Mingrui Jiang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Mingrui Jiang has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
- Can Li
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong SAR, China
10 shared publications - Ruibin Mao
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
6 shared publications - Jim Ignowski
Hewlett Packard Labs, Hewlett Packard Enterprise, Milpitas, USA
4 shared publications - Bo Wen
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
3 shared publications - Giacomo Pedretti
Hewlett Packard Labs, Hewlett Packard Enterprise, Milpitas, USA
3 shared publications Zhiyuan Du
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
3 shared publications
Latest Publications
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
METRIC: An MDS-Analog ECC with Minimum Two-Column Redundancy for Fault-Tolerant, Layer-Aware IMC Architecture
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessMETRIC: An MDS-Analog ECC with Minimum Two-Column Redundancy for Fault-Tolerant, Layer-Aware IMC Architecture
- 12 June 2026
- Published by Springer Nature in Nature Electronics
- Vol. 9 (7) , 811-824
Fault-free analogue computing with imperfect hardware
- 12 June 2026
- Published by Springer Nature in Nature Electronics
- Vol. 9 (7) , 811-824
Fault-free analogue computing with imperfect hardware
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (4) , 926-938
Improving Cyclic Positioning Strategy Accuracy of UWB System in Complex Underground Environments
The high-precision automatic positioning technology of the shearer is crucial for the automated and unmanned mining. However, the existing localization approaches commonly deliver inaccurate, unsatisfactory and unreliable results. In response to this challenge, we propose a novel cyclic positioning strategy to automatically migrate the anchor nodes group (ANG) after the completion of the current cutting operation, thereby achieving long-term period positioning. Meanwhile, we design an innovative technique that appropriately incorporates the shrinkage estimation method based on the minimum mean square error criterion (MMSEC), the improved diversity-guided quantum particle swarm optimization (QPSO), and the extended Kalman filter (EKF) to enhance the localization accuracy of the ultra-wideband (UWB) system in each cycle positioning. Firstly, the cycle positioning strategy and the calculation method of the ANG coordinate position are proposed at the end of fully mechanized mining face. Secondly, the MMSEC method is developed by taking into account of the large measurement noise in the underground environments, and the improved diversity-guided QPSO method is implemented to optimize the result of MMSEC approach. Subsequently, the EKF is executed to suppress the influence of distance residuals on the positioning accuracy and further refine the final estimation accuracy. Lastly, the experimental investigation is performed to evaluate the effectiveness of the proposed cyclic positioning strategy. The experimental results sufficiently demonstrate that the developed MMSEC-QPSO-EKF technique significantly enhances localization accuracy and substantially outperforms the conventional methods, which is capable of achieving better estimation accuracy in each cycle positioning. The proposed cyclic positioning strategy can be successfully implemented, and the achieved accuracies are less than 0.2 m as the ANG is migrated three times, showcasing outstanding localization performance and robustness.
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/CAA Journal of Automatica Sinica
- Vol. 13 (4) , 926-938
Improving Cyclic Positioning Strategy Accuracy of UWB System in Complex Underground Environments
The high-precision automatic positioning technology of the shearer is crucial for the automated and unmanned mining. However, the existing localization approaches commonly deliver inaccurate, unsatisfactory and unreliable results. In response to this challenge, we propose a novel cyclic positioning strategy to automatically migrate the anchor nodes group (ANG) after the completion of the current cutting operation, thereby achieving long-term period positioning. Meanwhile, we design an innovative technique that appropriately incorporates the shrinkage estimation method based on the minimum mean square error criterion (MMSEC), the improved diversity-guided quantum particle swarm optimization (QPSO), and the extended Kalman filter (EKF) to enhance the localization accuracy of the ultra-wideband (UWB) system in each cycle positioning. Firstly, the cycle positioning strategy and the calculation method of the ANG coordinate position are proposed at the end of fully mechanized mining face. Secondly, the MMSEC method is developed by taking into account of the large measurement noise in the underground environments, and the improved diversity-guided QPSO method is implemented to optimize the result of MMSEC approach. Subsequently, the EKF is executed to suppress the influence of distance residuals on the positioning accuracy and further refine the final estimation accuracy. Lastly, the experimental investigation is performed to evaluate the effectiveness of the proposed cyclic positioning strategy. The experimental results sufficiently demonstrate that the developed MMSEC-QPSO-EKF technique significantly enhances localization accuracy and substantially outperforms the conventional methods, which is capable of achieving better estimation accuracy in each cycle positioning. The proposed cyclic positioning strategy can be successfully implemented, and the achieved accuracies are less than 0.2 m as the ANG is migrated three times, showcasing outstanding localization performance and robustness.
