Aghiles DOUADI

Abstract: All security primitives, when implemented in hardware, are subject to physical attacks. Among those, fault injection attacks consist in disturbing the operation at run-time to obtain secret data or unauthorized access. One particularly powerful technique to inject faults in a device is laser fault injection. For instance, using an infrared laser, it is possible to alter the value stored in a flip-flop or to corrupt the instructions when they are fetched from the Flash memory.
While there are many possibilities offered by this technique, the associated so-called attacker model is quite restrictive. Indeed, it is necessary to have a direct access to the backside of the die to shoot the laser on it. This severely limits the applicability of the technique in a real-life attack scenario.
The aim of this thesis is to study the feasibility of laser fault injection attacks on power-off targets. One important aspect of the project is to design countermeasures that are fully effective at detecting that a power-off attack has been carried out. This implies that the countermeasure does not need to be constantly powered on to operate. Therefore, when powered on again, the proposed sensor should indicate if the structure has been attacked or not. Beyond detection, structures which are resistant to these attacks will be studied too.
The objectives of this PhD are:
• to understand/simulate the effect(s) of power-off laser fault injection on hardware security primitives,
• to investigate techniques to detect that a power-off attack happened,
• to design hardware security primitives that are resistant to power-off attacks,
• to design hardware security primitives that can be tested on-line.