Highly parallel simulation and optimization of photonic circuits in time and frequency domain based on the deep-learning framework PyTorch

Floris Laporte,Peter Bienstman,Joni Dambre

doi:10.1038/s41598-019-42408-2

Floris Laporte, Peter Bienstman + Show 1 more

Open Access

https://doi.org/10.1038/s41598-019-42408-2

Copy DOI

Journal: Scientific Reports	Publication Date: Apr 11, 2019
Citations: 24	License type: open-access

Affiliation: Ghent University

Abstract

We propose a new method for performing photonic circuit simulations based on the scatter matrix formalism. We leverage the popular deep-learning framework PyTorch to reimagine photonic circuits as sparsely connected complex-valued neural networks. This allows for highly parallel simulation of large photonic circuits on graphical processing units in time and frequency domain while all parameters of each individual component can easily be optimized with well-established machine learning algorithms such as backpropagation.

Highlights

Photonic circuit simulation software already exists, it definitely has not yet reached the same maturity as electronic circuit simulation software
We present Photontorch, a tool loosely based on the node-based approach of Caphe[2], which in itself is based on coupled-mode theory[5,6], but reduces the number of parameters by eliminating memory-less nodes that are independent on time before doing a simulation
It is this approach of treating a photonic circuit as essentially a sparsely connected recurrent neural network that may be a key ingredient in future photonic circuit design

Summary

Description of the framework

This network cannot be connected to other components, as it has no free ports left. For these kind of top-level networks, the number of ports can be reduced. Ports for which a custom action is defined will have access to the state x, as well as previous states through a buffer This allows for example the definition of a component action in the form of an ordinary differential equation (ODE). Where q represents the current time step such that x(q ⋅ dt) = xq, m represents the number of wavelengths (or modes), n represents the number of MC nodes and b represents the number of parallel simulations performed at once, i.e. the batch size. This enables a whole new way of optimizing the parameters of the photonic circuit

Performance metrics

Optimization results

Discussion

Findings

Additional Information