Chenglong Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Chenglong Li has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Shuisheng Lin
School of Information and Communication Engineering, University of Electronic Science and Technology of China (UESTC), Chengdu, China
3 shared publications - Jun Zhou
University of Electronic Science and Technology of China
2 shared publications - Liang Chang
University of Electronic Science and Technology of China
2 shared publications - Liming Xiu
TAF Circuits Inc., Plano, TX, USA
2 shared publications - Xiangye Wei
School of Software and Microelectronics, Peking University, Beijing, China
2 shared publications - Xin Zhao
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, China
2 shared publications
Latest Publications
- 13 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (5) , 1470-1483
General Purpose Deep Learning Accelerator Based on Bit Interleaving
Along with the rapid evolution of deep neural networks, the ever-increasing complexity imposes formidable computation intensity on the hardware accelerator. In this article, we propose a novel computing philosophy called “bit interleaving” and the associate accelerator couple called “Bitlet” and Bitlet-X to maximally exploit the bit-level sparsity. Apart from the existing bit-serial/parallel accelerators, Bitlet leverages the abundant “sparsity parallelism” in the parameters to enforce the inference acceleration. Bitlet is versatile by supporting diverse precisions on a single platform, including floating-point 32 and fixed-point from$1b$ to$24b$ . The versatility enables Bitlet feasible for both efficient inference and training. Besides, by updating the key compute engine in the accelerator, Bitlet-X could furthermore improve the peak power consumption and efficiency for the inference-only scenario, with competitive accuracy. Empirical studies on 12 domain-specific deep learning applications highlight the following results: 1) up to$81\times /21\times $ energy efficiency improvement for training/inference over recent high-performance GPUs; 2) up to$15\times /8\times $ higher speedup/efficiency over state-of-the-art fixed-point accelerators; 3)$1.5 {\mathrm{ mm}}^{2}$ area and scalable power consumption from 570 mW$(fp32)$ to 432 mW$(16b)$ and 365 mW$(8b) \text{@}28$ -nm TSMC; 4)$1.3\times $ improvement of the peak power efficiency for the Bitlet-X over Bitlet; and 5) highly configurable justified by the ablation and sensitivity studies.- 13 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Vol. 43 (5) , 1470-1483
General Purpose Deep Learning Accelerator Based on Bit Interleaving
Along with the rapid evolution of deep neural networks, the ever-increasing complexity imposes formidable computation intensity on the hardware accelerator. In this article, we propose a novel computing philosophy called “bit interleaving” and the associate accelerator couple called “Bitlet” and Bitlet-X to maximally exploit the bit-level sparsity. Apart from the existing bit-serial/parallel accelerators, Bitlet leverages the abundant “sparsity parallelism” in the parameters to enforce the inference acceleration. Bitlet is versatile by supporting diverse precisions on a single platform, including floating-point 32 and fixed-point from$1b$ to$24b$ . The versatility enables Bitlet feasible for both efficient inference and training. Besides, by updating the key compute engine in the accelerator, Bitlet-X could furthermore improve the peak power consumption and efficiency for the inference-only scenario, with competitive accuracy. Empirical studies on 12 domain-specific deep learning applications highlight the following results: 1) up to$81\times /21\times $ energy efficiency improvement for training/inference over recent high-performance GPUs; 2) up to$15\times /8\times $ higher speedup/efficiency over state-of-the-art fixed-point accelerators; 3)$1.5 {\mathrm{ mm}}^{2}$ area and scalable power consumption from 570 mW$(fp32)$ to 432 mW$(16b)$ and 365 mW$(8b) \text{@}28$ -nm TSMC; 4)$1.3\times $ improvement of the peak power efficiency for the Bitlet-X over Bitlet; and 5) highly configurable justified by the ablation and sensitivity studies.- 20 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (2) , 256-268
IPOCIM: Artificial Intelligent Architecture Design Space Exploration With Scalable Ping-Pong Computing-in-Memory Macro
