Neuromorphic Computing and High Accuracy (>94%) Pattern Recognition with Cu ₂ SrSnS ₄ -Based RRAM Devices

In the present era of artificial intelligence, devices for smart and fast data computation are crucial for reducing vast amounts of power consumption. The existing von Neumann-based memory devices have limitations due to the discrete storage and computing units; in contrast, RRAM-based memory devices are under consideration as they can potentially surmount the von Neumann bottleneck. In the present work, Cu2SrSnS4 (CSTS)-based RRAM devices, in a Cu/CSTS/ITO configuration, are investigated for their potential to mimic the biological human brain for neuromorphic computing applications. The RRAM device exhibits stable resistive switching for 103 current–voltage cycles and retains the resistive state for more than 9.5 × 104 seconds. Furthermore, the neuromorphic characteristics, such as synaptic weight, PPF, STDP, and LTP/LTD, have also been successfully investigated together with the impact of humidity (85%) and different temperatures (300–360 K) on device characteristics. The Cu/CSTS/ITO RRAM device exhibits a good postsynaptic current (PSC) under a voltage pulse, demonstrating its capability for analog/digital coding and potential suitability for future spiking neural network (SNN) applications. The devices also exhibited excellent pattern recognition performance for the MNIST data set, with a synaptic device-based accuracy of over 94%, demonstrating the potential of such RRAM devices for neuromorphic computing.