"A system that needs to be highly reliable in the presence of unavoidable transient hardware faults, such as uncorrectable errors caused by cosmic particle strikes, may require hardware support for checkpoint recovery so that upon detection the execution can be restarted from an earlier correct state."