Exploring Power Side-Channel Challenges in Embedded Systems Security

Pouya Narimani, Meng Wang, Ulysse Planta, Ali Abbasi

TL;DR

Researchers have developed a novel wavelet-based signal processing method that allows Power Side-Channel (PSC) attacks to work on real-world, "black-box" embedded devices without any hardware modifications. By shifting analysis to the frequency domain, they successfully fingerprinted code execution on everything from simple microcontrollers to high-end 1GHz processors like the Raspberry Pi.

What Problem Does This Paper Address?

Power Side-Channel (PSC) attacks are the "listening at the door" of the digital world. By measuring the minute fluctuations in power consumption as a chip operates, attackers can often deduce secret cryptographic keys.

However, there is a massive gap between academic PSC research and real-world application. Most lab studies use "white-box" conditions: they have the source code, they use custom boards with "shunt resistors" (components added specifically to make power measuring easier), and they operate in low-noise environments.

In the real world, an attacker faces "black-box" systems where the firmware is unknown, the hardware cannot be modified, and the power signal is drowned out by "noise" from other components (peripherals, OS interrupts, and background tasks). This paper tackles these hurdles, aiming to make PSC a viable tool for broader system security tasks like intrusion detection and malware analysis on off-the-shelf hardware.

Methods

The authors introduce a signal-processing pipeline designed to extract "fingerprints" of code execution from noisy power traces.

  1. Frequency Domain Analysis: Instead of looking at power usage over time (which is easily misaligned by OS jitter), they transform the signal into the frequency domain using Wavelet Transforms. Unlike standard Fourier Transforms, wavelets preserve some timing information while highlighting the high-frequency "snaps" caused by CPU instruction changes.
  2. Noise Suppression: They use a 5th-degree Butterworth high-pass filter to strip away low-frequency noise generated by board peripherals (like LEDs or radio components), leaving behind the sharp signals related to the CPU's internal logic.
  3. Control Flow Fingerprinting: By identifying "drastic" changes in the power signal—specifically those caused by branch instructions—they create a unique vector (a fingerprint) that represents the program's execution path.
  4. Non-Invasive Probing: They demonstrate that these signals can be captured by probing existing test pins or power lines on a board, completely bypassing the need to desolder components or add shunt resistors.

Datasets

The researchers evaluated their approach across a diverse set of targets:

  • Ideal/Dev Boards: STM32F3 and STM32F4 microcontrollers (used to establish baselines).
  • Black-Box Targets: A UBlox GNSS receiver (where firmware and ISA are unknown) and a Satellite RF-transceiver board.
  • High-End Target: A Raspberry Pi Zero 2 W featuring a 1GHz ARM Cortex-A53—a significant step up from the simple 8-bit or 32-bit MCUs usually found in PSC research.
  • Software: They tested against standard algorithms like AES-128, SHA-256, and LZ4, as well as real-world RTOS samples like the Zephyr Project’s MQTT sensor.

Key Insights and Ideas

  • The "Covariate Shift" Trap: The authors show that Machine Learning models trained on random assembly code fail miserably (dropping from 81% to 41% accuracy) when asked to identify real-world algorithms like AES. This highlights why "profiling" (training on a known device) is often a bottleneck for real-world attacks.
  • Software Noise is the Real Enemy: In complex systems, "software noise" (context switches, interrupts, and background processes) is more disruptive than electrical hardware noise. Their wavelet method is specifically designed to be robust against this temporal "jitter."
  • Frequency is King: By focusing on high-frequency components, the researchers could distinguish between different types of "hard-faults" (crashes) just by looking at the power pattern immediately preceding the crash.

Key Results

  • 100% Accuracy: The wavelet-based fingerprinting achieved 100% accuracy in identifying the execution of AES, SHA, and LZ4 on microcontrollers, with a false positive rate below 12%.
  • High-End Success: They successfully identified the distinct "rounds" of AES encryption running on a Raspberry Pi Zero 2 W, proving that PSC is viable even on high-speed, complex application processors.
  • SPI Decoding: On the satellite board, they were able to decode data being sent over the SPI bus (Serial Peripheral Interface) simply by analyzing the power peaks, effectively "seeing" data moving through internal PCB layers that were physically inaccessible.

Limitations and Future Work

The primary limitation is computational overhead. The method requires extremely high sampling rates (up to 3.125 Gigasamples per second) and significant processing power. Currently, the data must be transferred to a host PC with a powerful GPU (like an Nvidia 3070) to perform the wavelet transforms in a reasonable timeframe.

The authors suggest that moving the algorithm onto an FPGA (Field-Programmable Gate Array) would allow for true real-time processing, potentially enabling a "plug-and-play" security monitor that could detect intrusions or malware on an embedded system just by "feeling" its pulse through the power line.

Summary updated

Abstract

Power side-channel (PSC) attacks are widely used in embedded microcontrollers, particularly in cryptographic applications, to extract sensitive information. However, expanding the applications of PSC attacks to broader security contexts in the embedded systems domain faces significant challenges. These include the need for specialized hardware setups to manage high noise levels in real-world targets and assumptions regarding the attacker's knowledge and capabilities. This paper systematically analyzes these challenges and introduces a novel signal-processing method that addresses key limitations, enabling effective PSC attacks in real-world embedded systems without requiring hardware modifications. We validate the proposed approach through experiments on real-world black-box embedded devices, verifying its potential to expand its usage in various embedded systems security applications beyond traditional cryptographic applications.
BibTeX
@misc{narimani2024exploring,
  author = {Pouya Narimani and Meng Wang and Ulysse Planta and Ali Abbasi},
  title = {{Exploring Power Side-Channel Challenges in Embedded Systems Security}},
  year = {2024},
  eprint = {2410.11563},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2410.11563},
}

Loading related content...

Figure