Plasmonic photonic synapse with spectral bipolar weight control and ultra-low energy consumption using GST material

Abstract

Neuromorphic photonic systems have emerged as a promising platform for overcoming the bandwidth and energy limitations of conventional electronic neural networks, where the realization of compact, low-power, and non-volatile synaptic elements remains a key challenge. In this work, an all-optical plasmonic nano-synapse is proposed for energy-efficient neuromorphic photonic architectures. The device is based on a nanoscale plasmonic ring resonator orthogonally coupled to two independent waveguides, enabling physically separated data transmission and optical control. A nanometric Ge₂Sb₂Te₅ (GST) phase-change layer is integrated at the coupling region between the resonator and the data waveguide, allowing non-volatile, multi-level modulation of synaptic weights through resonance wavelength tuning. Unlike conventional photonic synapses that primarily support unipolar weight modulation, the proposed architecture enables bipolar synaptic weight control within a single compact device by operating on opposite sides of the resonance wavelength, corresponding to excitatory and inhibitory synaptic responses. This spectral-domain weight encoding provides a robust alternative to time-domain spiking schemes and enables spike-timing-dependent plasticity (STDP)-compatible weight updates while significantly reducing circuit complexity. Numerical simulations demonstrate an ultra-compact footprint of approximately 0.09 μm², an active GST volume below 5 × 10⁻⁵ μm³, and an ultra-low optical switching energy of about 5 pJ. The use of a low-loss silver-based plasmonic platform combined with air cladding results in strong field confinement and an optical contrast approaching 99%. Owing to its compact size, non-volatility, and spectral bipolar weight capability, the proposed plasmonic synapse offers a scalable and practical building block for high-density, low-power neuromorphic photonic processors.