王美琪

一、 个人简介
二、 研究领域
本人的主要研究领域为面向多样化人工智能应用需求的集成电路和智能系统设计。人工智能(AI)技术在当今社会应用广泛,近几年大模型等技术的爆发驱使数据规模和模型大小不断增长,对AI 算法高效运行的算力需求也急剧提高。本人的研究聚焦 AI 算法部署中面向多样化需求的重点问题,从“面向AI通用计算的微架构深度优化”,“面向成熟AI模型的数据-算法-硬件协同加速”,和“面向AI新兴需求的算法优化与硬件设计”三个方面展开研究,对AI专用硬件处理器的速度、功耗、时延、灵活性、鲁棒性等重要特性进行优化,推动AI算法在自动驾驶,机器人,AI内容生成(AIGC),数字孪生等领域中的落地应用。
在上述研究领域中,侧重沿着以下的技术路径展开研究:
-
软硬协同优化:通用计算领域的系统软硬件抽象分层设计限制了AI等应用在硬件部署时的效率提升,软硬协同优化通过多层次协同的优化思路,在算法设计中充分考虑硬件计算和访存特性的适配性,并针对优化后的算法进行专用硬件架构设计,可最大化提升AI系统的效率。本人的研究工作包括面向AI专用处理器的模型压缩与轻量化网络设计、高能效FPGA/ASIC架构及加速器设计,面向大模型高效推理/训练的软硬协同优化等。
-
存算一体:随着AI应用和云计算的发展,数据存取慢、存取能耗大等问题成为了系统的关键瓶颈,“存储墙”、“通信墙”等问题成为了AI算法高效部署的一大障碍。存算一体技术通过在存储器中嵌入计算能力,大幅减少非必要的数据搬移延迟和功耗,可显著提升AI计算能效,降低成本。本人的研究工作包括基于SRAM的存算一体硬件架构设计,面向新兴网络的存算一体架构设计等。
三、 教育背景
2014年 - 2023年, 南京大学,学士、博士;北京大学,访学
四、 工作经历
2023.06至今,中山大学,助理教授、硕士生导师
2019.11-2020.11,南京风兴科技有限公司,算法部/硬件部,负责AI模型压缩和加速器设计相关工作
五、 部分代表性成果
- [DATE 25] D. Zou, G. Zhang, K. Sun, W. Zhe, M. Wang*, Z. Wang*, "LLM4GEMMV: A Flexible Performance-Aware LLM-based Verilog Generation Framework For GEMM". in Design, Automation and Test in Europe Conference (DATE), 2025
- [ASPDAC 25] K. Sun, M. Wang*, J. Zhou and Z. Wang*, "UEDA: A Universal And Efficient Deformable Attention Accelerator For Various Vision Tasks," in Asia and South Pacific Design Automation Conference (ASP-DAC), 2025
- [TVLSI 24] D. Zou, G. Zhang, X. Zhang, M. Wang* and Z. Wang*, "An Efficient and Precision-Reconfigurable Digital CIM Macro for DNN Accelerators," in IEEE Transactions on Very Large Scale Integration (VLSI) Systems, doi: 10.1109/TVLSI.2024.3455091.
- [TCAS-II 23] X. Wu, S. Liang, M. Wang* and Z. Wang*, "ReAFM: A Reconfigurable Nonlinear Activation Function Module for Neural Networks," in IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 70, no. 7, pp. 2660-2664, July 2023.8.
- [TCAS-II 22] M. Wang, X. Cheng, D. Zou and Z. Wang, "FACCU: Enable Fast Accumulation for High-Speed DSP Systems," in IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 69, no. 12, pp. 4634-4638, Dec. 2022.
- [TII 22] M. Wang#, T. Su#, S. Chen, W. Yang, J. Liu and Z. Wang, "Automatic Model-Based Dataset Generation for High-Level Vision Tasks of Autonomous Driving in Haze Weather," in IEEE Transactions on Industrial Informatics (TII), doi: 10.1109/TII.2022.3224958
- [TNNLS 21] M. Wang, L. He, J. Lin and Z. Wang, "Rethinking Adaptive Computing: Building a Unified Model Complexity-Reduction Framework with Adversarial Robustness," in IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 4, pp. 1803-1810, Nov. 2021.
- [JETCAS 20] M. Wang, Z. Wang, J. Lu, J. Lin and Z. Wang, "E-LSTM: An Efficient Hardware Architecture for Long Short-Term Memory," in IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 9, no. 2, pp. 280-291, June 2019.
- [SiPS 19] M. Wang, J. Mo, J. Lin, and Z. Wang, "DynExit: A Dynamic Early-Exit Strategy for Deep Residual Networks", IEEE International Workshop on Signal Processing Systems (SiPS), 2019 (最佳论文奖)
- [SOCC 20] S. Lu, M. Wang, S. Liang, and Z. Wang, "Hardware Accelerator for Multi-Head Attention and Position-Wise Feed-Forward in the Transformer", IEEE International System-on-Chip Conference (SOCC), 2020 (最佳论文奖)