NJIT
Gen-acceleration: Pioneering work for hardware accelerator generation using large language models
Abstract
dc:description.abstract<p>Optimizing computational power is critical in the age of data-intensive applications and Artificial Intelligence (AI)/Machine Learning (ML). While facing challenging bottlenecks, conventional Von-Neumann architecture with implementing such huge tasks looks seemingly impossible. Hardware Accelerators are critical in efficiently deploying these technologies and have been vastly explored in edge devices. This study explores a state-of-the-art hardware accelerator; Gemmini is studied; we leveraged the open-sourced tool. Furthermore, we developed a Hardware Accelerator in the study we compared with the Non-Von-Neumann architecture. Gemmini is renowned for efficient matrix multiplication, but configuring it for specific tasks requires manual effort and expertise. We propose implementing it by reducing manual intervention and domain expertise, making it easy to develop and deploy hardware accelerators that are time-consuming and need expertise in the field; by leveraging the Large Language Models (LLMs), they enable data-informed decision-making, enhancing performance. This work introduces an innovative method for hardware accelerator generation by undertaking the Gemmini to generate optimizing hardware accelerators for AI/ML applications and paving the way for automation and customization in the field.</p>
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Electrical Engineering - (M.S.)
- Discipline thesis:degree_discipline
- Electrical and Computer Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Vungarala, Durga Lakshmi Venkata Deepak
- Contributors dc:contributor
-
- Shaahin Angizi
- Durgamadhab Misra
- Cong Wang
Subjects
dc:subject × 9Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.njit.edu/theses/2376