arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2305.00741physics.app-ph
收藏

具有内部记忆的人工纳米光子神经元,用于生物启发和储备池网络计算

Artificial Nanophotonic Neuron with Internal Memory for Biologically Inspired and Reservoir Network Computing

David Winge, Magnus Borgström, Erik Lind, Anders Mikkelsen

AI总结:

本文提出并建模了一种基于III-V族纳米线、具有电荷限时记忆的纳米光子神经节点,通过模拟储备池网络和昆虫脑中央复合网络,证明可变记忆衰减时间能显著提升网络性能并具备现实制造容差下的可行性。

摘要:

具有内部记忆的神经元已被提议用于生物和生物启发神经网络,从而增加了有趣的功能。我们提出并建模了一种纳米尺度光电神经节点,该节点具有基于电荷的限时记忆和信号评估能力。节点之间的连接通过发射和接收的加权光信号实现。该器件基于成熟的III-V族纳米线技术,该技术已展示出高光电转换效率、低能耗和亚波长光聚集能力。我们利用纳米线性能的实验值,建立了完整人工神经节点器件的灵活计算模型。该模型可以模拟具有不同硬件衍生属性和广泛可变光互连的节点组合。利用该模型,我们模拟了两种类型神经网络的硬件实现。首先,我们表明节点记忆衰减时间的有意变化可以显著提高储备池网络的性能。其次,我们模拟了纳米线节点实现昆虫脑中央复合网络的解剖学约束功能模型,并发现即使在包含现实器件制造中会出现的节点性能变化的情况下,该网络也能良好运行。我们的工作证明了一种具体的、可变的、具有记忆的纳米光子神经节点的可行性。使用可变记忆时间常数为网络性能开辟新机会是一种通用的硬件衍生特征,应适用于广泛的实现方式。

英文摘要:

Neurons with internal memory have been proposed for biological and bio-inspired neural networks, adding interesting functionality. We propose and model a nanoscale optoelectronic neural node with charge-based time-limited memory and signal evaluation. Connectivity is achieved by weighted light signals emitted and received by the nodes. The device is based on well-developed III-V nanowire technology, which has shown high photo-conversion efficiency, low energy consumption and sub-wavelength light concentration. We create a flexible computational model of the complete artificial neural node device using experimental values for wire performance. The model can simulate combinations of nodes with different hardware derived properties and widely variable light interconnects. Using this model, we simulate the hardware implementation for two types of neural networks. First, we show that intentional variations in the memory decay time of the nodes can significantly improve the performance of a reservoir network. Second, we simulate the nanowire node implementing an anatomically constrained functioning model of the central complex network of the insect brain and find that it functions well even including variations in the node performance as would be found in realistic device fabrication. Our work demonstrates the feasibility of a concrete, variable, nanophotonic neural node with a memory. The use of variable memory time constants to open new opportunities for network performance is a general hardware derived feature and should be applicable for a broad range of implementations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