Benchmarking QCi's EmuCore Device for Time Series Forecasting

Abstract

Reservoir computing (RC) has emerged as an efficient and scalable approach for processing time-dependent data, particularly in physical reservoir computing implementations. QCi's EmuCore platform, a novel photonic-inspired time-delayed reservoir system, offers a compact and energy-efficient solution for dynamic data transformations. In this study, we perform a comprehensive benchmark of EmuCore using the Mackey-Glass blood cell production model, a challenging nonlinear time series forecasting task. EmuCore's performance is compared against classical reservoir computing, RNN, and LSTM models implemented on both Apple M2 CPUs and Nvidia Jetson Nano GPUs.

Our evaluation spans three critical dimensions: runtime efficiency, forecasting accuracy, and power consumption. While EmuCore achieves competitive prediction accuracy with a mean absolute percentage error (MAPE) below 2.5\%, it significantly reduces training runtimes compared to LSTM and RNN models. Power measurements reveal EmuCore's advantages in edge computing applications, though opportunities for further energy optimizations remain. These findings underscore EmuCore's potential as a powerful and efficient solution for real-time, resource-constrained temporal pattern recognition tasks, particularly in automotive systems.