I think, functionally speaking, a software failure is no different to that of hardware-based control / instrumentation systems (think of all the odd behaviour that could be caused by a short circuit or a bad connection somewhere in the wiring). The difference is that software systems tend to be more complex, and failures are in theory less random and more systematic - it's more a case of stepping on an unlikely combination of inputs (i.e. a bug that occurs in an untested corner-case) than encountering a random component failure which is more likely the case for a hardware system.

Edit: actually I suppose I just made your point