We are the MAIS (Microarchitecture and Artificial Intelligence Security) Group led by Dr. Berk Gulmezoglu in Electrical and Computer Engineering (ECpE) Department at Iowa State University.
Our research is based on discovery of new microarchitectural leakages and analyzing side-channel data with Deep Learning algorithms. We are currently working on the broad range of microarchitectural attacks on AMD, Intel and ARM devices. New attacks and detection techniques are developed to increase the awareness of industry on the potential of side-channel attacks.
- Microarchitectural attack detection using Deep Learning based techniques: Since microarchitectural attacks were introduced, number and variety of the attacks have increased tremendously in the last decade. Hence, chip vendors such as Intel, AMD, ARM have difficulties to come up with patches in a short time. Until these patches become available to customers, attackers have a time period to exploit individual devices through novel side-channel attacks. This project aims to develop new low overhead detection mechanisms to protect the commercial servers and personal computers in real time. Especially, with the advanced learning capabilities of Deep Learning algorithms, it has become possible to train large-scale models to detect side-channel attacks. Therefore, we train various Deep Learning models to compare the efficiency of algorithms and discover new ways to identify ongoing side-channel attacks on the devices.
- Discovering new leakage sources in Intel and AMD architectures: Chip vendors are in a competition to establish the best performance to their customers for a long time. To increase their chip’s performance, they introduce new features such as larger cache sizes, speculative and out-of-order execution mechanisms and so on. While these features provided a large performance gain, they have also become targets for microarchitectural attacks. Especially with Spectre and Meltdown attacks, the potential destruction of microarchitectural attacks is more obvious. Thus, it is more important than before to identify the leakage sources and attacker capabilities before they are exploited by malicious people. This project aims to analyze the architectural components in Intel and AMD devices and discover new ways to leak personal information. Moreover, we focus on several hardware countermeasures to protect confidential information leakage.
- Profiling-based side-channel attacks on CPU and GPU systems: Once new attacks are discovered, side-channel data is collected to obtain a meaningful information about targeted victim. The information extraction mostly requires experts to implement side-channel data analysis techniques such as DPA, template attacks etc. This project focuses on new side-channel analysis techniques to extract maximum information from side-channel data. The analysis techniques can also be developed by implementing newest Deep Learning techniques.