All-Optical Doubly Resonant Cavities for ReLU Function in Nanophotonic Deep Learning

Abstract

We present a novel approach to implementing all-optical Rectified Linear Unit (ReLU) activation functions using compact doubly-resonant cavities with dimensions of approximately $10\mu \mathrm{m}$. Our design leverages ${\chi }^{(2)}$ nonlinear processes within carefully engineered photonic structures that simultaneously resonate at both fundamental and second-harmonic frequencies. By exploiting the phase-sensitive nature of second-harmonic generation, we demonstrate an optical analog to the ReLU function, achieving femtojoule-level activation energy-comparable to state-of-the-art approaches-while reducing device footprint by two orders of magnitude compared to previous implementations. We develop the theoretical framework using coupled-mode theory and validate it through rigorous finite-difference time-domain simulations. Beyond ReLU, we show that the same physical structure can implement alternative activation functions such as ELU and GELU through simple adjustments to input conditions. Neural network simulations demonstrate that our proposed optical activation functions achieve classification accuracy within $0.4\%$ of ideal electronic implementations while offering significant advantages in energy efficiency and processing speed. This work represents a significant advancement toward realizing energy-efficient, high-density optical neural networks for next-generation artificial intelligence hardware.

References in corpus (23)

• Deep Learning with Coherent Nanophotonic Circuits
• All-Optical Machine Learning Using Diffractive Deep Neural Networks
• Parallel convolution processing using an integrated photonic tensor core
• Outlook for inverse design in nanophotonics
• Photonics for artificial intelligence and neuromorphic computing
• 11 TeraFLOPs per second photonic convolutional accelerator for deep learning optical neural networks
• Continuous-variable optical quantum state tomography
• Inverse design and demonstration of a compact and broadband on-chip wavelength demultiplexer
• Neuromorphic Silicon Photonic Networks
• Attojoule Optoelectronics for Low-Energy Information Processing and Communications: a Tutorial Review
• Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit
• Modeling of octave-spanning Kerr frequency combs using a generalized mean-field Lugiato-Lefever model
• Self-Configuring Universal Linear Optical Component
• Periodically poled thin film lithium niobate microring resonators with a second-harmonic generation efficiency of 250,000%/W
• Reprogrammable Electro-Optic Nonlinear Activation Functions for Optical Neural Networks
• Nanophotonic Computational Design
• Photonic tensor cores for machine learning
• Chi2 and chi3 harmonic generation at a critical power in inhomogeneous doubly resonant cavities
• Quantum Dynamics of Kerr Optical Frequency Combs below and above Threshold: Spontaneous Four-Wave-Mixing, Entanglement and Squeezed States of Light
• Integrated Nonreciprocal Photonic Devices with Dynamic Modulation
• Nonlinear harmonic generation and devices in doubly-resonant Kerr cavities
• Dynamic Control of Spontaneous Emission Using Magnetized InSb Higher-Order-Mode Antennas
• Free-Space Optical Spiking Neural Network