作者：Zhong Chuyu, Hu Xiaoyong, Lin Hongtao；等
来源：NATURE COMMUNICATIONS 卷:14 期:1 页(号):6939 出版时间:OCT 2023
Artificial neural networks rely on nonlinear activation functions to solve complicated tasks. A novel graphene/silicon heterojunction device that allows reconfigurable optical activation functionalities with low-power consumption, low-power threshold, and low-time delay is developed. The nonlinear functions generated from the heterojunction device were applied for digital complex-valued ONNs to finish handwritten letters and image recognition tasks.
