Circuit complexities reduce overall reliability and mean-time-between-
failure rates of today's very large processing arrays. Our integrated,
three-level hierarchy of reconfiguration methods provides reasonable
levels of fault tolerance for such systems. Operating in a completely
distributed fashion, the hierarchy does not require that any component
s be fault free. It significantly improves array reliability by using
a combination of transient fault rollback techniques and local and glo
bal reconfiguration algorithms.