Youming Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Youming Yang has a total of 38 co-authors. Here are the top co-authors by number of shared publications:
- Yuan Wang
Peking University
11 shared publications - Yi Zhong
Key Laboratory of Microelectronic Devices and Circuits, MPW Center, School of Integrated Circuits, Peking University, Beijing, China
6 shared publications - Chang Xue
School of Integrated Circuits and the Beijing Advanced Innovation Center for Integrated Circuits, Peking University, Beijing, China
4 shared publications Song Jia
Key Laboratory of Microelectronic Devices and Circuits, MPW Center, School of Integrated Circuits, Peking University, Beijing, China
4 shared publications- Xiaoxin Cui
Peking University Shenzhen Graduate School
4 shared publications - Zilin Wang
Key Laboratory of Microelectronic Devices and Circuits, MPW Center, School of Integrated Circuits, Peking University, Beijing, China
4 shared publications
Latest Publications
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. PP (00189200) , 1-13
MixCIM: A Hybrid Computing-in-Memory Macro With Less Data-Movement and Better Memory-Reuse for Depthwise Separable Neural Networks
Computing-in-memory (CIM) architectures have demonstrated strong potential for edge artificial intelligence (AI) devices due to their enhanced parallelism and energy efficiency. With the growing complexity of AI tasks and the rapid increase in model size, computation and deployment costs have surged. Depthwise separable neural networks (DSNNs) have attracted interest for offering appreciable performance with far fewer parameters. However, the core operation in DSNNs, depthwise separable convolution (DSC), exhibits a different multiply-and-accumulate (MAC) pattern that conventional CIM macros are not optimized to handle efficiently. As a result, deploying DSC on conventional CIM platforms remains challenging. In this article, we propose MixCIM, a mixed macro that integrates computing-near-memory (CNM) and CIM. It achieves high energy efficiency for DSC computations through a mix-memory computing architecture. With a 2’s complement storage-sign magnitude computation (TCS-SMC) encoding, it reduces the activations’ movement and MAC energy. A slidable eDRAM array enhances the parallelism of DW convolution to improve throughput. A memory sharing scheme between input and output activations increases effective memory use. It optimizes complete DSC with both PW and DW convolutions, achieving superior energy efficiency. Measurement results show that the proposed 28-nm macro achieves 40.99 TOPS/W for pointwise convolution (PWC), 15.31 TOPS/W for depthwise convolution (DWC), and 17.2 TOPS/W for DSC, all at signed 8-bit inputs and weights. The accuracy on CIFAR-10 is 92.47% for MobileNet-V2.
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. PP (00189200) , 1-13
MixCIM: A Hybrid Computing-in-Memory Macro With Less Data-Movement and Better Memory-Reuse for Depthwise Separable Neural Networks
Computing-in-memory (CIM) architectures have demonstrated strong potential for edge artificial intelligence (AI) devices due to their enhanced parallelism and energy efficiency. With the growing complexity of AI tasks and the rapid increase in model size, computation and deployment costs have surged. Depthwise separable neural networks (DSNNs) have attracted interest for offering appreciable performance with far fewer parameters. However, the core operation in DSNNs, depthwise separable convolution (DSC), exhibits a different multiply-and-accumulate (MAC) pattern that conventional CIM macros are not optimized to handle efficiently. As a result, deploying DSC on conventional CIM platforms remains challenging. In this article, we propose MixCIM, a mixed macro that integrates computing-near-memory (CNM) and CIM. It achieves high energy efficiency for DSC computations through a mix-memory computing architecture. With a 2’s complement storage-sign magnitude computation (TCS-SMC) encoding, it reduces the activations’ movement and MAC energy. A slidable eDRAM array enhances the parallelism of DW convolution to improve throughput. A memory sharing scheme between input and output activations increases effective memory use. It optimizes complete DSC with both PW and DW convolutions, achieving superior energy efficiency. Measurement results show that the proposed 28-nm macro achieves 40.99 TOPS/W for pointwise convolution (PWC), 15.31 TOPS/W for depthwise convolution (DWC), and 17.2 TOPS/W for DSC, all at signed 8-bit inputs and weights. The accuracy on CIFAR-10 is 92.47% for MobileNet-V2.