- Open Access
- 24 April 2026
- Published by Springer Nature in Nature Communications
- Vol. 17
Trustworthy tree-based machine learning by MoS2 flash-based analog content-addressable memory with inherent soft boundaries
The rapid advancement of artificial intelligence has raised concerns regarding its trustworthiness, especially in terms of interpretability and robustness. Tree-based models such as Random Forest excel in interpretability and accuracy for tabular data, but scaling them remains computationally expensive due to poor data locality and high data dependence. Previous efforts to accelerate these models with analog content-addressable memory (CAM) have struggled because difficult-to-implement sharp decision boundaries are highly susceptible to device variations, leading to poor hardware performance and vulnerability to adversarial attacks. Here, we present a hardware-software co-design approach using MoS2 flash-based analog CAM with inherent soft boundaries, enabling efficient inference with soft tree-based models. Our fabricated analog CAM arrays achieve 96% accuracy on the Wisconsin Diagnostic Breast Cancer dataset, and our experimentally calibrated model shows only a 0.6% accuracy drop on MNIST under 10% device threshold variation, compared with 45.3% for traditional decision trees.
- 24 April 2026
- Published by Springer Nature in Nature Communications
- Vol. 17
Open AccessTrustworthy tree-based machine learning by MoS2 flash-based analog content-addressable memory with inherent soft boundaries
The rapid advancement of artificial intelligence has raised concerns regarding its trustworthiness, especially in terms of interpretability and robustness. Tree-based models such as Random Forest excel in interpretability and accuracy for tabular data, but scaling them remains computationally expensive due to poor data locality and high data dependence. Previous efforts to accelerate these models with analog content-addressable memory (CAM) have struggled because difficult-to-implement sharp decision boundaries are highly susceptible to device variations, leading to poor hardware performance and vulnerability to adversarial attacks. Here, we present a hardware-software co-design approach using MoS2 flash-based analog CAM with inherent soft boundaries, enabling efficient inference with soft tree-based models. Our fabricated analog CAM arrays achieve 96% accuracy on the Wisconsin Diagnostic Breast Cancer dataset, and our experimentally calibrated model shows only a 0.6% accuracy drop on MNIST under 10% device threshold variation, compared with 45.3% for traditional decision trees.
- Open Access
- 23 February 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
A hardware-adaptive learning algorithm for superlinear-capacity associative memory on memristor crossbars
The human brain recalls complete patterns from partial cues via associative memory, but Hopfield neural networks emulating this process are inefficient on conventional hardware, and prior memristor-based implementations are vulnerable to device defects and have limited capacity, particularly for continuous patterns. We introduce a hardware-adaptive learning algorithm that incorporates experimentally calibrated device constraints during training and validate it on an integrated memristor crossbar compute-in-memory platform. The approach improves defect tolerance and effective capacity, achieving threefold higher capacity than a pseudo-inverse baseline at 50% stuck-at faults. The same framework extends to scalable multilayer architectures supporting binary and continuous-valued patterns, where we observe superlinear capacity scaling on correlated data (∝N1.49 and ∝N1.74, respectively). Leveraging crossbar parallelism with synchronous updates, the implementation reduces energy by 8.8× and latency by 99.7% for 64-dimensional patterns versus asynchronous schemes. These results provide a practical algorithm-hardware co-design for robust, efficient Hopfield-style associative recall.
- 23 February 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessA hardware-adaptive learning algorithm for superlinear-capacity associative memory on memristor crossbars
The human brain recalls complete patterns from partial cues via associative memory, but Hopfield neural networks emulating this process are inefficient on conventional hardware, and prior memristor-based implementations are vulnerable to device defects and have limited capacity, particularly for continuous patterns. We introduce a hardware-adaptive learning algorithm that incorporates experimentally calibrated device constraints during training and validate it on an integrated memristor crossbar compute-in-memory platform. The approach improves defect tolerance and effective capacity, achieving threefold higher capacity than a pseudo-inverse baseline at 50% stuck-at faults. The same framework extends to scalable multilayer architectures supporting binary and continuous-valued patterns, where we observe superlinear capacity scaling on correlated data (∝N1.49 and ∝N1.74, respectively). Leveraging crossbar parallelism with synchronous updates, the implementation reduces energy by 8.8× and latency by 99.7% for 64-dimensional patterns versus asynchronous schemes. These results provide a practical algorithm-hardware co-design for robust, efficient Hopfield-style associative recall.
