High-performance all-optical photonic crystal synapse based on Mach-Zehnder interferometer and directional coupler utilizing GSST phase-change material

• A novel photonic crystal architecture enables efficient optical neuromorphic synapses. • Significant miniaturization and high integration density improve optical neuromorphic systems. • The embedded cavity in the MZI ensures single-mode signal transmission at the output. • GSST-PCM rods enable dynamic, phase-state-dependent modulation of signal transmission. • The proposed structure offers potential for next-generation optical neural networks. This paper presents a novel architecture for all-optical photonic crystals, leveraging the integration of Mach-Zehnder interferometers and directional couplers for advanced optical neuromorphic synapses. By utilizing photonic crystals and germanium-antimony-selenium-tellurium (GSST) phase-change materials, we achieve precise control over optical transmission. The use of photonic crystals enables a compact footprint and significantly reduces the device size compared to conventional silicon photonics, offering a key advantage in achieving higher integration density for optical neuromorphic systems. Comprehensive finite-difference time-domain (FDTD) simulations demonstrate that incorporating a photonic crystal cavity at the input port significantly enhances single-mode operation, leading to an output signal transmission of more than 99 %. Furthermore, variations in the crystallinity fraction of phase-change material (PCM) rods significantly influence the output signal transmission, enabling precise control of the signal dynamics. Under amorphous and fully crystalline conditions of the GSST-PCM rods, the signal transmission rate varies between −0.02 dB and −13.5 dB, highlighting the profound impact of phase-state changes on system performance. This innovative photonic crystal platform offers a promising avenue for the realization of next-generation optical synapses, paving the way for advanced optical neural networks and machine learning.