A 1.625 TOPS/W SOC for Deep CNN Training and Inference in 28nm CMOS.
Yu-Tung LiuChuKing KungMing-Hang HsiehHsiu-Wen WangChun-Pin LinChao-Yang YuChi-Shi ChenTzi-Dar ChiuehPublished in: ESSDERC (2021)
Keyphrases
- low power
- structured prediction
- probabilistic inference
- high speed
- training set
- cellular neural networks
- training process
- belief networks
- power consumption
- training examples
- training samples
- online learning
- hardware and software
- low cost
- training phase
- power supply
- cmos technology
- deep architectures
- artificial neural networks
- training data
- vlsi circuits
- belief nets