Intellectual Merit: This award produced several new computer codes that allowed us to explore how well we can use machine learning (or artificial intelligence) to reduce the time it takes to predict which chemical components will improve the performance of solar cell materials. The materials we chose to explore are particularly exciting because they can be made in a solution at room temperature using earth-abundant components. They use much less energy than the market leader, silicon. We created new computer codes that searched through an otherwise impossibly large set of possible components to consider if we had used a trial-and-error approach.  The machine learning code was trained against accurate computer-generated data that were accurate down to an atom-by-atom description of the material. These computer-generated predictions were tested in comparison to carefully designed complementary experiments. In this way, we were able to test how well our unimaginably small quantum-level models can produce information that can be measured at a far larger scale in the laboratory. We were able to show that, for several test cases for which we were able to determine the answer in advance, our machine learning code was able to find the best answer by considering only 1/7 of all possible candidate materials. We also showed that our code outperformed existing search methods. Along the way, we also uncovered a lot of fundamental information about the nature of these solar cell materials and how they form that is important for making them at scale. For instance, we found both experimental and computational ways to determine which is the best solvent to use to make the materials in the lab. 

Broader Impacts: The major impact of this award is likely to be its profound effect on human capital, in the career development of the 19 personnel associated with this project. One supported post-doc was hired as a university faculty member (but who is now leading Uber’s Bayesian Optimization group in San Francisco).  The other supported graduate students and postdocs are now helping companies such as Uber, Google, Yelp, United Display Corporation and GlaxoSmithKline with their specialized knowledge of materials and/or machine learning.  Six of the 11 undergraduates who helped us with this project have now graduated or matriculated in graduate programs and the remainder have also applied this year. Our team members reflected a strongly diverse group: Two of the three faculty PIs are women. Three of the five PhD students were women and two were from groups under-presented in STEM. Six of the eleven undergraduates were women and one was a URM. It is noteworthy that the award trained several women to become expert programmers in a field in which women are traditionally very scarce. All our team members gained an unusual perspective on combining information from multiple sources outside their own domain knowledge. Every student in the team improved their communication skills, which they used in interactions with grade-school students to prepare them to write code. We developed a new course to introduce freshmen to computational science and engineering, aimed at those who are typically under-represented in programming careers. We also created new Diversity and Inclusion programs at Cornell and Johns Hopkins Universities.   Beyond the career development aspects of this award, we also produced software codes that are readily available to the whole national and international communities of scientists using the open-source code repository, github.  It is important to note that the codes we produced are not specific to the materials class that we studied, but are generally applicable to essentially any materials system.   Our award has produced 29 peer-reviewed publications in journals and conference proceedings and made about 26 presentations at national venues.

All of us in this team are grateful for this opportunity to learn more about materials that could help us to achieve a more sustainable future.  This team of people with three different sets of skills which we have interwoven towards the single focus of this project to make better materials for solar cells; this made it a truly special program. 

Last Modified: 03/05/2020
Modified by: Paulette Clancy