CdS/CZTSSe heterojunction synaptic memristor: Enabling efficient handwritten digit recognition

Abstract

Owing to the outstanding performance of memristors in brain-like parallel computing and data processing, especially in the efficient recognition of handwritten digits and complex patterns, they are regarded as key components for next-generation artificial intelligence systems. This study developed a memristor based on the P–N heterostructure CdS/Cu2ZnSn(S,Se)4 (CdS/CZTSSe). The results indicate that the Ag/CdS/CZTSSe/Mo memristor exhibits stable non-volatile bipolar resistive switching. By investigating the conductivity mechanism of the device, a resistive switching model is established that regulates the conductive filaments of Cu ions in the heterojunction. This device not only demonstrates a concentrated Set/Reset voltage distribution, good durability (>200 cycles), and time retention characteristics (>104 s) but also features continuously adjustable conductance under electrical pulse square wave stimulation. Such a behavior enables the memristor to simulate important biological synaptic functions, including excitatory postsynaptic current, excitatory and inhibitory synaptic plasticity, and short-term/long-term plasticity and paired-pulse facilitation. Furthermore, neuromorphic simulations validated that the artificial neural network model using this memristor achieved a 94.1% recognition rate for Modified National Institute of Standards and Technology handwritten digits. These results significantly advance the development of heterojunction memristors in artificial synapses and lay a foundation for future neuromorphic applications.