UMass Researchers Contribute to Science Paper on High-precision Computing with Low-precision Devices
Content
![Qiangfei Xia](/engineering/sites/g/files/ijdqth166/files/styles/text_area_full_xl/public/2024-03/NEWS-Xia_Qiangfei.jpg?itok=VH6hCZa9)
Professors Qiangfei Xia and J. Joshua Yang of the UMass Amherst Electrical and Computer Engineering (ECE) Department have contributed to a trailblazing paper on in-memory computing published in the highly respected journal Science. Xia and Yang are essential members of a 21-person research team from UMass Amherst, the University of Southern California, and TetraMem Inc. (lead institution) that co-authored the Science paper. The paper explains a collaborative effort that has proposed and demonstrated a groundbreaking circuit architecture and programming protocol that enables low-precision analog devices to perform high-precision computing.
In-memory computing, which computes within the memory cells, represents an effective method for modeling complex physical systems that are typically challenging for conventional computing architectures. According to the Science paper, many complex physical systems are too complicated for analytical techniques, and direct numerical computation is hindered by the “curse of dimensionality,” which requires exponentially increasing resources as the size of the problem increases.
“These systems range from nanoscale problems in material modeling to large-scale problems in climate science,” explains the Science paper. “Although the need for accurate and high-performance computing solutions is growing, traditional von Neumann computing architectures are reaching their limit in terms of speed, energy consumption, and infrastructure…A promising alternative is in-memory computing that circumvents the memory-processor bottleneck inherent to von Neumann architectures.”
However, in-memory computing has been hindered by issues such as reading noise and writing variability that restrict scalability, accuracy, and precision in high-performance computations, which are the issues to be tackled by this new paper.
Accordingly, as the Science paper observes, “Analog devices have been primarily used for applications without high-precision requirements, such as machine learning, randomness-based processing like stochastic computing, and hardware security. To achieve high-precision solutions, innovations in architecture and algorithms, codesigned with analog devices, must be made.”
The Science paper describes an elegant solution to this challenge. As the paper notes, “In this work, we propose and demonstrate a new circuit architecture and programming protocol that can efficiently represent high-precision numbers using a weighted sum of multiple, relatively low-precision analog devices, such as memristors, with a greatly reduced overhead in circuitry, energy, and latency compared with existing quantization approaches.”
The researchers have experimentally demonstrated the concepts using the state-of-the-art “system on a chip” they built. As they explain, “We have demonstrated an innovative circuit architecture and programming protocol that can efficiently program inaccurate analog devices with arbitrarily high precision within the limit of digital peripheral circuits. This method enables us to execute partial-differential-equation solvers with high precision, energy efficiency, and throughput.”
The new method opens doors for computational applications previously considered infeasible for emerging analog memories, such as scientific computing, neural network training, and the modeling of complex physical systems.
Xia heads the Nanodevices and Integrated Systems Laboratory, which studies beyond-CMOS devices, integrated systems, and enabling technologies, with applications in machine intelligence, reconfigurable radio-frequency systems, and hardware security.
Yang, as an adjunct professor in the ECE department at UMass Amherst, researches nanoelectronics and nanoionics, especially for energy and computing applications. His research includes high-performance, non-volatile memories, analog computing using resistance analog switches, and neuromorphic/synaptic computing using memristive devices. He is currently a Professor of Electrical and Computer Engineering at the University of Southern California. (March 2024)