Computing-in-memory (CIM) architecture has become a possible solution to designing an energy-efficient artificial intelligent processor. Various CIM demonstrators indicated the computing efficiency of CIM macro and CIM-based processors. However, previous studies mainly focus on macro optimization and low CIM capacity without considering the weight update strategy of CIM architecture. The artificial intelligence (AI) processor with a CIM engine practically induces issues, including updating memory data and supporting different operators. For instance, AI-oriented applications usually contain various weight parameters. The weight stored in the CIM architecture should be reloaded for the considerable gap between the capacity of CIM and growing weight parameters. The computation efficiency of the CIM architecture is reduced by the weight updating and waiting. In addition, the natural parallelism of CIM leads to the mismatch of various convolution kernel sizes in different networks and layers, which reduces hardware utilization efficiency. In this work, we develop a CIM engine with a ping-pong computing strategy as an alternative to typical CIM macro and weight buffer, hiding the data update latency and improving the data reuse ratio. Based on the ping-pong engine, we propose a flexible CIM architecture adapting to different sizes of neural networks, namely, intelligent pong computing-in memory (IPOCIM), with a fine-grained data flow mapping strategy. Based on the evaluation, IPOCIM can achieve a 1.27–$6.27\times $ performance and 2.34–$5.30\times $ energy efficiency improvement compared to the state-of-the-art works.- 20 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
- Vol. 32 (2) , 256-268
IPOCIM: Artificial Intelligent Architecture Design Space Exploration With Scalable Ping-Pong Computing-in-Memory Macro
Computing-in-memory (CIM) architecture has become a possible solution to designing an energy-efficient artificial intelligent processor. Various CIM demonstrators indicated the computing efficiency of CIM macro and CIM-based processors. However, previous studies mainly focus on macro optimization and low CIM capacity without considering the weight update strategy of CIM architecture. The artificial intelligence (AI) processor with a CIM engine practically induces issues, including updating memory data and supporting different operators. For instance, AI-oriented applications usually contain various weight parameters. The weight stored in the CIM architecture should be reloaded for the considerable gap between the capacity of CIM and growing weight parameters. The computation efficiency of the CIM architecture is reduced by the weight updating and waiting. In addition, the natural parallelism of CIM leads to the mismatch of various convolution kernel sizes in different networks and layers, which reduces hardware utilization efficiency. In this work, we develop a CIM engine with a ping-pong computing strategy as an alternative to typical CIM macro and weight buffer, hiding the data update latency and improving the data reuse ratio. Based on the ping-pong engine, we propose a flexible CIM architecture adapting to different sizes of neural networks, namely, intelligent pong computing-in memory (IPOCIM), with a fine-grained data flow mapping strategy. Based on the evaluation, IPOCIM can achieve a 1.27–$6.27\times $ performance and 2.34–$5.30\times $ energy efficiency improvement compared to the state-of-the-art works.- 2 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 70 (10) , 9981-9991
Using All-Digital On-Chip Syntonistor to Compensate Clock Frequency Error in Network Time Synchronization for Accuracy Reaching to Sub-µs Range
The emergence of Internet of Everything has made time awareness a feature of high priority. One of the enabling factors is time synchronization. High-quality synchronization demands frequency source of high stability, which is pricey and bulky. When low-end source is used, synchronization quality would be severely impaired. This situation can be improved by the technique of time synchronization. In mainstream approaches, however, the frequency of clock pulse train is not tuned due to the lack of appropriate tool. As requirement for synchronization advances into microsecond or even nanosecond regime, frequency tuning in hardware is expected to be beneficial. Moreover, wireless networks do not have frequency transfer mechanism. This exacerbates the problem of frequency deviation. Those issues make syntonization (synchronization of frequency) become necessary. In this work, an all-digital frequency-tunable clock source is used as on-chip syntonistor to physically compensate the clock frequency errors of nodes in a network. Time discrepancies are hence alleviated. A wireless network has been built to validate this scheme. Time synchronization quality has reached to sub-µs range. The key contribution is the suggestion of adopting on-chip syntonistor for improving the quality of network synchronization. This work serves as a demonstration of this new scheme.
