Haoming Chu
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
Haoming Chu has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Yuxiang Huan
Guangdong Institute of Intelligence Science and Technology, Zhuhai, Guangdong, China
7 shared publications - Lirong Zheng
Fudan University
6 shared publications - Zhuo Zou
Fudan University
3 shared publications - Hao Jia
State Key Laboratory of ASIC and System, Fudan University, Shanghai, China
2 shared publications - Ning Ma
Guangdong Institue of Intelligence Science and Technology, Zhuhai, Guangdong, China
2 shared publications - Yulong Yan
School of Information Science and Technology, Fudan University, Shanghai, China
2 shared publications
Latest Publications
- Open Access
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
DAP: A Software/Hardware Co-design for Accelerating Multi-chip-based distributed Sparse-Dense Matrix Multiplication
- 26 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-7
Open AccessDAP: A Software/Hardware Co-design for Accelerating Multi-chip-based distributed Sparse-Dense Matrix Multiplication
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (6) , 3018-3031
AFP-CIM: All-Inclusive Floating-Point With Segmented Compute-in-Memory Macro
This article reports an all-inclusive floating-point (AFP) with the segmented 8T-static random access memory (SRAM) compute-in-memory (CIM) macro. It features: 1) a segmented read-wordline (SRWL) to efficiently support the AFP formats including the AFP4/6/8/16 with all of the exponent–mantissa ratios (EMRs); 2) a bit-wise accumulation first (BWAF) circuit structure with a segmented in-memory accumulator (SIMA) for computing the parallel processing of the exponent and mantissa (Parallel-EM) to double the throughput of AFP operations; and 3) a reconfigurable weight-activation ratio (WAR) in AFP-CIM for supporting heterogeneous dataflow within CIM (HD-CIM) to minimize the external memory access (EMA) among various layers and models. Prototyped in 28-nm CMOS, the proposed AFP-CIM exhibits a peak throughput density of 0.9 TFLOPS/mm2 and a peak energy efficiency of 22.7 TFLOPS/W with AFP4 operations. Evaluated with ResNet-50 and ViT-B for ImageNet classification, the AFP-CIM with HD reduces 90.2% and 43.3% EMA, respectively. It achieves the inference accuracy of 75.4% and 80.6% with the AFP4/6/8 operations, respectively.
- 7 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (6) , 3018-3031
AFP-CIM: All-Inclusive Floating-Point With Segmented Compute-in-Memory Macro
This article reports an all-inclusive floating-point (AFP) with the segmented 8T-static random access memory (SRAM) compute-in-memory (CIM) macro. It features: 1) a segmented read-wordline (SRWL) to efficiently support the AFP formats including the AFP4/6/8/16 with all of the exponent–mantissa ratios (EMRs); 2) a bit-wise accumulation first (BWAF) circuit structure with a segmented in-memory accumulator (SIMA) for computing the parallel processing of the exponent and mantissa (Parallel-EM) to double the throughput of AFP operations; and 3) a reconfigurable weight-activation ratio (WAR) in AFP-CIM for supporting heterogeneous dataflow within CIM (HD-CIM) to minimize the external memory access (EMA) among various layers and models. Prototyped in 28-nm CMOS, the proposed AFP-CIM exhibits a peak throughput density of 0.9 TFLOPS/mm2 and a peak energy efficiency of 22.7 TFLOPS/W with AFP4 operations. Evaluated with ResNet-50 and ViT-B for ImageNet classification, the AFP-CIM with HD reduces 90.2% and 43.3% EMA, respectively. It achieves the inference accuracy of 75.4% and 80.6% with the AFP4/6/8 operations, respectively.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (3) , 2681-2693
PicoSleepNet: An Ultra Lightweight Sleep Stage Classification by Spike Neural Network Using Single-Channel EEG Signal
This study introduces PicoSleepNet, an ultra-lightweight sleep stage classification method that utilizes a spiking neural network (SNN) with single-channel electroencephalogram (EEG) signals. Traditional methods use multi-bit Nyquist sampling and dense computing, which result in high complexity and power consumption, hindering their deployment on wearable devices. To address these limitations, we propose an innovative pipeline combining single-bit sub-Nyquist level-crossing sampling (LCS) and sparse computing based on SNN. First, LCS adaptively encodes EEG signals into event-driven spike sequences, reducing data volume by 6.98× while preserving essential signal characteristics compared to Nyquist sampling. Second, a sparse recurrent spiking neural network (RSNN) architecture, optimized by the masked backpropagation and sparse regularization (MaskedBPSR) technique, improves performance and reduces computational