significantly thinner and fainter grain boundary contrast, and there was a grading in grain boundary contrast depending on the height through the film and hence local selenium content. Our calculations of grain boundary recombination velocity showed that recombination at CST grain boundaries is at least 60-80% less than at the CdTe boundaries. Alongside the results for grain interior passivation, this explains the superior carrier lifetimes and performance of high efficiency

129 selenium-alloyed CdTe devices, and also helps explain how CdSeTe devices have achieved efficiencies almost as high as large-grained multi-crystalline silicon.

8.3 Future work

8.3.1 Chlorine in CdTe

In terms of determining the effects of chlorine in CdTe there is a lot more that can still be done. Our results showed that in addition to grain boundaries, chlorine permeates every region of the CdTe absorber layer, and is more highly concentrated in defective regions. In future, an ion-implanted chlorine sample could be used as a calibration standard so that absolute chlorine concentrations can be measured. Because of the importance of the front interface to device performance, it would be particularly valuable to determine chlorine concentrations here. Further atomic resolution studies of the interface could also be performed to ascertain to what extent the chlorine is present substitutionally, interstitially, or perhaps as a different phase of material [107]. In our work, we found that chlorine concentrations vary significantly from grain boundary to grain boundary, sometimes as much as doubling from one boundary to another. Correlative CL measurements could determine if this variation in concentration affects non-radiative recombination levels at each boundary. If found to be true, efforts would need to be made to determine the grain boundary type as well as its segregated chlorine concentration, since this would likely also affect recombination [35].

130

The main area for extension of the chlorine work presented in this thesis is in quantification of the electronic effects of the element in the different parts of the absorber, and on device efficiency as a whole. Since work on the thesis started, modelling of this sort has been performed by Moseley et al [100]. Estimates of bulk carrier lifetimes, bulk doping densities, and grain boundary recombination were made for both chlorine treated and untreated material. These were then fed into a 2-dimensionsal device model to see if the measured improvements in these material properties could account for the CdCl2-based efficiency improvements seen in experiment. Our results suggest that in future this work could be supplemented by including chlorine-based improvements at the front interface into the model, as well as a reduction in recombination at incoherent twins in the grain interiors.

Recent work has shown that chlorine from the CdCl2 treatment limits the maximum achievable doping density by causing carrier compensation, and so is hampering efforts to move towards 25% CdTe device efficiency [45], [108] (in addition chlorine adds significantly to the already complex point defect dynamics in bulk CdTe and CdSeTe, and can hamper the incorporation of arsenic and phosphorous for new group 5 doping of

CdTe). As such, there has been interest in how high efficiency devices might be made without the CdCl2 heat treatment process

Our results suggest that making high efficiency devices without CdCl2 will be extremely difficult with the current polycrystalline, high defect density CdTe-based absorber layer.

Chlorine not only significantly reduces recombination at grain boundaries, but also likely at the front interface, which is non-epitaxial, and at rod-shaped incoherent twins that span

131 grain interiors. In order to circumvent this problem it is likely that a method would need to be developed to fast-deposit CdTe films with significantly lower defect density, probably single crystal material so that there are no grain boundaries and so that the front interface can be more effectively passivated.

8.3.2 Selenium in CdTe

Work on the thesis has concentrated largely on the ways in which selenium improves

CST/CdTe device efficiency . We have found that selenium significantly reduces non- radiative recombination in both the grain bulk and at the grain boundaries of alloyed

CdSeTe material. These findings open three broad areas of future research:

1) Determining the mechanisms of selenium-based passivation in CdSeTe bulk and

grain boundaries

2) Device-level modelling of the effects of selenium alloying on cell efficiency

3) Experimental work on efficiency improvement through altering selenium

distributions within the absorber.

In terms of bulk selenium passivation mechanisms, it needs to be determined whether the reduced non-radiative recombination in alloyed grain interiors is due to: 1) direct interaction of harmful point defects with selenium (such as Cdv, or Tellurium-on-cadmium antisite defects), or 2) whether the CBM and VBM shifts caused by selenium alloying effectively passivate previously harmful defect levels by making them shallower relative to the band edges. This could be investigated with further DFT studies. Likewise, the same should be determined with grain boundary passivation by selenium. Here we have slightly

132 more information to work from, because our results show that there is not significant segregation of bulk selenium to the grain boundaries. This means that the ~10 at% selenium at the boundaries could only directly passivate roughly 10% of defects at the grain boundary. Passivation of only one in ten of the grain boundary defects seems unlikely to be sufficient to cause the reduction in non-radiative recombination seen in the work, which is at least a 60-80%. This reasoning cannot be applied in bulk material because the harmful point defects in the lattice, or the selenium atoms themselves, could move and then directly interact with one another, unbeknownst to us. A crucial question in defect passivation by selenium is why alloying with sulphur does not have any passivation effect even though it reduces the material band gap in the same way as selenium does, and is from the same group [34]. Future modelling work might therefore consider differences in the behaviour of selenium and sulphur in CdTe. Besides being useful for understanding of high efficiency CdTe photovoltaics, determining the passivation mechanisms of selenium alloying in CdTe might have implications for other CdTe-based technologies and for defect passivation in other II-VI materials.

