Memristors Based on Ferroelectric Cu‐Deficient Copper Indium Thiophosphate for Multilevel Storage and Neuromorphic Computing

It is essential to explore the interactions between intrinsic ferroelectricity and ionic activities in 2D ferroelectrics for theoretically understanding and experimentally modulating device performance. Due to the tendency of Cu⁺ migration in ferroelectric copper indium thiophosphate (CIPS) and formation of Cu conductive filaments, herein, Cu-deficient CIPS (CIPS^*) is employed to investigate resistive switching (RS). Different from CIPS with controllable threshold switching and write-once read-many-times (WORM) behaviors, CIPS^* shows stable non-volatile digital and analog RS behaviors by controlling the operation voltage. Owing to the formation of non-stoichiometric In_4/3P₂S₆ (IPS) with metallic phase at the low-resistance state, the fabricated memristors demonstrate high ON/OFF ratio up to 5 × 10⁵ and high endurance stability (>2000 cycles), which can be utilized to implement multilevel storage. And more intriguing, amplitude-dependent and polarity-independent long-term potentiation and depression can be simulated based on the analog memristors. Artificial neural network based on CIPS^* synaptic memristors can realize handwritten digit recognition with the accuracy of 91.15%. Even after considering the cycle-to-cycle and device-to-device variations of the synaptic functions, the accuracy remains as high as 90.71%. Such investigations pave the way toward highly reliable memristors base on 2D ferroelectrics with potential applications in multilevel storage and neuromorphic computing.