- 15 December 2025
- Published by Springer Nature in Nature Nanotechnology
- Vol. 21 (2) , 198-206
Sb-contacted MoS2 flash memory for analogue in-memory searches
The explosion of artificial intelligence and edge devices has exposed a critical bottleneck in traditional hardware: the slow data transfer between memory and processing. Content-addressable memories offer a promising solution by processing information directly within the memory, but existing implementations using static random-access memory and, more recently, those using emerging non-volatile memories are constrained by the performance of silicon transistors. Here we introduce an analogue content-addressable memory utilizing atomically thin two-dimensional MoS2 flash memories with semimetal antimony contacts. Our device achieves a high read-out current (60 μA μm−1) and large ON/OFF ratios (>109) in two-dimensional flash memories. These breakthroughs have led to very low energy consumption (under 0.1 fJ per search per cell) and latency (36 ps) during analogue in-memory search operations within our 8 × 16 analogue content-addressable memory array, featuring 256 MoS2 flash memory devices. We have also successfully demonstrated analogue Hamming distance computing for k-nearest neighbour classification, showcasing high accuracy, high energy efficiency and low latency for machine learning applications. This research highlights the transformative potential of two-dimensional materials in overcoming current hardware limitations, enabling more efficient and scalable computing solutions in intelligent edge devices.
- 15 December 2025
- Published by Springer Nature in Nature Nanotechnology
- Vol. 21 (2) , 198-206
Sb-contacted MoS2 flash memory for analogue in-memory searches
The explosion of artificial intelligence and edge devices has exposed a critical bottleneck in traditional hardware: the slow data transfer between memory and processing. Content-addressable memories offer a promising solution by processing information directly within the memory, but existing implementations using static random-access memory and, more recently, those using emerging non-volatile memories are constrained by the performance of silicon transistors. Here we introduce an analogue content-addressable memory utilizing atomically thin two-dimensional MoS2 flash memories with semimetal antimony contacts. Our device achieves a high read-out current (60 μA μm−1) and large ON/OFF ratios (>109) in two-dimensional flash memories. These breakthroughs have led to very low energy consumption (under 0.1 fJ per search per cell) and latency (36 ps) during analogue in-memory search operations within our 8 × 16 analogue content-addressable memory array, featuring 256 MoS2 flash memory devices. We have also successfully demonstrated analogue Hamming distance computing for k-nearest neighbour classification, showcasing high accuracy, high energy efficiency and low latency for machine learning applications. This research highlights the transformative potential of two-dimensional materials in overcoming current hardware limitations, enabling more efficient and scalable computing solutions in intelligent edge devices.
- Open Access
- 6 November 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-11
Memristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory
Compute-in-memory technology offers promising solutions for neural network acceleration but its potential is severely limited by inflexible and resource-intensive analog-to-digital converters. Here, we present a memristor-based analog-to-digital converter featuring adaptive quantization for diverse output distributions. Our design employs analog content-addressable memory cells with programmable overlapped boundaries to establish optimized quantization thresholds, demonstrating excellent integral and differential non-linearities. Extensive experiments validate the robustness of our approach by achieving 89.55% accuracy on CIFAR-10 (VGG8) at 5-bit adaptive quantized precision and maintaining competitive performance on ImageNet (ResNet18) through a proposed super-resolution strategy under experimental memristor variations. Compared to state-of-the-art designs, our converter achieves a 15.1× improvement in energy efficiency and a 12.9× reduction in area. Furthermore, integrating our converter into CIM systems reduces the energy and area overhead by up to 57.2% and 30.7%, respectively. This work establishes a paradigm for efficient and accurate signal quantization in practical compute-in-memory systems.