The second broad area for selenium related future work is in device modelling. In much the same way as described in the chlorine future work section above, this would involve quantification of the impact of selenium on bulk, grain boundary, and front interface recombination, and then inputting these numbers into a device-level model. TRPL measurements would need to be made to quantify bulk carrier lifetimes at a range of different alloying fractions between 0-50 at% selenium. Likewise, our work could be extended by quantifying grain boundary recombination velocities for the full range of selenium concentrations a typical grain boundary. A 2D device model would need to be

133 made that can simulate the effects of a range of different selenium concentrations, and hence different bulk carrier lifetimes and grain boundary recombination velocities, have on device efficiency. In addition to its effects on recombination, the model would ideally incorporate measured changes in carrier concentration caused by selenium alloying, and

CBM and VBM changes both through the depth of the device and laterally across grains, as shown by our NanoSIMS measurements of selenium interdiffusion. The model described above could then be used to inform changes in the absorber selenium distribution that would increase efficiency.

The third main area where there is opportunity for further work is in the experimental improvement of selenium-graded CdTe device efficiency. Our results suggest that slightly increasing the bulk concentration of selenium at the back of the absorber would reduce non-radiative recombination in the absorber layer and increase efficiency. Only a small amount of selenium, less than 1-2 at%, would be needed to improve the optoelectronic quality of the material, and this would enable a steep selenium concentration gradient at the front of the device to be maintained. Different gradients through the depth of the film could be modelled as described above and trialled experimentally using deposition techniques such as co-sublimation. We also know from our results that concentrations of selenium at grain boundaries in the CdTe layer, a result of interdiffusion from the CST, are very low and only fractions of a percent towards the back of the device. Our results clearly show that the more selenium that is present at and around a boundary, up to at least 10 at%, the lower the levels of non-radiative carrier recombination at the boundary. It therefore follows that recombination at grain boundaries in the CdTe part of the absorber could be reduced significantly if the selenium concentration there was higher. This might

134 change the lateral fields and hence lateral carrier transport in the device, so field effects would also need to be considered. Our results suggest it would be worthwhile to explore strategies for increasing grain boundary selenium concentrations in both the CST and CdTe layers of selenium graded CdTe solar cells.

8.4 The future of solar photovoltaics

Two factors have been key to the rise in deployment of solar photovoltaics up to this point.

These are: 1) increased efficiencies of silicon and CdTe modules, and 2) declining production costs of silicon and CdTe modules. Along with system-level improvements, these have driven rapid reductions in the levelised costs of electricity from solar PV, to the point where they are now lower than for most other electricity generation technologies in many regions of the world [3], [4].

At least in the short-term, these factors look set to continue for both silicon and CdTe photovoltaics.

High-end silicon module efficiencies are >20%, whereas the record silicon cell efficiency is

26.7%, and Hanergy have recently produced a one-off, non-production module with an efficiency of 24.9% [109]. This leaves plenty of headroom for further efficiency improvement of production modules. Given that efficiencies of almost 25% have been achieved for specially made modules, it is reasonable to assume that production silicon module efficiencies ~25% can be achieved within the next decade [12]. However, increases beyond 25% will be marginal because the single-junction thermodynamic

135 efficiency limit for 1.1 eV bandgap material is ~30%, and the practical module efficiency limit is 26-27%.

In addition to silicon performance improvements, production capacity looks set to increase further, and module costs are predicted to continue declining [12]. The April 2020

International Technology Roadmap for Photovoltaics (ITRPV) estimates that the total system costs from silicon solar farms in 2030 will be 63% of the 2019 value [12]. Silicon technology therefore looks likely to continue to dominate the market in the short-term, and to achieve significant further reductions in its LCOE.

