| Abstract Scope |
Numerical simulations have revolutionized material design. However, although simulations excel at mapping an input material to its output property, their direct application to inverse design has traditionally been limited by their high computing cost and lack of differentiability. Here, taking the example of the inverse design of a porous matrix featuring a target sorption isotherm, we introduce a computational inverse design framework that addresses these challenges. First, we adopt end-to-end differentiability to build a differentiable forward numerical simulation. Thanks to its differentiability, the forward simulation is used to directly train a backward deep generative model, which outputs an optimal porous matrix based on an arbitrary input sorption isotherm curve. Second, this inverse design pipeline leverages the power of tensor processing units (TPU)—an emerging family of dedicated chips, which, although they are specialized in deep learning, are flexible enough for intensive scientific simulations. |