SIMCom: Statistical sniffing of inter-module communications for runtime hardware trojan detection

Faiq Khalid, Syed Rafay Hasan, Osman Hasan, Muhammad Shafique

Introduction

The modern semiconductor supply chain is global and fragmented, forcing System-on-Chip (SoC) designers to rely heavily on third-party Intellectual Property (3PIP) blocks. These blocks are often treated as "black boxes," creating a significant security vulnerability. A malicious vendor or compromised foundry can insert a Hardware Trojan (HT)—stealthy, hidden circuits that remain dormant during standard testing but later activate to leak sensitive data, jam communications, or cause catastrophic system failure.

SoC Design Flow and Targeted Attack Model

Traditional detection methods, such as functional testing or side-channel analysis, often fail to identify these dormant threats. Functional tests rarely trigger the specific conditions required for Trojan activation, and side-channel analysis (monitoring power or EM emissions) requires precise "golden" signatures that are difficult to obtain for third-party components. To address this, the SIMCom framework introduces a runtime detection mechanism that identifies Trojans by monitoring statistical anomalies in inter-module communication traffic.

The SIMCom Framework

SIMCom (Statistical Sniffing of Inter-Module Communications) treats SoC traffic as a statistical time series. Rather than inspecting the data payload, it analyzes the "rhythm and flow" of communication to identify deviations from expected behavior. The methodology involves three primary stages:

  1. Statistical Modeling: At design time, the "normal" behavior of a trusted module is modeled using three parameters: the Hurst Exponent ($H$) for traffic burstiness, Spatial Hop Distribution ($P$) for packet travel distance, and Spatial Injection ($\sigma$) for packet injection rates.
  2. Assertion Generation: These models are converted into Property Specification Language (PSL) assertions, defining the boundaries of legitimate traffic.
  3. Runtime Monitoring: Lightweight hardware monitors are embedded within the SoC to calculate $H, P,$ and $\sigma$ in real-time. If live traffic violates the PSL assertions, an intrusion is flagged immediately.

Experimental Evaluation

The researchers evaluated SIMCom across three SoC environments, ranging from simple 8051-based systems to complex LEON3-based architectures. They utilized standard benchmark Trojans from Trust-Hub (MC8051-T200 through T800) to simulate realistic attack scenarios, including information leakage and communication jamming.

Statistical Abnormalities in Communication Behavior for Various Hardware Trojans

As shown in the experimental data, even Trojans that adhere to standard communication protocols inevitably alter the statistical signature of the traffic. For instance, leaking a 128-bit key creates a subtle burst that the Hurst exponent is uniquely capable of detecting. The framework proved robust across different traffic distributions, such as Gaussian and Exponential models.

Key Results and Discussion

SIMCom achieved a 100% detection rate for the implemented Trust-Hub benchmarks. Its most significant advantage is its efficiency; the hardware monitors are extremely lightweight, resulting in a total area and power overhead of less than 1% for complex SoCs.

While highly effective, the system does rely on the existence of at least one trusted IP to serve as a baseline for monitoring. Furthermore, future research is needed to explore the limits of "adversarial traffic shaping," where a sophisticated attacker might attempt to mask a Trojan's activity by mimicking the statistical properties of legitimate traffic. Nevertheless, SIMCom provides a powerful, low-overhead defense for securing modern SoCs against supply chain threats.

Summary updated

BibTeX
@article{khalid2020simcom,
  author = {Faiq Khalid and Syed Rafay Hasan and Osman Hasan and Muhammad Shafique},
  title = {{SIMCom: Statistical sniffing of inter-module communications for runtime hardware trojan detection}},
  journal = {Microprocess. Microsystems},
  year = {2020},
  doi = {10.1016/j.micpro.2020.103122},
}
Figure