|
" Machine Learning Augmentation Micro-Sensors for Smart Device Applications "
Hasan, Mohammad H.
Alsaleem, Fadi
| Document Type
|
:
|
Latin Dissertation
|
| Language of Document
|
:
|
English
|
| Record Number
|
:
|
1108638
|
| Doc. No
|
:
|
TLpq2467600671
|
| Main Entry
|
:
|
Alsaleem, Fadi
|
|
|
:
|
Hasan, Mohammad H.
|
| Title & Author
|
:
|
Machine Learning Augmentation Micro-Sensors for Smart Device Applications\ Hasan, Mohammad H.Alsaleem, Fadi
|
| College
|
:
|
The University of Nebraska - Lincoln
|
| Date
|
:
|
2020
|
| student score
|
:
|
2020
|
| Degree
|
:
|
Ph.D.
|
| Page No
|
:
|
143
|
| Abstract
|
:
|
Novel smart technologies such as wearable devices and unconventional robotics have been enabled by advancements in semiconductor technologies, which have miniaturized the sizes of transistors and sensors. These technologies promise great improvements to public health. However, current computational paradigms are ill-suited for use in novel smart technologies as they fail to meet their strict power and size requirements. In this dissertation, we present two bio-inspired colocalized sensing-and-computing schemes performed at the sensor level: continuous-time recurrent neural networks (CTRNNs) and reservoir computers (RCs). These schemes arise from the nonlinear dynamics of micro-electro-mechanical systems (MEMS), which facilitates computing, and the inherent ability of MEMS devices for sensing. Furthermore, this dissertation addresses the high-voltage requirements in electrostatically actuated MEMS devices using a passive amplification scheme. The CTRNN architecture is emulated using a network of bistable MEMS devices. This bistable behavior is shown in the pull-in, the snapthrough, and the feedback regimes, when excited around the electrical resonance frequency. In these regimes, MEMS devices exhibit key behaviors found in biological neuronal populations. When coupled, networks of MEMS are shown to be successful at classification and control tasks. Moreover, MEMS accelerometers are shown to be successful at acceleration waveform classification without the need for external processors. MEMS devices are additionally shown to perform computing by utilizing the RC architecture. Here, a delay-based RC scheme is studied, which uses one MEMS device to simulate the behavior of a large neural network through input modulation. We introduce a modulation scheme that enables colocalized sensing-and-computing by modulating the bias signal. The MEMS RC is tested to successfully perform pure computation and colocalized sensing-and-computing for both classification and regression tasks, even in noisy environments. Finally, we address the high-voltage requirements of electrostatically actuated MEMS devices by proposing a passive amplification scheme utilizing the mechanical and electrical resonances of MEMS devices simultaneously. Using this scheme, an order-of-magnitude of amplification is reported. Moreover, when only electrical resonance is used, we show that the MEMS device exhibits a computationally useful bistable response.
|
| Subject
|
:
|
Artificial intelligence
|
|
|
:
|
Computer science
|
|
|
:
|
Mathematics
|
|
|
:
|
Mechanical engineering
|
| |