Correlative X-ray microscopy studies of CuIn₁-xGaxSe₂ solar cells

It is well known that the overall performance of a solar cell is limited by the worst performing areas of the device. These areas are usually micro and nano-scale defects inhomogenously distributed throughout the material. Mitigating and/or engineering these effects is necessary to provide a path towards increasing the efficiency of state-of-the-art solar cells. The first big challenge is to identify the nature, origin and impact of such defects across length scales that span multiple orders of magnitude, and dimensions (time, temperature etc.). In this work, I present a framework based on correlative X-ray microscopy and big data analytics to identify micro and nanoscale defects and their impact on material properties in CuIn1-xGaxSe2 (CIGS) solar cells.

Synchrotron based X-ray Fluorescence (XRF) and X-ray Beam Induced Current (XBIC) are used to study the effect that compositional variations, between grains and at grain boundaries, have on CIGS device properties. An experimental approach is presented to correcting XRF and XBIC quantification of CIGS thin film solar cells. When applying XRF and XBIC to study low and high gallium CIGS devices, it was determined that increased copper and gallium at grain boundaries leads to increased collection efficiency at grain boundaries in low gallium absorbers. However, composition variations were not correlated with changes in collection efficiency in high gallium absorbers, despite the decreased collection efficiency observed at grain boundaries.

Understanding the nature and impact of these defects is only half the battle; controlling or mitigating their impact is the next challenge. This requires a thorough understanding of the origin of these defects and their kinetics. For such a study, a temperature and atmosphere controlled in situ stage was developed. The stage was utilized to study CIGS films during a rapid thermal growth process. Comparing composition variations across different acquisition times and growth temperatures required the implementation of machine learning techniques, including clustering and classification algorithms. From the analysis, copper was determined to segregate the faster than indium and gallium, and clustering techniques showed consistent elemental segregation into copper rich and copper poor regions. Ways to improve the current framework and new applications are also discussed.