Hillebrecht, TilTilHillebrechtAlfert, JohannesJohannesAlfertSchierholz, Christian MortenChristian MortenSchierholzHassab, YoucefYoucefHassabSchuster, ChristianChristianSchuster2024-11-292024-11-292024-09-18IEEE Transactions on Electromagnetic Compatibility 66 (6): 1967-1976 (2024)https://hdl.handle.net/11420/52205In this article, a physics inspired and data-driven analysis is performed using a very large dataset of printed circuit board (PCB) based high-speed interconnects. Three different subsets are defined having approximately 15000 PCB material and geometry variations. To the authors' knowledge, this dataset is the largest ever generated and applied dataset for signal integrity (SI) purposes. The dataset is generated using accurate physics based (PB) modeling up to 60 ,mathrm{G}mathrm{Hz} and provided for download in the SI/PI-Database. The first two subsets consist of a single-ended and a differential via array with up to 116 ports with arbitrary positions for signal, power, and ground vias. The third subset represents a link over differential transmission lines between two via arrays. A data and machine learning (ML) analysis is performed showing the dependencies between the PCB parameters and clock frequency of PCI Gen 6. The large and realistic defined design spaces reflect the high complexity and dimensionality of the SI design problems and allow multipurpose ML-based analysis.en0018-9375|||1558-187X202461967-19761967-1976IEEEMachine learning (ML) | SI/PI-Database | signal integrity (SI)MLE@TUHHComputer Science, Information and General Works::005: Computer Programming, Programs, Data and SecurityJournal Article10.1109/TEMC.2024.3450307Journal Article