RoNo: A novel way in generating reconfigurable on-chip nonlinear activation functions

TL;DR

This paper introduces RoNo, a reconfigurable on-chip nonlinear activation function generator for optical neural networks, using inverse design and control networks to achieve compact, high-quality nonlinear responses suitable for wave-based analog computing.

Contribution

It presents a novel optical design and integration method for reconfigurable nonlinear activation functions in ONNs, improving compactness and functionality over prior approaches.

Findings

Successfully generated two nonlinear responses with inverse design.

Achieved high accuracy on MNIST and CIFAR-10 datasets.

Demonstrated reconfigurability and compactness of the system.

Abstract

Due to the limitations of Moore's Law and the increasing demand of computing, optical neural network (ONNs) are gradually coming to the stage as an alternative to electrical neural networks. The control of nonlinear activation functions in optical environments, as an important component of neural networks, has always been a challenge. In this work, firstly, we use inverse design tools to design a optical patterned area in silicon-carbide-on-insulator. This patterned area could generate two different nonlinear responses of the amplitude. Secondly, the patterned region is integrated with a control network to form a reconfigurable on-chip nonlinear activation function generator for wave-based analog computing. Experiment shows that neural network that uses such a system as an activation function performs well in the MNIST handwritten dataset and CIFAR-10, respectively. Compared to previous works, we propose a novel approach to generate on-chip reconfigurable activation functions in optical neural networks, which achieves compact footprint and enables high-quality activation function generation.