ML/DL-based Signal Integrity Optimization for InFO Routing

Signal integrity (SI) has been a critical concern in IC/package routing due to the importance in timing and function. However, very few researches discuss related issues in advanced packaging solutions such as InFO (Integrated Fan-Out) structure. This work proposes a new InFO routing flow that improves SI and makes the InFO router SI-aware by integrating well-trained ML/DL models. First, we adopt multiple shielding structures to reduce the impact of crosstalk and obtain eye diagrams to verify the SI improvement. Next, we utilize eye diagram simulation results obtained from the previous step to train the models. Lastly, we integrate the ML/DL models into the InFO router. To evaluate the SI improvement, we apply our new InFO routing flow on several HBM3 testcases, and observe SI improvement in worst eye height ranging from 7.69% to 11.27% and in average eye height ranging from 8.52% to 9.54%.