A Digitally Controlled Analog kernel for Convolutional Neural Networks.

Convolutional Neural Networks benefit from higher accuracy at the cost of higher hardware requirements and power consumption. The multiply and accumulate unit ((MAC), which perform the convolution operation inside a CNN, consumes a significant amount of power consumption. In this work we propose a mixed-signal approach for implementing analog MAC unit that can replace the digital MAC units in CNNs. The Analog MAC unit architecture is constituted from binary weighted current steering DAC circuit and capacitors. A digital parallel interfaced is designed to provide input image and filter values to the MAC unit. To realize an input layer for the CNN a low power ADC is then employed at the output to convert the final value back to digital value. When a 3x3 convolution is performed the analog MAC unit offers area and power consumption benefits.