Robust Nonvolatile Memory and Synaptic Emulation via Dual-Mode Modulation in ReS2/h-BN/Ta2NiSe5 Floating-Gate Device.

The integration of optical sensing with neuromorphic computing offers a promising pathway to overcome the von Neumann bottleneck in data-intensive artificial intelligence applications. However, realizing robust nonvolatile storage with synergistic electrical and optical modulation in 2D devices remains a challenge. Here, we demonstrate a high-performance floating-gate memory based on a ReS2/h-BN/Ta2NiSe5 van der Waals heterostructure. Under precise photo-electrical cooperative modulation, the device exhibits excellent nonvolatile memory characteristics, including a large memory window of 96.2 V, long retention exceeding 104 s, and stable endurance over 1000 cycles. Additionally, it features multi-bit storage capabilities enabled by optical erasure and electrical writing. Beyond digital storage, the device effectively mimics biological synaptic plasticity, such as paired-pulse facilitation and learning processes. Validating its potential for neuromorphic vision systems, a simulated three-layer artificial neural network achieves a high recognition accuracy of 96.43% on the MNIST dataset. This work establishes a promising paradigm for future integrated sensing, storage, and computation applications based on 2D materials.