“…[1][2][3] Pathway of classic machine vision systems from the image sensors, analog-to-digital converter (ADC), a memory unit to computation units can be implemented based on conventional CMOS technology where the photo sensors convert the light shined on each pixel into a voltage signal, [4,5] ADCs binarize the data after amplification and de-noising, memory store the binary information and computation units perform final recognition and classification. [6,7] However, this pathway is in flat processing schemes where redundant sensory data were employed as input to be processed by computation units with machine-learning algorithms. [8][9][10] Owing to its high latency and inefficient power consumption originating from the data shuffling between memory and computation units, the general and robust resolutions to the major problems including visionbased navigation, object recognition, activity recognition, and motion detection, are still beyond the reach of current hardwareimplemented machine vision system.…”