A Hacker Guide To Deep-Learning Based Side Channel Attacks

Conference:  Defcon 27



Deep learning is a powerful tool for hardware side-channel attacks and has the potential to revolutionize hardware cryptography.
  • Deep learning eliminates the need for trace processing and human assumptions, making attacks more efficient and reusable.
  • Deep learning's ability to output probability allows for efficient probabilistic attacks.
  • Deep learning scales linearly with data and computing, making it a perfect fit for scaling attacks.
  • The speaker provides an anecdote about successfully recovering 100 keys using deep learning.
  • The speaker emphasizes the importance of automation and the potential for deep learning to improve hardware cryptography.
The speaker shares that they were able to recover 100% of the tiny AES key using the attack they showed, and were able to recover 81% of the key using a single trace. They emphasize that deep learning is more powerful than traditional techniques for side-channel attacks.


This talk explores how AI is revolutionizing hardware side-channel attacks and what this new wave of attacks mean for the future of hardware cryptography. Based on the lessons learned while successfully attacking many hardware AES implementations using deep-learning this talk discuss why those attacks are fundamentally more efficient and details how to conduct then in practice.