The main reason, I think, for blackbox extraction methodology, is to reduce the extraction time and size of a post-layout netlist.
This benefit is achieved if you have a cell that is instantiated multiple times in a hierarchical design (up to dozens or hundreds of times).
The idea is that you extract a cell once, then instantiate it multiple time.
The cell is extracted at its level, once.
Then, you do top-level extraction, with black boxing (or grayboxing) of extracted cells.
Then, in SPICE deck, or in hierarchy editor, you instantiate the top level block, and lower-level cells.
The drawback of this approach is that many parasitic effects get missed, or get messed up.
For example:
- coupling form upper level routing to inside the cell - will be decoupled (i.e. grounded), so that if there is a critical aggressor to victim coupling - it will be completely missed.
- capacitive interaction between the blackboxes cells.
- resistive effects - if long-range routing goes through the cell hierarchy level.
- many others.