Wei Mao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Wei Mao has a total of 140 co-authors. Here are the top co-authors by number of shared publications:
- Hao Yu
School of Microelectronics, Southern University of Science and Technology, Shenzhen, CHINA
14 shared publications - Genquan Han
Xidian University
7 shared publications - Kai Li
School of Microelectronics, Southern University of Science and Technology, Shenzhen, CHINA
6 shared publications - Yan Liu
Xidian University
5 shared publications - Mingqiang Huang
School of Microelectronics, Southern University of Science and Technology, Shenzhen, CHINA
4 shared publications - Yongfu Li
Shanghai Jiao Tong University
4 shared publications
Latest Publications
- Open Access
- 10 July 2026
- Published by Elsevier in SSRN Electronic Journal
TopoSeqNet: A Topology-Aware Spatial-Attentive Sequence Network for EEG Emotion Recognition
Electroencephalogram (EEG)-based emotion recognition remains challenging due to unstable signal patterns, insufficient exploitation of high-frequency neural information, and the loss of electrode topology during spatial-temporal modeling. To address these issues, this paper proposes TopoSeqNet, a topology-aware spatial-temporal learning framework for robust EEG emotion recognition. The proposed framework constructs a frequency-expanded representation over 1-75 Hz, enabling the exploitation of complementary high-frequency emotion-related information. Meanwhile, an adaptive temporal stabilization strategy is introduced to reduce local feature fluctuations and improve sequence continuity. The stabilized 62-channel EEG features are organized into 9×9 spatial topological maps, thereby preserving electrode-level spatial relationships. Based on these structured inputs, TopoSeqNet learns discriminative spatial representations through topology-aware local encoding, spatial refinement, and adaptive feature enhancement. Moreover, a topology-preserving dynamic learning module models consecutive EEG spatial maps without destroying their two-dimensional structure, enabling effective integration of spatial and temporal information. Experiments on the SEED and SEED-IV datasets demonstrate that TopoSeqNet achieves average accuracies of 99.37% and 99.12%, respectively. Further analyses verify the effectiveness of frequency-expanded representation, adaptive feature stabilization, topology-aware spatial learning, and topology-preserving temporal modeling. These results indicate that TopoSeqNet provides an accurate and stable solution for EEG emotion recognition.
- 10 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTopoSeqNet: A Topology-Aware Spatial-Attentive Sequence Network for EEG Emotion Recognition
Electroencephalogram (EEG)-based emotion recognition remains challenging due to unstable signal patterns, insufficient exploitation of high-frequency neural information, and the loss of electrode topology during spatial-temporal modeling. To address these issues, this paper proposes TopoSeqNet, a topology-aware spatial-temporal learning framework for robust EEG emotion recognition. The proposed framework constructs a frequency-expanded representation over 1-75 Hz, enabling the exploitation of complementary high-frequency emotion-related information. Meanwhile, an adaptive temporal stabilization strategy is introduced to reduce local feature fluctuations and improve sequence continuity. The stabilized 62-channel EEG features are organized into 9×9 spatial topological maps, thereby preserving electrode-level spatial relationships. Based on these structured inputs, TopoSeqNet learns discriminative spatial representations through topology-aware local encoding, spatial refinement, and adaptive feature enhancement. Moreover, a topology-preserving dynamic learning module models consecutive EEG spatial maps without destroying their two-dimensional structure, enabling effective integration of spatial and temporal information. Experiments on the SEED and SEED-IV datasets demonstrate that TopoSeqNet achieves average accuracies of 99.37% and 99.12%, respectively. Further analyses verify the effectiveness of frequency-expanded representation, adaptive feature stabilization, topology-aware spatial learning, and topology-preserving temporal modeling. These results indicate that TopoSeqNet provides an accurate and stable solution for EEG emotion recognition.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 610-615
Descriptor: Arc Image Dataset for Power Grid Fault Detection (PG-ArcSet)
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 610-615
Open AccessDescriptor: Arc Image Dataset for Power Grid Fault Detection (PG-ArcSet)
- 7 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 45 (3) , 1129-1133
Collaborative Design of FeRAM via a Joint Ferroelectric Device and Circuit Analysis
Ferroelectric random access memory (FeRAM) is a promising candidate to further dynamic random access memory (DRAM) scaling. However, the design of the FeRAM bit cell is nontrivial as the ferroelectric device model is not well supported by electronic design automation (EDA) tools. Modern integrated circuit design heavily depends on circuit-level SPICE simulators that integrate compact device models through modified nodal analysis (MNA) representation. This article presents a novel MNA-based SPICE simulation method for ferroelectric device models, targeted at the design space exploration of FeRAM bitcells. Furthermore, this article provides a co-design procedure for FeRAM bitcells and sense amplifiers via a comprehensive case study.
