Neuromorphic Photonics Based on Phase Change Materials

Abstract

Neuromorphic photonics devices based on phase change materials (PCMs) and silicon photonics technology have emerged as promising solutions for addressing the limitations of traditional spiking neural networks in terms of scalability, response delay, and energy consumption. In this review, we provide a comprehensive analysis of various PCMs used in neuromorphic devices, comparing their optical properties and discussing their applications. We explore materials such as GST (Ge₂Sb₂Te₅), GeTe-Sb₂Te₃, GSST (Ge₂Sb₂Se₄Te₁), Sb₂S₃/Sb₂Se₃, Sc_0.2Sb₂Te₃ (SST), and In₂Se₃, highlighting their advantages and challenges in terms of erasure power consumption, response rate, material lifetime, and on-chip insertion loss. By investigating the integration of different PCMs with silicon-based optoelectronics, this review aims to identify potential breakthroughs in computational performance and scalability of photonic spiking neural networks. Further research and development are essential to optimize these materials and overcome their limitations, paving the way for more efficient and high-performance photonic neuromorphic devices in artificial intelligence and high-performance computing applications.

Keywords: neuromorphic photonics; phase change materials; silicon photonics.

Figure 1

Nonlinear model of a neuron. The neuron is composed of a set of synapses (connecting links); an adder or linear combiner (performing a weighted sum of signals); and a nonlinear activation function [8].

Figure 2

Neuromorphic photonic devices utilizing PCMs and integrated waveguides. (a) A 3D schematic representation of the hybrid waveguide. (b) Simulated electric field profiles for the fundamental quasi-TE mode of the switching unit at 1550 nm in the crystalline state (n_eff = 2.68 − 0.05i), and (c) the amorphous state (n_eff = 2.60 − 1.06 × 10⁻³ i) [19]. p++: p-type heavily doped region, n++: n-type heavily doped region, cGST: crystalline GST, aGST: amorphous GST.

Figure 3

Photonic memory using optical phase change materials. (a) The schematic diagram of integrated on-chip memory. Information is stored in the phase state of the GST section atop the waveguide. Ultrashort optical pulses enable both reading and writing of the memory, as the guided light interacts with the GST through its evanescent field. During readout, data are encoded in the amount of optical transmission through the waveguide, as the two crystallographic states of GST display a high contrast in optical absorption [13]. (b) The multi-bit, multi-wavelength architecture realized by three different microrings coupled with the same waveguide. Using optical pulses close to resonance, each cell could be addressed selectively [13]. (c) A single memory cell within the 16 × 16 photonic matrix memory [22] (d) Schematic of multiplexed all-optical matrix-vector multiplication [23].

Figure 4

Neuromorphic devices based on GST. (a) Schematic of the integrated photonic synapse, which emulates the function of a neural synapse. The synapse comprises a tapered waveguide (dark blue) with discrete PCM islands on top, optically connecting the presynaptic (pre-neuron) and postsynaptic (post-neuron) signals [24]. (b) Schematic of a bipolar integrate-and-fire neuron based on GST-embedded ring resonator devices, illustrating the integration and firing unit [25]. (c) All-optical spiking neuronal circuits: Input spikes are weighted using PCM cells and summed up with a wavelength division multiplexer. When the integrated power of the postsynaptic spikes exceeds a certain threshold, the PCM cell on the ring resonator switches, generating an output pulse (neuronal spike) [26]. (d) The complete programmable metasurface mode converter device consists of an encapsulated GST phase gradient metasurface (red box) and a mode selector (yellow box) [27].

Figure 5

Phase change materials for in-memory computing. (a) Sketch of a waveguide crossing array illustrating the two-pulse addressing of individual phase-change cells. Only overlapping pulses provide sufficient power to switch the desired PCM cell [29]. (b) Optical micrograph of a studied crossed-waveguide photonic array [29]. (c) Schematic of the multiplication of two scalars (a and b), encoded in the device transmittance T and the energy of the read pulse P_in [30]. (d) The low-energy read pulse P_in, which propagates through the device without inducing phase change, is measured at the output with an amplitude modulated by the transmittance T [30].

Figure 6

Neuromorphic devices based on variety phase change materials. (a) Non-volatile integrated photonic switches composed by a Ge₂Sb₂Se₄Te₁ strip on top of the SiN waveguide [32]. (b) Graphene-assisted phase shifter based on Sb₂Se₃ in a microring [35]. (c) All-chalcogenide optical perceptron model composed of Sb₂S₃ and GST [36]. (d) Schematic diagram of integrated photonic memory device based on Sc_0.2Sb₂Te₃ [37].

Figure 7

Nonvolatile all-optical memory in epitaxial In₂Se₃–silicon microring resonators. (a) Normalized transmission spectra for the hybrid resonator with α- (black), β- (red), and retrieved α-state (dark red) In₂Se₃. The dots represent experimental data, and the curves are coupled-mode-theory fittings. (b) Scanning electron microscope image of the as-prepared molecular beam epitaxy (MBE) film transferred onto a silicon photonic single-mode waveguide. (c) Optical microscopy image of an In₂Se₃–Si microring device. (d) Atomic structure of α-In₂Se₃ at room temperature (semitransparent) and after thermally activated shear-glide (solid color). The outer Se atoms fall into the interstitial sites, leading to compression of the interlayer distance. (e) Subsequent intra-quintuple-layer (QL) rearrangement results in the β-In₂Se₃ structure. For clarity, only the middle two QLs are included here.