Artificial neural network-based resistive switching devices are a better alternative to the von Neumann bottleneck of emerging memory devices because of the integrated data and processing unit with lower energy consumption. Thus, it becomes essential to develop materials and devices that support these developments simultaneously. This work addresses an analog, cost-efficient, and asymmetric (CZTS = Cu2ZnSnS4) resistive random-access memory (RRAM) device for an artificial neural network application that mimics the human brain. Cu/CZTS/ITO RRAM device shows excellent analog resistive switching for 103 current–voltage cycles, resistance state retention characteristics for 5 × 103 s with RHRS/RLRS ∼ 3, and endurance for 1.2 × 103 cycles. The drift of top electrode-Cu ions under the high electric field is attributed to the resistive switching mechanism in this device. The neuromorphic evaluation on the same device shows significant characteristics such as synaptic weight, pulsed-pair facilitation for 10 ms pulse duration, Hebb learning for 2.5 s, and transition from short-term memory to long-term memory. The recalling probability of the device is ∼80% with a pulse width of 10 ms. Further, the Cu/CZTS/ITO RRAM device can distinguish different lights with different intensities, similar to the human brain. The device shows enhancement in conductance for light irradiation of 10 and 20 mW/cm2 with respect to dark conductance for all blue, green, and orange lights. These findings allow for a better understanding of the CZTS-based bioinspired synaptic device.
Yadav et al. (Thu,) studied this question.