A 28nm 38-to-102-TOPS/W 8b Multiply-Less Approximate Digital SRAM Compute-In-Memory Macro for Neural-Network Inference.
Yifan HeHaikang DiaoChen TangWenbin JiaXiyuan TangYuan WangJinshan YueXueqing LiHuazhong YangHongyang JiaYongpan LiuPublished in: ISSCC (2023)
Keyphrases
- neural network
- auto associative
- random access memory
- dynamic random access memory
- associative memory
- back propagation
- efficient computation
- power consumption
- inference in graphical models
- bayesian inference
- neural network model
- genetic algorithm
- provide an upper bound
- memory requirements
- bayesian networks
- bp neural network
- artificial neural networks
- embedded dram
- leakage current
- pattern recognition
- main memory
- self organizing maps
- memory usage
- junction tree
- fuzzy logic
- inference process
- fault diagnosis
- multilayer perceptron
- prediction model
- recurrent neural networks