Optoelectronic Synergy and Multimodal Neuromorphic Computing in SnS 2 /h‐BN/NbOCl 2 Van Der Waals Floating Gate Architecture

ABSTRACT The kinetic basis by which a single charge‐storage layer supports both short‐term plasticity (STP) and persistent long‐term plasticity (LTP) remains unresolved. Using a SnS 2 /h‐BN/NbOCl 2 van der Waals floating‐gate architecture, we provide three complementary lines of evidence. Removing NbOCl 2 reduces the memory window from 92.2 to ∼2 V, identifying NbOCl 2 as the dominant charge‐storage medium. Variable‐temperature post‐stimulus relaxation (310–330 K), analyzed by biexponential fitting and Arrhenius regression, resolves two kinetically distinct components: a weakly activated component (WAC, E a,1 ≈ 0.026 eV) with fast relaxation, consistent with transient STP conductance modulation, and a thermally activated component (TAC, E a,2 ≈ 0.111 eV) with much slower relaxation, consistent with persistent LTP retention. Their approximately 4‐fold effective activation‐energy difference provides quantitative kinetic evidence for two distinct temperature‐dependent relaxation behaviors associated with the same NbOCl 2 layer. Sigmoidal fitting of the memory index yields a crossover power density P c = 26.49 µW/cm 2 , separating a predominantly volatile response at P < P c (Path A) from a larger persistent contribution at P > P c (Path B). Transient electrostatic screening and tunneling‐assisted charge neutralization are proposed as the underlying processes. The device achieves 92.81% MNIST and 93.2% spoken‐digit recognition accuracy, demonstrating promise for multimodal neuromorphic computation.