A Posit Based Multiply-accumulate Unit with Small Quire Size for Deep Neural Networks

Posit is a numerical representation that especially focused on deep neural networks (DNNs). However, specific register called quire necessary for posit multiply-accumulate (MAC) unit to ensure the calculation accuracy. In this paper, we proposed based MAC can optimize size according target applications. We also develop DNN library explore of unit. Experimental result with ResNet-9 showed achieved same level accuracy as Deep Positron but area reduction 43%.