As discussed in Chapter 1, CdTe has similar module production costs to silicon, despite having only around 1/10th of the production scale. This demonstrates CdTe technology’s intrinsically low production costs. It also suggests that if the annual production capacity of the two technologies were similar, CdTe would have significantly lower $/W module costs than silicon. However, for this to happen in the next 5-10 years, CdTe production capacity needs to grow faster than that of silicon. As things stand, this seems unlikely since there is only one CdTe manufacturer with significant production capacity, which is already a large company, and there are significant barriers to entry into CdTe PV manufacturing (First

Solar Inc holds hundreds of patents related to cadmium telluride PV). In addition, CdTe module efficiencies are lower than silicon at ~18%. Although this means there is more headroom for module efficiency improvement, up to ~25%, the highest cell efficiency that has been demonstrated for CdTe is only 22%. It still remains to be seen whether CdTe will achieve 25% efficiency at the cell level, let alone at the module level. CdTe absorber chemistry is extremely complex, which makes device development more difficult – even

136 with sophisticated correlative characterisation techniques and atomistic modelling. Grain boundaries also remain a significant impediment to achieving >25% efficiency [97].

Another problem for CdTe is that as module prices decrease, the cost of buying the modules takes up a smaller fraction of total solar farm costs. In this situation, area-related balance of systems costs such as land, labour, and racking take on even more importance. The ability to use high efficiency modules to reduce these area-related costs therefore becomes critical, putting lower efficiency technologies like CdTe at a disadvantage moving forward.

Nevertheless, CdTe offers the potential for lower cost PV than wafer-based silicon, if similar module efficiencies and production volumes can eventually be achieved.

Besides the incumbent silicon and CdTe technologies, there are significant barriers to entry for other technologies, perhaps based on other materials, to make an impact on solar module market in the short term. These include barriers related to economies of scale and high capital costs. In addition, it is difficult for new products to prove (in only a few years) that they will be durable under field operation for 25+ years.

Despite these barriers, there are several companies attempting to break into the PV market. In particular, a number of companies are attempting to commercialise perovskite- based photovoltaics for utility-scale power production. Oxford PV is currently pursuing the strategy of stacking a perovskite top-cell on top of a silicon sub-cell, and have demonstrated > 28% efficiency for the tandem stack [7]. Their main challenge will be to maintain the perovskite-based efficiency benefits for as long as the >25 year lifetime of the underlying silicon device, and without the perovskite shading the silicon. Another approach is to make perovskite/perovskite tandem stacks, which is currently being pursued by Swift

137

Solar. This is a promising strategy and can perhaps circumvent the worst of the stability issues by periodically replacing degraded panels at relatively low cost [110].

Researchers are also looking into CdTe-based tandems [52], and into decreasing the production costs of GaAs-based multi-junctions [111]. Whichever way tandems and multi- junctions end up being implemented, they seem the most likely route for improving commercial module efficiency beyond the ~30% Shockley-Quiesser limit (they are so far the only way that the 30% limit has been broken for research cells). Yet again, the issue is how to produce the panels at low enough cost. As the number of layers in the device stack increases, developers will run into the issue of microstructural and compositional complexity – especially for those devices incorporating the fast-deposited polycrystalline layers that are essential for minimising production costs. Correlative high-resolution characterisation will therefore be even more important as the complexity of the devices increases with higher numbers of junctions and layers of different materials. Nevertheless, with continued effort in research and development, it seems likely that solar PV performance and costs will continue to decrease, and the technology will grow to provide a significant portion of our electricity needs and improve living standards as it does so.

8.5 References

[1] S. Oelz, R. Sims, and N. Kirchner, “Contribution of Renewables to Energy Security,” Technology, no. April, p. 74, 2007.

[2] “IEA (2019), SDG7: Data and Projections, IEA, Paris https://www.iea.org/reports/sdg7-data-and-projections.”

138

[3] “Lazard’s levelised costs of energy analysis-version 13.0,” no. November, 2019.

[4] “https://www.bloomberg.com/news/articles/2020-04-28/solar-and-wind- cheapest-sources-of-power-in-most-of-the-world.” . [5] “Best Research-Cell Efficiency Chart.” [Online]. Available: https://www.nrel.gov/pv/cell-efficiency.html. [6] J. F. Geisz et al., “Six-junction III–V solar cells with 47.1% conversion efficiency under 143 Suns concentration,” Nat. Energy, vol. 5, no. 4, pp. 326–335, Apr. 2020.

[7] M. A. Green, E. D. Dunlop, J. Hohl-Ebinger, M. Yoshita, N. Kopidakis, and A. W. Y. Ho- Baillie, “Solar cell efficiency tables (Version 55),” Prog. Photovoltaics Res. Appl., vol. 28, no. 1, pp. 3–15, 2020.

[8] V. Smill, Energy Transitions: Global and National Perspectives. 2017.