- 6 November 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-11
Open AccessMemristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory
Compute-in-memory technology offers promising solutions for neural network acceleration but its potential is severely limited by inflexible and resource-intensive analog-to-digital converters. Here, we present a memristor-based analog-to-digital converter featuring adaptive quantization for diverse output distributions. Our design employs analog content-addressable memory cells with programmable overlapped boundaries to establish optimized quantization thresholds, demonstrating excellent integral and differential non-linearities. Extensive experiments validate the robustness of our approach by achieving 89.55% accuracy on CIFAR-10 (VGG8) at 5-bit adaptive quantized precision and maintaining competitive performance on ImageNet (ResNet18) through a proposed super-resolution strategy under experimental memristor variations. Compared to state-of-the-art designs, our converter achieves a 15.1× improvement in energy efficiency and a 12.9× reduction in area. Furthermore, integrating our converter into CIM systems reduces the energy and area overhead by up to 57.2% and 30.7%, respectively. This work establishes a paradigm for efficient and accurate signal quantization in practical compute-in-memory systems.
- Open Access
- 25 March 2025
- Published by Springer Nature in Research Square
Memristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory
Compute-in-memory (CIM) technology offers promising solutions for neural network acceleration with high energy efficiency. However, the potential of existing CIM systems is severely limited by inflexible and resource-intensive analog-to-digital converters (ADCs). Here, we present a memristor-based ADC featuring adaptive quantization for diverse CIM output distributions. Our design employs analog content-addressable memory cells with programmable overlapped boundaries to establish optimized quantization thresholds, demonstrating excellent integral and differential non-linearity of 0.319 and 0.419 least significant bit, respectively. Extensive experiments validate the robustness of our approach by achieving 89.55% accuracy on CIFAR-10 (VGG8) at 5-bit adaptive quantized precision and maintaining competitive performance on ImageNet (ResNet18) through a proposed super-resolution strategy under memristor variations observed in our experiment. Compared to state-of-the-art ADCs, our design achieves a 15.1× improvement in energy efficiency and a 17.3× reduction in area. Furthermore, the integration of our ADC into CIM systems reduces the energy and area overhead by 57.2% and 34.4% respectively for the CIFAR-10 (VGG8) task, and by 56.9% and 27.8% for the ImageNet (ResNet18) task. This work could establish a new paradigm for efficient and accurate signal quantization in practical CIM systems.
- 25 March 2025
- Published by Springer Nature in Research Square
Open AccessMemristor-based adaptive analog-to-digital conversion for efficient and accurate compute-in-memory
Compute-in-memory (CIM) technology offers promising solutions for neural network acceleration with high energy efficiency. However, the potential of existing CIM systems is severely limited by inflexible and resource-intensive analog-to-digital converters (ADCs). Here, we present a memristor-based ADC featuring adaptive quantization for diverse CIM output distributions. Our design employs analog content-addressable memory cells with programmable overlapped boundaries to establish optimized quantization thresholds, demonstrating excellent integral and differential non-linearity of 0.319 and 0.419 least significant bit, respectively. Extensive experiments validate the robustness of our approach by achieving 89.55% accuracy on CIFAR-10 (VGG8) at 5-bit adaptive quantized precision and maintaining competitive performance on ImageNet (ResNet18) through a proposed super-resolution strategy under memristor variations observed in our experiment. Compared to state-of-the-art ADCs, our design achieves a 15.1× improvement in energy efficiency and a 17.3× reduction in area. Furthermore, the integration of our ADC into CIM systems reduces the energy and area overhead by 57.2% and 34.4% respectively for the CIFAR-10 (VGG8) task, and by 56.9% and 27.8% for the ImageNet (ResNet18) task. This work could establish a new paradigm for efficient and accurate signal quantization in practical CIM systems.
