Nonlinear robust codes and their applications for design of reliable and secure devices.

摘要

Linear codes are widely used for error detection and correction in modern digital systems. These codes concentrate their error detecting and correcting capabilities on what are considered to be the most probable errors, which are typically errors of a small multiplicity. The reliability and the security of systems protected by these codes largely depend on the accuracy of the targeted error model. In many applications where the error model is hard to predict, the performance of linear codes cannot be guaranteed.;This work is on the development of special classes of codes -- nonlinear robust codes with a given distance, multilinear codes and algebraic manipulation detection codes -- for the design of secure cryptographic devices resilient to fault injection attacks and for the build of reliable memories. The primary difference between the proposed codes and linear codes is that the proposed codes provide nearly equal protection against all non-zero error patterns. As a result, the reliability and the security of the protected devices can be guaranteed regardless of the accuracy of the error model. The advantages of the proposed codes over linear codes will become more significant if the same non-zero error pattern stays for several clock cycles.;The proposed codes are applied for various reliable and secure applications. The error detecting and correcting properties, the area, the power and the latency of the encoder and the decoder for designs based on the proposed codes are estimated and compared to those for designs based on linear codes. It is shown that adopting the proposed codes for the protection of modern digital devices can drastically reduce the number of errors undetected or miscorrected for all codewords thus increasing the reliability and the security of the system at the cost of a reasonable increase in the hardware overhead compared to protection mechanisms using linear codes.