Exploring Security Challenges and Opportunities in Emerging Memory and Computing Technologies
Open Access
- Author:
- Nagarajan, Karthikeyan
- Graduate Program:
- Computer Science and Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 15, 2022
- Committee Members:
- Swaroop Ghosh, Chair & Dissertation Advisor
Abhronil Sengupta, Major Field Member
Saptarshi Das, Outside Unit & Field Member
Mehdi Kiani, Major Field Member
Chitaranjan Das, Program Head/Chair - Keywords:
- Non Volatile Memory
Artificial Intelligence
In-Memory Computing
Spiking Neural Networks
Fault Injection
Information Leakage
Denial of Service
Hardware Trojan
Side Channel Attack - Abstract:
- The computing industry has progressed through the rapid development of PCs in the 1990s, the explosion of gaming in the 2000s, and the introduction of cloud and mobile computing in the 2010s. In response, the memory industry has always technologically kept up to enable the latest computing paradigms. As we enter the new age of Artificial Intelligence and Machine Learning, the emerging computing requirements are driving the need for better memory performance, power efficiency and memory capacity. The rapid and parallel innovation in the fields of AI and memory introduces novel security threats and vulnerabilities in both domains. Many of these memories and SNN architectures have already been deployed as discrete chips in full systems and are expected to become ubiquitous in future computing devices. For memory applications, emerging Non Volatile Memories (NVM) such as, Spin-Transfer Torque RAM (STTRAM) and Resistive RAM (RRAM), present improved speed, power consumption, scalability, retention times and enable In-Memory Computing (IMC). While exciting at first glance, they suffer from new sources of variability and security issues. This work identifies vulnerabilities is emerging NVMs and proposes attack vectors ranging through device level, circuit level, and system level points of view. It also presents circuit/system level countermeasures and security enhancements to render NVMs robust against security and privacy issues. Finally, it also presents the IMC implementations of NVM to improve security applications. In the field of AI, Spiking Neural Networks (SNN) are fast emerging as a promising option that is computationally more powerful and provides higher energy-efficiency than traditional Deep Neural Networks (DNN). These novel SNN implementations contain security-sensitive assets and vulnerabilities that can be exploited by adversaries. This work presents detailed analysis of global fault injection attacks through side channel attack vectors on commonly employed SNNs. It also explores the impact of various SNN design parameters and identifies design choices for robust and secure implementations. Finally, the work also presents circuit/system level countermeasures and fault detection mechanisms for secure SNN deployment.