【公開日：2023.08.01】【最終更新日：2023.05.22】
課題データ / Project Data
課題番号 / Project Issue Number
22KT1250
利用課題名 / Title
Reservoir Computing with Surface Acoustic Wave Resonators
利用した実施機関 / Support Institute
京都大学
機関外・機関内の利用 / External or Internal Use
内部利用（ARIM事業参画者以外）/Internal Use (by non ARIM members)
技術領域 / Technology Area
【横断技術領域 / Cross-Technology Area】（主 / Main）加工・デバイスプロセス/Nanofabrication（副 / Sub）-
【重要技術領域 / Important Technology Area】（主 / Main）マルチマテリアル化技術・次世代高分子マテリアル/Multi-material technologies / Next-generation high-molecular materials（副 / Sub）高度なデバイス機能の発現を可能とするマテリアル/Materials allowing high-level device functions to be performed
キーワード / Keywords
Reservoir Computing, Surface Acoustic Waves ,リソグラフィ/Lithography,CVD,3D積層技術/ 3D lamination technology
利用者と利用形態 / User and Support Type
利用者名（課題申請者）/ User Name (Project Applicant)
Meffan Robert Claude
所属名 / Affiliation
京都大学 大学院工学研究科
共同利用者氏名 / Names of Collaborators in Other Institutes Than Hub and Spoke Institutes
ARIM実施機関支援担当者 / Names of Collaborators in The Hub and Spoke Institutes
利用形態 / Support Type
（主 / Main）機器利用/Equipment Utilization（副 / Sub）,技術代行/Technology Substitution
利用した主な設備 / Equipment Used in This Project
KT-205：プラズマCVD装置
KT-154：両面マスクアライナー露光装置
KT-111：ウエハスピン洗浄装置
KT-110：レジスト現像装置
KT-103：レーザー直接描画装置
報告書データ / Report
概要（目的・用途・実施内容）/ Abstract (Aim, Use Applications and Contents)
Reservoir computing is a neural network algorithm that intends to improve the power efficiency of neuralnetworks by reducing the required training. Recently, non-linear MEMs sensors have been used as non-linear neurons in reservoir computers. This works aims to use a non-linear SAW resonator device as a reservoir computing node to facilitate transmission band radio frequency neural network processing. This work has applications in RF communication, but can also be applied to high speed preprocessing
実験 / Experimental
SAW resonators were fabricated on chemically reduced YZ-LiNbO3 (Yamaju Ceramics Co., Ltd.) using a standard image reversal lift-off procedure (AZ5214E, pre-bake at 110◦C, 50s, 8.66 mJ.cm−2 exposure, image reversal bake at 117◦C for 2 min, flood exposure). The wafers were then diced into individual SAW resonators using a dicing SAW (Disco, DAD3231) and glued onto a custom PCB for measurement. The non-linear processing capacity of the resonators was evaluated using the time delayed binary parity task. To perform this task, the basic resonator structure was supported by external electronic memory. In this case, provide by a small micro-computer (CY8CKIT-050B, Infineon). In future iterations this will be replaced with a SAW delay structure.
結果と考察 / Results and Discussion
The results of the binary parity task are shown in Figure 1, below. The reservoir was able to predict up to parity order 5 with a 95% accuracy. The high-performance parameter space was centered inside the bandwidth of the resonant cavity as anticipated. However, interestingly, driving frequencies near the edges of the SAW devices bandwidth seemed to perform better than those nearer to the center. When the total Memory Capacity of the reservoir was evaluated, it was found to have a Memory Capacity of ~4.8 Bits, this indicated that the neural network contained information from 5 time steps prior. In the future, we seek to further increase this memory capacity through parameter optimization.
図・表・数式 / Figures, Tables and Equations
Figure 1 The performance of the SAW reservoir computer on the binary parity task. a) The performance of the SAW-RC on binary parity tasks of order 2-7. The area of high performance is centered inside the bandwidth of the resonant peak. b) The prediction performance of the reservoir computer, as the order (difficulty) of the task is increased, the certainty of the prediction is decreased.
その他・特記事項（参考文献・謝辞等） / Remarks(References and Acknowledgements)
成果発表・成果利用 / Publication and Patents
論文・プロシーディング（DOIのあるもの） / DOI (Publication and Proceedings)
-
Robert Claude Meffan, Non-linear processing with a Surface Acoustic Wave reservoir computer, , , (2023).
DOI: 10.21203/rs.3.rs-2430313/v1
口頭発表、ポスター発表および、その他の論文 / Oral Presentations etc.
- (1) “A non-linear SAW resonator as a reservoir computing node” Claude Meffan, Amit Banjeree, Jun Hirotani and Toshiyuki Tsuchiya, 2022 JSME-IIP/ASME-ISPS Joint International Conference on Micromechatronics for Information and Precision Equipment (MIPE 2022), August 28-31, 2022, Nagoya University, Japan. (Hybrid event)
特許 / Patents
特許出願件数 / Number of Patent Applications：0件
特許登録件数 / Number of Registered Patents：0件