Detecting a malicious hardware modification is difficult when its trigger is designed to activate only under rare conditions. Conventional functional tests may never reach those conditions, leaving a Trojan dormant and its payload inactive. A backscattering side channel offers another way to look for such modifications: it measures how changes in a chip’s internal impedance affect a signal reflected from the integrated circuit (IC), rather than relying only on changes in external behavior.
For industrial automation, where controllers and other electronic systems underpin operational reliability and security, this approach is relevant to the broader challenge of establishing trust in hardware. It is a nondestructive, external measurement technique that seeks to detect a circuit modification even when the Trojan has not activated.
Why dormant hardware Trojans are difficult to detect
Hardware Trojans typically combine a trigger and a payload. The trigger monitors signals or operating conditions; once a specified condition occurs, the payload changes the circuit’s behavior. That change may affect functional outputs, power consumption, information leakage, or device lifetime. Before activation, however, the payload is usually inert, and the circuit can continue to behave as expected.
This makes ordinary functional testing a weak safeguard against rare trigger conditions. Destructive inspection can reveal physical changes, but it requires opening or removing layers of the chip and can destroy the device being examined. Nondestructive side-channel methods instead look for indirect evidence in properties such as power, temperature, current, or electromagnetic emissions. Each has limitations: some signals have limited bandwidth, while others can be difficult to distinguish from noise or interference.
How the backscattering side channel works
The technique begins by transmitting a continuous radio-frequency signal toward the IC. As transistors switch, the impedances associated with circuit nodes change. Those state-dependent impedance changes affect the IC’s reflection characteristics, modulating the signal that returns from the chip. The reflected signal therefore carries information about on-chip activity.
This differs from a conventional electromagnetic-emanation measurement. Rather than observing only electromagnetic effects associated with brief current changes, the backscattering approach observes how the circuit’s impedance state shapes a transmitted signal. Because the change in impedance can persist after a gate switches, it can contribute to measurable components associated with the circuit’s clock harmonics.
In the demonstrated method, the measurement system examines the amplitudes of selected frequency components beside the transmitted carrier. These components correspond to harmonics of the circuit clock. A set of amplitude measurements forms a trace that characterizes impedance-change activity across a clock cycle. Comparing traces can reveal differences introduced by additional circuitry, including a dormant Trojan trigger.
Turning measurements into a detection decision
Measurements can vary when the chip and probes are positioned slightly differently. Distance affects the received signal’s overall amplitude, potentially making direct comparisons misleading. The proposed method addresses this by comparing ratios between neighboring harmonic amplitudes. Since distance attenuates the trace in a similar way across its points, ratios reduce the effect of that shared attenuation.
The detection process has two stages:
- Training: Measure a known Trojan-free circuit multiple times. Calculate the mean and variation for each amplitude ratio, then establish a threshold based on the observed variation.
- Detection: Measure the circuit under test, calculate its amplitude ratios, and compare them with the Trojan-free reference. A sufficiently large overall deviation indicates a possible Trojan.
This approach depends on having a trusted, Trojan-free reference for training. The experimental evaluation used that assumption so the backscattering method could be compared with an electromagnetic-emanation approach under the same conditions.
What the experiments show
The researchers evaluated the method using FPGA implementations of an AES circuit and three hardware Trojan designs. The Trojans differed in their trigger and payload behavior, and measurements were taken with the payloads dormant. Training used one physical board, while testing used nine other boards, allowing the evaluation to account for variation between hardware instances.
Across the reported measurements, the method identified all tested Trojan-afflicted cases and produced no false positives on the Trojan-free test cases. The paper also reports that backscattering-based detection outperformed the compared electromagnetic-emanation method for dormant Trojans. The results support the principle that measuring impedance-related effects can expose circuit modifications that are difficult to identify through conventional current-related emissions.
Detection was not equally straightforward for every condition. Active Trojans were easier to distinguish than dormant ones, consistent with the additional switching activity created when a payload operates. The experiments also examined changes in Trojan size and physical placement. In particular, the trigger’s size mattered for dormant-Trojan detection because the trigger is the component that continues to monitor inputs before activation. Its switching behavior and connections to the original circuit can influence the measured signature even while the payload remains inactive.
Practical implications and limitations
Backscattering is an external, nondestructive measurement concept, so it does not require adding measurement circuitry to the chip itself. The transmitted carrier can also be selected to avoid some sources of interference, while its strength and frequency offer measurement flexibility. These characteristics make the technique a potential complement to other hardware assurance methods, not a replacement for design review, testing, or supply-chain controls.
The reported results are proof-of-concept findings from particular FPGA circuits and measurement conditions. They do not establish universal detection performance across all IC designs, manufacturing processes, or operating environments. Practical use would need to account for the availability and quality of trusted reference data, measurement setup consistency, circuit variation, and the characteristics of the suspected modification.
Conclusion
A backscattering side channel uses reflected-signal changes to reveal impedance variations caused by on-chip switching. By comparing harmonic amplitude ratios against a Trojan-free reference, the demonstrated method detected dormant hardware Trojans across multiple physical test boards. For engineers and technical decision-makers, the central implication is that impedance-based measurements may provide a useful additional route to hardware assurance when a malicious payload remains inactive and functional testing alone is unlikely to expose it.