Voltage-tunable plasmonic metamaterial networks for neuromorphic optical computing at telecommunication wavelengths

Abstract

We propose a modeling framework for voltage-tunable plasmonic neuromorphic processors operating at telecommunication wavelengths. The architecture combines split-ring resonator (SRR) artificial synapses with InGaAsP/InP photodetector neurons, validated through multiphysics simulations and preliminary experimental characterization. Device-level results demonstrate SRR quality factors of 28 and photodetector bandwidths of 18 GHz. Electrostatic gap tuning (±15 nm) enables synaptic weight modulation with measured transmission changes of ±12 dB. System-level simulations project neuron densities of 10⁶ cm⁻², MNIST accuracy of 99.7% in software simulations of the idealized network, and inference latencies of 8.5 ns, with training performed offline at sub-MHz rates. The theoretically estimated energy cost per synaptic update is 0.75 fJ under quasistatic assumptions. Practical implementation requires addressing optical I/O interfaces, waveguide propagation losses, thermal management, and device-to-device fabrication variability before reaching the projected performance levels. While full network-level experimental validation remains future work, the demonstrated component performance and validated simulation methodology establish clear design guidelines for scalable plasmonic neuromorphic hardware. This work positions voltage-tunable plasmonic metamaterials as a promising candidate for energy-efficient photonic neural accelerators.