[9] M. A. Green, “How Did Solar Cells Get So Cheap?,” Joule, vol. 3, no. 3. pp. 631–633, 2019. [10] “No Title.” [Online]. Available: https://www.energy- storage.news/news/bloombergnef-lcoe-of-battery-storage-has-fallen-faster-than- solar-or-wind-i. [11] “Viridian Solar.” [Online]. Available: https://www.viridiansolar.co.uk/resources-4-2- photovoltaic-solar-cells.html. [12] VDMA, “International Technology Roadmap for Photovoltaic 2020,” Itrpv, no. April, p. 76, 2020.

[13] D. S. Philips and W. Warmuth, “Fraunhofer Photovoltaics Report,” no. March, pp. 1– 43, 2016. [14] B. M. Basol and B. McCandless, “Brief review of cadmium telluride-based photovoltaic technologies,” J. Photonics Energy, vol. 4, no. 1, p. 040996, 2014.

[15] M. A. Green, Y. Hishikawa, E. D. Dunlop, D. H. Levi, J. Hohl-Ebinger, and A. W. Y. Ho- Baillie, “Solar cell efficiency tables (version 51),” Prog. Photovoltaics Res. Appl., vol. 26, no. 1, pp. 3–12, 2018. [16] J. Benick et al., “High-Efficiency n-Type HP mc Silicon Solar Cells,” IEEE J. Photovoltaics, vol. 7, no. 5, pp. 1171–1175, Sep. 2017.

[17] R. Venkatasubramanian et al., “18.2% (AM1.5) efficient GaAs solar cell on optical- grade polycrystalline Ge substrate,” in Conference Record of the Twenty Fifth IEEE Photovoltaic Specialists Conference - 1996, 1996, pp. 31–36.

[18] C. Hagenorf, M. Ebert, M. Raugei, D. Lincot, J. Bengoechea, and M. Rodriguez, “Cener Report: Assessment of performance, environmental, health and safety aspects of First Solar’s CdTe PV technology,” 2016.

139

[19] J. D. Major, “Grain boundaries in CdTe thin film solar cells: a review,” Semicond. Sci. Technol., vol. 31, no. 9, p. 093001, 2016.

[20] “Clean Energy Reviews.” [Online]. Available: https://www.cleanenergyreviews.info/blog/solar-panel-components-construction.

[21] S. Takeuchi and K. Suzuki, “Stacking Fault Energies of Tetrahedrally Coordinated Crystals,” Phys. Status Solidi, vol. 171, no. 1, pp. 99–103, 1999. [22] M. Emziane, K. Durose, N. Romeo, A. Bosio, and D. P. Halliday, “A combined SIMS and ICPMS investigation of the origin and distribution of potentially electrically active impurities in CdTe/CdS solar cell structures,” Semicond. Sci. Technol., vol. 20, no. 5, pp. 434–442, 2005. [23] X. Wu, “High-efficiency polycrystalline CdTe thin-film solar cells,” Sol. Energy, vol. 77, no. 6, pp. 803–814, 2004.

[24] A. H. Munshi et al., “Polycrystalline CdTe photovoltaics with efficiency over 18% through improved absorber passivation and current collection,” Sol. Energy Mater. Sol. Cells, vol. 176, no. November 2017, pp. 9–18, 2018. [25] A. Munshi, “INVESTIGATIONS OF PROCESSING, MICROSTRUCTURES AND EFFICIENCIES OF POLYCRYSTALLINE CDTE PHOTOVOLTAIC FILMS AND DEVICES,” 2016.

[26] H. Hutter, “Dynamic Secondary Ion Mass Spectrometry (SIMS),” Surf. Thin Film Anal. A Compend. Princ. Instrumentation, Appl. Second Ed., pp. 141–159, 2011. [27] P. Hoppe, S. Cohen, and A. Meibom, “NanoSIMS: Technical Aspects and Applications in Cosmochemistry and Biological Geochemistry,” Geostand. Geoanalytical Res., vol. 37, no. 2, pp. 111–154, 2013. [28] D. Mao, G. Blatz, C. E. Wickersham, and M. Gloeckler, “Correlative impurity distribution analysis in cadmium telluride (CdTe) thin-film solar cells by ToF-SIMS 2D imaging,” Sol. Energy Mater. Sol. Cells, vol. 157, pp. 65–73, 2016. [29] D. Mao, C. E. Wickersham, and M. Gloeckler, “Measurement of chlorine concentrations at CdTe grain boundaries,” IEEE J. Photovoltaics, vol. 4, no. 6, pp. 1655–1658, 2014. [30] and A. N. T. Lukas Kranz, Christina Gretener, Julian Perrenoud, Dominik Jaeger, Stephan S. A. Gerstl, Rafael Schmitt, Stephan Buecheler, “Tailoring Impurity Distribution in Polycrystalline CdTe Solar Cells for Enhanced Minority Carrier Lifetime,” Adv. Energy Mater., 2014.

