Approximated computing for low power neural networks

Gian Carlo Cardarilli (University of Rome Tor Vergata)
Luca Di Nunzio (University of Rome Tor Vergata)
Rocco Fazzolari (University of Rome Tor Vergata)
Daniele Giardino (University of Rome Tor Vergata)
Marco Matta (University of Rome Tor Vergata)
Mario Patetta (University of Rome Tor Vergata)
Marco Re (University of Rome Tor Vergata)
Sergio Spanò (University of Rome Tor Vergata)

TELKOMNIKA (Telecommunication Computing Electronics and Control)
Vol 17, No 3: June 2019

This paper investigates about the possibility to reduce power consumption in Neural Network using approximated computing techniques. Authors compare a traditional fixed-point neuron with an approximated neuron composed of approximated multipliers and adder. Experiments show that in the proposed case of study (a wine classifier) the approximated neuron allows to save up to the 43% of the area, a power consumption saving of 35% and an improvement in the maximum clock frequency of 20%.