Neuromorphic Phototransistors with Built-in Heterojunction for Efficient and Accurate Adaptive Sensing

Neuromorphic vision systems have the capacity to simulate the perception and processing of visual information by visual cells in the retina. However, when confronted with the challenges posed by a substantial volume of complex data and complex environments, traditional neuromorphic vision systems are unable to handle redundant signals of overenhancement and suppression. These systems are required to face the considerable challenges posed by complicated circuits and algorithms. In this paper, we present an adaptive synaptic transistor with a built-in heterojunction that can switch between three modes of synaptic excitation-inhibition effect, excitation-adaptive, and inhibition-adaptive photoconductivity effect by utilizing light switching and wavelength change. The device can complete the entire adaptation process from excitation sensitization to self-adaptation to the initial current in 1 s, and from excitation sensitization to adaptation in 3.2 s. The adaptation speed is superior to that of the human eye (5 min). The combination of convolutional neural networks (CNNs) with adaptive synaptic transistors has yielded the development of an advanced neuromorphic vision system. This system exhibits fast self-adaptation and static image recognition and classification capabilities, with a recognition rate that exceeds 90%, thereby facilitating the advancement of next-generation neuromorphic vision systems.