[31] S. P. Harvey, G. Teeter, H. Moutinho, and M. M. Al-Jassim, “Direct evidence of enhanced chlorine segregation at grain boundaries in polycrystalline CdTe thin films via three-dimensional TOF-SIMS imaging,” Prog. Photovoltaics Res. Appl., vol. 23, no. 7, pp. 838–846, Jul. 2015.

140

[32] B. G. Yacobi and D. B. Holt, Cathodoluminescence Microscopy of Inorganic Solids, no. 1. Boston, MA: Springer US, 1990.

[33] S. J. Bauman, Q. Yan, M. Benamara, and J. B. Herzog, “Plasmonic nanogap structures studied via cathodoluminescence imaging,” vol. 1034607, no. August 2017, p. 6, 2017. [34] J. Moseley et al., “Cathodoluminescence spectrum imaging analysis of CdTe thin-film bevels,” J. Appl. Phys., vol. 120, no. 10, 2016.

[35] J. Moseley et al., “Recombination by grain-boundary type in CdTe,” J. Appl. Phys., vol. 118, 2015. [36] J. Moseley et al., “Opto-electronic characterization of CdTe solar cells from TCO to back contact with nano-scale CL probe,” 2015 IEEE 42nd Photovolt. Spec. Conf. PVSC 2015, pp. 2–5, 2015.

[37] H. R. Moutinho et al., “Grain Boundary Character and Recombination Properties in CdTe Thin Films,” pp. 3249–3254, 2013. [38] H. Guthrey et al., “Spatial luminescence imaging of dopant incorporation in CdTe Films,” J. Appl. Phys., vol. 121, no. 4, p. 045304, 2017. [39] J. Moseley et al., “Cathodoluminescence analysis of grain boundaries and grain interiors in thin-film CdTe,” IEEE J. Photovoltaics, vol. 4, no. 6, pp. 1671–1679, 2014. [40] H. R. Moutinho et al., “Grain boundary character and recombination properties in CdTe thin films,” Conf. Rec. IEEE Photovolt. Spec. Conf., pp. 3249–3254, 2013.

[41] G. Stechmann et al., “A correlative investigation of grain boundary crystallography and electronic properties in CdTe thin film solar cells,” Sol. Energy Mater. Sol. Cells, vol. 166, no. January, pp. 108–120, 2017. [42] B. G. Mendis, D. Gachet, J. D. Major, and K. Durose, “Long lifetime hole traps at grain boundaries in CdTe thin-film photovoltaics,” Phys. Rev. Lett., vol. 115, no. 21, pp. 1–5, 2015.

[43] “Cathodoluminescence from insulators, metals, and plasmonics.” [Online]. Available: https://www.gatan.com/cathodoluminescence-insulators-metals-and-plasmonics. [44] M. Kociak and L. F. Zagonel, “Cathodoluminescence in the scanning transmission electron microscope,” Ultramicroscopy, vol. 176, pp. 112–131, 2017. [45] D. Krasikov and I. Sankin, “Defect interactions and the role of complexes in CdTe solar cell absorber,” J. Mater. Chem. A, vol. 5, pp. 3503–3513, 2017. [46] C. Li et al., “Grain-Boundary-Enhanced Carrier Collection in CdTe Solar Cells,” Phys. Rev. Lett., vol. 112, no. 15, p. 156103, Apr. 2014.

[47] A. A. Taylor et al., “A comparative study of microstructural stability and sulphur

141

diffusion in CdS/CdTe photovoltaic devices,” Sol. Energy Mater. Sol. Cells, vol. 141, pp. 341–349, 2015.

[48] A. Abbas et al., “The Effect of a Post-Activation Annealing Treatment on Thin Film CdTe Device Performance,” 2015.

[49] D. E. Swanson et al., “Single vacuum chamber with multiple close space sublimation sources to fabricate CdTe solar cells,” J. Vac. Sci. Technol. A Vacuum, Surfaces, Film., vol. 34, no. 2, p. 021202, 2016.

[50] J. D. Poplawsky, C. Li, N. R. Paudel, W. Guo, Y. Yan, and S. J. Pennycook, “Nanoscale doping profiles within CdTe grain boundaries and at the CdS/CdTe interface revealed by atom probe tomography and STEM EBIC,” Sol. Energy Mater. Sol. Cells, vol. 150, no. March, pp. 95–101, 2016. [51] L. G. Harrison, “Influence of dislocations on diffusion kinetics in solids with particular reference to the alkali halides,” Trans. Faraday Soc., vol. 57, p. 1191, 1961.