- 15 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (12) , 7755-7768
A Bit-Partitioned Floating-Point 6T SRAM Computing-in-Memory Macro Based on Dual-Edge Time-Domain Structure
In the computing-in-memory (CIM) field, floating-point (FP) CIM is afflicted with high computing latency and energy consumption due to the intricate procedures involved in exponent computation and processing. In this work, an 8Kb FP time-domain (TD) static-random-access-memory (SRAM) CIM macro is presented. Fabricated with a 180nm process, this macro exhibits low computational latency and high energy efficiency. A novel FP computing architecture is proposed, which is capable of concurrently executing exponent summation, maximum value finding, difference generation, and mantissa shifting. This architecture effectively reduces the overall delay in exponent computation and processing, thereby enhancing the throughput. Furthermore, a bit-partitioned computing concept and an exponent sparsity scheme are introduced. In this scheme, sparsity judgment is made solely by processing the high 4 bits of the exponent, which significantly reduces power consumption in the remaining exponent computation and processing steps. Additionally, based on the bit-partitioned concept, a dual-edge TD exponent summation and mantissa multiplication-and-accumulation (MAC) circuit is devised. This circuit not only suppresses nonlinear errors during multi-bit computation but also exploits both the rising and falling edges of pulses for computation, thus accelerating the macro’s operation speed. Compared to previous approaches, an extra 24% power reduction is achieved. At a sparsity level of 90%, a normalized energy efficiency of 14.418 TFLOPS/W and a normalized area efficiency of 0.041 TFLOPS/mm2 are attained. When this work is applied to the ResNet-18 model with BF16 format for input, weight, and output, the accuracy loss on the CIFAR-100 dataset is merely −0.16%.
- 15 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 72 (12) , 7755-7768
A Bit-Partitioned Floating-Point 6T SRAM Computing-in-Memory Macro Based on Dual-Edge Time-Domain Structure
In the computing-in-memory (CIM) field, floating-point (FP) CIM is afflicted with high computing latency and energy consumption due to the intricate procedures involved in exponent computation and processing. In this work, an 8Kb FP time-domain (TD) static-random-access-memory (SRAM) CIM macro is presented. Fabricated with a 180nm process, this macro exhibits low computational latency and high energy efficiency. A novel FP computing architecture is proposed, which is capable of concurrently executing exponent summation, maximum value finding, difference generation, and mantissa shifting. This architecture effectively reduces the overall delay in exponent computation and processing, thereby enhancing the throughput. Furthermore, a bit-partitioned computing concept and an exponent sparsity scheme are introduced. In this scheme, sparsity judgment is made solely by processing the high 4 bits of the exponent, which significantly reduces power consumption in the remaining exponent computation and processing steps. Additionally, based on the bit-partitioned concept, a dual-edge TD exponent summation and mantissa multiplication-and-accumulation (MAC) circuit is devised. This circuit not only suppresses nonlinear errors during multi-bit computation but also exploits both the rising and falling edges of pulses for computation, thus accelerating the macro’s operation speed. Compared to previous approaches, an extra 24% power reduction is achieved. At a sparsity level of 90%, a normalized energy efficiency of 14.418 TFLOPS/W and a normalized area efficiency of 0.041 TFLOPS/mm2 are attained. When this work is applied to the ResNet-18 model with BF16 format for input, weight, and output, the accuracy loss on the CIFAR-100 dataset is merely −0.16%.
