Reproducible Reservoir Computing with Thermally Driven Superparamagnets: Controlling Temperature Sensitivity

This paper simulates how ambient temperature changes affect the magnetization dynamics and task performance of superparamagnetic nanodot ensembles used as reservoir-computing substrates. The authors study heterogeneous patterns containing nanodots of different sizes and thermal-activation timescales as a mitigation strategy. On the NARMA-10 benchmark, optimized heterogeneity stabilizes reservoir performance across 5–35°C while causing only a small loss in ultimate performance. The study also examines the trade-off between peak task performance and temperature robustness through reservoir hyperparameters. The results are simulation-based and target a practical limitation of thermally activated, ultra-low-energy unconventional computing devices.