Energy-Efficient Neuromorphic Cores Using 2d Material Synapses

Abstract

ABSTRACT: Neuromorphic computing demands devices that mimic synaptic plasticity while operating at ultralow energy levels. This paper investigates energy-efficient neuromorphic VLSI cores that employ 2D-material-based synapses such as MoS?, WS?, and h-BN heterostructures. The architectures leverage chargetrapping dynamics and low-voltage switching characteristics to implement biologically plausible learning rules. An integrated array of 64×64 synapses demonstrates analog weight modulation with picojoule programming energies, while the associated neuron circuits maintain stable operation under device variability. System-level evaluations on edge-AI benchmarks show up to 55% energy savings compared to SRAM-based ANN accelerators. The proposed design also includes a reliability model addressing endurance degradation and layer-interface defects. Overall, 2D synaptic materials show exceptional potential for creating scalable, low-power neuromorphic platforms for sensoredge intelligence.

KEYWORDS: Neuromorphic VLSI, 2D materials, Synaptic devices, Edge computing, Low power