Excitonic neuromorphic computing in vdW heterostructures

Neuromorphic computing systems that integrate memory and processing overcome the limitations of the conventional von Neumann approach, delivering significant improvements in speed and energy efficiency for advanced artificial intelligence (AI) applications. Here, we introduce a nanoscale exciton-modulated platform with both electrical and optical control for programmable multilevel synaptic memory arrays. Our device incorporates a WSe₂/Mo_0.5W_0.5Se₂ heterobilayer interfaced with a nanoscale hBN capacitor on a gold substrate. By applying an electric field with a plasmonic gold tip, we reversibly switch each bilayer region between p- and n-type doping and modulate exciton and interlayer exciton emission. Under varying bias, individual sites exhibit stable, multilevel optical states spanning a wide dynamic range. These excitonic responses are incorporated into a neural-network simulation for temporal-pattern classification, achieving high accuracy by exploiting the intrinsic state diversity. Operating at switching rates up to 0.1 GHz, the device converts nanosecond electrical pulses directly into optical outputs, enabling low‑latency, on-chip optoelectronic communication compatible with fiber‑optic systems. By combining localized field modulation with two-dimensional heterostructures, this platform realizes programmable heterosynaptic arrays with adjustable weight resolution and sub‑10 ns reconfiguration. Our results establish a paradigm for scaling excitonic memristive networks and integrating neuromorphic-photonic hardware, guiding the development of energy-efficient, high-performance AI accelerators.