Unidirectional error correction by crypto functions

Asymmetric errors are observed in many communication channels or memory systems. Some non-volatile memory systems and optical media exhibit such error behavior in a unidirectional fashion. The lifetime of such units depends often on aging, and especially on the number of access cycles performed resulting with such unidirectional error behavior. In many applications, error correcting codes are not deployed in order to save redundancy, costs and complexity. However in many applications, there are security operations such as encryption or hash functions providing strict rules which can serve to detect and even correct errors. Error correcting codes use basically linear mappings, whereas crypto functions involve highly nonlinear confusion and diffusion randomizing functions. In this paper, some hash functions used for authentication are investigated to check experimentally their symmetric and especially unidirectional error correctability. Good hash functions and ciphers are designed to attain a high degree of pseudo-randomizing mappings. Our experimental results on such hash functions showed that, it is possible to efficiently correct a class of selected unidirectional errors by deploying a portion of the hash bits which are typically used solely for authentication. The proposed correction procedure is a hard decision trade-off operation which would improve the system-survivability by slightly reducing its authenticity level without changing the system configuration. This procedure allows a sort of best-effort salvation in some existing systems. The procedure simply modifies the corrupted data experimentally until the hash value becomes valid. This procedure may however turn out to become equivalent to a successful authentication attack. This is however less probable if only a part of the hash bits is deployed in the error correction procedure and if the corrected error-class is kept as small as possible.