Dissertation
Hardware-software co-design for neuromorphic computing
Doctor of Philosophy (Ph.D.), Drexel University
May 2022
DOI:
https://doi.org/10.17918/00001077
Abstract
Neuromorphic devices represent an attempt to mimic the computational dynamics and distributed architecture of mammalian brains. Mixed-signal neuromorphic platforms executing Spiking Neural Networks (SNNs) show significant gains in computational speed and energy consumption, when compared to recent deep learning accelerators. However, with growing model size and complexity of SNN-based applications and learning algorithms, allocating resources efficiently on neuromorphic hardware, while ensuring optimal performance, in terms of algorithm accuracy, latency and energy consumption, is becoming increasingly challenging. Therefore, there is a growing need for an extensible simulation framework that can (1) perform architectural explorations with SNNs, including both platform-based design of today's hardware, and (2) hardware-software co-design and design-technology co-optimization of the future. In this thesis, a hardware-software co-design framework is presented with an aim to (1) design large scale SNNs (2) compile SNN-based applications for neuromorphic hardware, (3) optimize the allocation of resources on the neuromorphic platform to execute the compiled SNN-based application, and (4) design a novel communication architecture (CA) for future neuromorphic platforms, with an aim to address and ensure the scalability of crossbar-based platforms. The proposed framework will significantly improve the performance of existing neuromorphic platforms, in terms of algorithm accuracy, latency, energy consumption and reliability, and explore architectural improvements to address the limitations of existing platforms.
Metrics
3182 File views/ downloads
105 Record Views
Details
- Title
- Hardware-software co-design for neuromorphic computing
- Creators
- Adarsha Balaji
- Contributors
- Anup Das (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Doctor of Philosophy (Ph.D.)
- Publisher
- Drexel University; Philadelphia, Pennsylvania
- Number of pages
- xx, 101 pages
- Resource Type
- Dissertation
- Language
- English
- Academic Unit
- College of Engineering (1970-2026); Electrical (and Computer) Engineering (1970-2026); Drexel University
- Other Identifier
- 991018021331304721