Xiaoyu Feng
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
Xiaoyu Feng has a total of 60 co-authors. Here are the top co-authors by number of shared publications:
- Yongpan Liu
Tsinghua University
8 shared publications - Huazhong Yang
Tsinghua University
7 shared publications - Jinshan Yue
Institute of Microelectronics of the Chinese Academy of Sciences, Beijing, China
4 shared publications - Chen Tang
Department of Electronic Engineering, Tsinghua University, Beijing, China
2 shared publications - Hedi Wang
Tsinghua University
2 shared publications - Jingyu Wang
Department of Electronic Engineering, Tsinghua University, Beijing, China
2 shared publications
Latest Publications
- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (1) , 197-210
An Edge 3-D Gaussian Splatting Processor Based on Shape-Aware Computation Architecture and Spatio-Temporal Gaussian Cache
Three-dimensional Gaussian splatting (3D GS) is capturing attention due to its widespread applications in cutting-edge fields. Compared with the traditional neural radiance field (NeRF)-based methods, 3D GS demonstrates superior rendering realism and speed on high-performance graphics processing units (GPUs). However, 3D GS still faces challenges on edge devices. First, rasterization in 3D GS requires a large number of operators and a high bitwidth. This resource-intensive dataflow leads to significant power consumption and area overhead. Moreover, the extensive input-dependent ineffective computation in 3D GS causes unnecessary computational delays. Finally, the irregular and repeated access pattern of off-chip Gaussian parameters presents challenges for external memory access (EMA). To address these challenges, this work presents an energy-efficient edge 3D GS processor featuring: 1) a shape-aware reconfigurable rasterization–interpolation computation architecture reducing area and power overhead; 2) an online shape-aware early skipping controller (SESC) which efficiently detects ineffective computation; and 3) a spatio-temporal Gaussian cache to alleviate EMA overhead. The fabricated silicon prototype achieves a maximum throughput of 373 fps and a minimum rendering energy of$0.58~\text {mJ}/\text {frame}$ . Compared with state-of-the-art (SOTA) rendering accelerators, it achieves$1.96\times $ higher performance and$6.72\times $ lower energy consumption.- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 61 (1) , 197-210
An Edge 3-D Gaussian Splatting Processor Based on Shape-Aware Computation Architecture and Spatio-Temporal Gaussian Cache
Three-dimensional Gaussian splatting (3D GS) is capturing attention due to its widespread applications in cutting-edge fields. Compared with the traditional neural radiance field (NeRF)-based methods, 3D GS demonstrates superior rendering realism and speed on high-performance graphics processing units (GPUs). However, 3D GS still faces challenges on edge devices. First, rasterization in 3D GS requires a large number of operators and a high bitwidth. This resource-intensive dataflow leads to significant power consumption and area overhead. Moreover, the extensive input-dependent ineffective computation in 3D GS causes unnecessary computational delays. Finally, the irregular and repeated access pattern of off-chip Gaussian parameters presents challenges for external memory access (EMA). To address these challenges, this work presents an energy-efficient edge 3D GS processor featuring: 1) a shape-aware reconfigurable rasterization–interpolation computation architecture reducing area and power overhead; 2) an online shape-aware early skipping controller (SESC) which efficiently detects ineffective computation; and 3) a spatio-temporal Gaussian cache to alleviate EMA overhead. The fabricated silicon prototype achieves a maximum throughput of 373 fps and a minimum rendering energy of$0.58~\text {mJ}/\text {frame}$ . Compared with state-of-the-art (SOTA) rendering accelerators, it achieves$1.96\times $ higher performance and$6.72\times $ lower energy consumption.- Open Access
- 8 August 2025
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 138 (6)
Transforming photodiode into avalanche photodiode: An efficient method improving the sensitivity performance of foundry-fabricated Si/Ge receiver
Avalanche photodiodes (APDs) are essential semiconductor optoelectronic devices that could effectively improve the optical receiver sensitivity due to the internal gain. Based on Si/Ge photodiodes (PDs) from standard foundry process design kits, this work demonstrates a simple tuning method to transform PDs into APDs and further improve optical receiver sensitivity by carefully increasing the bias voltage before breakdown. The tuning principle is demonstrated theoretically and experimentally. A useful method distinguishing whether a foundry-fabricated PD can be tuned to the APD is proposed via confirming if the tunneling breakdown happens before avalanche. Subsequently, based on the foundry-fabricated Si/Ge PD, the receiver sensitivity is improved by 5 dB via increasing bias voltage from 2 to 12 V under the non-return-to-zero bit rate of 10 Gbps, at the incident power of −16 dBm and gain of 3 the bit error rate is suppressed by eight orders of magnitude.
