Electromigration hardware Trojans exploit a normal wear-out process in semiconductor interconnects to create delayed, difficult-to-detect changes in circuit behavior. The risk is especially relevant when chip design and fabrication are handled by separate organizations: modifications made during manufacturing may not be visible to the design team. Understanding how electromigration (EM) affects metal wires—and how wire structure influences failure—can help engineers and decision-makers assess this supply-chain security concern.
Why electromigration matters to hardware security
Electromigration occurs when momentum from conducting electrons moves metal atoms within an interconnect. This movement builds hydrostatic stress in the wire. If stress reaches a critical level, atoms can diffuse and form voids or hillocks, eventually changing the wire’s resistance or interrupting its current path.
In an EM-based hardware Trojan, the physical layout of an interconnect is altered so that aging causes a targeted wire to fail earlier or behave differently. Unlike a modification that immediately changes logic, a reliability-based Trojan may remain dormant until the chip has accumulated sufficient operating time. That delayed activation makes conventional functional testing and side-channel checks more challenging: before the triggering wear-out occurs, the circuit may appear to operate as intended.
The concern is heightened by outsourced fabrication. A foundry with access to physical design data or masks could potentially alter interconnect structures during manufacturing. The risk applies to power and ground networks, clock trees, and signal nets, although the available current and activity differ across these types of wiring.
Why wire topology changes EM failure behavior
Traditional EM reliability estimates often use Black’s equation, an empirical model fitted to test data. It is widely used for reliability sign-off and can provide a conservative estimate, but it has limitations when estimating time to failure across varied current densities and complex, multi-segment wire structures.
Physics-based approaches, including models derived from Korhonen’s equation, instead describe how stress develops over time. A three-phase model separates the process into:
- Nucleation: Stress accumulates until a void forms.
- Incubation: The void grows, with little immediate effect on wire resistance.
- Growth: In applicable structures, the void causes resistance to increase and can lead to parametric failure.
This approach can account for interactions among connected wire segments. Their geometry and current direction influence stress and the movement of atoms, so neighboring segments can change a wire’s predicted lifetime. A passive reservoir may help relieve stress and extend lifetime; a sink can increase stress at a wire end. These structural effects create possible failure pathways that a current-density-only view may not capture.
Early and late failure modes
Interconnect configuration also affects what happens after a void grows. In a via-above arrangement, a void can block the via and interrupt current, producing an early, abrupt failure. In a via-below arrangement, current may continue through a higher-resistance liner after the via is blocked, creating a more gradual, late failure. The distinction matters because changing a wire’s layer relationship can alter its failure mode without relying only on a change in current density.
How topology-based attacks can reduce visibility
Earlier EM attack concepts focused on changing wire width to increase current density. But narrowing a conductor can affect resistance, IR drop, and interconnect delay before the Trojan activates, potentially disrupting operation or creating measurable anomalies.
Topology-based methods described in physics-based EM research take a different approach: they modify structures around a target wire while leaving its current density unchanged. At a high level, the strategies include reducing a passive reservoir, adding a passive sink, or changing a wire’s metal-layer configuration to shift its failure behavior. Because these changes do not depend on narrowing the current-carrying wire, they may have less immediate effect on circuit performance. Their feasibility still depends on the layout, available routing space, current conditions, and the wire’s role.
An EM failure can also serve as a delayed trigger for separate malicious logic. In that concept, the circuit behaves normally until a stressed wire fails; the resulting change in current flow then alters a logic signal or enables a payload. Such a trigger must have an adequate stress source, which can be a design challenge—particularly for signal nets that do not carry enough current or activity to drive the intended aging effect.
Mitigation: reduce exposure and test for aging risks
Two mitigation approaches address different parts of the problem:
- Split fabrication: Separating front-end device manufacturing from back-end interconnect fabrication can reduce the opportunity for an untrusted foundry to alter the interconnects. A trusted organization completes the back-end process. This approach can also limit the design information available to a manufacturing partner, but it requires changes to the fabrication workflow.
- Burn-in testing: Subjecting chips to elevated stress conditions can accelerate aging and may reveal Trojans designed to fail early. Combining burn-in with functional tests and side-channel analysis can improve the chance of detecting a resulting fault. It will not necessarily expose every Trojan: designs targeting longer failure times may remain dormant through the test.
These measures complement, rather than replace, reliability analysis and supply-chain controls. Because topology can influence EM behavior, assessment should consider interconnect structure and multi-segment effects, not only current density. Physics-based analysis can help examine those effects, although it may require more computation and specialized methods than conventional sign-off flows.
Conclusion
Electromigration hardware Trojans exploit the gap between a chip’s initial functional behavior and its long-term reliability. Wire topology, via configuration, and neighboring segments can influence how stress builds and how failure unfolds, making some risks difficult to identify through checks aimed only at immediate logic behavior. For engineering and security teams, the practical response is to treat interconnect integrity as part of hardware assurance: evaluate aging behavior, consider fabrication trust boundaries, and use burn-in alongside other testing methods while recognizing that no single measure guarantees detection.