Automating Logic Rectification by Approximate SPFDs

In the digital VLSI cycle, a netlist is often modified to correct design errors, perform small specification changes or implement incremental rewiring-based optimization operations. Most existing automated logic rectification tools use a small set of predefined logic transformations when they perform such modifications. This paper first shows that a small set of predefined transformations may not allow rectification to exploit the full potential of the design. Then, it proposes an automated simulation-based methodology to "approximate" sets of pairs of functions to be distinguished (SPFDs) and avoid the memory/time explosion problem. This representation is used by a SAT-based algorithm that devises appropriate logic transformations to fix a design. The SAT method is later complemented by a greedy one that improves on runtime performance. An extensive suite of experiments documents the added potential of the proposed rectification methodology.