End-to-end differentiability and tensor processing unit computing to accelerate materials’ inverse design

Numerical simulations have revolutionized material design. However, their direct application to inverse design has traditionally been limited by their high computing cost and lack of differentiability. This paper introduces a computational inverse design framework that addresses these challenges by programming differentiable simulation on the TensorFlow platform, leveraging automated end-to-end differentiation. The differentiable simulation directly trains a deep generative model, outputting an optimal porous matrix based on an input sorption isotherm curve. This pipeline leverages Tensor Processing Units (TPUs) for intensive scientific simulations, promising to accelerate inverse materials design.

Han Liu, Yuhan Liu, Kevin Li, Zhangji Zhao, Samuel S. Schoenholz, Ekin D. Cubuk, Puneet Gupta, Mathieu Bauchy