On the Effect of Clock Frequency on Voltage and Electromagnetic Fault Injection

Stefanos Koffas, Praveen Kumar Vadnala

TL;DR

This paper investigates how clock frequency impacts the success rate of voltage and electromagnetic fault injection (FI) attacks on System-on-Chips (SoCs). The authors found that higher clock frequencies consistently lead to a greater success rate for both types of FI attacks, suggesting that as processors get faster, they may become more vulnerable to these physical attacks.

What Problem Does This Paper Address?

Fault injection (FI) attacks are a significant threat to the security of embedded systems and cryptographic implementations. These attacks involve physically manipulating a device's behavior, often by introducing transient errors, to bypass security mechanisms, extract sensitive data, or cause unintended operations. While the general principles of FI are well-understood, the specific environmental factors that influence their success rate, particularly in modern, complex Systems-on-Chips (SoCs), are not fully characterized. SoCs often operate at variable clock frequencies, and it's unclear how this dynamic behavior affects the effectiveness of common FI techniques like voltage glitching and electromagnetic pulse injection. This paper aims to fill this gap by systematically studying the relationship between clock frequency and FI success rates on a RISC-V SoC.

Methods

The researchers developed three distinct test applications designed to probe different components of a System-on-Chip (SoC):

  • Register-based Loop: This test focuses on CPU registers, minimizing reliance on caches. It involves incrementing and decrementing registers to monitor for unintended modifications.
  • Memory-based Loop: This test involves loading and storing data from memory in each iteration, thus exercising the data cache and memory transfer mechanisms.
  • Unrolled Loop: This test is designed to stress the instruction cache and the CPU-to-memory bus. Depending on the loop's unrolling factor, it can either fit entirely within the instruction cache or require fetching instructions from main memory.

These tests were implemented in assembly to ensure precise control over execution and avoid compiler optimizations. The experiments were conducted on a SiFive HiFivel development board featuring the FE310-G000 RISC-V SoC.

Two primary fault injection techniques were employed:

  • Voltage Fault Injection (VFI): This involved manipulating the SoC's supply voltage. Modifications to the HiFivel board were made to allow direct connection of a glitch amplifier to the CPU's power supply pins, bypassing filter capacitors for more effective glitching.
  • Electromagnetic Fault Injection (EMFI): This involved using an EMFI transient probe to generate localized electromagnetic pulses near the chip.

The experiments systematically varied the clock frequency of the SoC across three settings: 16MHz (slow), 90MHz (medium), and 320MHz/240MHz (fast, depending on the attack type). For each clock frequency and test application, a comprehensive parameter space scan was performed for both VFI and EMFI, exploring various glitch voltages, lengths, and delays. The success of a fault was categorized as "Expected" (no fault effect), "Crash/Mute" (device unresponsive), or "Successful" (unintended but non-crashing behavior). The success rate was calculated as the number of successful attempts divided by the total number of attempts.

Datasets

Not discussed in the paper. The experiments utilized custom-written test applications and a specific hardware target (SiFive HiFivel board).

Key Insights and Ideas

  • Clock Frequency is a Critical Factor: The most significant finding is the clear correlation between higher clock frequencies and increased fault injection success rates for both VFI and EMFI. This suggests that as processors become faster, they may inherently become more susceptible to physical attacks.
  • VFI and Timing Constraints: VFI attacks exploit timing constraint violations. As clock frequencies increase, clock periods decrease, making it easier to induce these violations and thus increasing the success rate of VFI.
  • EMFI and Charge-Based Model: The EMFI results align more closely with the charge-based fault model, which suggests that fault susceptibility windows can be influenced by operating frequency. This contrasts with the sampling fault model, which posits frequency independence.
  • Test Application Sensitivity: Different test applications exhibited varying sensitivities to fault injection. The memory-based loop, in particular, showed a higher percentage of successful faults, likely due to the inherent vulnerability of memory operations (loads and stores) to errors.
  • SPI Flash Vulnerability: In lower clock frequency tests (16MHz and 90MHz), successful EMFI glitches were often observed near the SPI flash memory. This suggests that the unprotected external SPI flash was the target, rather than the core SoC logic, and these results were deemed less relevant for understanding SoC vulnerability.

Key Results

  • Increased Success Rate with Frequency: Across all three test applications and both VFI and EMFI, the success rate of fault injection attacks consistently increased with higher clock frequencies. For instance, VFI experiments showed the largest number of successful glitches at 240MHz and the smallest at 16MHz.
  • VFI at 16MHz: At 16MHz, VFI could induce branch instruction skips in register-based and memory-based loops, and instruction manipulation in the unrolled loop.
  • VFI at 90MHz: At 90MHz, similar faults were observed, with branch skipping and instruction manipulation being prevalent.
  • VFI at 240MHz: At 240MHz, the highest frequency tested for VFI, the most successful faults were observed, including branch skipping, memory corruption, and instruction manipulation.
  • EMFI at 320MHz: At 320MHz, EMFI attacks showed successful faults across all three tests, with results indicating instruction manipulation and skipping. The paper notes that at this frequency, the results were less related to the SPI flash and more indicative of charge-based fault models.
  • EMFI vs. VFI: The paper provides detailed tables (Table 2 for EMFI, Table 3 for VFI) listing specific successful fault outcomes (register values, comments on fault type) for various clock frequencies and test applications.

Limitations and Future Work

  • Specific SoC Architecture: The study is conducted on a specific RISC-V SoC (FE310-G000). While providing valuable insights, the results might not be directly generalizable to all SoC architectures or instruction set architectures without further investigation.
  • Test Application Scope: While three test applications were developed to cover different SoC components, a more exhaustive set of tests could provide an even broader understanding of fault impact.
  • Parameter Space Exploration: The paper mentions that the parameter space was explored randomly within defined ranges. A more exhaustive grid search or optimization algorithm could potentially uncover additional fault injection windows.
  • Future Work: The authors suggest that further research could explore the impact of different fault models, investigate the susceptibility of specific instructions, and develop more robust countermeasures against these frequency-dependent fault injection attacks. The need for a profiling phase to understand an application's susceptibility to EMFI is also highlighted.

Summary updated

Abstract

We investigate the influence of clock frequency on the success rate of a fault injection attack. In particular, we examine the success rate of voltage and electromagnetic fault attacks for varying clock frequencies. Using three different tests that cover different components of a System-on-Chip, we perform fault injection while its CPU operates at different clock frequencies. Our results show that the attack’s success rate increases with an increase in clock frequency for both voltage and EM fault injection attacks. As the technology advances push the clock frequency further, these results can help assess the impact of fault injection attacks more accurately and develop appropriate countermeasures to address them.
BibTeX
@inproceedings{koffas2022effect,
  author = {Stefanos Koffas and Praveen Kumar Vadnala},
  title = {{On the Effect of Clock Frequency on Voltage and Electromagnetic Fault Injection}},
  booktitle = {ACNS Workshops},
  year = {2022},
  url = {https://openreview.net/forum?id=4yyPenhx5d},
}

Loading related content...

Figure