UNPU: A 50.6TOPS/W unified deep neural network accelerator with 1b-to-16b fully-variable weight bit-precision.
Jinmook LeeChanghyeon KimSanghoon KangDongjoo ShinSangyeob KimHoi-Jun YooPublished in: ISSCC (2018)
Keyphrases
- neural network
- weight update
- artificial neural networks
- back propagation
- pattern recognition
- neural nets
- weight assignment
- fault diagnosis
- high precision
- neural network model
- parallel implementation
- average precision
- genetic algorithm
- high recall
- deep learning
- precision and recall
- network model
- neural network is trained
- multilayer perceptron
- feed forward
- random access memory
- radial basis function
- image reconstruction from projections
- knn
- random number generator
- self organizing maps
- total weight
- auto associative
- prediction model
- network architecture
- hidden layer
- recurrent neural networks