We envision a scientific laboratory where the process of materials discovery continues without disruptions, aided by computational power augmenting the human mind, and freeing the latter to perform research closer to the speed of imagination, addressing societal challenges in market-relevant timeframes. — “Accelerating Materials Development via Automation, Machine Learning, and High-Performance Computing,” Joule 2, 1410 (2018)
In this talk, I’ll describe how machine learning (ML) is ripe to disrupt three research tasks: diagnosis, process optimization, and discovery. “Diagnosis” in this context refers to the purposeful application of characterization to identify underlying performance-limiting physics, an essential step toward improving early-stage prototypes. ML can increase diagnosis speed and efficiency. As an example, I’ll demonstrate how Bayesian inference identifies underlying bulk and interface properties limiting the performance of early-stage photovoltaic devices, ten times faster than traditional spectroscopy methods. Accurate diagnosis directs “process optimization,” which can help determine the performance ceiling of novel materials, and when to abandon unsuccessful candidates quickly. Lastly, materials screening, in combination with high-throughput experimentation enabled by ML, can lead to novel materials “discovery,” as we illustrate with the case of novel lead-free perovskite-inspired photovoltaic materials with promising optoelectronic properties. In conclusion, I’ll illustrate how these principles generalize to other systems, and promise to accelerate the cycle of learning by ≥10x across a range of chemistry and materials disciplines.
Bio: Tonio Buonassisi is Associate Professor of Mechanical Engineering at the Massachusetts Institute of Technology, and director of the A*STAR Programme on Accelerated Materials Development for Manufacturing. His research interests are at the intersection of machine learning, high-throughput experimentation, and high-performance computing. He founded the MIT PVLab and co-founded the Fraunhofer Center for Sustainable Energy Systems in Boston, USA, and is known for his work in the field of photovoltaics, predictive process simulation, defects, multiscale characterization (synchrotron), photoelectrochemistry, system design and modeling, and technoeconomic analysis. His work was recognized with a Google Faculty Award, Presidential Early Career Award for Scientists and Engineers (PECASE), MIT Everett Moore Baker Memorial Award for Excellence in Undergraduate Teaching, National Science Foundation CAREER Award, among others.