Ultra-compact non-volatile GSST-MIM plasmonic memory with in-circuit optical programming as a binary weighting element for neuromorphic networks

This paper presents a numerical investigation of an ultra-compact, non-volatile plasmonic memory cell based on the low-loss Ge 2 Sb 2 Se 4 Te 1 (GSST) phase-change material platform integrated within a metal-insulator-metal (MIM) waveguide cavity. The optical dispersion of the silver (Ag) metallic substrate is modeled using the Drude-Lorentz formulation, with air and GSST serving as the dielectric media. Numerical simulations conducted via the two-dimensional Finite-Difference Time-Domain (2D-FDTD) method demonstrate that at the operating resonance wavelength of 1362 nm, the proposed architecture yields an optical readout contrast of 73% between the amorphous and crystalline states, alongside a high extinction ratio (ER) of 23.24 dB. The signal attenuation during state read and write operations is rigorously evaluated: during the State '1' write and read operations, the insertion loss (IL) is 4.15 dB and 3.7 dB, respectively, while for the high-attenuation State '0', the writing and reading insertion losses reach 24 dB and 18.18 dB. Boasting an ultra-compact footprint of 0.004 and an integration density of 250 bits/ , this architecture addresses operational alignment risks of external lasers by incorporating an auxiliary 90°-bent MIM control waveguide for in-circuit optical pulse routing. Transient time-domain response analysis confirms that readout discrimination relies on ultrafast optical propagation delay ( = 7 fs), distinctly decoupled from the thermal phase-transition kinetics. Given these quantitative characteristics, the memory cell offers strong potential application as a reconfigurable binary optical weighting element in photonic synapses for integrated neuromorphic computing.