Morphology-Guided Dual-Hypermutation Immune Algorithm for Multichip PDN Decoupling in Space-Constrained High-Density SiP

Abstract

To address the coupled challenges of wideband impedance control, ultralow power-noise tolerance, and strict spatial limitations in power distribution networks (PDNs) for high-density system-in-package (SiP) designs, this study develops a morphology-guided dual-hypermutation immune algorithm (DHIA). The developed method incorporates a physics-aware violation-mapping mechanism based on impedance morphology, together with morphology-guided directed hypermutation, probability-modulated stochastic hypermutation, and concentration-controlled competitive selection. This hybrid evolutionary framework tackles two critical issues that often hinder conventional optimization approaches under multidimensional constraints and high-dimensional decision spaces: the low overall value of feasible solutions and the poor optimization efficiency caused by severe constraint coupling. Compared with five representative baseline methods, DHIA achieves the lowest final fitness under weak-reward weighting while retaining the minimum capacitor count and reducing the occupied area of high-value layout real estate by approximately 9.9% to 14.1%. Under strong-reward weighting, DHIA further reduces the occupancy of this region by approximately 9 percentage points while avoiding excessive reward-driven component accumulation. Ablation, repeatability, and parameter sensitivity analyses further demonstrate the effectiveness and stability of the proposed evolutionary mechanisms. Finally, the calculated impedance results are validated through electromagnetic comparison in the SiP scenario and further verified against simulation and measurement using a scaled PCB prototype, supporting the engineering applicability of the proposed framework.