A 0.76 mm2 0.22 nJ/Pixel DL-Assisted 4K Video Encoder LSI for Quality-of-Experience Over Smartphones

This letter proposes the world’s first deep learning (DL)-assisted video encoder LSI fabricated in a 10-nm process with a core area of 0.76 mm² to integrate quad-core DL accelerators and <inline-formula> <tex-math notation="LaTeX">$4\text{K}\times 2\text{K}$ </tex-math></inline-formula> H.264/H.265 video encoders. A visual-contact-field network (VCFNet) DL model is newly designed to predict human focus information with extraordinary reduction of encoding complexity, leading to 82.3% power reduction. Moreover, input channel reduction and layer merging approaches reduce VCFNet complexity by 69%. Operated at 0.9 V and 504 MHz, the proposed DL-assisted 4K video encoder LSI consumes 56.54 mW to achieve 0.22 nJ/pixel of energy efficiency, cutting 0.1-14 nJ/pixel compared to conventional designs.