A 0.8 V Intelligent Vision Sensor With Tiny Convolutional Neural Network and Programmable Weights Using Mixed-Mode Processing-in-Sensor Technique for Image Classification

This article presents an intelligent vision sensor (IVS) with embedded tiny convolutional neural network (CNN) model and programmable processing-in-sensor (PIS) circuit for real-time inference applications of low-power edge devices. The proposed imager realizes the full computing functions of a customized three-layers tiny network, which includes a <tex-math notation="LaTeX">3 × 3 </tex-math> convolution layer (stride <tex-math notation="LaTeX">$=$ </tex-math> 3) with activation function of rectified linear unit (ReLU), a <tex-math notation="LaTeX">2 × 2 </tex-math> maximum pooling (MP) layer (stride <tex-math notation="LaTeX">$=$ </tex-math> 2), and a <tex-math notation="LaTeX">1 × 1 </tex-math> fully connected (FC) layer for inference. A 0.8 V <tex-math notation="LaTeX">128 × 128 </tex-math> IVS prototype was fabricated and verified in TSMC 0.18 <tex-math notation="LaTeX">μ m </tex-math> standard CMOS technology. In normal image mode, it consumed 76.4 <tex-math notation="LaTeX">μ W </tex-math> with full-resolution ( <tex-math notation="LaTeX">126 × 126 </tex-math> active resolution) image output at 125 f/s. In CNN mode, it consumed 134.5 <tex-math notation="LaTeX">μ W </tex-math> at 250 f/s and an achieved iFoMs of 33.8 pJ/pixel <tex-math notation="LaTeX">· </tex-math> frame. Using the proposed mixed-mode PIS circuits, the prototype is configured to demonstrate a “human face or not detection” task with an achieved accuracy of 93.6%.