Your FPGA's Tamper Sensors Are the Tamper Surface

FPGA anti-tamper sensors — the last line of defense against physical attacks — can be bypassed without invasive tampering. This paper maps the attack surface nobody modeled.

Imagine a bank vault with motion sensors on every wall. You'd assume an intruder has to deal with the sensors before cracking the safe. But what if the sensors have blind spots that shift with the weather, degrade with age, and can be steered around by someone who understands the building's electrical wiring? That's what this paper demonstrates for FPGA anti-tamper sensors — the very components trusted to detect and respond to physical attacks. The core claim is direct: state-of-the-art FPGA anti-tamper sensors, deployed as both hard-IP blocks and soft-IP cores, are a fragile and underexplored attack surface. Both their detection mechanisms and their response mechanisms can be bypassed without invasive tampering. This isn't a theoretical concern — the authors draw on recent experimental results to show these bypasses are practical. The attack surface splits into three categories that standard threat models miss. First, the spatial and frequency-dependent behavior of the chip's power delivery network (PDN) means a carefully shaped voltage glitch can hit a target circuit while the nearby sensor sees nothing — the PDN itself acts as a frequency-selective shield. Second, environmental temperature shifts move detection thresholds, widening or narrowing the window an attacker can exploit. Third, long-term transistor aging (bias temperature instability, hot-carrier injection) drifts sensor calibration over months and years, potentially opening silent detection gaps that no one recalibrates for. Architecturally, this is an analog-security paper grounded in the physics of on-chip power distribution and semiconductor aging, not a digital-logic or cryptographic attack. The methods rely on understanding PDN impedance profiles, thermal sensitivity curves, and NBTI/HCI degradation models. The attacker's toolkit is an EM probe or a laser fault injector plus knowledge of the chip's physical layout — not a billion-dollar fab. The integrity picture is honest but limited. The paper synthesizes recent experimental results rather than presenting a single new end-to-end attack; it functions more as a systematization-of-knowledge (SoK) contribution. The cited bypass demonstrations are real, but the aging and temperature arguments are supported by known semiconductor physics rather than new long-duration experiments on production sensors. No independent replication of the combined attack chain exists yet. The proposed countermeasures — hardening both detection and response paths — point toward the milestone the field needs: sensors whose security guarantees hold across the chip's full operational lifetime, not just at room temperature on day one. The gap between 'works on the test bench' and 'works after 10 years in a desert avionics bay' is where the real engineering challenge lives. The obvious experiment not run: a multi-year accelerated-aging campaign on commercial FPGA sensor blocks, measuring detection-threshold drift under realistic mission profiles. The likely reason is resource constraints — accelerated aging studies on real silicon are expensive, slow, and require access to commercial hard-IP internals that vendors don't share willingly. This is the experiment that would convert the paper's physics-based argument into a field-deployable vulnerability disclosure.