Monolithic 2D Synapse‐Neuron Integration via Defect‐Engineered h‐BN for Energy‐Efficient Neuromorphic Computing

ABSTRACT Neuromorphic computing offers a route to overcome the von Neumann bottleneck by emulating coupled synaptic and neuronal functions. However, monolithic integration of these components remains challenging due to mismatches in material platforms, operating mechanisms, and device architectures. Here, we exploit defect‐engineered hexagonal boron nitride (h‐BN) to realize a monolithic synapse–neuron platform within a unified 2D materials system. An optoelectronic synapse based on a MoS 2 /h‐BN heterostructure and a filamentary neuron formed in an Ag/h‐BN/graphite stack are spatially separated yet electrically coupled through a shared h‐BN dielectric. Under ultraviolet illumination and gate bias, defects in h‐BN and at the h‐BN/SiO 2 interface are activated, enabling tunable, reversible, and retained multilevel conductance that emulates both short‐ and long‐term plasticity. The synapse supports 8‐bit‐equivalent accessible conductance updating and 6‐bit retained state separability, enabling CIFAR‐10 inference comparable to software baselines. Volatile h‐BN filaments provide threshold switching and leaky integrate‐and‐fire dynamics. In a one‐neuron–one‐synapse configuration, gate‐tunable current compliance and intrinsic neuronal thresholding suppress redundant low‐weight synaptic signals, reducing the estimated synaptic read‐current contribution by >80% with negligible accuracy degradation. This work establishes a proof‐of‐concept shared‐h‐BN platform for physically coupled synaptic storage and neuronal activation, providing a device‐level route toward energy‐efficient neuromorphic hardware.