Fan Cui
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Fan Cui has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Kexing Zhou
School of Integrated Circuits, Peking University, Beijing, China
2 shared publications - Youwei Xiao
School of Integrated Circuits, Peking University, Beijing, China
2 shared publications - Yun Liang
School of Integrated Circuits, Peking University, Beijing, China
2 shared publications - Chenyang Yin
Peking University, Beijing, China
1 shared publication - Dahua Lin
The Chinese University of Hong Kong, Hong Kong, China
1 shared publication - Demin Song
Shanghai AI Laboratory, Shanghai, China
1 shared publication
Latest Publications
- Open Access
- 27 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 53
An Empirical Comparision of LLM-based Hardware Design and High-level Synthesis
Field-Programmable Gate Arrays (FPGAs) are increasingly used for accelerating diverse applications due to their reconfigurability and ability to implement custom hardware architectures. However, programming FPGAs remains challenging, traditionally relying on low-level Hardware Description Languages (HDLs) like Verilog, which are intricate and time-consuming. High-Level Synthesis (HLS) tools, such as Vitis HLS, have emerged to address these issues by allowing hardware functionality description in high-level languages like C/C++, but they come with their own limitations, including less efficient hardware implementations, delay overhead caused by conservative scheduling strategies, and unpredictable solutions due to semantic differences between software and hardware. This paper explores the potential of Large Language Models (LLMs) in FPGA design, particularly for generating complex Verilog kernels. We present a novel approach that guides LLMs to generate synthesizable and efficient Verilog code for complex FPGA kernels. Our method addresses key challenges in LLM-based hardware design. Through a case study on the PolyBench suite, we demonstrate that our LLM-guided approach can generate HDL implementations that surpass HLS tools in performance and resource utilization. The experimental results show that our approach reduces latency by an average of 28.88% compared to HLS, with a maximum reduction of 66.94% in pipelined designs and an average reduction of 8.41%, peaking at 55.76% in sequential stages. Furthermore, it decreases LUT usage by an average of 25.18% and flip-flop usage by 57.23% compared to HLS.
- 27 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 53
Open AccessAn Empirical Comparision of LLM-based Hardware Design and High-level Synthesis
Field-Programmable Gate Arrays (FPGAs) are increasingly used for accelerating diverse applications due to their reconfigurability and ability to implement custom hardware architectures. However, programming FPGAs remains challenging, traditionally relying on low-level Hardware Description Languages (HDLs) like Verilog, which are intricate and time-consuming. High-Level Synthesis (HLS) tools, such as Vitis HLS, have emerged to address these issues by allowing hardware functionality description in high-level languages like C/C++, but they come with their own limitations, including less efficient hardware implementations, delay overhead caused by conservative scheduling strategies, and unpredictable solutions due to semantic differences between software and hardware. This paper explores the potential of Large Language Models (LLMs) in FPGA design, particularly for generating complex Verilog kernels. We present a novel approach that guides LLMs to generate synthesizable and efficient Verilog code for complex FPGA kernels. Our method addresses key challenges in LLM-based hardware design. Through a case study on the PolyBench suite, we demonstrate that our LLM-guided approach can generate HDL implementations that surpass HLS tools in performance and resource utilization. The experimental results show that our approach reduces latency by an average of 28.88% compared to HLS, with a maximum reduction of 66.94% in pipelined designs and an average reduction of 8.41%, peaking at 55.76% in sequential stages. Furthermore, it decreases LUT usage by an average of 25.18% and flip-flop usage by 57.23% compared to HLS.
- Open Access
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-9
OriGen: Enhancing RTL Code Generation with Code-to-Code Augmentation and Self-Reflection
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-9
Open AccessOriGen: Enhancing RTL Code Generation with Code-to-Code Augmentation and Self-Reflection