Adv Sci (Weinh). 2025 Sep 14:e10063. doi: 10.1002/advs.202510063. Online ahead of print.
ABSTRACT
Artificial intelligence (AI) is constrained by the high energy consumption of von Neumann architectures and the limited scalability of traditional silicon-based synapses. Two-dimensional (2D) van der Waals (vdW) materials, with their atomic-scale thickness, tunable electronic properties, and ease of heterogeneous integration, offer a promising platform for next-generation neuromorphic hardware. Here, the authors report a vdW floating-gate transistor (BP/POx/WSe2) with a high on-off current ratio (≈105) and a large memory window (73 V), benefitting from the optimized interface band alignment via 2D heterostructure engineering. Key synaptic functionalities are demonstrated, including short-term plasticity (STP), long-term plasticity (LTP), and electro-optical dependent plasticity, short-term paired-pulse facilitation (PPF), and long-term potentiation/depression (LTP/D). Notably, the device mimics human visual memory under optical stimuli while achieving ultralow energy consumption (10 pJ per synaptic event), outperforming most reported photoelectronic synaptic devices. Furthermore, a two-path convolutional neural network (CNN) is introduced that synergistically merges optical and electronic inputs, which enables efficient feature extraction and weight updating, and achieves 96.9% accuracy in the Labeled Faces in the Wild (LFW) face recognition task. The work presents a promising approach for neuromorphic electronics, paving the way for energy-efficient vision processing in edge AI applications.
PMID:40946193 | DOI:10.1002/advs.202510063
Recent Comments