Deep learning-based profiling side-channel attacks in SPECK cipher.

Area of Science:

• Cryptography
• Computer Science
• Machine Learning

Background:

• Deep learning profiling side-channel analysis is advancing, particularly for classical ciphers like AES.
• Adaptation of these advanced techniques to lightweight cryptography, such as the SPECK cipher, is underexplored.
• SPECK is crucial for Internet of Things (IoT) devices, necessitating strong defenses against side-channel attacks.

Purpose of the Study:

• To introduce a novel deep learning-based profiling side-channel analysis for the SPECK lightweight cipher.
• To address the gap in research applying advanced deep learning attacks to lightweight primitives.
• To evaluate the effectiveness of deep learning ensemble methods against SPECK implementations.

Main Methods:

• Utilized a sequential divide and conquer ensemble of deep learning models.
• Applied the technique to software implementations of the SPECK-32/64 cipher.
• Focused on profiling side-channel analysis to recover the secret key.

Main Results:

• Successfully recovered the 8-byte secret key of the SPECK-32/64 cipher.
• Achieved key recovery in fewer than 250 traces, demonstrating high efficiency.
• This represents the first known deep learning-based profiling attack on both unprotected and protected SPECK implementations.

Conclusions:

• Deep learning ensemble methods are effective for profiling side-channel attacks on lightweight ciphers like SPECK.
• The proposed method offers a significant advancement in securing resource-constrained devices in IoT.
• Further research is warranted to explore these techniques on other lightweight cryptographic primitives.