- 7 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 45 (3) , 1129-1133
Collaborative Design of FeRAM via a Joint Ferroelectric Device and Circuit Analysis
Ferroelectric random access memory (FeRAM) is a promising candidate to further dynamic random access memory (DRAM) scaling. However, the design of the FeRAM bit cell is nontrivial as the ferroelectric device model is not well supported by electronic design automation (EDA) tools. Modern integrated circuit design heavily depends on circuit-level SPICE simulators that integrate compact device models through modified nodal analysis (MNA) representation. This article presents a novel MNA-based SPICE simulation method for ferroelectric device models, targeted at the design space exploration of FeRAM bitcells. Furthermore, this article provides a co-design procedure for FeRAM bitcells and sense amplifiers via a comprehensive case study.
- Open Access
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Integrated Circuits and Systems
- Vol. 2 (2) , 81-92
Analog-to-Digital Converter Design for Diverse Performance Computing-in-Memory Systems: A Comprehensive Review
Computing-in-Memory (CIM) architectures have emerged as a pivotal technology for next-generation artificial intelligence (AI) and edge computing applications. By enabling computations directly within memory cells, CIM architectures effectively minimize data movement and significantly enhance energy efficiency. In the CIM system, the analog-to-digital converter (ADC) bridges the gap between efficient analog computation and general digital processing, while influencing the overall accuracy, speed and energy efficiency of the system. This review presents theoretical analyses and practical case studies on the performance requirements of ADCs and their optimization methods in CIM systems, aiming to provide ideas and references for the design and optimization of CIM systems. The review comprehensively explores the relationship between the design of CIM architectures and ADC optimization, and raises the issue of design trade-offs between low power consumption, high speed operation and compact integration design. On this basis, novel customized ADC optimization methods are discussed in depth, and a large number of current CIM systems and their ADC optimization examples are reviewed, with optimization methods summarized and classified in terms of power consumption, speed, and area. In the final part, this review analyzes energy efficiency, ENOB, and frequency scaling trends, demonstrating how advanced processes enable ADCs to balance speed, power, and area trade-offs, guiding ADC optimization for next-gen CIM systems.
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Integrated Circuits and Systems
- Vol. 2 (2) , 81-92
Open AccessAnalog-to-Digital Converter Design for Diverse Performance Computing-in-Memory Systems: A Comprehensive Review
Computing-in-Memory (CIM) architectures have emerged as a pivotal technology for next-generation artificial intelligence (AI) and edge computing applications. By enabling computations directly within memory cells, CIM architectures effectively minimize data movement and significantly enhance energy efficiency. In the CIM system, the analog-to-digital converter (ADC) bridges the gap between efficient analog computation and general digital processing, while influencing the overall accuracy, speed and energy efficiency of the system. This review presents theoretical analyses and practical case studies on the performance requirements of ADCs and their optimization methods in CIM systems, aiming to provide ideas and references for the design and optimization of CIM systems. The review comprehensively explores the relationship between the design of CIM architectures and ADC optimization, and raises the issue of design trade-offs between low power consumption, high speed operation and compact integration design. On this basis, novel customized ADC optimization methods are discussed in depth, and a large number of current CIM systems and their ADC optimization examples are reviewed, with optimization methods summarized and classified in terms of power consumption, speed, and area. In the final part, this review analyzes energy efficiency, ENOB, and frequency scaling trends, demonstrating how advanced processes enable ADCs to balance speed, power, and area trade-offs, guiding ADC optimization for next-gen CIM systems.
- Open Access
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Integrated Circuits and Systems
- Vol. 2 (2) , 49-57
Enhancing LLM Inference Performance on ARM CPUs Through Software and Hardware Co-Optimization Strategies
Large language models (LLMs) have exhibited remarkable performance across a broad spectrum of tasks, yet their extensive computational and memory requirements present substantial challenges for deployment in resource-constrained scenarios. To address the challenges, this work introduces software and hardware co-optimization strategies aimed at enhancing the inference performance of LLMs on ARM CPU-based platforms. A mixed-precision quantization technique is employed, preserving the precision of critical weights to maintain model accuracy while quantizing non-essential weights to INT8, thereby reducing the model's memory footprint. This work also capitalizes on the SIMD instruction set of ARM CPUs to efficiently process model data. Furthermore, the inference framework is optimized by fusing components of the attention computation and streamlining the dequantization process through modifications to the scaling factor. These enhancements result in a significant reduction in model memory usage and improved throughput during the prefill and decode stages. The efficacy of the proposed approach is demonstrated through the optimization of the Qwen-1.8B model on Armv9, with only a 0.66% decrease in accuracy and a reduction in memory usage to 58.8% of the baseline, while achieving a 4.09× and 15.23× increase in inference performance for the prefill and decode stages over the baseline, respectively.
