<italic>DefScan</italic>: Provably Defeating Scan Attack on AES-Like Ciphers

Abstract

Scan-based Design-for-testability (DfT) is the de facto standard in the semiconductor testing industry to guarantee the functional and structural correctness of chips. It provides improved observability and controllability, leading to enhanced fault coverage. However, owing to widespread usage, attackers devise techniques to misuse this method to steal secret keys embedded in a security-critical chip. A vast majority of off-the-shelf defense mechanisms are based on either randomizing the scan output or restricting access to the scan. However, none of these defense mechanisms leverage the fundamental properties of the scan attack

thus, they tend to be complex and incur high area and computation overhead. In this paper, we propose a defense mechanism by preventing the vulnerabilities of fundamental properties from being exploited in scan attacks. The paper first pinpoints the ultimate condition of the scan attack on Advanced Encryption Standard (AES). This attack condition is inherent to the cryptographic property of the cipher, which, when violated, thwarts the attack. To implement our defense, we interchange the AES round outputs by applying pre-computed masks. The designer chooses inputs with a fixed difference to swap the outputs. A modified incorrect key is recovered instead of an actual key if a scan-based attack is launched. To show the generality of the proposed defense, it is extended to another AES-like cipher, Light Encryption Device (LED). To the best of our knowledge, this is the first defense against a scan attack wherein the complete testing process, including structural and functional tests, can be outsourced to untrusted third parties without compromising the actual key. In comparison, logic locking techniques limaye2020thwarting can outsource only structural testing to untrusted third parties. IEEE