Noise in analog programmable-photonic computation

TL;DR

This paper analyzes noise effects in analog programmable-photonic computation (APC) using a geometric model, experimentally validating noise maps and providing design criteria for robust optical computing systems.

Contribution

It introduces a comprehensive noise analysis framework for APC based on the Generalized Bloch Sphere, validated through experiments on silicon photonic chips.

Findings

Noise sources can be mapped onto the GBS for APC.

Experimental validation confirms the noise model's accuracy.

Design criteria for noise-resilient analog constellations are proposed.

Abstract

Analog Programmable-Photonic Computation (APC) leverages programmable integrated photonics (PIP) to perform high-speed matrix operations using optical waves. However, the continuous nature of optical waves that implement the analog bits or anbits - the fundamental unit of information in APC - makes computational results intrinsically sensitive to physical noise. Here, we establish and experimentally validate a comprehensive noise analysis in APC platforms using the geometric representation of the anbit in the Generalized Bloch Sphere (GBS). By modeling the physical noise sources in PIP circuits as random photocurrent fluctuations at the output of the opto-electrical (O/E) converter, and using error propagation theory, the noise statistics can be projected onto the GBS. This approach leads to specific noise maps in the GBS for each noise source, enabling the identification of the dominant noise sources within the APC system. Analytical predictions are numerically and experimentally validated on a fabricated silicon PIP chip. Beyond statistical characterization of system noise, the proposed model provides quantitative design criteria for noise-adapted analog constellations in the GBS, advancing APC towards scalable and robust optical computing systems, with potential applications in emerging paradigms such as photonic neuromorphic computing.