Hi Hao,
The Berkeley ABC implementation of IC3 (referred to as PDR) uses Minisat
and is much faster than the original implementation of IC3. The
implementation of PDR has the option of using a monolithic SAT instance,
but by default uses different SAT instances for each time frame. Part
of the reason for this is to only include logic relevant for proving a
proof obligation (or the property) in each frame. Most of the power of
PDR (IC3) comes from how well the generalization step is performed.
Perhaps there are some optimizations in your implementation that you are
not using? e.g., ternary valued simulation, generalization by MUC
inspection, restarting the instances corresponding to each from when the
number of learned clauses reaches a threshold, etc ...?
- John