costs. Third, quantization-aware training (QAT) ensures that the model maintains high accuracy with low-bitwidth quantization, significantly reducing computational power consumption and enabling hardware-friendly deployment. Compared with current state-of-the-art sleep staging approaches, PicoSleepNet achieves competitive performance on three public datasets (SleepEDF-20, Sleep-EDF-78, and ISRUC-Sleep) with accuracies of 83.5%, 77.9%, 79.4% and macro-F1 scores of 75.2%, 68.1%, 77.2%, respectively. Meanwhile, by leveraging the computational sparsity design of RSNN and the joint optimization of MaskedBPSR and QAT, PicoSleepNet achieves an ultra-lightweight model with only 14.0-25.8K parameters (reduced by nearly 2×) and 681.4-842.0K operations (reduced by 27×), reducing computational power consumption by 1480×. This approach demonstrates the feasibility of deploying ultra-lightweight sleep staging systems in wearable devices and neuromorphic hardware, paving the way for broader applications in real-time health monitoring.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (3) , 2681-2693
PicoSleepNet: An Ultra Lightweight Sleep Stage Classification by Spike Neural Network Using Single-Channel EEG Signal
This study introduces PicoSleepNet, an ultra-lightweight sleep stage classification method that utilizes a spiking neural network (SNN) with single-channel electroencephalogram (EEG) signals. Traditional methods use multi-bit Nyquist sampling and dense computing, which result in high complexity and power consumption, hindering their deployment on wearable devices. To address these limitations, we propose an innovative pipeline combining single-bit sub-Nyquist level-crossing sampling (LCS) and sparse computing based on SNN. First, LCS adaptively encodes EEG signals into event-driven spike sequences, reducing data volume by 6.98× while preserving essential signal characteristics compared to Nyquist sampling. Second, a sparse recurrent spiking neural network (RSNN) architecture, optimized by the masked backpropagation and sparse regularization (MaskedBPSR) technique, improves performance and reduces computational costs. Third, quantization-aware training (QAT) ensures that the model maintains high accuracy with low-bitwidth quantization, significantly reducing computational power consumption and enabling hardware-friendly deployment. Compared with current state-of-the-art sleep staging approaches, PicoSleepNet achieves competitive performance on three public datasets (SleepEDF-20, Sleep-EDF-78, and ISRUC-Sleep) with accuracies of 83.5%, 77.9%, 79.4% and macro-F1 scores of 75.2%, 68.1%, 77.2%, respectively. Meanwhile, by leveraging the computational sparsity design of RSNN and the joint optimization of MaskedBPSR and QAT, PicoSleepNet achieves an ultra-lightweight model with only 14.0-25.8K parameters (reduced by nearly 2×) and 681.4-842.0K operations (reduced by 27×), reducing computational power consumption by 1480×. This approach demonstrates the feasibility of deploying ultra-lightweight sleep staging systems in wearable devices and neuromorphic hardware, paving the way for broader applications in real-time health monitoring.
- 24 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 19 (6) , 1187-1201
A 0.66-mm 2 0.49 pJ/SOP SNN Processor With Temporal-Spatial Post-Neuron-Processing and Model-Adaptive Crossbar in 40-nm CMOS
This paper presents a Spiking Neural Network (SNN) processor specifically designed to overcome the limitations of existing parallel architectures in maintaining high energy efficiency and model adaptability in a compact area footprint for Artificial Intelligence of Things (AIoT). This is achieved through two key design features: a Temporal-Spatial Post-Neuron Processing (PoNP) scheme that efficiently reuses membrane potential, maximizes parallelism, and reduces memory bank requirements; and a Model-Adaptive Crossbar design with preconfigured parameters and a dynamic switching mechanism enables processing of various SNN models through operation orchestration without efficiency degradation. Using an 8-way parallel pipeline design, the processor achieves a throughput of 128 Synaptic Operations (SOPs) per cycle, resulting in a 2.8× enhancement in energy efficiency. Fabricated in a 40-nm CMOS process, the chip occupies a compact core area of 0.66 mm2. It achieves a power consumption of 6.26 mW, an energy efficiency of 0.49 pJ/SOP, and a throughput of 12.8 GSOPS/s at 0.75 V, 100 MHz. The chip is evaluated using typical spatial, temporal, and temporal-spatial datasets, including MIT-BIH, MNIST, N-MNIST, NavGesture, and SHD. These results demonstrate that our chip achieves best-in-class in terms of energy efficiency and latency compared to state-of-the-art architectures.
