SaTC: CORE: Small: Microarchitectural side channel attacks and defenses in integrated CPU-GPU systems

点击翻译按钮获取中文摘要

In modern computing systems, Graphics Processing Units (GPUs) are key components to provide multimedia and computing capabilities. Modern mobile phones, tablets, laptops, and desktops are heterogeneous, comprised of conventional Central Processing Units (CPUs), GPUs, and other accelerators that are tightly integrated on the same chip and share some hardware resources to provide high performance and energy efficiency. The investigator has discovered that a malicious application running on a processor (e.g., GPU) within these computing devices can launch an attack on a victim application running on another processor (e.g., the CPU), allowing the attacker to gain sensitive information about the victim application’s activities. These attacks, which span different components in the computing system, are extremely dangerous as they bypass proposed mitigations that focus on a single component (either the CPU or the GPU). This project assesses the feasibility of cross-component attacks and develops defense mechanisms to secure computing systems against these attacks, which is important for the computer chip industry. The project also includes activities to broaden the participation of underrepresented groups in computing.The project explores and characterizes the threat model and develops cross-component side-channel attacks that leak sensitive data through unintended side-effects of program execution observed through shared microarchitectural resources, such as cache and interconnect in integrated heterogeneous systems. In addition, the project develops both hardware and software solutions to mitigate the attacks with emphasis on security, high performance, and low complexity. Finally, the project characterizes how modern and evolving web standards to task accelerators can be used to launch these attacks remotely and proposes approaches to secure them, at a critical time when these standards are being developed and before they are widely deployed. The project furthers the knowledge and scope of microarchitectural attacks beyond the prior focus on CPUs, which is a threat model increasing in importance as we move increasingly towards heterogeneous computing platforms.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.