“Yes, The Same Device Twice”—Toward Improved Reproducibility in Halide Perovskite Research

Assuming reproducibility is essential for scientific progress (and manufacturing scale-up!), it is surprising how few studies make “reproducibility” a key focus area. This talk is a response to this status quo.<br/><br/>First, we give space to the discussion of “hidden” variables affecting reproducibility. Goetz and Vaynzof’s seminal contribution [1] states: “it is clear that [halide perovskite] properties are incredibly sensitive to numerous factors, many of which remain unknown even after a decade of extensive research.” Recent efforts in the field to elucidate, measure, and control these variables will be summarized, as well as spotlights of recent research occurring throughout the ADDEPT Center to make perovskites more reproducible [2].<br/><br/>Second, we give space to the concepts of “determinism” and “probability” in materials research. While indeed the underlying laws governing perovskite film formation and degradation appear to be deterministic, the outsized role of discrete events in stimulating phase transitions invites a probabilistic perspective. In this context, we discuss a growing body of evidence that suggests that process control through automation and end-to-end impurity management can reduce the probability of negative uncontrolled variables, thus enabling the reproducibility needed to perform high-quality science.<br/><br/>Third, we discuss practical challenges of managing some degree of irreproducibility when establishing a process-recipe baseline or conducting an optimization campaign. In one example of an efficiency-optimization campaign, moderate run-to-run variance prompted us to consider another approach to straight-up Bayesian optimization: establishing “trust regions” believed to contain the noise-free optimum, and performing Latin hypercube searches therein, dynamically adjusting as experimental conditions changed [3].<br/><br/><br/>[1] Katelyn P. Goetz and Yana Vaynzof, “The Challenge of Making the Same Device Twice in Perovskite Photovoltaics,” <i>ACS Energy Letters</i> <b>7</b>, 1750–1757 (2022).<br/>[2] https://news.mit.edu/2023/moving-perovskite-advancements-lab-manufacturing-floor-0420<br/><br/>[3] Z. Liu <i>et al</i>., “Machine learning with knowledge constraints for process optimization of open-air perovskite solar cell manufacturing,” <i>Joule</i> <b>6</b>, 834–849 (2022).