Non-linear processing with a surface acoustic wave reservoir computer
Non-linear processing with a surface acoustic wave reservoir computer
复制标题
DOI：
10.1007/s00542-023-05463-4
复制
发表时间：
2023-05
期刊：
影响因子：
--
通讯作者：
Claude Meffan;Taiki Ijima;Amit Banerjee;J. Hirotani;T. Tsuchiya
中科院分区：
文献类型：
--
作者：
Claude Meffan;Taiki Ijima;Amit Banerjee;J. Hirotani;T. Tsuchiya
水库计算是一种神经网络算法，它减少了神经网络功能所需的训练。最近，已经使用具有普遍的非线性动力学的MEMS设备来实现水库计算，以执行非线性处理任务。虽然在过去进行了部分探索，但人们对使用声表面波器件作为低能量射频处理器重新产生了兴趣。然而，他们还没有被探索的水库计算框架。本文设计并测量了一个基于化学还原YZ铌酸锂的39.16MHz双端口声表面波谐振器。测量了器件的品质因数、插入损耗、线性传输和非线性传输，并讨论了这些特性与水库计算的关系。SAW谐振器，然后配置为一个时间复用水库，它的非线性处理能力进行了讨论，使用时间延迟奇偶校验基准。
Reservoir computing is a neural network algorithm that reduces the training needed for a neural network to be function. Recently, reservoir computing has been implemented using MEMs devices with prevalent non-linear dynamics to perform non-linear processing tasks. While partially explored in the past, there has been renewed interest in using Surface Acoustic Wave devices as low energy radio-frequency processors. However they have yet to be explored in the reservoir computing framework. In this work, a 39.16 MHz two-port SAW resonator on chemically reduced YZ Lithium Niobate is design and measured. The quality factor, insertion loss, linear transmission, and non-linear transmission of the devices is measured, and the relationship of these properties to reservoir computing is discussed. The SAW resonator is then configured as a time-multiplexed reservoir, and it’s non-linear processing capabilities are discussed using the time-delayed parity benchmark.