- 24 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 19 (6) , 1187-1201
A 0.66-mm 2 0.49 pJ/SOP SNN Processor With Temporal-Spatial Post-Neuron-Processing and Model-Adaptive Crossbar in 40-nm CMOS
This paper presents a Spiking Neural Network (SNN) processor specifically designed to overcome the limitations of existing parallel architectures in maintaining high energy efficiency and model adaptability in a compact area footprint for Artificial Intelligence of Things (AIoT). This is achieved through two key design features: a Temporal-Spatial Post-Neuron Processing (PoNP) scheme that efficiently reuses membrane potential, maximizes parallelism, and reduces memory bank requirements; and a Model-Adaptive Crossbar design with preconfigured parameters and a dynamic switching mechanism enables processing of various SNN models through operation orchestration without efficiency degradation. Using an 8-way parallel pipeline design, the processor achieves a throughput of 128 Synaptic Operations (SOPs) per cycle, resulting in a 2.8× enhancement in energy efficiency. Fabricated in a 40-nm CMOS process, the chip occupies a compact core area of 0.66 mm2. It achieves a power consumption of 6.26 mW, an energy efficiency of 0.49 pJ/SOP, and a throughput of 12.8 GSOPS/s at 0.75 V, 100 MHz. The chip is evaluated using typical spatial, temporal, and temporal-spatial datasets, including MIT-BIH, MNIST, N-MNIST, NavGesture, and SHD. These results demonstrate that our chip achieves best-in-class in terms of energy efficiency and latency compared to state-of-the-art architectures.
- 15 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 291-296
Improved EEG Decoding Method Based on Spiking Neural Network for SSVEPs’ Classification
This paper presents an electroencephalogram (EEG) signals decoding method based on an event-driven spiking neural network (SNN) model design for steady-state visual evoked potentials (SSVEP) tasks’ classification and recognition. Adopting the approach of data-driven ensemble empirical mode decomposition (EEMD) of little additional input parameters, the original signals is enhanced to obtain relatively accurate EEG data with valid SSVEP characteristics, which strengthen the signal-to-noise (SNR) at the stimulation frequency. A canonical correlation analysis (CCA) method is considered to be applied to extract frequency characteristics from the multi-lead EEG data, declining the amount of data and training overhead. The training samples with correlation coefficient vectors are sampled into sparse spike sequences through the level crossing (LC) coding and input directly to a resembled excitation–inhibition (E–I) balanced SNN model in the way of event-driven processing. With the sparse data representation, the SNN model adopts the approaches of sparse training and neuromorphic computing to perform pattern recognition for the SNN-compatible spikes. On the Wearable SSVEP-BCI dataset and BETA database provided by Tsinghua University for 12-target and 40-target EEG classification, this SNN method can achieve 96.95% and 89.72% classification accuracy, with 99.66 bits/min and 130.26 bits/min information transfer rate (ITR), respectively.
- 15 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 291-296
Improved EEG Decoding Method Based on Spiking Neural Network for SSVEPs’ Classification
This paper presents an electroencephalogram (EEG) signals decoding method based on an event-driven spiking neural network (SNN) model design for steady-state visual evoked potentials (SSVEP) tasks’ classification and recognition. Adopting the approach of data-driven ensemble empirical mode decomposition (EEMD) of little additional input parameters, the original signals is enhanced to obtain relatively accurate EEG data with valid SSVEP characteristics, which strengthen the signal-to-noise (SNR) at the stimulation frequency. A canonical correlation analysis (CCA) method is considered to be applied to extract frequency characteristics from the multi-lead EEG data, declining the amount of data and training overhead. The training samples with correlation coefficient vectors are sampled into sparse spike sequences through the level crossing (LC) coding and input directly to a resembled excitation–inhibition (E–I) balanced SNN model in the way of event-driven processing. With the sparse data representation, the SNN model adopts the approaches of sparse training and neuromorphic computing to perform pattern recognition for the SNN-compatible spikes. On the Wearable SSVEP-BCI dataset and BETA database provided by Tsinghua University for 12-target and 40-target EEG classification, this SNN method can achieve 96.95% and 89.72% classification accuracy, with 99.66 bits/min and 130.26 bits/min information transfer rate (ITR), respectively.
