Fingerprint-Based Detection and Diagnosis of Malicious Programs in Hardware

Fingerprint-Based Detection and Diagnosis of Malicious Programs in Hardware In today’s Integrated Circuit industry, a foundry, an Intellectual Property provider, a design house, or a Computer Aided Design vendor may install a hardware Trojan on a chip which executes a malicious program such as one providing an information leaking back door. In this paper, we propose a fingerprint-based method to detect any malicious program in hardware. We propose a tamper-evident architecture (TEA) which samples runtime signals in a hardware system during the performance of a computation, and generates a cryptographic hash-based fingerprint that uniquely identifies a sequence of sampled signals. A hardware Trojan cannot tamper with any sampled signal without leaving tamper evidence such as a missing or incorrect fingerprint. We further verify fingerprints off-chip such that ahardware Trojan cannot tamper with the verification process. As a case study, we detect hardware-based code injection attacks in a SPARC V8 architecture LEON2 processor. Based on a lightweight block cipher called PRESENT, a TEA requires only a 4.5% area increase, while avoiding being detected by the TEA increases the area of a code injection hardware Trojan with a 1 KB ROM from 2.5% to 36.1% of a LEON2 processor. Such a low cost further enables more advanced tamper diagnosis techniques based on a concurrent generation of multiple fingerprints.