26th Australian Institute of Physics Congress and the Australian and New Zealand Conference on Optics and Photonics

Hybrid photonic synapse for on-chip neuromorphic hardware

Not scheduled

20m

Belinda Hutchinson Building (The University of Sydney )

Contributed Oral ANZOS | Photonics and Optics (ANZCOP)

Description

The rapidly increasing data-intensive AI workloads and large-scale data centers are driving a fundamental shift toward energy-efficient, scalable, and compact computational hardware design. Neuromorphic computing addresses this need by integrating neurobiological principles with physical hardware to enable fast, low-power, and low-latency processing.
Photonic platform is well-suited for neuromorphic hardware, offering parallel, high-speed, and low-energy data processing with sub-nanosecond latency married with inherently broad optical bandwidth. Integrating phase-change materials (PCMs) into nanophotonic devices enables a unique on-chip system for neuromorphic architectures.

We developed an on-chip hybrid photonic artificial synapse based on reconfigurable broadband nanophotonic switch with an integrated PCM (Ge2Sb2Te5 (GST)) memory unit assisted with tailored nanocrystalline graphene microheaters. One of key element of our nanophotonic device (full in-house fabrication yield >95%) is the specially-engineered, transparent, and cavity-compatible 2D nanocrystalline graphene microheaters in contrast to the standard absorptive metallic electrodes.

We experimentally demonstrated dual-mode short- and long-term synaptic plasticity triggered optically (via evanescent coupling) as well as electrically (via microheaters), presenting coexistence volatile and non-volatile memory dynamics of the developed artificial synapse on-chip, achieving optical weight contrasts up to 75% and 50%, respectively. Crucially, further we achieved hybrid switching of PCM enabling reconfiguration of synaptic weight through both optical and electrical stimuli in a synergistic manner.
Importantly, we experimentally realized an unsupervised learning rule with unique bio-plausible feature of our photonic artificial synapse, enabled by, hybrid PCM switching dynamics governed directly by the intrinsic device physics — a behavior of photonic synapse demonstrated, to the best of our knowledge, for the first time.
The established on-chip bio-realistic hybrid photonic synapse exhibits dual-mode plasticity with reconfigurable and continuously tunable weights. It inherently supports an unsupervised learning rule with bio-plausible unique self-limiting mechanism, enabling stable weight adaptation. The multifunctionality of developed synapse paves the way for scalable photonic neuromorphic networks.