学位论文简介

随着人工智能与自主系统的发展，高维动态系统控制正面临计算复杂度高与实时性强的双重挑战。传统数字计算受限于存储与计算分离架构，存在数据搬移开销大、延迟高和能效低等问题，难以满足复杂控制任务需求。基于忆阻器的模拟计算电路可在模拟域内直接实现高效计算，具有快速求解和免数据搬运等优势，但仍面临通用设计不足、系统应用不够及非理想因素影响等问题。为此，本文从基础矩阵算子、核心控制算法及完整控制系统三个层面展开了系统研究，主要创新成果如下：

(1) 针对传统数字架构求解高维矩阵方程时面临的存储墙严重、搬运开销大、能效低等问题，提出了面向矩阵计算的模拟电路设计方法，构建了由数学定义、物理方程到电路架构逐层贯通的实现路径，为大规模矩阵计算提供了高能效、可扩展的硬件设计方案。

(2) 针对高维动态系统中Kalman滤波、线性二次型调节器（LQR）及模型预测控制（MPC）等核心控制算法在边缘计算场景下面临的延迟与能效瓶颈，提出了控制算法全流程的模拟域执行电路架构。经系统评估表明，在不同维度控制系统下，所设计的电路在器件编程误差、输入信号噪声及线电阻等非理想性因素干扰下仍保持90%以上计算精度，展现了强鲁棒性。

(3) 针对传统控制架构在未知干扰下适应性不足、缺乏实时约束等问题，本文提出了融合强化学习与线性约束MPC的模拟计算控制框架。该方法利用强化学习智能体实时修正参考轨迹，并结合线性约束MPC实现轨迹跟踪；同时将策略网络与MPC求解器软硬协同设计部署于忆阻阵列和模拟电路中，实现算法的全模拟域执行。实验表明，该框架在六自由度无人机轨迹高频跟踪任务中具有良好的鲁棒性与高效性。

主要学术成果

[1] Pingdan Xiao, Zhengmiao Wei, Sichun Du, Wanli Chang, Qinghui Hong. ACIM-QMM: Efficient analog computing-in-memory accelerator for QC-MDPC McEliece cryptosystem [C]. Proceedings of the 62nd ACM/IEEE Design Automation Conference (DAC), 2025: 1–6. (CCF A类会议，第一作者)

[2] Pingdan Xiao, Du Sichun, Gu Yiliu, Bingqian Zhang, Wanli Chang, Qinghui Hong. RM-CIMA: An analog computing-in-memory accelerator for a robust UAV trajectory tracking framework [C]. Proceedings of the 63rd ACM/IEEE Design Automation Conference (DAC), 2026. (已接收，CCF A类会议，共同第一作者)

[3] Sen Zhang⁺, Pingdan Xiao⁺, Qinghui Hong, Lin Tang, Zhengdao Xie, Rui He, Bei Jiang, Xitong Hong, Xinjie Li, Haodi Zhu, Ruohao Hong, Chang Liu, Xingqiang Liu, Yawei Lv, Yang Chai, Lei Liao, Xuming Zou. Rational design of a vision fusion system with visible and near-infrared spectral integration for improved environmental perception [J]. National Science Review, 2025, 12(7): nwaf204. (中科院一区，共同第一作者)

[4] Pingdan Xiao⁺, Yiliu Gu⁺, Haoyou Jiang, Zhen Huan, Sichun Du, Qinghui Hong. Memristive neural network circuit implementation of model predictive control for trajectory tracking [J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2026, 45(2): 955–968.(中科院一区，共同第一作者)

[5] Pingdan Xiao⁺, Meimei Ma⁺, Haoyou Jiang, Zhen Huan, Sichun Du, Qinghui Hong. Analog solver design of LU decomposition algorithm for accelerating public color watermarking [J]. IEEE Transactions on Industrial Informatics, 2026, 22(2): 969–980. (中科院一区，共同第一作者)

[6] Pingdan Xiao⁺, Qinghui Hong, Sichun Du, Jiliang Zhang. Analog-in-memory accelerator design based on memristive arrays for opposite directional interference alignment algorithm [J]. IEEE Transactions on Industrial Informatics, 2024, 20(3): 3628–3638. (中科院一区，第一作者)

[7] Pingdan Xiao, Qinghui Hong, Jingru Sun, Sichun Du, Jiliang Zhang. Design and application of programmable analog circuit for solving Lyapunov matrix equation based on memristors [J]. IEEE Transactions on Industrial Electronics, 2024, 71(7): 7786–7795. (中科院一区，第一作者)

[8] Pingdan Xiao, Junjie Fang, Zhengmiao Wei, Yu Dong, Sichun Du, Shiping Wen, Qinghui Hong. A Riccati matrix equation solver design based neurodynamics method and its application [J]. IEEE Transactions on Automation Science and Engineering, 2025, 22: 15163–15176. (中科院二区，高被引论文，第一作者)

[9] Pingdan Xiao, Qinghui Hong, Sichun Du, Jiliang Zhang. CIM-KF: Efficient computing-in-memory circuits for full-process execution of Kalman Filter algorithm [C]. Proceedings of the 53rd International Conference on Parallel Processing (ICPP), 2024: 1032–1041. (CCF B类会议，第一作者)

[10] Pingdan Xiao, Sichun Du, Zhengmiao Wei, Qinghui Hong, Shiping Wen. Design and application of programmable analog computing circuit for Kalman Filter algorithm based on memristive array [J]. IEEE Transactions on Circuits and Systems for Artificial Intelligence, 2024, 1(2): 272–284. (第一作者)

[11] 肖平旦, 洪庆辉, 杜四春, 孙辰, 黎俊, 张吉良. 基于忆阻器的可编程矩阵行列式求解电路设计及应用 [J]. 中国科学: 信息科学, 2023, 53(5): 1008–1025. (中文CCF A类期刊，第一作者)

[12] Qinghui Hong, Pingdan Xiao, Ruijia Fan, Sichun Du. Memristive neural network circuit design based on locally competitive algorithm for sparse coding application [J]. Neurocomputing, 2024, 578: 127369. (中科院二区，导师一作学生二作)

[13] Pingdan Xiao, Bingqian Zhang, Sichun Du, Qinghui Hong. An Analog Matrix Computing Scheme for QC-MDPC McEliece Cryptosystem based on Memristive Array [J]. IEEE Transactions on Computers. (CCF A类论文，第一作者，大修)

[14] Pingdan Xiao, Bingqian Zhang, Pengcheng Zeng, Haoyou Jiang, Sichun Du, Qinghui Hong. Real-Time DC-OPF Solution via a Memristive Neural Network Circuit Approach[J]. IEEE Transactions on Circuits and Systems for Artificial Intelligence. (第一作者，审稿中)

[15] Pingdan Xiao, Yiliu Gu, Hegan Chen, Bingqian Zhang, Sichun Du, Qinghui Hong. PDE-ACiM: A Cross-Layer Co-Design of Physics-informed Analog Computing-in-Memory Accelerator for Efficient PDE Solving. (第一作者，拟提交)