Abstract
We propose a new n-dimensional fault-tolerant hypercube architecture. We use (n-1) switching networks to connect the N=2/sup n/ active processors and k spare processors to form an n-dimensional fault-tolerant hypercube. The k spare processors can be used to back-up any k processor failures in the fault-tolerant hypercube. We call such a method global sparing and it is optimal in terms of the number of processor failures that can be back-up by k spare processors. The new architecture can achieve a higher level of reliability using less hardware compared to previously proposed schemes.