spectrapepper: A Python toolbox for advanced analysis of spectroscopic data for materials and devices.

Abstract

In recent years, the complexity of novel high-tech materials and devices has increased considerably. This complexity is primarily in the form of increasing numbers of components and broader ranges of applications. An example of the latter is the last generation of thin-film solar cells, which comprise several functional microand nanolayers including back contact, absorber, buffer, and transparent front contact. Most of these layers are complex multicomponent compounds (Cu(In,Ga)Se2, Sb2Se3, CdTe, CdS, Zn(O,S), ZnO:Al, etc.) that require fine-tuning of their physicochemical properties to ensure functionality and high peformance (Chopra et al., 2004; Powalla et al., 2018). This embedded complexity means that further development of such devices requires advanced characterization and methodologies that allow correlating the physicochemical data of the different layers (chemical composition, structural properties, defect concentration, etc.) with the performance of the final devices in a fast, precise, and reliable way. In this regard, non-destructive methodologies based on spectroscopic characterization techniques (Raman, photoluminescence, X-ray fluorescence, reflectance, transmittance, etc.) have already been demonstrated to possess a high versatility and potential for this type of analyses (Dimitrievska et al., 2019; Guc et al., 2017; Oliva et al., 2016). These spectroscopybased methodologies can provide deep information that encompasses the complexity of novel materials and devices in a non-destructive way, providing a profound understanding of their properties, failure mechanisms, and possible improvements (Grau-Luque et al., 2021). The latest advances in the application of spectroscopic methodologies for complex materials and devices include the implementation of combinatorial analysis (CA), artificial intelligence (AI) and machine learning (ML), that have been already used in few studies and are slowly becoming more common (Chen et al., 2020). Furthermore, the widespread use of this kind of tools in both laboratory environments and on-line/in-line monitoring of production lines is predicted to shorten development times by a factor of 10, from 10 to 20 years to just a few years (Aspuru-Guzik & Persson, 2018; Correa-Baena et al., 2018; Maine & Garnsey, 2006; Mueller et al., 2016). Unfortunately, several barriers for researchers to implement CA, AI, and ML remain (Gu et al., 2019; Mahmood & Wang, 2021). One of them is the proper pre-processing of spectroscopic data that allows not only to emphasize the relevant changes in the spectra, but also to combine data obtained from different techniques and instruments. Also, the use of ML requires substantial amounts of high-quality data for a precise analysis of the physicochemical parameters of new materials and devices, which necessitates the use of automated systems for massive characterization measurements. In other words, the implementation of automated high-throughput experiments and the capability to perform big-data pre-processing to enhance features of spectroscopic data for ML, and subsequent CA, requires deep theoretical, statistical, analytical, and programming knowledge.