Sensing–Computing Fusion in Machine Vision: A Review on Integrated Systems and Emerging Devices

The rising demand for high‐speed, energy‐efficient, multifunctional machine vision systems encounters fundamental bottlenecks in latency and energy consumption within conventional architectures, arising from the redundant massive data access and processing. To overcome these limitations, a transformative paradigm toward sensing–computing fusion is emerging, drawing inspiration from the neural adaptation, sparse coding, and hierarchical processing of the biological visual system. This work reviews two principal technological pathways: complementary metal‐oxide‐semiconductor (CMOS) vision chips and emerging bio‐inspired devices. CMOS vision chips integrate processing circuits from pixel to chip‐level, computing in parallel across digital, analog, or mixed‐signal domains and achieving functions such as early filtering, feature extraction, convolutional operations, or neural network inference. The physics‐based fusion through emerging bio‐inspired devices leverages novel materials like perovskites and 2D semiconductors to exploit phenomena such as ion migration and trap‐state dynamics, intrinsically embedding adaptive, event‐driven sensing, in‐sensor convolution, and feature extraction. While CMOS fusion provides engineering flexibility and scalability, and bio‐inspired devices promise ultrahigh efficiency and adaptation, their combined heterogeneous integration is prospected to offer a synergistic route toward next‐generation compact, high‐speed, and low‐power visual systems.