Answers are grounded in this paper and cite their sources.
Neuromorphic photonic hardware promises energy‐efficient and massively parallel computing, yet practical photonic synapses remain limited by volatility, thermal instability, and the lack of embedded learning mechanisms for continuous‐time adaptation. Here, a silicon photonic microring resonator integrated with 2D ferroelectric CuInP 2 S 6 (CIPS) is demonstrated as a compact photonic synapse with bidirectional analog programmability, multiday optical nonvolatility, intrinsic thermal stabilization, and low optical attenuation of 0.030 dB µm −1 . Electric field‐driven Cu + ion migration enables continuous resonance tuning with an efficiency of 27 pm V −1 , a resonance shift rate of 50.54 pm min −1 at +7 V, and a VπL product of 1.1 V cm. Programmed resonance states persist for days at zero bias, eliminating static power consumption. The hybrid Si–MRR/CIPS platform further suppresses silicon thermo‐optic drift by ~40%, stabilizing synaptic weights against temperature fluctuations. Using experimentally measured device dynamics, a device‐calibrated learning scheme is implemented for continuous‐time EEG seizure detection, achieving an AUC of 0.95, sensitivity of 0.92, and specificity of 0.88. These results establish Si–MRR/CIPS as a scalable platform for nonvolatile in‐memory photonic learning.
No takes yet. Share an insight, caveat, or question.
Tamalampudi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: