Description
Since Android 10, access to hardware identifiers has either been completely blocked or
restricted to system apps only. The urgent need for accurate device recognition lead to
the fact that some hardware vendors already started to provide specific frameworks
and APIs themselves to ship around this problem. In this thesis, the internal Graphics
Processing Unit (GPU) of Android devices is used as a potential new marker for device
identification and recognition. After an introduction into the topic of mobile device
fingerprinting, existing solutions for device identification are presented while high-
lighting their current limits. The third chapter familiarizes with required background
knowledge about GPUs and Machine Learning classifiers, followed by an analysis of
related work. In the following chapters, a new transparent and fast GPU fingerprinting
approach based on compute shaders is proposed and then implemented alongside
another state-of-the-art GPU hardware fingerprinting method which was modified to
enhance its performance on smartphones. Their identification effectiveness is evaluated
with various classifiers for a pool of Android smartphones, showing promising results
for both tested methods up to a point where two Android devices of the same model
and software configuration can be confidentially distinguished based on their GPU
behavior alone.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching