Proceedings of International Conference on Hybrid and Organic Photovoltaics (HOPV24)
DOI: https://doi.org/10.29363/nanoge.hopv.2024.023
Publication date: 6th February 2024
Organic or perovskite photovoltaics poses a multi-objective optimization problem in a large multi-dimensional parameter space. Massive progress was achieved in developing methods to accelerate solving such complex optimization tasks. We have demonstrated for both types of semiconductors, that the combination of Gaussian Process Regression (GPR) and Bayesian Optimization (BO) are most efficient in predicting new materials, identify optimized processing conditions or invent alternative device architectures in larger parameter rooms. For a 4 dim space (solvent, donor-acceptor ratio, spin speed, concentration) with about 1000 variations in a 10 % grid space, 30 samples are sufficient to find the optimum. For 5 & 6 dimensional spaces, the possible variations go into the millions. Nevertheless, our automated lines, being operated in an autonomous optimization mode, were able to identify globalized optima within several hundred´s of experiments. That raises the question whether these large material spaces as well hold the promise for discoveries. We extended the BO concept towards the discovery of new molecules that can be integrated into the device optimization cycle. The research campaign found molecular semiconductors that had not been published before but yielded performance values bypassing the current state of the art materials.