Low-Latency, Energy-Efficient In-DRAM CNN Acceleration with Bit-Parallel Unary Computing

With the rapidly growing use of convolutional neural networks (CNNs) in real-world applications related to machine learning and artificial intelligence (AI), several hardware accelerator designs for CNN inference and training have been proposed recently. In this chapter, we present ATRIA, a novel bit-parallel rate-coded unary-computing-based in-DRAM accelerator for energy-efficient and high-speed inference of CNNs. ATRIA employs light-weight modifications in DRAM cell arrays to implement bit-parallel rate-coded unary-computing (i.e., stochastic computing)-based acceleration of multiply-accumulate (MAC) operations inside DRAM. ATRIA significantly improves the latency, throughput, and efficiency of processing CNN inferences by enabling 16 MAC operations to be performed in only two consecutive memory operation cycles. We mapped four benchmark CNNs on ATRIA to compare its performance with five state-of-the-art in-DRAM accelerators from prior work. The results of our analysis show that ATRIA exhibits only 3.5% drop in CNN inference accuracy and still achieves improvements of up to 3.2 × in frames per second (FPS) and up to 10 × in efficiency (FPS/W/mm²), compared to the best-performing in-DRAM accelerator from prior work.