Year: 2024
Type: dissertation
Abstract: GPUs are important hardware accelerators for modern applications, particularly AI-based ones.They offer a high degree of parallelism, allowing multiple data to be processed simultaneously in a single chip.This is made possible by continuous technology scaling, resulting in higher transistor densities (e.g., 80 billion transistors for NVIDIA Hopper GPUs and 100 billion transistors for Intel GPUs).GPU devices offer great computational performance but can suffer from reliability issues associated with modern semiconductor technologies.As technology scales, several threats, such as accelerated wear-out, premature degradation, and high-temperature conditions, can increase the risk of hardware defects that lead to permanent faults.GPUs are now used in critical applications such as autonomous driving systems, aerospace, and avionics, among others.The reliability of GPUs is paramount, especially when used in such scenarios where they are required to operate correctly for longer lifetimes than typical consumer applications.
Source: PORTO@Iris
Author: GUERRERO BALAGUERA, JUAN DAVID
Institution: Simon Fraser University
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7218268134
