Abstract
With a rapid increase in operating frequency and package complexity, conventional analysis methods cannot efficiently cope with current complex electromagnetic interference (EMI) issues. In this paper, a new method built on a deep neural network (DNN) model is proposed to accurately and rapidly predict the maximum radiated electric field at 3-meters of a wire-bond ball grid array (WB-BGA) package. The key hyper-parameters of the DNN model, such as the learning rate and type of optimizers, are discussed in depth so as to attain optimal performance. Predicted radiation results by the DNN model and the results of the full-wave simulation show good agreement. Once DNN is trained, the prediction time is in the order of milliseconds and the model size is in megabytes, which can acquire the predicted radiation quickly and accurately and save storage space. Furthermore, to prevent the radiation from exceeding requirements, package structures are optimized by adjusting those parameters sensitive to radiation and disregarding insensitive parameters. The sensitivity of these WB-BGA package structural parameters to EMI radiation can be analyzed quickly based on data with different deviations generated by the trained DNN model. The sensitive parameters are adjusted according to their correlation with EMI radiation. The effectiveness and feasibility of the optimization method are verified by the WB-BGA package with two sets of different structural parameters.
Introduction
As high-speed electronic devices become more structurally complex and operate at higher frequencies, electromagnetic interference (EMI) has become a prominent bottleneck in the design process. Before a device can be sold, it must pass strict regulatory standards, such as the Federal Communications Commission (FCC) requirement that the maximum radiated electric field at 3 meters must not exceed 54 dBuV/m.
Traditionally, engineers use commercial full-wave electromagnetic solvers to calculate this radiated emission. However, these physics-based simulations are computationally expensive and time-consuming. Because designing a package involves tweaking multiple structural parameters to find a compliant design, engineers are forced to run these slow simulations repeatedly. This makes iterative structural optimization computationally intractable and heavily reliant on prior experience rather than systematic exploration.
Methodology
The authors propose replacing the slow full-wave solver with a fully connected Deep Neural Network (DNN) to act as a surrogate model. The DNN uses 7 hidden layers to map 7 key structural parameters of a WB-BGA package—including lid height ($h_1$), dielectric thickness ($h_2$), bonding wire height ($h_3$), dielectric permittivities ($\varepsilon_1, \varepsilon_2$), signal via radius ($r$), and the number of ground vias ($N$)—to the maximum radiated electric field across a 0.2 GHz to 20 GHz frequency band.
Training and Optimization Framework
The training process utilized 2,000 datasets generated via the Halton sequence to ensure broad coverage of the parameter space. The authors found that the Adam optimizer paired with a $10^{-3}$ learning rate provided the most stable convergence. Once trained, the DNN is used to rapidly generate data for a sensitivity analysis. By calculating a "pass rate" for FCC limits across varying parameter distributions, the authors isolate which parameters actually drive EMI. They then use Pearson correlation coefficients to determine exactly how to adjust these sensitive parameters to suppress radiation, creating a deterministic optimization loop.
Key Results and Discussion
The trained DNN achieved a Root Mean Square Error (RMSE) of just 1.87 dBuV/m on the test set, with an average relative error below 2% across the entire frequency band. Crucially, prediction time dropped from hours to milliseconds, and the entire model occupies only 3 MB of storage.
Sensitivity Analysis
The analysis revealed that out of the 7 parameters, only three—the relative permittivity of the top/bottom dielectric ($\varepsilon_2$), the signal via radius ($r$), and the number of ground vias ($N$)—are highly sensitive to EMI radiation. For example, $r$ was found to be positively correlated with EMI, while $N$ is negatively correlated. This provides a clear mathematical directive for designers: to reduce radiation, one should decrease the via radius and increase the number of ground vias.
Conclusion
This research demonstrates that deep learning can effectively replace physics-heavy solvers for iterative design tasks in electromagnetic engineering. By identifying sensitive parameters and their directional correlations with EMI, the proposed framework allows for rapid, automated optimization of WB-BGA packages to meet strict regulatory standards. Future work will focus on applying these methods to the ongoing trend of device miniaturization, where balancing ground via density and EMI performance is increasingly critical.
Summary updated