Ag-Ag2Sコアシェル型ナノ粒子を用いたリザーバーコンピューティングデバイスの作製

The performance of the von Neumann computer was greatly improved by miniaturizing transistors and increasing the density according to Moore\u27s Law. However, in recent years, the maximum permissible number of CPU transistors has remained constant, and further performance improvements have not been possible. In today\u27s nanoscale era, scaling to smaller sizes represents a major challenge in device manufacturing, circuit, and system design and integration. On the other hand, nanoscale technology has the potential to develop new materials and devices with unique properties. Memristors exhibit nonlinear current-voltage characteristics and have unique memory characteristics. That is, such a new nanoscale device whose current state depends on the past. It has the potential to create new computing paradigms for both non-linear and memory characteristics of Memristors. The purpose of this paper is to investigate the possibility of using wet chemical synthesis and Ag-Ag2S core-shell nanoparticles to develop a new computing paradigm called “Reservoir Computing” (RC) which belong to such a new paradigm. However, it differs from the traditional Recurrent Neural Network (RNN) method in that the pre-processor (ie, the reservoir) is composed of nonlinear elements that are randomly connected repeatedly. This greatly reduces the complexity of learning. In this thesis, we reported RC devices with low power consumption. The synthesis conditions of Ag-Ag2S core-shell nanoparticles operating at low voltage were searched. Next, synthesis parameters such as Ag / S molar ratio were examined, to control the particle size. We confirmed that the nanoparticle agglomerates have nonlinear electrical conductivity necessary for the development of RC computations, such as constantly exhibiting hysteresis in the current-voltage (I-V) curve, and investigated other conditions necessary for RC hardware. Since the linear regression of the output channel was trained to fit the target waveform, the potential of the nanoparticle-based RC device was shown.九州工業大学博士学位論文 学位記番号:生工博甲第359号 学位授与年月日:令和元年12月27日1 Introduction and Literature Review|2 Methodology|3 Effect of various synthesis procedure to electrical characteristics of the nanoparticles-based device|4 Effect of the Ag-Ag2S volume ratio to the electrical properties|5 Switching mechanism of Ag-Ag2S nanoparticles-based device and neuromorphic learning properties|6 Recurrent neural network properties of Ag-Ag2S nanoparticles-based device and its application as reservoir computing|7 Conclusions and Suggestions九州工業大学令和元年