IBM Research has been investigating ways to reinvent the way that AI is computed. Analog in-memory computing, or simply analog AI, is a promising approach to address the challenge by borrowing key features of how neural networks run in biological brains. In our brains, and those of many other animals, the strength of synapses (which are the “weights” in this case) determine communication between neurons. For analog AI systems, we store these synaptic weights locally in the conductance values of nanoscale resistive memory devices such as phase change memory (PCM) and perform multiply-accumulate (MAC) operations, the dominant compute operation in DNNs by exploiting circuit laws and mitigating the need to constantly send data between memory and processor.
Paper
Abstract
Analogue in-memory computing (AIMC) with resistive memory devices could reduce the latency and energy consumption of deep neural network inference tasks by directly performing computations within memory. However, to achieve end-to-end improvements in latency and energy consumption, AIMC must be combined with on-chip digital operations and on-chip communication. Here we report a multicore AIMC chip designed and fabricated in 14 nm complementary metal–oxide–semiconductor technology with backend-integrated phase-change memory. The fully integrated chip features 64 AIMC cores interconnected via an on-chip communication network. It also implements the digital activation functions and additional processing involved in individual convolutional layers and long short-term memory units. With this approach, we demonstrate near-software-equivalent inference accuracy with ResNet and long short-term memory networks, while implementing all the computations associated with the weight layers and the activation functions on the chip. For 8-bit input/output matrix–vector multiplications, in the four-phase (high-precision) or one-phase (low-precision) operational read mode, the chip can achieve a maximum throughput of 16.1 or 63.1 tera-operations per second at an energy efficiency of 2.48 or 9.76 tera-operations per second per watt, respectively.