[52] D. E. Swanson et al., “CdCl 2 passivation of polycrystalline CdMgTe and CdZnTe absorbers for tandem photovoltaic cells,” J. Appl. Phys., vol. 123, no. 20, p. 203101, May 2018.

[53] P. Capper and J. Garland, Mercury Cadmium Telluride: Growth, Properties and Application, 1. 2010.

[54] K. Durose, P. R. Edwards, and D. P. Halliday, “Materials aspects of CdTe/CdS solar cells,” J. Cryst. Growth, vol. 197, no. 3, pp. 733–742, 1999. [55] Y. Yan, R. G. Dhere, K. M. Jones, and M. M. Al-Jassim, “Influence of substrate structure on the growth of CdTe thin films,” J. Appl. Phys., vol. 89, no. 11 I, pp. 5944–5948, 2001. [56] A. H. Munshi et al., “Polycrystalline CdSeTe/CdTe Absorber Cells With 28 mA/cm 2 Short-Circuit Current,” IEEE J. Photovoltaics, vol. 8, no. 1, pp. 310–314, Jan. 2018. [57] R. M. Geisthardt, M. Topic, and J. R. Sites, “Status and Potential of CdTe Solar-Cell Efficiency,” IEEE J. Photovoltaics, vol. 5, no. 4, pp. 1217–1221, 2015.

[58] W. Shockley and H. J. Queisser, “Detailed Balance Limit of Efficiency of p-n Junction Solar Cells,” J. Appl. Phys., vol. 32, no. 3, p. 510, 1961. [59] J. M. Burst et al., “CdTe solar cells with open-circuit voltage greater than 1 V,” Nat. Energy, vol. 1, no. February, p. in press, 2016. [60] A. Abbas et al., “The effect of cadmium chloride treatment on close-spaced sublimated cadmium telluride thin-film solar cells,” IEEE J. Photovoltaics, vol. 3, no. 4, pp. 1361–1366, 2013. [61] T. Fiducia, A. Abbas, K. Barth, W. Sampath, and M. Walls, “Intragranular Defects in As- Deposited and Cadmium Chloride-Treated Polycrystalline Cadmium Telluride Solar Cells,” in 2016 IEEE 43rd Photovoltaic Specialist Conference (PVSC), 2016.

142

[62] T. A. M. Fiducia et al., “3D Distributions of Chlorine and Sulphur Impurities in a Thin- Film Cadmium Telluride Solar Cell,” MRS Adv., vol. 3, no. 56, pp. 3287–3292, May 2018.

[63] S. Yoo, K. T. Butler, A. Soon, A. Abbas, J. M. Walls, and A. Walsh, “Identification of critical stacking faults in thin-film CdTe solar cells,” Appl. Phys. Lett., vol. 105, no. 6, p. 062104, Aug. 2014. [64] Y. Yan, M. M. Al-Jassim, and K. M. Jones, “Structure and effects of double-positioning twin boundaries in CdTe,” J. Appl. Phys., vol. 94, no. 5, pp. 2976–2979, 2003.

[65] Y. Yan, M. M. Al-Jassim, and K. M. Jones, “Passivation of double-positioning twin boundaries in CdTe,” J. Appl. Phys., vol. 96, no. 1, pp. 320–326, 2004. [66] C. Li, J. Poplawsky, Y. Yan, and S. J. Pennycook, “Understanding individual defects in CdTe thin-film solar cells via STEM: From atomic structure to electrical activity,” Mater. Sci. Semicond. Process., vol. 65, no. June 2017, pp. 64–76, 2016.

[67] M. Gloeckler, I. Sankin, and Z. Zhao, “CdTe Solar Cells at the Threshold to 20% Efficiency,” IEEE J. Photovoltaics, vol. 3, no. 4, pp. 1389–1393, Oct. 2013.

[68] “Lazard’s Levelised Cost of Energy Analysis - Version 11.0,” no. 11, pp. 0–21, 2017.

[69] J. Balderelli, “Public Utilities Comission of Nevada Electronic Filing,” 2015.

