Achieving fault-tolerance by shifted and rotated operands in TMR non-diverse ALUs

A novel approach to the implementation of majority voting ALUs is presented. Differently from classical TMR systems, it allows to achieve high fault-tolerance ability with respect to common-mode failures, without requiring the use of diverse designs of the replicated ALUs. Diversity is obtained in a logical meaning by properly re-arranging the input operands of the three replicated ALUs. Compared to traditional TMR systems, our method offers significant savings in terms of design costs and, therefore, is suitable to allow a broader use of fault-tolerance (still limited by its significant costs), for instance to solve the reliability problems of next generation, very deep submicron systems.