Security-Oriented Encoding of Robust Codes for Non-Uniformly Distributed Words
The security of cryptographic devices is threatened by fault injection attacks on the hardware. By injecting faults an adversary can obtain secret information stored in the device. Modern fault injection techniques allow an adversary to introduce faults at any physical point of the circuitry. A fault can flip bits, cause a gate stuck-at effect, or change data on wires. Fault injection attacks can be detected using robust and partially robust codes. The codes are usually designed for uniformly distributed codewords. In practice, however, there are codewords that are much more likely to appear than others. A judicious attacker can use this information to choose an error that is hardly (if ever) detected.
In this talk we address two questions:
a) How good are the known robust codes, and in particular the optimum Quadratic Sum codes and codes derived from the quadratic code, against an adversary that knows the distribution of the codewords?
b) Is it possible to reduce the error masking probability of the code without adding more redundancy?
* The research presented in the seminar was carried out towards the M.Sc. degree, under the supervision of Dr. Osnat Keren