A differentiable solar cell simulator, newly developed by researchers at MIT and Google Brain, tells scientists which changes will provide the improvements they wish to make in a solar cell configuration. The new simulator computes the power conversion efficiency (PCE) of an input photovoltaic (PV) design, and the derivative of the PCE with respect to any input parameters. Further, it enables efficient materials optimization of PV cells, and it can be used with standard optimization methods and machine learning algorithms. The simulator is considered end-to-end because it computes the sensitivity of the cell’s efficiency while taking into account light absorption. Solar cells have many variables that can be adjusted to improve performance, including material type, thickness, and geometric arrangement. Such a system, which enables the joint optimization of multiple design variables, could speed the discovery of new, improved solar cell configurations. Traditional simulators allow scientists to evaluate potential design changes without having to actually build each new variation for testing, but the process of simulating one variation at a time is a slow one. Traditional simulators can predict the efficiency of a solar cell based on its configuration, but the differentiable simulator takes this capability a step further. It predicts the efficiency — that is, the percentage of the energy from incoming sunlight that is converted to an electric current — and it also shows how much the cell’s efficiency is affected by any one of the input parameters. “It tells you directly what happens to the efficiency if we make this layer a little bit thicker, or what happens to the efficiency if we, for example, change the property of the material,” MIT researcher Giuseppe Romano said. The new simulator shows scientists how changes to the relative thickness of the cell’s different layers will affect its output. “There is a strong interplay between light propagation and the thickness of each layer and the absorption of each layer,” researcher Sean Mann said. In addition to layer thickness, the differentiable solar cell simulator can evaluate the amount of doping that each layer receives, the dielectric constant of the insulating layers, and the bandgap, providing direction on how to change the cell for best results. “That makes the process much faster, because instead of exploring the entire space of opportunities, you can just follow a single path,” Mann said. The differentiable solar cell simulator is a 1D simulation tool that is programmed in Python and that solves drift-diffusion equations using JAX, an automatic differentiation library for scientific computing. Currently, the simulator is based on a 1D version of a solar cell, but the research team plans to expand the simulator’s capabilities to include 2D and 3D configurations. The researchers used the simulator to demonstrate perovskite solar cell optimization and multiparameter discovery, and compared the results with random search and finite differences. The team has made its simulator available as an open-source tool that is available now to help guide research in the photovoltaics field. The simulator’s computations can be coupled with an optimization algorithm or a machine learning system to rapidly assess a range of possible changes to a solar cell and the most promising ways to modify the cell to improve efficiency. The simulator’s code is open source, too, and the photovoltaics community will be able to make enhancements and improvements to the code as well as introduce new capabilities to the simulation tool. Additionally, the simulator can be used with most types of cells. It is not yet able to directly simulate tandem cells that use different materials, but the researchers said that a tandem solar cell could be approximated by simulating each cell individually. A team at MIT has introduced a system that both predicts the efficiency of new photovoltaic solar cell materials and shows how much different input parameters affect output. Courtesy of MIT News, iStockphoto. “We are decreasing the number of times that we need to run a simulator to give quicker access to a wider space of optimized structures,” Romano said. In addition, he said, the new simulation tool can identify unusual material parameters that were previously hidden from view. “Differentiable physics is going to provide new capabilities for the simulations of engineered systems,” Carnegie Mellon University professor Venkat Viswanathan said. “The differentiable solar cell simulator is an incredible example of differentiable physics that can now provide new capabilities to optimize solar cell device performance.” The research was published in Computer Physics Communications (www.doi.org/10.1016/j.cpc.2021.108232)