Dai Shilei, Wang Dingchen, Liu Xu, Cui Binbin, Tian Xinyu, Liu Dingyao, Hu Huawei, Guo Pu, Sun Tongrui, Zhang Junyao, Huang Jia, Zhang Shiming, Wang Zhongrui
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong 999077, China.
School of Microelectronics, Southern University of Science and Technology, Shenzhen 518055, China.
Natl Sci Rev. 2025 Jun 4;12(8):nwaf224. doi: 10.1093/nsr/nwaf224. eCollection 2025 Aug.
The scale-up of AI models and data makes smart edge devices (e.g. smartphones and AR/VR glasses) face increasing challenges in energy consumption, latency and hardware costs. These challenges are due to their physically separated memory, processing and display units. Collocating the functions of these units into a single device for in-display neuromorphic computing (IDNC) shows great promise in overcoming these challenges. Herein, we report an all-in-one electrically programmable IDNC (EP-IDNC) device that leverages the electrochromism of organic semiconductors. The electrochromism empowers the EP-IDNC device with synaptic and neural behaviors (e.g. multi-terminal operability, multi-level weight update, spatiotemporal information encoding), as well as the capability to visually display computing results through color changes. The IDNC functions are validated on tasks such as noise reduction and motion object perception using a prototype 3 × 3 EP-IDNC device array. We ultimately showcase the integration of neuromorphic and display functions for car steering reminders.
人工智能模型和数据的扩展使智能边缘设备(如智能手机和AR/VR眼镜)在能耗、延迟和硬件成本上面临越来越大的挑战。这些挑战源于它们物理上分离的内存、处理和显示单元。将这些单元的功能整合到单个设备中以实现显示内神经形态计算(IDNC),在克服这些挑战方面显示出巨大的潜力。在此,我们报告了一种利用有机半导体的电致变色特性的一体化电可编程IDNC(EP-IDNC)设备。电致变色特性赋予EP-IDNC设备突触和神经行为(如多终端可操作性、多级权重更新、时空信息编码),以及通过颜色变化直观显示计算结果的能力。使用3×3 EP-IDNC设备原型阵列,在降噪和运动物体感知等任务上验证了IDNC功能。我们最终展示了用于汽车转向提醒的神经形态和显示功能的集成。