- 23 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72 (8) , 1088-1092
UniPRE: An SNN-ANN Accelerator With Unified Max-Pooling Prediction and Redundancy Elimination
The integration of Spiking Neural Networks (SNNs) and Artificial Neural Networks (ANNs) for specific tasks has attracted considerable interest due to their potential for high energy efficiency and accuracy. In SNN-ANN fused hardware, many works focus on neuron-level fusion of operators. Though some have explored optimizations at the dataflow level, they are restricted to only one kind of networks. This paper introduces a dataflow-level unified predicting method to eliminate redundant computations resulted from max-pooling operations for both SNN and ANN by exploiting Channel-wise Importance (CI). An accelerator with online sorting of Channel-wise Importance (CI) to support this optimization is also proposed, named as UniPRE. Results show that UniPRE reduces 44.77% and 31.85% overall computations with negligible accuracy loss for SNN and ANN using 37.5% channels for prediction. Implemented in the standard 28-nm CMOS technology, UniPRE can reach an energy efficiency of 19.32 TSOPS/W and 4.26 TOPS/W, with an area efficiency of 370.10 GSOPS/mm2 and 92.52 GOPS/mm2 for SNN and ANN paradigms of 8-bit weight precision, respectively. In layer-wise evaluation of real networks, up to 1.79× energy reduction is achieved with 25% channels used for prediction.
- 23 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 72 (8) , 1088-1092
UniPRE: An SNN-ANN Accelerator With Unified Max-Pooling Prediction and Redundancy Elimination
The integration of Spiking Neural Networks (SNNs) and Artificial Neural Networks (ANNs) for specific tasks has attracted considerable interest due to their potential for high energy efficiency and accuracy. In SNN-ANN fused hardware, many works focus on neuron-level fusion of operators. Though some have explored optimizations at the dataflow level, they are restricted to only one kind of networks. This paper introduces a dataflow-level unified predicting method to eliminate redundant computations resulted from max-pooling operations for both SNN and ANN by exploiting Channel-wise Importance (CI). An accelerator with online sorting of Channel-wise Importance (CI) to support this optimization is also proposed, named as UniPRE. Results show that UniPRE reduces 44.77% and 31.85% overall computations with negligible accuracy loss for SNN and ANN using 37.5% channels for prediction. Implemented in the standard 28-nm CMOS technology, UniPRE can reach an energy efficiency of 19.32 TSOPS/W and 4.26 TOPS/W, with an area efficiency of 370.10 GSOPS/mm2 and 92.52 GOPS/mm2 for SNN and ANN paradigms of 8-bit weight precision, respectively. In layer-wise evaluation of real networks, up to 1.79× energy reduction is achieved with 25% channels used for prediction.
- 8 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 8, 117-120
A Segmented Precision Configurable Computing-in-Memory Macro With Dual-Edge Time-Domain Structure
In computing-in-memory (CIM) architecture, it is necessary to reliably adjust the precision according to the specific demands of the application, enabling a tradeoff between high precision and high energy efficiency. In addition, when performing multibit computations, nonlinearity errors between different bits can adversely affect the network’s accuracy. Therefore, this work proposes an 8Kb dual-edge time-domain CIM macro, which incorporates a segmented precision configuration scheme. By mapping the high and low 4 bits of the 8-bit input to the rising and falling edges of the pulse for independent computation, this design mitigates nonlinearity errors between high and low bits. The precision of multiplication-and-accumulation (MAC) operations for both high and low bits can be independently adjusted, ensuring sufficient accuracy while enhancing energy efficiency. This work attains an energy efficiency ranging from 8.03 to 13.20 TOPS/W in the end. For the CIFAR-10 dataset, when the inputs and weights are of 8-bit precision, this work reaches an inference accuracy of 90.27%–91.92%.
- 8 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Solid-State Circuits Letters
- Vol. 8, 117-120
A Segmented Precision Configurable Computing-in-Memory Macro With Dual-Edge Time-Domain Structure
In computing-in-memory (CIM) architecture, it is necessary to reliably adjust the precision according to the specific demands of the application, enabling a tradeoff between high precision and high energy efficiency. In addition, when performing multibit computations, nonlinearity errors between different bits can adversely affect the network’s accuracy. Therefore, this work proposes an 8Kb dual-edge time-domain CIM macro, which incorporates a segmented precision configuration scheme. By mapping the high and low 4 bits of the 8-bit input to the rising and falling edges of the pulse for independent computation, this design mitigates nonlinearity errors between high and low bits. The precision of multiplication-and-accumulation (MAC) operations for both high and low bits can be independently adjusted, ensuring sufficient accuracy while enhancing energy efficiency. This work attains an energy efficiency ranging from 8.03 to 13.20 TOPS/W in the end. For the CIFAR-10 dataset, when the inputs and weights are of 8-bit precision, this work reaches an inference accuracy of 90.27%–91.92%.