[70] N. R. Paudel and Y. Yan, “Enhancing the photo-currents of CdTe thin-film solar cells in both short and long wavelength regions,” Appl. Phys. Lett., vol. 105, no. 18, 2014. [71] D. E. Swanson, J. R. Sites, and W. S. Sampath, “Co-sublimation of CdSe x Te 1 À x layers for CdTe solar cells,” Sol. Energy Mater. Sol. Cells, vol. 159, pp. 389–394, 2017. [72] J. D. Poplawsky et al., “Structural and compositional dependence of the CdTexSe1−x alloy layer photoactivity in CdTe-based solar cells,” Nature Communications, vol. 7. p. 12537, 2016. [73] J. M. Kephart et al., “Sputter-Deposited Oxides for Interface Passivation of CdTe Photovoltaics,” IEEE J. Photovoltaics, pp. 1–7, 2018. [74] D. Kuciauskas, J. M. Kephart, J. Moseley, W. K. Metzger, W. S. Sampath, and P. Dippo, “Recombination velocity less than 100 cm/s at polycrystalline Al 2 O 3 /CdSeTe interfaces,” Appl. Phys. Lett., vol. 112, no. 26, p. 263901, Jun. 2018. [75] A. H. Munshi et al., “Electron Reflector Behavior of CdTe for CdSeTe Absorber Devices,” Sol. Energy Mater. Sol. Cells, no. In-press. [76] I. Dharmadasa, “Review of the CdCl2 Treatment Used in CdS/CdTe Thin Film Solar Cell Development and New Evidence towards Improved Understanding,” Coatings, vol. 4, no. 2, pp. 282–307, 2014.

[77] E. S. Barnard et al., “3D Lifetime Tomography Reveals How CdCl2 Improves

143

Recombination Throughout CdTe Solar Cells,” Adv. Mater., vol. 29, no. 3, 2017.

[78] Amit H. Munshi, “Polycrystalline CdTe Photovoltaics with Efficiency Over 18% through Improved Absorber Passivation and Current Collection.” [79] A. H. Munshi et al., “Effect of CdCl 2 passivation treatment on microstructure and performance of CdSeTe/CdTe thin-film photovoltaic devices,” Sol. Energy Mater. Sol. Cells, vol. 186, no. May, pp. 259–265, 2018. [80] D. W. Lane, “A review of the optical band gap of thin film CdSxTe1-x,” Sol. Energy Mater. Sol. Cells, vol. 90, no. 9, pp. 1169–1175, 2006. [81] B. G. Mendis, L. Bowen, and Q. Z. Jiang, “A contactless method for measuring the recombination velocity of an individual grain boundary in thin-film photovoltaics,” Appl. Phys. Lett., vol. 97, no. 9, pp. 0–3, 2010. [82] B. G. Mendis and L. Bowen, “Cathodoluminescence measurement of grain boundary recombination velocity in vapour grown p-CdTe,” J. Phys. Conf. Ser., vol. 326, no. 1, 2011. [83] J. Lee, N. C. Giles, D. Rajavel, and C. J. Summers, “Room-temperature band-edge photoluminescence from cadmium telluride,” Phys. Rev. B, vol. 49, no. 3, p. 1668, 1994. [84] J.-H. Yang, W.-J. Yin, J.-S. Park, J. Ma, and S.-H. Wei, “Review on first-principles study of defect properties of CdTe as a solar cell absorber,” Semicond. Sci. Technol., vol. 31, no. 8, p. 083002, 2016. [85] J. H. Yang, L. Shi, L. W. Wang, and S. H. Wei, “Non-Radiative Carrier Recombination Enhanced by Two-Level Process: A First-Principles Study,” Sci. Rep., vol. 6, no. February, pp. 1–10, 2016. [86] A. Abbas, “The microstructure of thin film cadmium telluride photovoltaic materials,” Loughborough University, 2014. [87] T. A. M. Fiducia et al., “Understanding the role of selenium in defect passivation for highly efficient selenium-alloyed cadmium telluride solar cells,” Nat. Energy, vol. 4, no. 6, pp. 504–511, Jun. 2019. [88] T. A. M. Fiducia et al., “Defect Tolerance in as-deposited Selenium-alloyed Cadmium Telluride Solar Cells,” in World Conference on Photovoltaic Energy Conversion (WCPEC-7) Waikoloa, Hawaii, 2018. [89] X. Zheng et al., “Recombination and bandgap engineering in CdSeTe/CdTe solar cells,” APL Mater., vol. 7, no. 7, p. 071112, 2019. [90] D. Krasikov, “Selenium lowers bulk recombination,” Nat. Energy, May 2019.

[91] J. D. Poplawsky et al., “Structural and compositional dependence of the CdTeSe alloy layer photoactivity in CdTe based solar cells,” Nat. Commun., vol. 7, p. 12537, Jul.

144

2016.

[92] T. A. M. Fiducia et al., “Large Area 3D Elemental Mapping of a MgZnO/CdTe Solar Cell with Correlative EBSD Measurements,” in WCPEC 2018 - A Joint Conference of 45th IEEE PVSC, 28th PVSEC and 34th EU PVSEC, 2018, pp. 1702–1706.

