Homogeneous integration of two-dimensional material-based optoelectronic neurons and ferroelectric synapses for neuromorphic vision

Dynamic vision processing at the edge requires in-sensor spiking neural networks (SNNs) to achieve high energy efficiency and rapid processing. Although optoelectronic leaky integrate-and-fire (LIF) neurons are essential for optical sensing and sparse coding, their practical utility has been hindered by incomplete emulation of biological behaviors and integration difficulties with synaptic devices. Here, we show an optoelectronic LIF neuron based on a MoS2 phototransistor that reproduces key neuronal features, including multispectral sensing, capacitor-less integration, and threshold-triggered spiking. This neuron supports complementary rate and time-to-first-spike coding, enabling versatile visual information processing at the hardware level. Furthermore, we achieve the homogeneous integration of these neurons with MoS2 ferroelectric synapses on a single substrate, unifying volatile optical encoding with non-volatile weight storage. The integrated SNN system attains recognition accuracies of 91.7% for color recognition and 93.5% for object detection, indicating its potential for scalable, high-performance next-generation neuromorphic vision systems. Integrating volatile optical sensing with non-volatile memory is crucial for neuromorphic vision applications. Wang et al. propose a homogeneous integration scheme that combines optoelectronic neurons and ferroelectric synapses on a single substrate for color recognition and object detection tasks.