- 30 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (5) , 1793-1804
A Variation-Tolerant Continuous-Time Ising Machine With eDRAM-Based Spin Interaction and Leaked Negative Feedback Annealing
Combinatorial optimization problems (COPs) are essential in various real-world decision-making scenarios, yet most COPs are classified as non-deterministic polynomial-time (NP) hard or complete, posing significant computational challenges. Traditional von Neumann computing architectures often incur high energy costs and prolonged computational latencies when tackling COPs. Recently, researchers have used continuous-time (CT) Ising machines as low-cost COP solvers due to their attractive solving speed. However, prior CT works suffer from inadequate coefficient precision, the lack of annealing mechanisms, and susceptibility to variations. In light of this, this work presents an in-eDRAM mixed-signal CT Ising machine featuring eDRAM-based compute-in-memory (CIM) architecture. It introduces a current-based multi-level interaction topology for updating spins and presents a CT annealing mechanism to access the ground energy landscape. To enhance its process, voltage, and temperature (PVT) robustness, the two-step current-programming technique is used to eliminate mismatches caused by PVT variations. The prototype chip with a$32\times 33$ spin array is fabricated using a 65-nm CMOS process. Each spin occupies an area of 848$\mu $ m2, including 20 3T1C current-programmed eDRAMs and eight 1T1C eDRAMs. Measurement results demonstrate that the proposed CT Ising machine has excellent PVT robustness, the lowest normalized spin area ($13.6~\mu $ m2), and the highest coefficient level of 15. It can solve multi-level max-cut problems on a nanosecond timescale and consumes only 0.33–1.05 nJ at a 1.0–1.2-V core supply voltage.- 30 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (5) , 1793-1804
A Variation-Tolerant Continuous-Time Ising Machine With eDRAM-Based Spin Interaction and Leaked Negative Feedback Annealing
Combinatorial optimization problems (COPs) are essential in various real-world decision-making scenarios, yet most COPs are classified as non-deterministic polynomial-time (NP) hard or complete, posing significant computational challenges. Traditional von Neumann computing architectures often incur high energy costs and prolonged computational latencies when tackling COPs. Recently, researchers have used continuous-time (CT) Ising machines as low-cost COP solvers due to their attractive solving speed. However, prior CT works suffer from inadequate coefficient precision, the lack of annealing mechanisms, and susceptibility to variations. In light of this, this work presents an in-eDRAM mixed-signal CT Ising machine featuring eDRAM-based compute-in-memory (CIM) architecture. It introduces a current-based multi-level interaction topology for updating spins and presents a CT annealing mechanism to access the ground energy landscape. To enhance its process, voltage, and temperature (PVT) robustness, the two-step current-programming technique is used to eliminate mismatches caused by PVT variations. The prototype chip with a$32\times 33$ spin array is fabricated using a 65-nm CMOS process. Each spin occupies an area of 848$\mu $ m2, including 20 3T1C current-programmed eDRAMs and eight 1T1C eDRAMs. Measurement results demonstrate that the proposed CT Ising machine has excellent PVT robustness, the lowest normalized spin area ($13.6~\mu $ m2), and the highest coefficient level of 15. It can solve multi-level max-cut problems on a nanosecond timescale and consumes only 0.33–1.05 nJ at a 1.0–1.2-V core supply voltage.- 18 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (12) , 2293-2302
Marmotini: A Weight Density Adaptation Architecture With Hybrid Compression Method for Spiking Neural Network
Brain-inspired spiking neural network (SNN) has recently attracted widespread interest owing to its event-driven nature and relatively low-power hardware for transmitting highly sparse binary spikes. To further improve energy efficiency, some matrix compression algorithms are used for weight storage. However, the weight sparsity of different layers varies greatly. For a multicore neuromorphic system, it is difficult for the same compression algorithm to adapt to all the layers of SNN model. In this work, we propose a weight density adaptation architecture with hybrid compression method for SNN, named Marmotini. It is a multicore heterogeneous design, including three types of cores to complete computation of different weight sparsity. Benefiting from the hybrid compression method, Marmotini minimizes the waste of neurons and weights as much as possible. Besides, for better flexibility, a reconfigurable core that can be configured to compute convolutional layer or fully connected layer is proposed. Implemented on Xilinx Kintex UltraScale XCKU115 field-programmable gate array (FPGA) board, Marmotini can operate at 150-MHz frequency, achieving 244.6-GSOP/s peak performance and 54.1-GSOP/W energy efficiency at 0% spike sparsity.
