Neuromorphic Codesign Through the Stack
Date Issued
May 2026
Author(s)
Patel, Karan
The University of Tennessee
Advisor(s)
Catherine D. Schuman
Additional Advisor(s)
James Plank
Garret Rose
Andy Sarles
Abstract
With the decline of Moore's Law and Dennard Scaling, traditional Von Neumann architectures face critical limitations in performance and energy efficiency, primarily due to the separation of processing and memory units and the resulting communication bottleneck. Neuromorphic computing has emerged as a promising alternative, drawing inspiration from the brain’s structure to enable efficient, low-power computation. These systems, often built on spiking neural networks (SNNs), integrate memory and computation using neuron-synapse models and event-driven processing, allowing for complex, robust behavior across a range of applications, from classification to control and anomaly detection. However, the design of neuromorphic systems is inherently challenging, requiring innovation across the full computing stack, from materials and circuits to algorithms and hardware integration. A codesign approach, which emphasizes collaboration across these layers, is essential to unlocking the full potential of neuromorphic computing. This body of work moves up the stack from exploring and characterizing low-level devices and materials for neuromorphic systems, designing optimization frameworks to tune these devices/properties to meet target constraints and objectives, then training these systems using algorithms and online-learning frameworks built from the ground up for target hardware platforms, ultimately leading to complete neuromorphic systems deployed for real-world applications.
Degree
Doctor of Philosophy
Major
Computer Engineering
File(s)![Thumbnail Image]()
Name
PATEL-PRIMARY-2026.pdf
Size
33.41 MB
Format
Adobe PDF
Checksum (MD5)
5f74721b502ecadb944a5e6e043753b6