Polarization-controlled memristive switching in semiconductor–ferroelectric heterostructures for neuromorphic computing

Abstract

Artificial synapses are essential hardware components for realizing energy-efficient, brain-inspired neuromorphic computing systems capable of overcoming the von Neumann memory bottleneck through computation-in-memory architectures. Here, we report a polarization-controlled memristive device based on semiconductor-ferroelectric heterojunction, in which MoS₂ nanosheets are incorporated into the PVDF matrix to engineer the interfacial electronic structure and modulate carrier transport. The remnant polarization of the PVDF layer generates an internal electrostatic field that dynamically controls interfacial charge trapping and detrapping, enabling polarization-assisted resistive switching in MoS₂/PVDF heterostructures. The device successfully emulates essential biological synaptic behaviors, including long-term potentiation, long-term depression, and paired-pulse facilitation, exhibiting highly stable and repeatable potentiation–depression characteristics over multiple cycles. The coupling between ferroelectric polarization and interfacial electronic transport leads to enhanced switching stability and reliable learning capability. The device exhibits a low areal programming energy of approximately 28.8 nJ mm⁻² per synaptic spike and demonstrates pulse-width-dependent conductance modulation. Leveraging these synaptic properties, handwritten pattern recognition using an artificial neural network (ANN) achieves a learning accuracy of 92.82%. Furthermore, system-level evaluation using a simulated spiking neural network (SNN), where experimentally derived conductance states are mapped as synaptic weights, yields a classification accuracy of 96% on the Modified National Institute of Standards and Technology (MNIST) dataset, confirming reliable learning capability and efficient neuromorphic computation. These results establish the heterostructures as an attractive platform for polarization-engineered memristive synapses and highlight their strong potential for energy-efficient neuromorphic hardware enabling brain-inspired computing.