Chen Jiajia, Xu Jiacheng, Gu Jiani, Chen Bowen, Zhang Hongrui, Qian Haoji, Liu Huan, Shen Rongzong, Lin Gaobo, Yu Xiao, Zhang Miaomiao, Ding Yi'an, Liu Yan, Tang Jianshi, Wu Huaqiang, Jin Chengji, Han Genquan
Hangzhou Institute of Technology, Xidian University, Hangzhou, 311231, China.
Faculty of Integrated Circuits, Xidian University, Xi'an, 710071, China.
Nat Commun. 2025 Jan 10;16(1):565. doi: 10.1038/s41467-024-55224-8.
Edge detection is one of the most essential research hotspots in computer vision and has a wide variety of applications, such as image segmentation, target detection, and other high-level image processing technologies. However, efficient edge detection is difficult in a resource-constrained environment, especially edge-computing hardware. Here, we report a low-power edge detection hardware system based on HfO-based ferroelectric field-effect transistor, which is one of the most potential non-volatile memories for energy-efficient computing. Different from the conventional edge detectors requiring sophisticated hardware for the complex operation such as convolution and gradient, the proposed edge detector is analogue-to-digital converter free and loaded into a multi-bit content addressable memory, which only needs one 4 × 4 ferroelectric field-effect transistor NAND array. The experimental results show that the proposed hardware system is able to achieve efficient image edge detection at low power consumption (~10 fJ/per operation), realizing no-accuracy-loss, low-power and analogue-to-digital-converter-free hardware system, providing a feasible solution for edge computing.
边缘检测是计算机视觉中最重要的研究热点之一，具有广泛的应用，如图像分割、目标检测和其他高级图像处理技术。然而，在资源受限的环境中，尤其是边缘计算硬件中，高效的边缘检测是困难的。在此，我们报告了一种基于基于HfO的铁电场效应晶体管的低功耗边缘检测硬件系统，该晶体管是用于节能计算的最具潜力的非易失性存储器之一。与传统的边缘检测器不同，传统边缘检测器需要复杂的硬件来进行诸如卷积和梯度等复杂操作，而所提出的边缘检测器无需模数转换器，并加载到多位内容可寻址存储器中，该存储器仅需要一个4×4铁电场效应晶体管与非阵列。实验结果表明，所提出的硬件系统能够在低功耗（约10 fJ/每次操作）下实现高效的图像边缘检测，实现无精度损失、低功耗且无需模数转换器的硬件系统，为边缘计算提供了一种可行的解决方案。