- 8 August 2025
- Published by AIP Publishing in Journal of Applied Physics
- Vol. 138 (6)
Open AccessTransforming photodiode into avalanche photodiode: An efficient method improving the sensitivity performance of foundry-fabricated Si/Ge receiver
Avalanche photodiodes (APDs) are essential semiconductor optoelectronic devices that could effectively improve the optical receiver sensitivity due to the internal gain. Based on Si/Ge photodiodes (PDs) from standard foundry process design kits, this work demonstrates a simple tuning method to transform PDs into APDs and further improve optical receiver sensitivity by carefully increasing the bias voltage before breakdown. The tuning principle is demonstrated theoretically and experimentally. A useful method distinguishing whether a foundry-fabricated PD can be tuned to the APD is proposed via confirming if the tunneling breakdown happens before avalanche. Subsequently, based on the foundry-fabricated Si/Ge PD, the receiver sensitivity is improved by 5 dB via increasing bias voltage from 2 to 12 V under the non-return-to-zero bit rate of 10 Gbps, at the incident power of −16 dBm and gain of 3 the bit error rate is suppressed by eight orders of magnitude.
- 18 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (3) , 1002-1013
A Scalable BEV Perception Processor for Image/Point Cloud Fusion Applications Using CAM-Based Universal Mapping Unit
The integration of multi-sensor data like image and point cloud for information complementarity is crucial for 3-D perception scenarios like autonomous driving. Recently, bird’s eye view (BEV)-based sensor fusion is attracting more and more attention but the significant computational overhead constrains their widespread application at the edge. First, there are numerous irregular memory access operations in BEV fusion networks. For example, sparse convolutions (SCONVs) in the point cloud branch and irregular BEV plane mapping result in significant memory addressing and mapping overhead. Furthermore, multi-sensor fusion leads to rapid expansion of model size, making it difficult and expensive for single-chip solutions to meet the demands. Based on the above challenges, this work proposes an image and point cloud fusion processor with two highlights: a content addressable memory (CAM)-based deep fusion core to accelerate a variety of irregular BEV operations and chip-level parallelism design supporting flexible interconnect topology. The proposed chip is fabricated in 28-nm CMOS technology. Compared with existing image or point cloud accelerators, the proposed chip achieves higher frequency, $2\times$ higher area efficiency, and $2.61\times$ higher energy efficiency for sparse point cloud processing. To the best of authors’ knowledge, this work is the first accelerator for BEV-based multi-modal fusion networks.
- 18 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 60 (3) , 1002-1013
A Scalable BEV Perception Processor for Image/Point Cloud Fusion Applications Using CAM-Based Universal Mapping Unit
The integration of multi-sensor data like image and point cloud for information complementarity is crucial for 3-D perception scenarios like autonomous driving. Recently, bird’s eye view (BEV)-based sensor fusion is attracting more and more attention but the significant computational overhead constrains their widespread application at the edge. First, there are numerous irregular memory access operations in BEV fusion networks. For example, sparse convolutions (SCONVs) in the point cloud branch and irregular BEV plane mapping result in significant memory addressing and mapping overhead. Furthermore, multi-sensor fusion leads to rapid expansion of model size, making it difficult and expensive for single-chip solutions to meet the demands. Based on the above challenges, this work proposes an image and point cloud fusion processor with two highlights: a content addressable memory (CAM)-based deep fusion core to accelerate a variety of irregular BEV operations and chip-level parallelism design supporting flexible interconnect topology. The proposed chip is fabricated in 28-nm CMOS technology. Compared with existing image or point cloud accelerators, the proposed chip achieves higher frequency, $2\times$ higher area efficiency, and $2.61\times$ higher energy efficiency for sparse point cloud processing. To the best of authors’ knowledge, this work is the first accelerator for BEV-based multi-modal fusion networks.
- 26 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (9) , 3070-3081
A 28-nm Energy-Efficient Sparse Neural Network Processor for Point Cloud Applications Using Block-Wise Online Neighbor Searching
Voxel-based point cloud networks composed of multiple kinds of sparse convolutions (SCONVs) play an essential role in emerging applications such as autonomous driving and visual navigation. Many researchers have proposed sparse processors for image applications. However, they cannot properly deal with three problems in the point cloud, including low efficiency of random memory access, non-parallel neighbor search and area overhead of supporting hybrid operators, and unbalanced workload among multiple cores. In this work, a 2-D/3-D unified SCONV accelerator is proposed with three key features: a block-wise sparse data storage format supporting out-of-order memory allocation and continuous memory access; a high-throughput and reconfigurable SCONV core providing unified support for multiple kinds of sparse CNNs; an asynchronous and synchronous hybrid scheduler for multiple cores with dynamic on-chip memory router to maximize data reusing and core utilization. This chip is fabricated in 28-nm CMOS technology and achieves 4.68-TOPS/W peak energy efficiency, 2 $\times $ higher than the previous accelerator. It is also the first accelerator to provide unified 2-D/3-D support and end-to-end inference ability for voxel-based point cloud networks.
