近日，北京大学杨玉超团队报道了局部有源忆阻振荡器实现可控的复杂行为和频域提取。2025年12月8日，《国家科学评论》杂志发表了这一成果。

莫特跃迁附近的物理非线性显示出神经形态计算的巨大潜力。复杂的计算行为源于其固有的局部活动特性。大多数研究集中于衰变动力学或规则振荡，将莫特器件主要视为简单的阈值元件。通过统一的理论模型将可测量的材料特性与更复杂的器件动力学及其控制方法联系起来仍然存在挑战。

研究组建立了一个基于电测量和局部有源原理的VO₂器件的热力学紧凑模型。利用莫特跃迁附近的非线性，研究组提出了一种基于注入的控制方法来调节非线性振荡器的行为，如分频、随机振荡和频率锁定。最后，在混沌边缘运行的单个设备展示了在物理计算框架内提取频域信息的卓越能力，在相同任务上实现了相当于两层卷积神经网络的性能。这项工作促进了从传统的局部无源器件到局部有源器件的范式转变，弥合了物理非线性、电路动力学和计算理论，以推进动态神经形态计算。

附：英文原文

Title: Local active memristive oscillator enables controllable complex behaviours and frequency domain extraction

Author: Wang, Yanghao, Tiw, Pek Jun, Liu, Yuheng, Tao, Yaoyu, Zhang, Teng, Yang, Yuchao

Issue&Volume: 2025-12-08

Abstract: Physical nonlinearities near the Mott transition exhibit substantial potential for neuromorphic computing. The complex computational behaviour stems from their intrinsic local active characteristics. Most studies focus on decay dynamics or regular oscillations, treating Mott devices primarily as simple threshold elements. Challenges remain in connecting measurable material properties to more complex device dynamics and their control methods through a unified theoretical model. Here, we develop a thermodynamic compact model for VO2 devices based on electrical measurements and the local active principle. Utilizing the nonlinearities near the Mott transition, we propose an injection-based control method to regulate behaviours of nonlinear oscillators, such as frequency division, stochastic oscillations, and frequency locking. Finally, a single device operating at the edge of chaos demonstrates exceptional capability of extracting information in frequency domain within a physical computing framework, achieving performance equivalent to a two-layers convolutional neural network on the same task. This work facilitates a paradigm shift from traditional local passive devices to local active devices, bridging the physical nonlinearities, circuit dynamics and computational theory to advance dynamic neuromorphic computing.

DOI: 10.1093/nsr/nwaf546

Source: https://dx.doi.org/10.1093/nsr/nwaf546