面向固定时间控制的忆阻模糊神经形态系统的分析与设计

The computing capability of current microprocessors faces the bottleneck of the Von-Neumann architecture and the restriction of Moore's Law, which makes traditional chips approach the theoretical performance limits. The emergence of neuromorphic chips provides novel methods to solve these two problems. Memristor has state memory characteristics and nonlinear electrical properties, it is an ideal device to realize the synaptic function of a neuron. This project attempts to employ theories of differential inclusion, the comparison strategy, and fuzzy analysis to investigate fixed-time stabilization of memristive fuzzy neuromorphic systems and fixed-time synchronization of coupled memristive fuzzy neuromorphic systems, considering the effect of fixed-time behavior from hybrid time-varying delays and topology structure, it provides theoretical approaches and new techniques for fixed-time analysis of memristor-based fuzzy neuromorphic systems. Meanwhile, by utilizing the nonvolatility and the feature of continuous multi-resistance of memristor, the multi-valued dynamical storage technology of neuromorphic systems based upon memristive crossbar arrays and fuzzy logics is designed, the system model and storage technology are optimized. Combining with simulation software, the neuron synapse is designed based on memristor, via incorporating fuzzy logic inference, memristive fuzzy neuromorphic circuit can be built to validate the theoretical results. The research of this project will play positive roles in complex system analysis, fixed-time control, multi-value storage, and optimization design.

现代微处理器计算能力面临冯诺依曼架构的瓶颈和摩尔定律的限制，使得传统芯片已经接近理论性能极限。神经形态芯片的出现为解决这两方面难题提供了新思路。忆阻器具有状态记忆特性和非线性电学性能，是实现神经元突触功能的理想器件。本项目拟利用微分包含理论、比较原理和模糊分析，研究忆阻模糊神经形态系统的固定时间镇定和耦合忆阻模糊神经形态系统的固定时间同步，考虑混合多时滞与拓扑结构对系统固定时间行为的影响，为基于忆阻的模糊神经形态系统的固定时间分析提供理论方法与新技术。同时，利用忆阻的非易失性和连续多阻值特性，设计基于忆阻交叉阵列和模糊逻辑的神经形态系统的多值动态存储技术，并优化系统模型与存储技术。结合仿真软件，设计基于忆阻的神经元突触，融合模糊逻辑推理，搭建忆阻模糊神经形态电路以验证理论结果。本项目的研究将对复杂系统分析、固定时间控制、多值存储与优化设计等研究方向产生积极作用。

本项目从时滞神经网络出发，利用比较方法和不等式技巧研究了忆阻神经网络的全局吸引性和全局指数镇定，分析了双向联想记忆神经网络的全局稳定性。设计模糊间歇控制策略，在Filippov微分包含意义下讨论了模糊忆阻神经网络的全局指数镇定。通过构建有限时间比较函数考虑了时滞神经网络的有限时间控制，给出了收敛时间估计，与现有结果相比简化了控制器设计与证明过程。依托本项目共发表6篇IEEE汇刊常文，其中3篇IEEE Transactions on Cybernetics，2篇IEEE Transactions on Neural Networks and Learning Systems，1篇IEEE Transactions on Systems, Man, and Cybernetics: Systems。项目主持人2020年11月入职华中科技大学人工智能与自动化学院，担任副教授和硕士生导师，2022年9月获批国家自然科学基金面上项目，2022年11月获亚太神经网络学会青年研究学者奖。