Photonic In-Memory Computing Primitive for Spiking Neural Networks Using Phase-Change Materials

The recent demonstration of ultrafast photonic computing devices exhibiting optical switching of phase-change materials has piqued interest in their viability for neuromorphic computing, but scaling these standalone devices to parallel computing platforms poses a major challenge. The authors leverage the parallelism offered by wavelength-division multiplexing (WDM) to propose a photonic ``in-memory'' platform, which can be used to emulate a spiking neural network. This solution could potentially bridge the gap between isolated computing devices and large-scale implementations of neuromorphic systems.