A self-checking ALU design with efficient codes

Abstract

Recently, a self-testing ALU design has been proposed that uses Berger codes and compares the check value of the ALU output to a predicted check value that is calculated based on the input operand check values. Berger codes have the property of being able to detect all unidirectional errors. More efficient codes exist for detecting up to t unidirectional errors. This paper examines applying these codes to self-testing ALU designs and shows that the potential savings in check circuitry over Berger codes is up to 61%, depending on the code and the information word length.