Architecture-Level Simplification and Nonlinearity Enhancement of Photonic Reservoir Computing with Only Two MZMs
Baoqin Ding, Li Pei, Jianshuai Wang, Bing Bai, Tigang Ning, Zhouyi Hu, Bowen Bai · ACS Photonics · 2026
Photonic reservoir computing (PRC) has emerged as a promising framework for ultrafast, low-power information processing. While increasing the number of physical or virtual nodes can improve performance, it also substantially raises hardware and computational complexity. In this work, we propose a simplified PRC architecture that uses only two Mach–Zehnder modulators (MZMs) and photodetectors. By exploiting the intrinsic sinusoidal response of MZMs, our system replaces conventional quadratic functions with sine-squared modulation, enabling up to seventh-order nonlinear transformations at the input layer. Compared to Volterra-based next-generation PRC schemes, our design achieves comparable computational accuracy with significantly reduced structural complexity. Experimental results demonstrate symbol error rates of 5.56 × 10 –4 and a normalized mean square error of 0.155 on the nonlinear channel equalization (NCE) and tenth-order Nonlinear Autoregressive Moving Average (NARMA10) benchmarks, respectively, using only 16 and 22 feature dimensions. These findings underscore the potential of our architecture to simplify physical reservoir computing implementations while boosting computational efficiency and scalability for integrated photonic platforms.