Photonic KAN: A Kolmogorov-Arnold Network Inspired Efficient Photonic Neuromorphic Architecture

Abstract

Photonic analog accelerators offer a promising shift in AI hardware, potentially improving inference bandwidth, latency, and power consumption by several orders of magnitude over digital counterparts. Recently, Kolmogorov-Arnold Networks (KAN) models were introduced, demonstrating enhanced parameter scaling and interpretability compared to traditional multilayer perceptron (MLP) models. Inspired by the KAN architecture, we propose the Photonic KAN -- an integrated all-optical neuromorphic platform leveraging highly parametric nonlinear transfer functions along KAN edges to overcome key limitations in photonic neural networks. In this work, we implement such nonlinearities in the form of cascaded ring-assisted Mach-Zehnder Interferometer (RAMZI) devices. In our test cases, the Photonic KAN showcases enhanced parameter scaling and interpretability compared to existing photonic neural networks. The Photonic KAN achieves approximately 2300× reduction in footprint-energy efficiency, alongside a 7× reduction in latency in function-fitting tasks compared to previous MZI based photonic accelerators. This breakthrough presents a promising new avenue for expanding the scalability and efficiency of neuromorphic hardware platforms.