Abstract:Shape optimization approaches to inverse design offer low-dimensional, physically-guided parameterizations of structures by representing them as combinations of shape primitives. However, on discretized rectilinear simulation grids, computing the gradient of a user objective via the adjoint variables method requires a sum reduction of the forward/adjoint field solutions and the Jacobian of the simulation material distribution with respect to the structural shape parameters. These shape parameters often perturb large or global parts of the simulation grid resulting in many non-zero Jacobian entries, which are typically computed by finite-difference in practice. Consequently, the gradient calculation can be non-trivial. In this work we propose to accelerate the gradient calculation by invoking automatic differentiation (AutoDiff) in instantiations of structural material distributions. In doing so, we develop extensible differentiable mappings from shape parameters to shape primitives and differentiable effective logic operations (denoted AutoDiffGeo). These AutoDiffGeo definitions may introduce some additional discretization error into the field solutions because they relax notions of sub-pixel smoothing along shape boundaries. However, we show that some mappings (e.g. simple cuboids) can achieve zero error with respect to volumetric averaging strategies. We demonstrate AutoDiff enhanced shape optimization using three integrated photonic examples: a multi-etch blazed grating coupler, a non-adiabatic waveguide transition taper, and a polarization-splitting grating coupler. We find accelerations of the gradient calculation by AutoDiff relative to finite-difference often exceed 50x, resulting in total wall time accelerations of 4x or more on the same hardware with little or no compromise to final device performance. Our code is available open source at <a class="link-external link-https" href="https://github.com/smhooten/emopt" rel="external noopener nofollow">this https URL</a>

A Gradient-Based Optimization Approach to the Inverse Problem for Multi-Layered Structures

Conjugate Adjoint Gradient-Based Inverse Design Method for Aperiodic Frequency-Selective Surface with Weighted Loss

Adversarial-Network Regularized Inverse Design of Frequency-Selective Surface with Frequency-Temporal Deep Learning

Simulation of Locally Inhomogeneous Multilayer Planar Structure by Hybrid Method

Design Multiple-Layer Gradient Coils Using Least-Squares Finite Element Method

Solving Electromagnetic Inverse Problem Using Adaptive Gradient Descent Algorithm.

Merging automatic differentiation and the adjoint method for photonic inverse design

An Optimization Approach to the Frequency-Domain Inverse Problem for a Nonuniform LCRG Transmission Line

Electrical performance simulation of inhomogeneous multilayer LTCC structure by hybrid method

Gradient-based topology optimization of soft dielectrics as tunable phononic crystals

Optimisation Approach To The Time-Domain Electromagnetic Inverse Problem For A Bi-Isotreopic Slab

Efficient inverse design of optical multilayer nano-thin films using neural network principles: backpropagation and gradient descent

Automatic differentiation accelerated shape optimization approaches to photonic inverse design on rectilinear simulation grids

Multilayered Media Parameter Inversion Based on Reflected Trajectory Fitting Method

Inverse Design for Waveguide Dispersion with a Differentiable Mode Solver

Inverse design of nanophotonic structures using complementary convex optimization

Transformation Inverse Design

Reconstruction of the 2-Dimensional Inhomogeneous Conductivity Profile Using Iterative and Bi-Conjugate Gradient Method

An Optimization Approach to a Three-Dimensional Acoustic Inverse Problem in the Time Domain

Parameter Inversion of Layered Media Covered by a Rough Surface

Optical Multilayer Thin Film Structure Inverse Design: From Optimization to Deep Learning