- 18 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (12) , 2293-2302
Marmotini: A Weight Density Adaptation Architecture With Hybrid Compression Method for Spiking Neural Network
Brain-inspired spiking neural network (SNN) has recently attracted widespread interest owing to its event-driven nature and relatively low-power hardware for transmitting highly sparse binary spikes. To further improve energy efficiency, some matrix compression algorithms are used for weight storage. However, the weight sparsity of different layers varies greatly. For a multicore neuromorphic system, it is difficult for the same compression algorithm to adapt to all the layers of SNN model. In this work, we propose a weight density adaptation architecture with hybrid compression method for SNN, named Marmotini. It is a multicore heterogeneous design, including three types of cores to complete computation of different weight sparsity. Benefiting from the hybrid compression method, Marmotini minimizes the waste of neurons and weights as much as possible. Besides, for better flexibility, a reconfigurable core that can be configured to compute convolutional layer or fully connected layer is proposed. Implemented on Xilinx Kintex UltraScale XCKU115 field-programmable gate array (FPGA) board, Marmotini can operate at 150-MHz frequency, achieving 244.6-GSOP/s peak performance and 54.1-GSOP/W energy efficiency at 0% spike sparsity.
- 30 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (2) , 651-671
PAICORE: A 1.9-Million-Neuron 5.181-TSOPS/W Digital Neuromorphic Processor With Unified SNN-ANN and On-Chip Learning Paradigm
The neuromorphic approach of fulfilling brain-like edge intelligence is confronted with three paramount challenges: 1) ever-increasing application demands versus insufficient on-chip resources; 2) diverse data sources and models versus heterogeneous computing paradigms; and 3) adaptation to real scenarios versus absence of on-chip learning. To bridge those gaps in the current research community, this article presents PAICORE, a 537.98-mm $^{2}$ digital neuromorphic processor with a unified computing and learning paradigm. PAICORE is a scalable 1024-core design, integrating over 1.919 million neurons and 4.773 billion synapses on a single chip. PAICORE is implemented in globally asynchronous locally synchronous (GALS) style with a five-level fat up–down quadtree as dedicated network-on-chip (NoC) infrastructure and tiled 2-D chip array as inter-chip architecture. The distributed processing cores fuse the hybrid spiking neural network (SNN), artificial neural network (ANN), and on-chip spike-timing-dependent plasticity (STDP) learning models into a unified description by modular-level reconfiguration. PAICORE achieves the peak performance of 20.74 TSOPS, 41.49 TOPS, and 115.25 GSOPS, with the best energy efficiency of 5.181 TSOPS/W, 10.372 TOPS/W, and 1.222 TSOPS/W for SNN, ANN, and on-chip learning paradigms, respectively. Benefiting from the PAIFLOW software framework, PAICORE hardware platform can be equivalently simulated and efficiently programmed in a large-scale deployment, in accordance with specific optimizing demands for its target multi-paradigm tasks.
