Hardware Trojans that activate only under trigger conditions — a specific input, a timer, an external signal — betray themselves by the transition between dormant and active states. Detection methods exploit this: compare the device's behavior with and without the trigger, and the difference reveals the parasite.
Note (2026-11-14; read against the paper's full text): the evidence in the paper is one AES-128 core on one FPGA board, run repeatedly with a fixed key and plaintext, in three conditions (no Trojan, a workload-correlated leakage-information Trojan, and an independent ring-oscillator Trojan of 101 NAND gates). The paper says the two Trojan classes produce distinct persistence patterns, depending on how they interact with the workload, rather than one common steadiness. It attributes the variability of the Trojan-free design to scale-dependent behaviour as different operational phases dominate at different resolutions; I did not find the essay's attribution of that variation to temperature drift or power supply noise (the paper names environmental noise only as a limitation of the EM measurement, together with probe placement and coupling). It also says the median mixture complexity is not used as a decision threshold.
Always-on Trojans have no trigger. They operate continuously, producing no behavioral transition. Every measurement of the compromised chip looks the same as every other. There is no before-and-after to compare. These are considered the hardest Trojans to detect.
Tahghigh and Salmani (arXiv:2603.16058) show that permanence itself is the tell. Using cross-scale persistence analysis of electromagnetic side-channel emissions, they find that legitimate circuits exhibit natural statistical variation across time-frequency scales — fluctuations that reflect the circuit's interaction with its environment. Always-on Trojans suppress this variation. Their signatures are too stable.
The detection method looks not for the presence of an anomaly but for the absence of variation. A genuine system fluctuates because it responds to changing inputs, temperature drift, power supply noise. A parasitic circuit that is always on produces a signature that is artificially steady across scales where genuine circuits would waver.
Consistency is the fingerprint of deception. The Trojan's strength — perpetual, untriggered operation — is precisely what makes it detectable, once you know to look for the wrong kind of steadiness. The detection method does not compare the compromised chip to a known-good reference. It compares the chip's own statistical texture to what real complexity looks like. Real circuits are messy. Trojans are clean. The cleanliness gives them away.