You need to look at the herarchical partitioning flow. For doing hierarchical P&R the top level needs to be partitioned into sub-blocks. A separate work area directory is created for each sub-block. The sub-blocks are placed&routed independently and a timing model (ILM or .lib) needs to be created. When the sub-blocks are completed, there are imported back into top level with all information including timing and DEF.
I've attached a Cadence app noteon the subject. It's a few years old so some details may be different but I think the general flow is the same.