Advisory Committee Chair
Mohammad R Haider
Advisory Committee Members
Dalton S Nelson
Date of Award
Degree Name by School
Doctor of Philosophy (PhD) School of Engineering
The technological achievements made with silicon-based transistors have led to nearly all the computing seen today, complete with benefits such as maximized operating speeds, parallel/tensor processing, minimized power overhead, etc. The silicon electronics industry is foundational to scientific progression with its profound impact on modeling, simulations, and fast/reliable computing. However, silicon-based transistors have evolved to meet their physical limitations and have saturated fields of innovation. New fields in computing outside of silicon-based methods have emerged as exotic materials enter the research domain. The suite of super-performing materials like Graphene, Carbon Nanotubes, and Molybdenum Disulfide have chaotic, non-linear behaviors that make utilization as a computing element possible. With silicon technology well-established, exploration into alternative approaches has proliferated with important outputs such as quantum-dot, spintronics, and physical machine learning. The appeal of these are speed-of-light response times (photonics), nonvolatile memory (spintronics/quantum dot), and for physical machine learning, benefits include offline computing, irregular/flexible PCBs, low manufacturing costs, fast fabrication turnaround times, and eco-friendliness. The inherent limitations of silicon technology are addressed with alternative computing, applied as a supplemental/hybrid computing paradigm that perform in ways silicon-only approaches cannot. This work explores the newly emerged field of inkjet-printed electronics for alternative computing due to its compatibility with iv usage of unique nanoparticle materials, low fabrication costs, fast production times, environmental friendliness, and substrate variability. An inkjet-printed non-linear graphene element (IJPAN) is shown to compute within a well-known machine learning architecture as one of the first studies into machine learning systems using inkjet-printed neural networks as the hidden layer. Multiple non-linear elements are simulated as a network of identical nodes to replace the standard sigmoid layer (reservoir) of a Recurrent Neural Network called the Echo State Network. Inkjet-printer fabrication methods, error minimization, and evolution to the final artificial neuron are included for reproducibility. The impact to the scientific community is its application to edge-computing, cyber security, high-volume sensor networks, multi-physics perception, and irregular surfaces. Long term benefits to society include cost-effective, biodegradable, large-area, flexible, and offline electronics in telehealth, entertainment, safety equipment, home hobbies, and more.
Gardner, Steven D., "Inkjet-Printed Reseroir Computing Network for In Situ Sensor Predictions" (2023). All ETDs from UAB. 433.