- 30 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (2) , 651-671
PAICORE: A 1.9-Million-Neuron 5.181-TSOPS/W Digital Neuromorphic Processor With Unified SNN-ANN and On-Chip Learning Paradigm
The neuromorphic approach of fulfilling brain-like edge intelligence is confronted with three paramount challenges: 1) ever-increasing application demands versus insufficient on-chip resources; 2) diverse data sources and models versus heterogeneous computing paradigms; and 3) adaptation to real scenarios versus absence of on-chip learning. To bridge those gaps in the current research community, this article presents PAICORE, a 537.98-mm $^{2}$ digital neuromorphic processor with a unified computing and learning paradigm. PAICORE is a scalable 1024-core design, integrating over 1.919 million neurons and 4.773 billion synapses on a single chip. PAICORE is implemented in globally asynchronous locally synchronous (GALS) style with a five-level fat up–down quadtree as dedicated network-on-chip (NoC) infrastructure and tiled 2-D chip array as inter-chip architecture. The distributed processing cores fuse the hybrid spiking neural network (SNN), artificial neural network (ANN), and on-chip spike-timing-dependent plasticity (STDP) learning models into a unified description by modular-level reconfiguration. PAICORE achieves the peak performance of 20.74 TSOPS, 41.49 TOPS, and 115.25 GSOPS, with the best energy efficiency of 5.181 TSOPS/W, 10.372 TOPS/W, and 1.222 TSOPS/W for SNN, ANN, and on-chip learning paradigms, respectively. Benefiting from the PAIFLOW software framework, PAICORE hardware platform can be equivalently simulated and efficiently programmed in a large-scale deployment, in accordance with specific optimizing demands for its target multi-paradigm tasks.
- 25 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (12) , 6215-6228
NeuroREC: A 28-nm Efficient Neuromorphic Processor for Radar Emitter Classification
Radar emitter classification (REC) plays an important role in modern warfare. Traditional REC methods have difficulty identifying complex radar signals in the present day. Inspired by biology, spiking neural networks (SNNs) have gradually gained widespread attention due to their low power characteristics. Compared with convolutional neural networks (CNNs), SNNs are more suitable for application in the field of REC. The reason is that SNN can not only maintain higher accuracy in the presence of noise interference, but also reduce the power consumption of mobile devices. However, it is challenging to make full use of the input sparsity of radar emitter signals and the weight sparsity of pruned SNN models. In this paper, a 28-nm neuromorphic processor for REC named NeuroREC is proposed. It uses matrix compression algorithms to store sparse weights on chip, and designs corresponding spike detection circuits for this purpose. As a single-core design, we propose a ping-pong running mechanism to alleviate the imbalance between IO throughput and peak performance. Two SNN models for classifying RadioML2016.b and RadioML2018.a datasets are deployed on the chip, achieving competitive accuracy with only 8 timesteps, and demonstrating better robustness than CNN. Fabricated in 28-nm CMOS process, NeuroREC runs at frequencies ranging from 22.5MHz to 744MHz. Under specific sparsity conditions, it can reach an energy efficiency of 7.22TSOP/W for 8-bit weight.
- 25 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems I: Regular Papers
- Vol. 71 (12) , 6215-6228
NeuroREC: A 28-nm Efficient Neuromorphic Processor for Radar Emitter Classification
Radar emitter classification (REC) plays an important role in modern warfare. Traditional REC methods have difficulty identifying complex radar signals in the present day. Inspired by biology, spiking neural networks (SNNs) have gradually gained widespread attention due to their low power characteristics. Compared with convolutional neural networks (CNNs), SNNs are more suitable for application in the field of REC. The reason is that SNN can not only maintain higher accuracy in the presence of noise interference, but also reduce the power consumption of mobile devices. However, it is challenging to make full use of the input sparsity of radar emitter signals and the weight sparsity of pruned SNN models. In this paper, a 28-nm neuromorphic processor for REC named NeuroREC is proposed. It uses matrix compression algorithms to store sparse weights on chip, and designs corresponding spike detection circuits for this purpose. As a single-core design, we propose a ping-pong running mechanism to alleviate the imbalance between IO throughput and peak performance. Two SNN models for classifying RadioML2016.b and RadioML2018.a datasets are deployed on the chip, achieving competitive accuracy with only 8 timesteps, and demonstrating better robustness than CNN. Fabricated in 28-nm CMOS process, NeuroREC runs at frequencies ranging from 22.5MHz to 744MHz. Under specific sparsity conditions, it can reach an energy efficiency of 7.22TSOP/W for 8-bit weight.