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Integrated Circuits and Systems
- Vol. 2 (2) , 49-57
Open AccessEnhancing LLM Inference Performance on ARM CPUs Through Software and Hardware Co-Optimization Strategies
Large language models (LLMs) have exhibited remarkable performance across a broad spectrum of tasks, yet their extensive computational and memory requirements present substantial challenges for deployment in resource-constrained scenarios. To address the challenges, this work introduces software and hardware co-optimization strategies aimed at enhancing the inference performance of LLMs on ARM CPU-based platforms. A mixed-precision quantization technique is employed, preserving the precision of critical weights to maintain model accuracy while quantizing non-essential weights to INT8, thereby reducing the model's memory footprint. This work also capitalizes on the SIMD instruction set of ARM CPUs to efficiently process model data. Furthermore, the inference framework is optimized by fusing components of the attention computation and streamlining the dequantization process through modifications to the scaling factor. These enhancements result in a significant reduction in model memory usage and improved throughput during the prefill and decode stages. The efficacy of the proposed approach is demonstrated through the optimization of the Qwen-1.8B model on Armv9, with only a 0.66% decrease in accuracy and a reduction in memory usage to 58.8% of the baseline, while achieving a 4.09× and 15.23× increase in inference performance for the prefill and decode stages over the baseline, respectively.
- Open Access
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 720-726
A Layer-wised Mixed-Precision CIM Accelerator with Bit-level Sparsity-aware ADCs for NAS-Optimized CNNs
Exploring multiple precisions as well as sparsities for a computingin-memory (CIM) based convolutional accelerators is challenging. To further improve energy efficiency with minimal accuracy loss, this paper develops a neural architecture search (NAS) method to identify precision for each layer of the CNN and further leverages bit-level sparsity. The results indicate that following this approach, ResNet-18 and VGG-16 not only maintain their accuracy but also implement layer-wised mixed-precision effectively. Furthermore, there is a substantial enhancement in the bit-level sparsity of weights within each layer, with an average bit-level sparsity exceeding 90% per bit, thus providing broader possibilities for hardware-level sparsity optimization. In terms of hardware design, a mixed-precision (2/4/8-bit) readout circuit as well as a bit-level sparsity-aware Analog-to-Digital Converter (ADC) are both proposed to reduce system power consumption. Based on bit-level sparsity mixed-precision CNNs benchmarks, post-layout simulation results in 28nm reveal that the proposed accelerator achieves up to 245.72 TOPS/W energy efficiency, which shows about 2.52 -- 6.57× improvement compared to the state-of-the-art SRAM-based CIM accelerators.
- 20 January 2025
- Published by Association for Computing Machinery (ACM)
- p. 720-726
Open AccessA Layer-wised Mixed-Precision CIM Accelerator with Bit-level Sparsity-aware ADCs for NAS-Optimized CNNs
Exploring multiple precisions as well as sparsities for a computingin-memory (CIM) based convolutional accelerators is challenging. To further improve energy efficiency with minimal accuracy loss, this paper develops a neural architecture search (NAS) method to identify precision for each layer of the CNN and further leverages bit-level sparsity. The results indicate that following this approach, ResNet-18 and VGG-16 not only maintain their accuracy but also implement layer-wised mixed-precision effectively. Furthermore, there is a substantial enhancement in the bit-level sparsity of weights within each layer, with an average bit-level sparsity exceeding 90% per bit, thus providing broader possibilities for hardware-level sparsity optimization. In terms of hardware design, a mixed-precision (2/4/8-bit) readout circuit as well as a bit-level sparsity-aware Analog-to-Digital Converter (ADC) are both proposed to reduce system power consumption. Based on bit-level sparsity mixed-precision CNNs benchmarks, post-layout simulation results in 28nm reveal that the proposed accelerator achieves up to 245.72 TOPS/W energy efficiency, which shows about 2.52 -- 6.57× improvement compared to the state-of-the-art SRAM-based CIM accelerators.
