DOPP

DOPP improved PPA over Open3DBench while using many fewer evaluations than exhaustive search. DOPP can select candidates for different final PPA preferences rather than returning one fixed proxy-selected solution. DOPP generates proxy-diverse candidates, evaluates an informative subset, fits a surrogate, verifies top predictions, and selects the best evaluated candidate. DOPP selects the final solution from both coreset evaluations and verified top surrogate predictions. DOPP uses PPA evaluation as a selective learning and ranking signal for 3D-IC partition selection.