Transferable Learning on Analog Hardware

TL;DR

This paper introduces one-time error-aware training methods for analog neural networks, especially photonic interferometers, enabling robust performance despite significant static hardware errors, thus improving practicality and scalability.

Contribution

It presents novel training techniques that produce error-robust neural networks transferable to faulty hardware without re-training.

Findings

Networks maintain performance with errors up to 5x current tolerances.

One-time training achieves robustness comparable to ideal hardware.

Methods eliminate need for hardware-specific re-training or high-quality components.

Abstract

While analog neural network (NN) accelerators promise massive energy and time savings, an important challenge is to make them robust to static fabrication error. Present-day training methods for programmable photonic interferometer circuits, a leading analog NN platform, do not produce networks that perform well in the presence of static hardware errors. Moreover, existing hardware error correction techniques either require individual re-training of every analog NN (which is impractical in an edge setting with millions of devices), place stringent demands on component quality, or introduce hardware overhead. We solve all three problems by introducing one-time error-aware training techniques that produce robust NNs that match the performance of ideal hardware and can be exactly transferred to arbitrary highly faulty photonic NNs with hardware errors up to 5x larger than present-day fabrication tolerances.