Researchers from the Department of Electrical and Electronic Engineering at The University of Hong Kong (HKU), in collaboration with Xidian University and the Hong Kong University of Science and Technology (HKUST), have made a pioneering advancement in compute-in-memory (CIM) technology. Their study, titled “Memristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory,” was lately published in Nature Communications.
This cross-disciplinary research team, led by Professor Ngai WONG, Professor Can LI, and Dr. Zhengwu LIU, comprising Haiqiao HONG, Zhiyuan DU, Mingrui JIANG, Ruibin MAO, Yuan REN, Fuyi LI, Wei MAO, Muyuan PENG, Wei ZHANG, developed a memristor-based adaptive analog-to-digital converter (ADC) that enables efficient and precise signal processing for next-generation AI hardware.
Conventional AI accelerators suffer from energy and area constraints due to inflexible ADCs, essential components that translate analog signals into digital data. The HKU team introduced an adaptive quantisation architecture that leverages programmable memristors to dynamically adjust quantisation thresholds according to data distribution. This design achieves a 15.1× improvement in energy efficiency and a 12.9× reduction in circuit area compared with state-of-the-art solutions.
Beyond hardware performance, the proposed adaptive ADC demonstrates robust neural network inference, maintaining high accuracy across diverse workloads and network structures. By integrating this adaptive converter into CIM systems, the researchers further reduced overall energy and area overhead by 57.2% and 30.7%, paving the way for scalable and efficient AI chips.
The work represents a significant milestone in bridging the gap between algorithmic intelligence and hardware adaptability, showcasing how memristive computing can revolutionise the design of next-generation AI chips. It also highlights HKU EEE’s leadership in cross-domain research that unites device physics, circuit design, and machine learning systems.
The project received support from the Theme-based Research Scheme (TRS) project T45-701/22-R, the National Natural Science Foundation of China (62404187, 62122005), Croucher Foundation, and the General Research Fund (GRF) Project (17200925, 17203224, 17207925) of the Research Grants Council (RGC), Hong Kong SAR.
Link to the paper: https://doi.org/10.1038/s41467-025-65233-w
Figure 1: Challenges in CIM systems for neural network computation.
Figure 2: Design and benchmarking of the memristor-based ADC.