Table 2 shows the number of 5/3 CLBs for various strategies. Second line is number of levels. Number in parentheses is the number of EXOR decompositions. For comparison, CLB count and not DFC values are given. Table 3 shows the DFC-optimized solutions with their times (SPARCstation 5).
3pt
We introduced a new representation and a new general decomposition approach for strongly unspecified multi-output functions. Similarly, extensions for mv logic decomposition have been done. This approach opens several new research areas: input variable re-encoding problem to simplify DDs; using new decompositions in machine learning; efficient solving of combinatorial problems (such as graph coloring); bound set encoding and variable partitioning. Although the preliminary results are very good, we believe we will be able to further improve them with more sophisticated bound set partitioning and encoding algorithms. The method should be also compared with other DFC-based approaches to ML that use SOPs, trees, Curtis decompositions, and ESOPs [14].