Design of an 8-bit non-volatile plasmonic memory for synaptic weighting in optical neuromorphic architectures

In this paper, a non-volatile plasmonic memory based on a metal–insulator-metal (MIM) waveguide and phase-change material is proposed to enhance storage density and integration compatibility in photonic integrated circuits (PICs). Silver is utilized as the metal, with its dielectric function characterized by the Drude model, while air and G e 2 S b 2 T e 5 (GST) serve as the dielectric media. Two-dimensional Finite-Difference Time-Domain (2D-FDTD) simulations demonstrate an operating resonance at λ = 1814 nm, exhibiting an optical contrast of 87 % between phase states, an extinction ratio of ER = 44.04 dB, and an ultra-low insertion loss of 0.60 dB for logic state ’1′ (45.60 dB for logic ’0′). The cell features all-optical external programmability, non-destructive readout, and ultrafast temporal dynamics in the femtosecond regime ( τ r = 62 fs). By mapping GST phase transitions to quantized optical transmission levels, the proposed architecture functions effectively as a photonic weighting element in optical neural networks to alleviate the von Neumann bottleneck. Generalizing the unit cell yields scalable multi-bit configurations up to 8 bits (256 states), achieving an ultra-high storage density of 4.56 bits/ μ m 2 within a compact footprint of 0.219 μ m 2 . Furthermore, a 5-stage BEOL CMOS-compatible fabrication workflow (≤200 °C) is presented, and structural sensitivity analysis confirms high fabrication tolerance (ΔER} < 1 dB within a ± 5 nm window).