EMI characterization for power supplies and machine learning based modeling in EMC/SI

Signal integrity (SI) and electromagnetic interference (EMI) are essential for consumer electronics and high-speed server applications. It is necessary to do EMI and SI modeling at the design stage. In this research, several modeling approaches for EMI and SI problems are proposed. By using measurement-based modeling method, the mechanism of conducted emissions (CE) on ac to dc power supplies in an LED TV is analyzed. A system-level transient simulation model is built to predict the conducted emission (CE). To characterize the equivalent dipole moment sources in RFI and EMI problems, a dipole source reconstruction method based on machine learning techniques is proposed. The picture of the electromagnetic field is fed to the convolutional neural network, and the CNN performs a multi-label classification to determine all types of dominant dipole moments. The CNN also generates a class activation map, which indicates the locations of each type of present dipole moment. By using the integer programming method, a PCB stack-up design method is proposed. In high-speed PCB designs, a design with 30 layers or more is not uncommon and there are many logical design constraints that need to be considered. The constraints are converted to mathematical inequalities using the integer programming technique. Then an integer program solver is called to get all possible combinations. The number of possible combinations gets large as the number of layers increases, and the proposed method is much more efficient than brute-force searching. After a nominal design is picked, a searching algorithm based on integer programming is further used to find corner cases by considering the manufacturing variations --Abstract, page iii