- 9 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Electron Device Letters
- Vol. 46 (2) , 163-166
Ferroelectric Reconfigurable FET Memory on Fully Depleted Silicon-on-Insulator Platform
We experimentally demonstrated ferroelectric reconfigurable field-effect transistor (Fe-RFET) memories on a fully depleted silicon-on-insulator (FDSOI) platform. These devices boast an innovative architecture comprising an electrostatic doping selective gate (SG), source/drain Schottky contacts, and a ferroelectric memory gate (MG). This design enables their dynamic switching between N-type and P-type memory configurations, showcasing substantial readout current ratios peaking at 10 3 /10 5 for N/P-type operations. Moreover, our Fe-RFET memories exhibit good endurance and retention characteristics while inherently implementing reconfigurable in-memory logic functions of “AND” and “NOR.” Such an effective yet easy-to-implemented device strategy presents a promising technological avenue for developing next-generation high-density and energy-efficient memory chips.
- 9 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Electron Device Letters
- Vol. 46 (2) , 163-166
Ferroelectric Reconfigurable FET Memory on Fully Depleted Silicon-on-Insulator Platform
We experimentally demonstrated ferroelectric reconfigurable field-effect transistor (Fe-RFET) memories on a fully depleted silicon-on-insulator (FDSOI) platform. These devices boast an innovative architecture comprising an electrostatic doping selective gate (SG), source/drain Schottky contacts, and a ferroelectric memory gate (MG). This design enables their dynamic switching between N-type and P-type memory configurations, showcasing substantial readout current ratios peaking at 10 3 /10 5 for N/P-type operations. Moreover, our Fe-RFET memories exhibit good endurance and retention characteristics while inherently implementing reconfigurable in-memory logic functions of “AND” and “NOR.” Such an effective yet easy-to-implemented device strategy presents a promising technological avenue for developing next-generation high-density and energy-efficient memory chips.
- 4 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (1) , 370-375
Impact of Scaling Thickness on the Ferroelectric Properties of Pt/Al₀.₈Sc₀.₂N/Pt Capacitors
In this work, the effects of reducing the thickness of Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N films from 100 to 40 nm on the ferroelectric properties of Pt/Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N/Pt capacitors were investigated. As Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N film thickness decreases, remanent polarization significantly drops, and coercive field ( $\textit{E}_{\text{c}}$ ) increases, which origins from the diminished c -axis orientation and reduced crystalline size. Piezoresponse force microscopy results confirm polarization switching in all films, with thinner films exhibiting larger $\textit{E}_{\text{c}}$ and lower initial polarization. Notably, the thinner films show the lower leakage current, attributed to a decrease in nitrogen vacancies. Moreover, thinner Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N films exhibit better endurance performance, with a maximum switching cycle of 10 $^{\text{5}}$ cycles achieved for the 40-nm Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N film. These results underscore the critical role of film thickness in optimizing aluminum scandium nitride (AlScN) thin films for high-performance ferroelectric applications, providing crucial insights for future device development.
- 4 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Electron Devices
- Vol. 72 (1) , 370-375
Impact of Scaling Thickness on the Ferroelectric Properties of Pt/Al₀.₈Sc₀.₂N/Pt Capacitors
In this work, the effects of reducing the thickness of Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N films from 100 to 40 nm on the ferroelectric properties of Pt/Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N/Pt capacitors were investigated. As Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N film thickness decreases, remanent polarization significantly drops, and coercive field ( $\textit{E}_{\text{c}}$ ) increases, which origins from the diminished c -axis orientation and reduced crystalline size. Piezoresponse force microscopy results confirm polarization switching in all films, with thinner films exhibiting larger $\textit{E}_{\text{c}}$ and lower initial polarization. Notably, the thinner films show the lower leakage current, attributed to a decrease in nitrogen vacancies. Moreover, thinner Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N films exhibit better endurance performance, with a maximum switching cycle of 10 $^{\text{5}}$ cycles achieved for the 40-nm Al $_{\text{0.8}}$ Sc $_{\text{0.2}}$ N film. These results underscore the critical role of film thickness in optimizing aluminum scandium nitride (AlScN) thin films for high-performance ferroelectric applications, providing crucial insights for future device development.
