Defining a channel failure in an optical systolic array processor leads to a simple test for failure and an adjustment procedure that will correct minor failures. Noncorrectable, or major, failures must be bypassed when possible or limited in effect when no further bypassing is possible. This analysis leads to the concept of hierarchies of modules in which failures can be detected and either healed or bypassed. Software to detect instantaneous failures is described. Such a detection triggers additional correction by bypassing.