Impact of ADC Bits on CNNs in CIM Architecture

This paper investigates the impact of analog-to-digital converter (ADC) bit precision on the energy efficiency and accuracy of convolutional neural networks (CNNs) within compute-in-memory (CIM) architectures. By adjusting the ADC bit precision layer by layer, we delve into its effects on the energy consumption and accuracy. Experimental results indicate that the influence of ADC bit precision adjustment varies across CNN layers, one-size-fits-all approach is not optimal and layer by layer ADC bits adjustment is helpful to enhance the energy efficiency of CNNs in the CIM architecture.