Meters Closer, Miles Faster: HKUST Engineering Researchers Introduce Novel Cryogenic In-Memory Computing Scheme to Bridge AI with Quantum Computing
Scholars at the School of Engineering of the Hong Kong University of Science and Technology (HKUST) have unveiled an innovation that brings artificial intelligence (AI) closer to quantum computing – both physically and technologically.
Led by Prof. SHAO Qiming, Assistant Professor at the Department of Electronic and Computer Engineering, the research team has developed a new computing scheme that works at extremely low temperatures. As a critical advancement in quantum computing, it can significantly reduce latency between artificial intelligence (AI) agents and quantum processors while boosting energy efficiency. The solution was made possible by utilizing a special technology known as magnetic topological insulator Hall-bar devices.
This latest invention addresses a major challenge concerning the operation environment and hardware requirements of quantum computers, amid growing interest in the amalgamation of quantum computing – widely seen as the future of high-speed and high-efficiency computing, with artificial intelligence – a fast-evolving technology.
Prof. Shao said, “Quantum computers, which leverage thousands of quantum bits (qubits) for complex calculations, are widely regarded as the future of rapid and energy-efficient computation. To fully realize this potential, researchers have recently turned to machine learning techniques for their important role in enhancing quantum computing capabilities, particularly in error corrections.”
As quantum processors typically operate at milli-Kelvin temperatures (around -273°C), they are often installed several meters away from room-temperature graphics processing units (GPUs), causing substantial latency (See Figure 1a). To tackle this issue, Prof. Shao and his team proposed a novel cryogenic in-memory computing scheme that allows AI accelerators to work much closer to quantum processors (See Figure 1b). Now they can operate just tens of centimeters apart, enhancing speed and efficiency.
The researchers recognized that magnetic topological insulators had emerged as promising materials for this application. These materials have not only large bulk energy band gaps like those of insulators, but also conducting states on surfaces or edges. They have been found to exhibit unique phenomena like large spin current generation efficiency related to spin-momentum locking of surface states (a phenomenon which constrains spin orientation perpendicular to electron momentum) and the quantum anomalous Hall effect due to chiral edge states (a phenomenon that constrains electrons moving along the edge according to their momentum directions in the absence of a magnetic field).
For this study, the team selected one specific type of magnetic topological insulator known as chromium-doped bismuth-antimony-telluride (Cr-BST). The material stood out for its proven ability to provide giant quantum anomalous Hall resistance and efficient current-induced magnetization switching, which could enhance the performance of Hall devices.
Prof. Shao highlighted, “This breakthrough marks the first demonstration of Hall current summation scheme for low-power in-memory computing, especially at cryogenic temperatures. Our magnetic topological insulator Hall-bar array enables efficient implementation of reinforcement learning algorithms, such as quantum state preparation, near quantum processors.”
While previous designs using conventional ferromagnets like iron-cobalt-boron alloys faced challenges such as weak signals and sneak paths, the Hall-bar array circuit design in the present study has shown remarkable success. In proof-of-concept classification tasks, four Cr-BST Hall-bar devices achieved high accuracy while simulations of 512 × 512 neural networks indicated a performance level of 724 tera-operations per second per Watt (TOPS/W) for tasks related to image recognition and quantum state preparation at 2 K (300 K for ambient room temperature).
The research, titled “Cryogenic In-Memory Computing Using Magnetic Topological Insulators” and recently published in Nature Materials, not only highlights the potential of magnetic topological insulators but also opens new avenues for topological quantum-physics-based computing schemes.
Looking forward, Prof. Shao and his team will strive to further cut down the latency for both inference and online training by integrating AI agents with training units (See Figure 1c), and hope to open up more efficient quantum computing applications.
This project was a collaborative effort among HKUST, the University of California, Los Angeles (UCLA), the Institute of Physics of the Chinese Academy of Sciences, City University of Hong Kong, and Southern University of Science and Technology. The three co-first authors were Dr. LIU Yuting, former postdoctoral fellow at HKUST, Dr. Albert LEE, PhD alumnus at UCLA, and QIAN Kun, HKUST PhD student.
For media inquiries, please contact:
Celia Lee
Tel: 2358 8982 / Email: celialee@ust.hk
Dorothy Yip
Tel: 2358 5917 / Email: egkkyip@ust.hk
(This article was originally published on EurekAlert on March 24, 2025.)
