MiniFloat-NN and ExSdotp: An ISA Extension and a Modular Open Hardware Unit for Low-Precision Training on RISC-V cores.
Luca BertacciniGianna PaulinTim FischerStefan MachLuca BeniniPublished in: CoRR (2022)
Keyphrases
- training process
- neural network
- modular neural networks
- hardware architecture
- neural network training
- nearest neighbor
- feed forward neural networks
- low cost
- hardware and software
- training algorithm
- multi core processors
- computer systems
- back propagation
- control unit
- instruction set
- real time
- feedforward neural networks
- multi layer perceptron
- computing systems
- training set
- multi layer
- massively parallel
- processing units
- parallel architectures
- hardware implementation
- feed forward
- supervised learning
- low power consumption
- artificial neural networks
- computer architecture
- single chip
- efficient implementation
- input output