ELight: Enabling Efficient Photonic In-Memory Neurocomputing with Life Enhancement

TL;DR

ELight significantly reduces write efforts and power consumption in photonic in-memory neurocomputing, enhancing reliability and longevity while maintaining accuracy for practical machine learning applications.

Contribution

The paper introduces a novel optimization framework, ELight, combining write-aware training and post-training techniques to minimize PCM writes in photonic neural networks.

Findings

Over 20X reduction in total writes and dynamic power

Maintains comparable neural network accuracy

Extends PCM lifetime and reduces programming energy

Abstract

With the recent advances in optical phase change material (PCM), photonic in-memory neurocomputing has demonstrated its superiority in optical neural network (ONN) designs with near-zero static power consumption, time-of-light latency, and compact footprint. However, photonic tensor cores require massive hardware reuse to implement large matrix multiplication due to the limited single-core scale. The resultant large number of PCM writes leads to serious dynamic power and overwhelms the fragile PCM with limited write endurance. In this work, we propose a synergistic optimization framework, ELight, to minimize the overall write efforts for efficient and reliable optical in-memory neurocomputing. We first propose write-aware training to encourage the similarity among weight blocks, and combine it with a post-training optimization method to reduce programming efforts by eliminating redundant writes. Experiments show that ELight can achieve over 20X reduction in the total number of writes and dynamic power with comparable accuracy. With our ELight, photonic in-memory neurocomputing will step forward towards viable applications in machine learning with preserved accuracy, order-of-magnitude longer lifetime, and lower programming energy.