The concurrent simulation of nearly identical digital networks

• Authors: E. G. Ulrich;T. Baker
• Venue: DAC '73 Proceedings of the 10th Design Automation Workshop
• Year: 1973

Injecting a single fault into a fault-free digital network creates a “bad” network which is only slightly dissimilar from the original. Injecting the same stimuli (signals) into both of these networks will produce activity sequences which are often identical, normally almost identical, and rarely substantially different from each other. This similarity between good and bad networks and their activities suggests a method of simulation which avoids the customary duplication of essentially identical good and bad simulations. This method consists of simulating good network activity, and of initiating and performing a concurrent bad network simulation only if bad network activity actually differs from good activity. The run time savings inherent in this method are substantial if hundreds or thousands of bad networks can be simulated in concurrence with a single good network. FANSSIM II is a digital logic simulator under development capable of simulating a 2500 gate network in concurrence with approximately 10,000 single-fault networks. The storage requirements for this simulation are estimated to remain under 450,000 bytes. The effective simulation rate is expected to be above a million signals/dollar, exceeding the real simulation rate of 20,000 signals/dollar for the IBM 360-50 by a factor of 50:1. Some of the techniques and features used are the following: • Fault sources are detected during good network activity and trigger the initiation of concurrent bad network activity. • Fault effects are transmitted piggyback via good signals or separately as independent bad signals. • Fault effects arriving at good gates cause the divergence of bad gates. • Bad gates disappear, or converge, as soon as their inputs and outputs are again in agreement with the associated good gate. • The passage of time is simulated precisely by using assignable rise and fall gate delays. • Feedback, reconvergent fanout, and race detection are handled without special mechanisms. • Economical event handling, desirable here due to accumulation of events of many bad networks, is achieved by using the time mapping6event scheduling technique.