A Gate-Tunable Thermal Persistent Photocurrent Device for In-Sensor Spiking Neural Networks

Persistent photocurrent is widely observed in van der Waals (vdW) heterostructures and is often attributed to trap-assisted photogating, yet its microscopic origin remains unclear. Here, we clarify the mechanism in a gate-tunable MoS₂/black phosphorus (BP) p-n heterojunction by combining DC and lock-in measurements with time-resolved decay. We simultaneously measure the general photocurrent (DC difference between illuminated and dark currents) and the net photocurrent extracted by the lock-in detection of a modulated laser. The net photocurrent is small and short-lived, whereas the DC photocurrent shows decay lifetimes (τ) exceeding hundreds of seconds at negative back-gate voltage (V_g) but collapses rapidly for positive V_g or under reverse bias. This τ-V_g and bias dependence is incompatible with a purely trap-dominated picture. We show that the long-lived response is dominated by majority-carrier recombination-induced self-heating in the forward-biased p-n junction, which drives thermoelectric and bolometric currents. By tuning the heterostructure between a p-n and an n-n configuration, the back gate effectively switches this thermal channel on and off. Using the experimentally extracted τ(V_g) in a leaky integrate-and-fire model, we further demonstrate an in-sensor spiking neural network with 91.95% accuracy on MNIST, highlighting the potential of thermally engineered persistent-photocurrent devices for neuromorphic vision.