Answers are grounded in this paper and cite their sources.
Three‐factor learning, in which synaptic weight updates are regulated by local activity and an additional modulatory signal, provides a powerful framework for adaptive neuromorphic computing but remains difficult to implement in compact hardware. Here, a fully van der Waals (vdW) double‐floating‐gate (DFG) memory based on an α‐In 2 Se 3 /hexagonal boron nitride (h‐BN)/multilayer graphene (MLG)/h‐BN/MLG heterostructure is demonstrated. The coupled floating gates provide an additional electrostatic degree of freedom for state‐dependent conductance modulation. The device exhibits a threshold‐voltage shift exceeding 65 V, an on/off current ratio approaching 10 6 , and 48.7% and 55.1% reductions in LTP and LTD nonlinearity, respectively. Over the single‐floating‐gate counterpart, enabling ∼92.5% recognition accuracy on the Modified National Institute of Standards and Technology (MNIST) handwritten‐digit dataset. Gate‐programmable broadband optoelectronic responses extend to the 1310 nm optical communication band. Under 1310 nm illumination, persistent conductance changes are jointly governed by the initial conductance state, optical input, and gate‐programmed modulation of the learning rate, thereby realizing a hardware‐level analog of three‐factor learning. These results highlight the potential of this multifunctional vdW memory for adaptive optoelectronic in‐memory computing.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: