| 研究実績の概要 |
The goal of this work is to examine if the time-multiplexed reservoir computing strategy can enable non-linear processing for SAW devices. If this is shown to be possible, it would allow high performance real-time learning to be performed on a substrate 10^5 times smaller than competing optical implementations. This would allow high performance implementations of reservoir computer to be added to lightweight and portable devices.
To evaluate the research proposal, custom Surface Acoustic Wave resonators are to be fabricated, and new experimental apparatus designed and implemented. A range of SAW resonators, and other SAW devices are designed, and intended to be tested. The fabrication is possible through use Kyoto University and the Nano-hub’s equipment.
Following successful fabrication of the SAW devices, the reservoir computing system can be tested by designing and building supporting apparatus.
The resulting reservoir computer can then be compared to other implementations through the application of well-established reservoir computing benchmark tasks, which will allow direct comparison of one technology to another. In this case, we select the binary parity task.
A SAW resonator reservoir computer has been successfully implemented, using electronic memory in place of a SAW delay line. This is the first demonstration of a reservoir computer using a SAW devices, and can pave the way to a fully integrated delay line reservoir computer at a physical size which is practical for use in portable electronics.
|
