Thesis
An End-To-End Learning Framework for Supporting Green Neuromorphic Computing: From RRAM Design and Microfabrication to Applications
Washington State University
Master of Science (MS), Washington State University
2023
DOI:
https://doi.org/10.7273/000006382
Abstract
Conventional computing systems are commonly organized with a separate memory and CPU units, leaving them vulnerable to the von Neumann bottleneck. This stems from the data transfer rate between the CPU and memory, which is approximately one hundred times slower than the CPU processing speed \cite{pacheco}. The von Neumann bottleneck can significantly hinder computing or memory intensive executions that require processing large amounts of data, such as machine learning or deep learning algorithms. The increasing demand of data processing requirements and extensive use of computer-intensive applications has driven the research and development of energy-efficient hardware implementations for nonvolatile memory. Fabricating efficient memory devices for high-performance computing systems can be a costly and time-consuming process, particularly when relying on traditional trial-and-error methods that involve repetitive testing and modification until successful designs have been developed. The manufacturing process of optimal memory units can be potentially facilitated through the use of advanced evaluation methods and predictive analysis. Machine learning and deep Learning methods have emerged as powerful and effective tools for constructing predictive models that can be used to optimize the fabrication process for various devices and technologies,including non-volatile memory. In this research, we propose a general learning framework for predicting the behavioral properties and potential learning capabilities of memory devices used in neuromorphic computing, by employing machine learning and deep learning algorithms. This study focuses on resistive random access memory (RRAM) as a case study, however, the proposed workflow can be generally applied to other device fabrication analysis aiming to investigate variability and distributions. RRAM is of particular interest as it has demonstrated the potential for high energy and computational efficiency in computing vector-matrix multiplications, a key operation in neural network algorithms. The proposed workflow demonstrates how the application of learning models can facilitate the development of optimal devices for neuromorphic computing, ultimately leading to more efficient and cost-effective fabrication and manufacturing processes.
Metrics
4 File views/ downloads
30 Record Views
Details
- Title
- An End-To-End Learning Framework for Supporting Green Neuromorphic Computing
- Creators
- Abdi Yamil Vicenciodelmoral
- Contributors
- Xinghui Zhao (Advisor)Feng Zhao (Committee Member)Scott Wallace (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- School of Engineering and Computer Science (VANC)
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University
- Number of pages
- 147
- Identifiers
- 99901087515701842
- Language
- English
- Resource Type
- Thesis