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (7) , 3236-3246
A 28-nm 135.19 TOPS/W Bootstrapped-SRAM Compute-in-Memory Accelerator With Layer-Wise Precision and Sparsity
Artificial intelligence (AI) edge devices demand high energy efficiency as well as inference accuracy. SRAM-based compute-in-memory (CIM) accelerators have great potential for power reduction but still need to exploit higher throughput and better linearity performance. To meet edge-AI computing demands by CIM works, it is crucial to optimize algorithms and parameters for specific circuit systems to achieve hardware acceleration. This work firstly employs neural network search (NAS) method to find out the layer-wise optimized precisions and sparsities for convolutional neural networks (CNNs). Then, a 144-Kb charge-domain signed mixed-precision (2/4/8-bit) CIM accelerator employing bootstrapped SRAM cells with 9-transistors and 1-capacitor (9T1C) structure is proposed that incorporates a bit-level sparsity-aware analog-to-digital converter (ADC). This work not only achieves highly linear parallel accumulation operations to meet AI computing demands but also implements a hardware and software co-optimization system tailored to specific data characteristics. The design is verified on NAS-optimized networks VGG-16 and ResNet-18 using Cifar-10 dataset, which could achieve an equivalent accuracy at 4-bit of 68.68% while maintaining a high energy efficiency at 2-bit of 135.19TOPS/W by measurements.
- 15 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (7) , 3236-3246
A 28-nm 135.19 TOPS/W Bootstrapped-SRAM Compute-in-Memory Accelerator With Layer-Wise Precision and Sparsity
Artificial intelligence (AI) edge devices demand high energy efficiency as well as inference accuracy. SRAM-based compute-in-memory (CIM) accelerators have great potential for power reduction but still need to exploit higher throughput and better linearity performance. To meet edge-AI computing demands by CIM works, it is crucial to optimize algorithms and parameters for specific circuit systems to achieve hardware acceleration. This work firstly employs neural network search (NAS) method to find out the layer-wise optimized precisions and sparsities for convolutional neural networks (CNNs). Then, a 144-Kb charge-domain signed mixed-precision (2/4/8-bit) CIM accelerator employing bootstrapped SRAM cells with 9-transistors and 1-capacitor (9T1C) structure is proposed that incorporates a bit-level sparsity-aware analog-to-digital converter (ADC). This work not only achieves highly linear parallel accumulation operations to meet AI computing demands but also implements a hardware and software co-optimization system tailored to specific data characteristics. The design is verified on NAS-optimized networks VGG-16 and ResNet-18 using Cifar-10 dataset, which could achieve an equivalent accuracy at 4-bit of 68.68% while maintaining a high energy efficiency at 2-bit of 135.19TOPS/W by measurements.
- Open Access
- 16 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Circuits and Systems
- Vol. 5, 57-68
A Low-Power DNN Accelerator With Mean-Error-Minimized Approximate Signed Multiplier
Approximate computing is an emerging and effective method for reducing energy consumption in digital circuits, which is critical for energy-efficient performance improvement of edge-computing devices. In this paper, we propose a low-power DNN accelerator with novel signed approximate multiplier based on probability-optimized compressor and error compensation. The probability-optimized compressor is customized for partial product matrix (PPM) of signed operands, which gets the optimal logic circuit after probabilistic analysis and optimization. At the same time, we explored the PPM truncation method, found out the impact of different partial product (PP) truncation numbers on circuit benefit and error, and achieved a more ideal performance-error tradeoff through a reasonable error compensation method. In the optimal case of 8 bits, the proposed approximate multiplier saves 49.84% power, 46.41% area and 24.65% delay compared to the exact multiplier. We employed the proposed approximate multiplier in the vector systolic array as the processing element (PE). Under the VGG-16 evaluation, the proposed accelerator achieves performance improvement of energy efficiency $1.96\times $ , while the error loss was only 0.95%.
- 16 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Circuits and Systems
- Vol. 5, 57-68
Open AccessA Low-Power DNN Accelerator With Mean-Error-Minimized Approximate Signed Multiplier
Approximate computing is an emerging and effective method for reducing energy consumption in digital circuits, which is critical for energy-efficient performance improvement of edge-computing devices. In this paper, we propose a low-power DNN accelerator with novel signed approximate multiplier based on probability-optimized compressor and error compensation. The probability-optimized compressor is customized for partial product matrix (PPM) of signed operands, which gets the optimal logic circuit after probabilistic analysis and optimization. At the same time, we explored the PPM truncation method, found out the impact of different partial product (PP) truncation numbers on circuit benefit and error, and achieved a more ideal performance-error tradeoff through a reasonable error compensation method. In the optimal case of 8 bits, the proposed approximate multiplier saves 49.84% power, 46.41% area and 24.65% delay compared to the exact multiplier. We employed the proposed approximate multiplier in the vector systolic array as the processing element (PE). Under the VGG-16 evaluation, the proposed accelerator achieves performance improvement of energy efficiency $1.96\times $ , while the error loss was only 0.95%.