An analog-AI chip for energy-efficient speech recognition and transcription

Stefano Ambrogio, Pritish Narayanan, Atsuya Okazaki, Andrea Fasoli, Charles Mackin, Kohji Hosokawa, Akiyo Nomura, Takeo Yasuda, An Chen, Alexander M. Friz, Masatoshi Ishii, Jose Luquin, Yasuteru Kohda, Nicole Saulnier, Kevin Brew, Samuel Choi, Injo Ok, Timothy Philip, Victor Chan, Mary Claire Silvestre et al. (4 additional authors not shown)

Abstract: Models of artificial intelligence (AI) that have billions of parameters can achieve high accuracy across a range of tasks1,2, but they exacerbate the poor energy efficiency of conventional general-purpose processors, such as graphics processing units or central processing units. Analog in-memory computing (analog-AI)3–7 can provide better energy efficiency by performing matrix–vector multiplications in parallel on ‘memory tiles’. However, analog-AI has yet to demonstrate software-equivalent (SWeq) accuracy on models that require many such tiles and efficient communication of neural-network activations between the tiles. Here we present an analog-AI chip that combines 35 million phase-change memory devices across 34 tiles, massively parallel inter-tile communication and analog, low-power peripheral circuitry that can achieve up to 12.4 tera-operations per second per watt (TOPS/W) chip-sustained performance. We demonstrate fully end-to-end SWeq accuracy for a small keyword-spotting network and near-SWeq accuracy on the much larger MLPerf8 recurrent neural-network transducer (RNNT), with more than 45 million weights mapped onto more than 140 million phase-change memory devices across five chips. A low-power chip that runs AI models using analog rather than digital computation shows comparable accuracy on speech-recognition tasks but is more than 14 times as energy efficient.