- Open Access
- 29 January 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-15
Efficient nonlinear function approximation in analog resistive crossbars for recurrent neural networks
Analog In-memory Computing (IMC) has demonstrated energy-efficient and low latency implementation of convolution and fully-connected layers in deep neural networks (DNN) by using physics for computing in parallel resistive memory arrays. However, recurrent neural networks (RNN) that are widely used for speech-recognition and natural language processing have tasted limited success with this approach. This can be attributed to the significant time and energy penalties incurred in implementing nonlinear activation functions that are abundant in such models. In this work, we experimentally demonstrate the implementation of a non-linear activation function integrated with a ramp analog-to-digital conversion (ADC) at the periphery of the memory to improve in-memory implementation of RNNs. Our approach uses an extra column of memristors to produce an appropriately pre-distorted ramp voltage such that the comparator output directly approximates the desired nonlinear function. We experimentally demonstrate programming different nonlinear functions using a memristive array and simulate its incorporation in RNNs to solve keyword spotting and language modelling tasks. Compared to other approaches, we demonstrate manifold increase in area-efficiency, energy-efficiency and throughput due to the in-memory, programmable ramp generator that removes digital processing overhead.
- 29 January 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-15
Open AccessEfficient nonlinear function approximation in analog resistive crossbars for recurrent neural networks
Analog In-memory Computing (IMC) has demonstrated energy-efficient and low latency implementation of convolution and fully-connected layers in deep neural networks (DNN) by using physics for computing in parallel resistive memory arrays. However, recurrent neural networks (RNN) that are widely used for speech-recognition and natural language processing have tasted limited success with this approach. This can be attributed to the significant time and energy penalties incurred in implementing nonlinear activation functions that are abundant in such models. In this work, we experimentally demonstrate the implementation of a non-linear activation function integrated with a ramp analog-to-digital conversion (ADC) at the periphery of the memory to improve in-memory implementation of RNNs. Our approach uses an extra column of memristors to produce an appropriately pre-distorted ramp voltage such that the comparator output directly approximates the desired nonlinear function. We experimentally demonstrate programming different nonlinear functions using a memristive array and simulate its incorporation in RNNs to solve keyword spotting and language modelling tasks. Compared to other approaches, we demonstrate manifold increase in area-efficiency, energy-efficiency and throughput due to the in-memory, programmable ramp generator that removes digital processing overhead.
- Open Access
- 22 September 2023
- Published by Springer Nature in Nature Communications
- Vol. 14 (1) , 1-11
Efficient combinatorial optimization by quantum-inspired parallel annealing in analogue memristor crossbar
Combinatorial optimization problems are prevalent in various fields, but obtaining exact solutions remains challenging due to the combinatorial explosion with increasing problem size. Special-purpose hardware such as Ising machines, particularly memristor-based analog Ising machines, have emerged as promising solutions. However, existing simulate-annealing-based implementations have not fully exploited the inherent parallelism and analog storage/processing features of memristor crossbar arrays. This work proposes a quantum-inspired parallel annealing method that enables full parallelism and improves solution quality, resulting in significant speed and energy improvement when implemented in analog memristor crossbars. We experimentally solved tasks, including unweighted and weighted Max-Cut and traveling salesman problem, using our integrated memristor chip. The quantum-inspired parallel annealing method implemented in memristor-based hardware has demonstrated significant improvements in time- and energy-efficiency compared to previously reported simulated annealing and Ising machine implemented on other technologies. This is because our approach effectively exploits the natural parallelism, analog conductance states, and all-to-all connection provided by memristor technology, promising its potential for solving complex optimization problems with greater efficiency.
- 22 September 2023
- Published by Springer Nature in Nature Communications
- Vol. 14 (1) , 1-11
Open AccessEfficient combinatorial optimization by quantum-inspired parallel annealing in analogue memristor crossbar
Combinatorial optimization problems are prevalent in various fields, but obtaining exact solutions remains challenging due to the combinatorial explosion with increasing problem size. Special-purpose hardware such as Ising machines, particularly memristor-based analog Ising machines, have emerged as promising solutions. However, existing simulate-annealing-based implementations have not fully exploited the inherent parallelism and analog storage/processing features of memristor crossbar arrays. This work proposes a quantum-inspired parallel annealing method that enables full parallelism and improves solution quality, resulting in significant speed and energy improvement when implemented in analog memristor crossbars. We experimentally solved tasks, including unweighted and weighted Max-Cut and traveling salesman problem, using our integrated memristor chip. The quantum-inspired parallel annealing method implemented in memristor-based hardware has demonstrated significant improvements in time- and energy-efficiency compared to previously reported simulated annealing and Ising machine implemented on other technologies. This is because our approach effectively exploits the natural parallelism, analog conductance states, and all-to-all connection provided by memristor technology, promising its potential for solving complex optimization problems with greater efficiency.