14.3 A 28nm SoC with a 1.2GHz 568nJ/prediction sparse deep-neural-network engine with >0.1 timing error rate tolerance for IoT applications.
Paul N. WhatmoughSae Kyu LeeHyunkwang LeeSaketh RamaDavid M. BrooksGu-Yeon WeiPublished in: ISSCC (2017)
Keyphrases
- error rate
- neural network
- prediction model
- test set
- prediction accuracy
- artificial neural networks
- lower error rates
- misclassification rate
- pattern recognition
- management system
- neural network model
- word error rate
- equal error rate
- cost sensitive classification
- low power
- rule sets
- back propagation
- cloud computing
- training data
- rejection rate
- feed forward
- low cost
- decision trees