Wolfger Peelaers is a senior research scientist at Hewlett Packard Labs, the research division of Hewlett Packard Enterprise. He holds a PhD in theoretical high-energy physics from Stony Brook University. His current research focuses on leveraging machine learning methodologies to design analog devices, circuits, and systems that can serve as accelerators for optimization or machine learning workloads.
Over the past decade, more automation has trickled into the workflow of designers of photonic devices. Specifically, inverse design techniques based on the adjoint method have allowed designers to find low-loss passive integrated devices based on local search techniques. Lately, to avoid the limitations of inherently local search techniques, which might get stuck in local minima if sub-optimal initial conditions are provided, the field has explored global optimization techniques, inspired by recent progress in physics-informed neural networks and related trends in the machine learning community. In this talk, we will contrast and compare the latest progress in both local and global inverse design techniques, with a particular focus on their performance for integrated photonic devices and circuits, as well as ease-of-use.