APP下载

纯克尔非线性动力学在光学神经元中的应用研究

2024-07-08龙洪亮沈微宏SCHOENHARDTSteffen张启明

光学仪器 2024年3期

龙洪亮 沈微宏 SCHOENHARDT Steffen 张启明

摘要:采用尖峰神经元对神经形态计算进行并行处理,可以克服数字计算机的一些局限性,从而提高计算速度、计算性能和能量效率,实现更高效、更智能和更自适应的计算。超快克尔效应在光子神经形态计算中具有重要的意义和应用。利用改进的归一化耦合模理论(coupledmode theory, CMT)模型,对钙钛矿材料的光学微谐振腔中由超快克尔效应引起的非线性动力学特征进行了计算和分析,观察到自脉动行为;模拟并实现了光学神经元的兴奋性行为、泄露积分动力学和不应期现象。钙钛矿材料具有超快的克尔响应时间,可以将神经元的不应期控制在皮秒量级,为实现快速的脉冲神经网络提供了新思路。

关键词:克尔效应;耦合模式理论;微谐振腔;光学神经元

中图分类号: O 436.1 文献标志码: A

Research on the application of pure Kerr nonlineardynamics in optical neurons

LONG Hongliang1,2,SHEN Weihong1,SCHOENHARDT Steffen1,ZHANG Qiming1

(1. Institute of Photonic Chips, University of Shanghai for Science and Technology, Shanghai 200093, China;

2. School of Optical-Electrical and Computer Engineering, University of Shanghai forScience and Technology, Shanghai 200093, China)

Abstract: Neuromorphiccomputingusesspikeneuronsforparallelprocessing,whichcan overcome limitations of digital computers, thereby improving computing speed, performance, and energy efficiency to achieve more efficient, intelligent, and adaptive computing. The ultrafast Kerr effect has important significance and application in neuromorphic computing. An improved coupled mode theory (CMT) model was employed to analyzed the nonlinear dynamics behaviors in a micro- resonator based on the ultra-fast Kerr response of perovskite materials. Self-pulsation behavior was observed in the optical cavity. Based on this property, the excitation, leaky integrating dynamicsand refractory time of optical neurons were simulated and implemented. Because of the ultra-fast Kerr response of the perovskite material, the refractory time of the optical neuron can be reduced to the order of picoseconds, which paves the way for ultra-fast spiking neural networks.

Keywords: Kerr effect; coupled mode theory; micro-resonator; optical neurons

引言

自从1943年 McCulloch 和 Pitts 正式提出神经元的设计以来,神经形态计算一直是人工智能领域中的一个重要研究方向,它旨在模仿生物神经网络结构和功能[1]。神经形态计算的目标是建立一种类似于生物大脑的系统,包括架构、数据处理方法和功能,并实现比传统计算机更快速、准确的分析数据集的能力,同时使用更少的计算资源[2]。在实现神经形态计算过程中,设计出模拟神经元的硬件具有特别重要的作用。而尖峰神经元则为实现神经形态计算提供了重要的手段,因为尖峰神经元具有快速响应、低功耗等优势,所以它能够更加高效地模拟生物神经元的行为,从而实现更为精确和快速的神经形态计算[3]。……

登录APP查看全文