- 2 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 70 (10) , 9981-9991
Using All-Digital On-Chip Syntonistor to Compensate Clock Frequency Error in Network Time Synchronization for Accuracy Reaching to Sub-µs Range
The emergence of Internet of Everything has made time awareness a feature of high priority. One of the enabling factors is time synchronization. High-quality synchronization demands frequency source of high stability, which is pricey and bulky. When low-end source is used, synchronization quality would be severely impaired. This situation can be improved by the technique of time synchronization. In mainstream approaches, however, the frequency of clock pulse train is not tuned due to the lack of appropriate tool. As requirement for synchronization advances into microsecond or even nanosecond regime, frequency tuning in hardware is expected to be beneficial. Moreover, wireless networks do not have frequency transfer mechanism. This exacerbates the problem of frequency deviation. Those issues make syntonization (synchronization of frequency) become necessary. In this work, an all-digital frequency-tunable clock source is used as on-chip syntonistor to physically compensate the clock frequency errors of nodes in a network. Time discrepancies are hence alleviated. A wireless network has been built to validate this scheme. Time synchronization quality has reached to sub-µs range. The key contribution is the suggestion of adopting on-chip syntonistor for improving the quality of network synchronization. This work serves as a demonstration of this new scheme.
- 24 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-10
A Method of Low-Cost Pure-Digital GPS Disciplined Clock for Improving Frequency Accuracy and Steering Frequency Through TAF-DPS Frequency Synthesizer
One of the key characteristics of modern devices is time-awareness, which demands frequency source of good accuracy, precision, and stability. Unfortunately, all high-quality sources are expensive and bulky. An alternative is to add a companion clocking circuit to the primary source for performing frequency compensation in real time, under the guidance of a good time or frequency reference. The companion circuit aims at counteracting the primary source’s systematic error and environment-induced variations. In this work, a frequency synthesizer having the features of small frequency granularity and fast frequency switching is adopted as such companion circuit. It functions as a tuning tool for periodically adjusting device’s output frequency under the supervision of 1 PPS signal locked to global positioning system (GPS). This clock disciplining method can also be used for steering output frequency away from the specified value and toward a user-defined target. The main focus of this method is on improving frequency accuracy. The key advantage of the proposed method is low-cost and high-accessibility. It provides a new perspective, or devises a new instrument, for dealing with the sophisticated task of real-time frequency compensation. Our primary goal is to significantly simplify the engineering effort of this task without sacrificing the quality of clock signal noticeably. An field-programmable gate array (FPGA)-based system is created to demonstrate its capability of frequency compensation and steering. Further, 20 such systems are used to study the precision of a multi-node organization, as an example of using this instrument as a syntonization tool. The handy implementation on FPGA illustrates the advantage of DIY (do it yourself), making this method conveniently applicable for a variety of user scenarios.
- 24 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-10
A Method of Low-Cost Pure-Digital GPS Disciplined Clock for Improving Frequency Accuracy and Steering Frequency Through TAF-DPS Frequency Synthesizer
One of the key characteristics of modern devices is time-awareness, which demands frequency source of good accuracy, precision, and stability. Unfortunately, all high-quality sources are expensive and bulky. An alternative is to add a companion clocking circuit to the primary source for performing frequency compensation in real time, under the guidance of a good time or frequency reference. The companion circuit aims at counteracting the primary source’s systematic error and environment-induced variations. In this work, a frequency synthesizer having the features of small frequency granularity and fast frequency switching is adopted as such companion circuit. It functions as a tuning tool for periodically adjusting device’s output frequency under the supervision of 1 PPS signal locked to global positioning system (GPS). This clock disciplining method can also be used for steering output frequency away from the specified value and toward a user-defined target. The main focus of this method is on improving frequency accuracy. The key advantage of the proposed method is low-cost and high-accessibility. It provides a new perspective, or devises a new instrument, for dealing with the sophisticated task of real-time frequency compensation. Our primary goal is to significantly simplify the engineering effort of this task without sacrificing the quality of clock signal noticeably. An field-programmable gate array (FPGA)-based system is created to demonstrate its capability of frequency compensation and steering. Further, 20 such systems are used to study the precision of a multi-node organization, as an example of using this instrument as a syntonization tool. The handy implementation on FPGA illustrates the advantage of DIY (do it yourself), making this method conveniently applicable for a variety of user scenarios.