Artificial Synapses Based on Lead-Free Perovskite Floating-Gate Organic Field-Effect Transistors for Supervised and Unsupervised Learning

Abstract

Synaptic devices are expected to overcome von Neumann's bottleneck and served as one of the foundations for future neuromorphic computing. Lead halide perovskites are considered as promising photoactive materials but limited by the toxicity of lead. Herein, lead-free perovskite CsBi₃I₁₀ is utilized as a photoactive material to fabricate organic synaptic transistors with a floating-gate structure for the first time. The devices can maintain the <i>I</i>_light/<i>I</i>_dark ratio of 10³ for 4 h and have excellent stability within the 30 days test even without encapsulation. Synaptic functions are successfully simulated. Notably, by combining the decent charge transport property of the organic semiconductor and the excellent photoelectronic property of CsBi₃I₁₀, synaptic performance can be realized even with an operating voltage as low as -0.01 V, which is rare among floating-gate synaptic transistors. Furthermore, artificial neural networks are constructed. We propose a new method that can simulate the synaptic weight value in multiple digit form to achieve complete gradient descent. The image recognition test exhibits thrilling recognition accuracy for both supervised (91%) and unsupervised (81%) classifications. These results demonstrate the great potential of floating-gate organic synaptic transistors in neuromorphic computing.