- 26 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (9) , 3070-3081
A 28-nm Energy-Efficient Sparse Neural Network Processor for Point Cloud Applications Using Block-Wise Online Neighbor Searching
Voxel-based point cloud networks composed of multiple kinds of sparse convolutions (SCONVs) play an essential role in emerging applications such as autonomous driving and visual navigation. Many researchers have proposed sparse processors for image applications. However, they cannot properly deal with three problems in the point cloud, including low efficiency of random memory access, non-parallel neighbor search and area overhead of supporting hybrid operators, and unbalanced workload among multiple cores. In this work, a 2-D/3-D unified SCONV accelerator is proposed with three key features: a block-wise sparse data storage format supporting out-of-order memory allocation and continuous memory access; a high-throughput and reconfigurable SCONV core providing unified support for multiple kinds of sparse CNNs; an asynchronous and synchronous hybrid scheduler for multiple cores with dynamic on-chip memory router to maximize data reusing and core utilization. This chip is fabricated in 28-nm CMOS technology and achieves 4.68-TOPS/W peak energy efficiency, 2 $\times $ higher than the previous accelerator. It is also the first accelerator to provide unified 2-D/3-D support and end-to-end inference ability for voxel-based point cloud networks.
- 30 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (5) , 1612-1627
An Energy-Efficient Computing-in-Memory NN Processor With Set-Associate Blockwise Sparsity and Ping-Pong Weight Update
Computing-in-memory (CIM) chips have demonstrated the potential high energy efficiency for low-power neural network (NN) processors. Even with energy-efficient CIM macros, the existing system-level CIM chips still lack deep exploration on sparsity and large models, which prevents a higher system energy efficiency. This work presents a CIM NN processor with more sufficient support of sparsity and higher utilization rate. Three key innovations are proposed. First, a set-associate blockwise sparsity strategy is designed, which simultaneously saves execution time, power, and storage space. Second, a ping-pong weight update mechanism is proposed for a higher utilization rate, enabling simultaneous execution of CIM and write operations. Third, an efficient CIM macro is implemented with adaptive analog-digital converter (ADC) precision for better sparsity utilization and performance-accuracy trade-off. The 65-nm fabricated chip shows 9.5-TOPS/W system energy efficiency at 4-bit precision, with 6.25 $\times $ actual improvement compared with a state-of-the-art CIM chip. Besides, this work supports high CIM execution accuracy on the ImageNet dataset.
- 30 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Solid-State Circuits
- Vol. 59 (5) , 1612-1627
An Energy-Efficient Computing-in-Memory NN Processor With Set-Associate Blockwise Sparsity and Ping-Pong Weight Update
Computing-in-memory (CIM) chips have demonstrated the potential high energy efficiency for low-power neural network (NN) processors. Even with energy-efficient CIM macros, the existing system-level CIM chips still lack deep exploration on sparsity and large models, which prevents a higher system energy efficiency. This work presents a CIM NN processor with more sufficient support of sparsity and higher utilization rate. Three key innovations are proposed. First, a set-associate blockwise sparsity strategy is designed, which simultaneously saves execution time, power, and storage space. Second, a ping-pong weight update mechanism is proposed for a higher utilization rate, enabling simultaneous execution of CIM and write operations. Third, an efficient CIM macro is implemented with adaptive analog-digital converter (ADC) precision for better sparsity utilization and performance-accuracy trade-off. The 65-nm fabricated chip shows 9.5-TOPS/W system energy efficiency at 4-bit precision, with 6.25 $\times $ actual improvement compared with a state-of-the-art CIM chip. Besides, this work supports high CIM execution accuracy on the ImageNet dataset.
- 11 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-2
A Demonstration Platform for Large-Scaled Point Cloud Network Based on 28nm 2D/3D Unified Sparse Convolution Accelerator
3D point cloud processing plays an important role in many emerging applications such as autonomous driving, visual navigation, and virtual reality. It calls for hardware acceleration of multiple key operations, including 3D Submanifold SCONV, 3D non-Submanifold SCONV, and 2D SCONV. This work presents a 2D/3D unified sparse convolution accelerator for large-scale voxel-based point cloud networks. The chip is fabricated in TSMC 28nm CMOS technology to achieve 3.3-16.9 FPS running from 60-400MHz when computing the SECOND network on KITTI dataset. This work has been included by ISSCC2023 [1]. A demonstration is given to show the real-time 3D processing with a lidar sensor.
