Neuromorphic devices based on carrier relaxation have demonstrated significant potential in simplifying the integration of sensory-computing neural network chips. However, typical relaxation designs rely on defect sustained-release mechanisms and interaction with bulk channel carriers, which compromise sensing performance and hinder the effective coordination of gains between computing and sensing functions. Herein, the limitation is addressed by confining the interaction between defects and carriers to a 2D interface involving Tellurium (Te) and the Hafnium Dioxide (HfO2)/hexagonal Boron Nitride (h-BN). This approach enables high-fidelity perception and computation of optical communication timing sequences through precise defect modulation to allocate modal weights. The device exhibits competitive photo-electric conversion performance for communication digital signals under zero gate bias. More notably, under applied gate bias, the device can function as an optical synapse with a rise time 𝝉rise of 1.04 ms, the synaptic response processes can be directly utilized as a sequence detector rather than merely contributing to the final conductance. By simultaneously performing sensing and computing functionalities within a single device, high-precision timing recovery is achieved for optical communication using a spiking neural network (SNN) algorithm. This work presents a pioneering verification and offers a potential high-performance architecture reference for compact clock-free optical communication modules.
Loading....