Projet de recherche doctoral numero :3695


Date depot: 1 janvier 1900
Titre: Detection and Prevention of Hardware Trojans in Integrated Circuits
Directeur de thèse: Jean-Luc DANGER (LTCI (EDMH))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini

Resumé: Nowadays the manufacturing of integrated circuits is generaly made by specialized companies called 'founders' and not by the company who designed the circuit [3]. During the foundry stage, the design can be modi ed and internal data or secret could become accessible. This malevolent insertion is called 'Hardware Trojan' (HT). This potential threat has been taken very seriously by DARPA who launched the  Trust in IC  program in 2007 [8] whose objectives was to design efficient procedures to detect Hardware Trojans. The impact of HT can be very large, the list below describes some potential e ects which can wreak havoc on the trust of the device: -* To switch o remotely a device, as a missile launcher as described in [1]. -* To alter internal nodes to speed up the aging of the device [16]. -* To retrieve a secret like a ciphering key in cryptographic implementation [13]. -* To provide a backdoor which allows a malware to access protected resources [11]. The Hardware Tojans are activated by a trigger which corresponds to a speci c state or sequence of states. Then they can change a net value or activate a parasitic element like a resistor to increase the power consumption. The current methods to detect HT are based either on the insertion of speci c detection block (invasive method) or by using a reference model to compare with the manufactured circuit. For instance the invasive methods consist in adding obfuscating blocks [6], using the Q and Q of all the Flip-Flops [5], or inserting hidden Flip-Flops [12, 15, 9]. The non invasives method can be employed during the test phase, as proposed in [10] and improved in [7] with a higher detection rate. Another non-ivasive method consists in using side-channels and a reference model as presented in [2]. Other stastical techniques have been proposed which used many circuits to characterize its behaviour when the HT is activated, as in [14, 4]. All these techniques are perfectible, as it is impossible to get a detection rate of 100%, the detection is complex and statisticals methods are sensitive to process variation. This PhD subject is part of the FUI14 collaborative project 'HOMERE'. The goal of this thesis is to find efficient ways to avoid malevolent actions from HT.

Doctorant.e: Ngo Xuan Thuy