Low-depth, compact and error-tolerant photonic matrix-vector multiplication beyond the unitary group

TL;DR

This paper presents a new low-depth, error-tolerant photonic circuit architecture capable of implementing non-unitary matrices, enhancing the efficiency and practicality of photonic neural networks and quantum applications.

Contribution

The authors introduce a compact, low-depth photonic circuit design that supports non-unitary matrices, improving integration and error tolerance over previous architectures.

Findings

Lower circuit depth compared to standard designs

High tolerance to hardware errors

Supports non-unitary matrix implementation

Abstract

Large-scale programmable photonic circuits are opening up new possibilities for information processing providing fast and energy-efficient means for matrix-vector multiplication. Here, we introduce a novel architecture of photonic circuits capable of implementing non-unitary transfer matrices, usually required by photonic neural networks, iterative equation solvers or quantum samplers. Our architecture exploits compact low-depth beam-splitter meshes rather than bulky fully connected mixing blocks used in previous designs, making it more compatible with planar integrated photonics technology. We have shown that photonic circuits designed with our architecture have lower depth than their standard counterparts and are extremely tolerant to hardware errors.