- 11 June 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-2
A Demonstration Platform for Large-Scaled Point Cloud Network Based on 28nm 2D/3D Unified Sparse Convolution Accelerator
3D point cloud processing plays an important role in many emerging applications such as autonomous driving, visual navigation, and virtual reality. It calls for hardware acceleration of multiple key operations, including 3D Submanifold SCONV, 3D non-Submanifold SCONV, and 2D SCONV. This work presents a 2D/3D unified sparse convolution accelerator for large-scale voxel-based point cloud networks. The chip is fabricated in TSMC 28nm CMOS technology to achieve 3.3-16.9 FPS running from 60-400MHz when computing the SECOND network on KITTI dataset. This work has been included by ISSCC2023 [1]. A demonstration is given to show the real-time 3D processing with a lidar sensor.
- 27 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (2) , 416-420
An RRAM-Based Digital Computing-in-Memory Macro With Dynamic Voltage Sense Amplifier and Sparse-Aware Approximate Adder Tree
RRAM is a promising candidate to implement large-capacity in-memory computing on edge AI devices due to its high density. However, the efficiency and accuracy of RRAM-based computing-in-memory (CIM) works are limited by large accumulation currents and device variations. The SRAM-based digital CIM achieves superior performance and efficiency while the adder tree dominates the area. In this paper, a digital RRAM CIM macro is proposed to achieve a better trade-off between accuracy, energy, and performance by three techniques. First, a dynamic voltage sense amplifier is designed to reduce >90% read currents of low resistance state (LRS) cell. Second, an OR gate approximate adder tree is proposed to reduce the area of the adder tree by 40%. Third, a sparse-aware finetuning algorithm is proposed to reduce the accuracy loss of approximate arithmetic to 0.4% on Cifar-10 dataset. The proposed design achieves 1966TOPS/W energy efficiency and 51.2GOPS/Kb normalized throughput at 1-bit precision which is 1.2× and 1.8× higher than previous RRAM-based designs. This work demonstrates the advantage of digital CIM using RRAM devices.
- 27 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 70 (2) , 416-420
An RRAM-Based Digital Computing-in-Memory Macro With Dynamic Voltage Sense Amplifier and Sparse-Aware Approximate Adder Tree
RRAM is a promising candidate to implement large-capacity in-memory computing on edge AI devices due to its high density. However, the efficiency and accuracy of RRAM-based computing-in-memory (CIM) works are limited by large accumulation currents and device variations. The SRAM-based digital CIM achieves superior performance and efficiency while the adder tree dominates the area. In this paper, a digital RRAM CIM macro is proposed to achieve a better trade-off between accuracy, energy, and performance by three techniques. First, a dynamic voltage sense amplifier is designed to reduce >90% read currents of low resistance state (LRS) cell. Second, an OR gate approximate adder tree is proposed to reduce the area of the adder tree by 40%. Third, a sparse-aware finetuning algorithm is proposed to reduce the accuracy loss of approximate arithmetic to 0.4% on Cifar-10 dataset. The proposed design achieves 1966TOPS/W energy efficiency and 51.2GOPS/Kb normalized throughput at 1-bit precision which is 1.2× and 1.8× higher than previous RRAM-based designs. This work demonstrates the advantage of digital CIM using RRAM devices.
- 1 February 2018
- Published by Elsevier in Biochemical and Biophysical Research Communications
- Vol. 497 (1) , 272-278
Dietary nitrate protects submandibular gland from hyposalivation in ovariectomized rats via suppressing cell apoptosis
- 1 February 2018
- Published by Elsevier in Biochemical and Biophysical Research Communications
- Vol. 497 (1) , 272-278
Dietary nitrate protects submandibular gland from hyposalivation in ovariectomized rats via suppressing cell apoptosis
- 1 August 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 819-822
Data Acquisition and Transmission Reliability for Wireless Image Sensor Networks
Wireless image sensor networks acquire and transmit image data through self-organized method. Considering processing ability, energy consumption and cost, this paper implements a practical image sensor node with low-power sleeping mechanism in order to achieve image data's gathering, processing and transmission. To ensure transmission reliability of image data in wireless channel, this paper improves the weakness of retransmission mechanism in traditional Zigbee protocol, and proposes an efficient Zigbee-based retransmission and acknowledgment mechanism. Experimental results show the relationships among the total retransmission time, packet loss rate and the maximum number of retransmission.
- 1 August 2014
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 819-822
Data Acquisition and Transmission Reliability for Wireless Image Sensor Networks
Wireless image sensor networks acquire and transmit image data through self-organized method. Considering processing ability, energy consumption and cost, this paper implements a practical image sensor node with low-power sleeping mechanism in order to achieve image data's gathering, processing and transmission. To ensure transmission reliability of image data in wireless channel, this paper improves the weakness of retransmission mechanism in traditional Zigbee protocol, and proposes an efficient Zigbee-based retransmission and acknowledgment mechanism. Experimental results show the relationships among the total retransmission time, packet loss rate and the maximum number of retransmission.