[93] J. Yang and S.-H. Wei, “First-principles study of the band gap tuning and doping control in CdSexTe 1 − x alloy for high eficiency solar cell,” vol. 28, no. 8, 2019. [94] F. G. Sen et al., “Efficient CdTe photovoltaics by co-passivating grain boundaries,” in World Conference on Photovoltaic Energy Conversion (WCPEC-7) Waikoloa, Hawaii, 2018. [95] J. Guo et al., “Effect of selenium and chlorine co-passivation in polycrystalline CdSeTe devices,” Appl. Phys. Lett., vol. 115, no. 15, 2019. [96] B. I. Macdonald, A. Martucci, S. Rubanov, S. E. Watkins, P. Mulvaney, and J. J. Jasieniak, “Layer-by-Layer Assembly of Sintered CdSexTe1-x Nanocrystal Solar Cells,” ACS Nano, vol. 6, no. 7, pp. 5995–6004, 2012. [97] D. Kuciauskas, J. Moseley, P. Ščajev, and D. Albin, “Radiative Efficiency and Charge- Carrier Lifetimes and Diffusion Length in Polycrystalline CdSeTe Heterostructures,” Phys. Status Solidi - Rapid Res. Lett., vol. 1900606, pp. 1–6, 2019. [98] B. G. Mendis, A. Howkins, D. Stowe, J. D. Major, and K. Durose, “The role of transition radiation in cathodoluminescence imaging and spectroscopy of thin-foils,” Ultramicroscopy, vol. 167, pp. 31–42, 2016. [99] T. Bidaud, J. Moseley, M. Al-jassim, M. Amarasinghe, and W. K. Metzger, “Spatially and spectrally resolved defects in polycrystalline CdTe thin films revealed by quantitative cathodoluminescence,” in 46TH IEEE PHOTOVOLTAIC SPECIALISTS CONFERENCE, Chicago, 2019. [100] J. Moseley et al., “Luminescence methodology to determine grain-boundary, grain- interior, and surface recombination in thin-film solar cells,” J. Appl. Phys., vol. 124, no. 11, 2018.

[101] T. A. M. Fiducia et al., “Three-Dimensional Imaging of Selenium and Chlorine Distributions in Highly Efficient Selenium-Graded Cadmium Telluride Solar Cells,” IEEE J. Photovoltaics, vol. PP, pp. 1–5, 2019. [102] Y. Yoon et al., “Unveiling Defect-Mediated Charge-Carrier Recombination at the Nanometer Scale in Polycrystalline Solar Cells,” ACS Appl. Mater. Interfaces, vol. 11, no. 50, pp. 47037–47046, 2019.

[103] B. Bissig et al., “On a better estimate of the charge collection function in CdTe solar cells: Al2O3 enhanced electron beam induced current measurements,” Thin Solid Films, vol. 633. pp. 218–221, 2017.

145

[104] M. J. Watts, T. A. M. Fiducia, B. Sanyal, R. Smith, J. M. Walls, and P. Goddard, “Enhancement of photovoltaic efficiency in CdSe x Te 1− x (where 0 ⩽ x ⩽ 1): insights from density functional theory,” J. Phys. Condens. Matter, vol. 32, no. 12, p. 125702, Mar. 2020. [105] J. Guo et al., “Study of Effects of Cl and Se in CdSeTe Solar Cells Using Scanning Transmission Electron Microscopy,” Microsc. Microanal., vol. 25, no. S2, pp. 2150– 2151, 2019. [106] R. F. Klie, “Improving reliability and reducing cost in CdTe photovoltaics via grain boundary engineering, DOE award number DE-EE0007545,” 2019. [107] D. L. McGott et al., “Thermomechanical Lift-Off and Recontacting of CdTe Solar Cells,” ACS Appl. Mater. Interfaces, vol. 10, no. 51, pp. 44854–44861, 2018. [108] W. Orellana, E. Menéndez-Proupin, and M. A. Flores, “Self-compensation in chlorine- doped CdTe,” Sci. Rep., vol. 9, no. 1, 2019.

[109] “Hanergy sets new efficiency record for heterojunction cell.” [Online]. Available: https://www.pv-magazine.com/2019/08/08/hanergy-sets-new-efficiency-record- for-heterojunction-module/.

[110] J. Jean, M. Woodhouse, and V. Bulović, “Accelerating Photovoltaic Market Entry with Module Replacement,” Joule, pp. 1–18, 2019.

[111] K. T. Vansant et al., “Toward Low-Cost 4-Terminal GaAs//Si Tandem Solar Cells,” ACS Appl. Energy Mater., vol. 2, no. 4, pp. 2375–2380, 2019.

146