In Frontiers in computational neuroscience
Recent advances in deep learning have been driven by ever-increasing model sizes, with networks growing to millions or even billions of parameters. Such enormous models call for fast and energy-efficient hardware accelerators. We study the potential of Analog AI accelerators based on Non-Volatile Memory, in particular Phase Change Memory (PCM), for software-equivalent accurate inference of natural language processing applications. We demonstrate a path to software-equivalent accuracy for the GLUE benchmark on BERT (Bidirectional Encoder Representations from Transformers), by combining noise-aware training to combat inherent PCM drift and noise sources, together with reduced-precision digital attention-block computation down to INT6.
Spoon Katie, Tsai Hsinyu, Chen An, Rasch Malte J, Ambrogio Stefano, Mackin Charles, Fasoli Andrea, Friz Alexander M, Narayanan Pritish, Stanisavljevic Milos, Burr Geoffrey W
BERT, DNN, PCM, RRAM, Transformer, analog accelerators, in-memory computing