Bachelor- and Master-Seminar

Dear All,

the next seminar takes place on 23.12. at 14:00.

Session A:
Yu-De Lin - Amir Heinisch

https://cispa-de.zoom.us/j/96786205841?pwd=M3FOQ3dSczRabDNLb3F1czVXVUpvdz09

Meeting-ID: 967 8620 5841
Kenncode: BT!u5=
 

14:00-14:30 

Speaker: Yu-De Lin
Type of talk: Master intro
Advisor: Dr. Nils Ole Tippenhauer
Title: Extracting DNN Models from Embedded Devices via the Power Side-channel

Abstract: As hardware becomes more and more powerful, running deep neural networks on embedded devices are more popular than ever. Many hardware vendors have developed their own AI chips inside their own products or selling development kits for their AI-powered boards. The era of edge computing has arrived. However, the new threats on these AI devices are emerging too. Since there is only little security protection for embedded devices where AI models are used, the intellectual property is under high risks of being stolen or leaked.
In our project, we are investigating how the attacker can extract AI models on the chip in the physical layer. In particular, we focus on the power-side channel, which is less complicated to setup compared to EM-channels in prior work . In addition, we design a framework with the aid of machine learning that the attacker can use to automate the attack to reverse engineer the achitecture and the weights of models.

14:30-15:00

Speaker: Amir Heinisch
Type of talk: Bachelor Thesis Intro Talk
Advisor: Dr. Nils Ole Tippenhauer
Title: Leveraging Trusted Execution Environments to Implement Trustworthy Motor Controls

Abstract:

Every modern vehicle nowadays contains a large scale of digital controllers. Many new attack
vectors arise. For physical attackers (especially tuners) there are many new ways to
manipulate vehicles (e.g. modify firmware to increase performance) and with that violate the law.
Protecting against these threats requires technical mechanisms to verify regulations.

In this thesis we implement and evaluate a method to detect, prevent and even report
violations with the help of Trusted Execution Environments. In particullar we want
to run code in a secure environment which is able to verify control decisions made by
complex motor control software. For this we will build a proof of concept using the
ARM TrustZone technology on the latest STM32L5 microcontroller. This allows us to
evaluate such an approach and show its feasibility.

 

15:00-15:30

No talk this week.