Kraken: Higher-order EM Side-Channel Attacks on DNNs in Near and Far Field

TL;DR

This paper demonstrates novel side-channel attacks on GPU Tensor Cores for model theft, including near and far field electromagnetic analysis, revealing vulnerabilities in modern ML hardware.

Contribution

It introduces the first parameter extraction attack on GPU Tensor Cores using physical side-channel analysis, expanding the scope of model stealing threats.

Findings

Tensor Cores can be exploited for model extraction via side-channel attacks

Electromagnetic radiation leaks can be detected from 100 cm away through glass

Energy consumption estimates enable efficient correlation power analysis attacks

Abstract

The multi-million dollar investment required for modern machine learning (ML) has made large ML models a prime target for theft. In response, the field of model stealing has emerged. Attacks based on physical side-channel information have shown that DNN model extraction is feasible, even on CUDA Cores in a GPU. For the first time, our work demonstrates parameter extraction on the specialized GPU's Tensor Core units, most commonly used GPU units nowadays due to their superior performance, via near-field physical side-channel attacks. Previous work targeted only the general-purpose CUDA Cores in the GPU, the functional units that have been part of the GPU since its inception. Our method is tailored to the GPU architecture to accurately estimate energy consumption and derive efficient attacks via Correlation Power Analysis (CPA). Furthermore, we provide an exploratory analysis of hyperparameter and weight leakage from LLMs in far field and demonstrate that the GPU's electromagnetic radiation leaks even 100 cm away through a glass obstacle.