- 21 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 70 (10) , 4028-4039
A Low-Power Hybrid-Precision Neuromorphic Processor With INT8 Inference and INT16 Online Learning in 40-nm CMOS
In this work, we present a neuromorphic processor for artificial intelligence of things (AIoT) applications featuring low-power consumption, a small footprint, STDP-based online learning, and the ability to adapt to multiple applications. Hybrid precision, i.e., INT8 for inference and INT16 for training, is suggested to achieve balanced accuracy and energy efficiency. A precision-configurable leaky integrate-and-fire(LIF) neuron unit and a unified memory architecture are designed to maximize datapath reuse. A dynamic pruning technique is proposed to exploit the temporal sparsity, yielding synaptic operations reduction by 3.68x in training and 1.63x in inference, respectively. The design is implemented and fabricated in a 40-nm CMOS process, with a core area of 0.87 mm $^2$ . It is measured to consume a minimal power of 680 $\mu$ W at 70 MHz under a 0.75 V power supply, corresponding to 9.9 pJ per synaptic operation. Evaluated with typical spatial, temporal, and spatiotemporal datasets (MNIST, MIT-BIH, and N-MNIST), the proposed design achieve energy efficiency comparable to the best-in-class solutions with handcrafted training and customized ASICs, while demonstrating improved versatility across multiple applications with balanced accuracy, power consumption, and model adaptability.
- 21 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 70 (10) , 4028-4039
A Low-Power Hybrid-Precision Neuromorphic Processor With INT8 Inference and INT16 Online Learning in 40-nm CMOS
In this work, we present a neuromorphic processor for artificial intelligence of things (AIoT) applications featuring low-power consumption, a small footprint, STDP-based online learning, and the ability to adapt to multiple applications. Hybrid precision, i.e., INT8 for inference and INT16 for training, is suggested to achieve balanced accuracy and energy efficiency. A precision-configurable leaky integrate-and-fire(LIF) neuron unit and a unified memory architecture are designed to maximize datapath reuse. A dynamic pruning technique is proposed to exploit the temporal sparsity, yielding synaptic operations reduction by 3.68x in training and 1.63x in inference, respectively. The design is implemented and fabricated in a 40-nm CMOS process, with a core area of 0.87 mm $^2$ . It is measured to consume a minimal power of 680 $\mu$ W at 70 MHz under a 0.75 V power supply, corresponding to 9.9 pJ per synaptic operation. Evaluated with typical spatial, temporal, and spatiotemporal datasets (MNIST, MIT-BIH, and N-MNIST), the proposed design achieve energy efficiency comparable to the best-in-class solutions with handcrafted training and customized ASICs, while demonstrating improved versatility across multiple applications with balanced accuracy, power consumption, and model adaptability.
- Top Cited Papers
- 8 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 16 (4) , 511-523
A Neuromorphic Processing System With Spike-Driven SNN Processor for Wearable ECG Classification
This paper presents a neuromorphic processing system with a spike-driven spiking neural network (SNN) processor design for always-on wearable electrocardiogram (ECG) classification. In the proposed system, the ECG signal is captured by level crossing (LC) sampling, achieving native temporal coding with single-bit data representation, which is directly fed into an SNN in an event-driven manner. A hardware-aware spatio-temporal backpropagation (STBP) is suggested as the training scheme to adapt to the LC-based data representation and to generate lightweight SNN models. Such a training scheme diminishes the firing rate of the network with little plenty of classification accuracy loss, thus reducing the switching activity of the circuits for low-power operation. A specialized SNN processor is designed with the spike-driven processing flow and hierarchical memory access scheme. Validated with field programmable gate arrays (FPGA) and evaluated in 40 nm CMOS technology for application-specific integrated circuit (ASIC) design, the SNN processor can achieve 98.22% classification accuracy on the MIT-BIH database for 5-category classification, with an energy efficiency of 0.75 $\mu$ J/classification.
- 8 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Circuits and Systems
- Vol. 16 (4) , 511-523
Top Cited PapersA Neuromorphic Processing System With Spike-Driven SNN Processor for Wearable ECG Classification
This paper presents a neuromorphic processing system with a spike-driven spiking neural network (SNN) processor design for always-on wearable electrocardiogram (ECG) classification. In the proposed system, the ECG signal is captured by level crossing (LC) sampling, achieving native temporal coding with single-bit data representation, which is directly fed into an SNN in an event-driven manner. A hardware-aware spatio-temporal backpropagation (STBP) is suggested as the training scheme to adapt to the LC-based data representation and to generate lightweight SNN models. Such a training scheme diminishes the firing rate of the network with little plenty of classification accuracy loss, thus reducing the switching activity of the circuits for low-power operation. A specialized SNN processor is designed with the spike-driven processing flow and hierarchical memory access scheme. Validated with field programmable gate arrays (FPGA) and evaluated in 40 nm CMOS technology for application-specific integrated circuit (ASIC) design, the SNN processor can achieve 98.22% classification accuracy on the MIT-BIH database for 5-category classification, with an energy efficiency of 0.75 $\mu$ J/classification.