PSpice-MATLAB Based Evolutionary Algorithm for Automatic Extraction of Parasitics in SiC-MOSFET Test Circuits
Authors
Scholten, Stefan Harmen ; Jensen, Thomas Broberg
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-02
Pages
86
Abstract
This thesis presents a PSpice- and MATLAB-based evolutionary algorithm that automatically extracts parasitic elements in SiC MOSFET test circuits (silicon carbide transistors). The model estimates these unintended circuit effects from DPT measurement data (a common switching test). The extraction combines two optimization methods (PSO and PS), and the objective function uses WT (a time-frequency method) to capture how the frequency content evolves over time. To evaluate how well the simulations match the measurements, the Euclidean distance is used as the metric. Results are compared with a manufacturer's reference model, which shows errors in voltage overshoot and switching energy. The simulations demonstrate that the proposed algorithm achieves a much better match for overshoot, electromagnetic interference (EMI), and switching energies than the reference model. The model aligns well with experimental data, improves EMI fitting below 30 MHz, and better captures switching energy during turn-on and turn-off.
Denne afhandling præsenterer en PSpice- og MATLAB-baseret evolutionær algoritme, der automatisk udtrækker parasitære elementer i SiC-MOSFET testkredsløb (siliciumcarbid-transistorer). Modellen udtrækker de parasitære elementer fra data målt i en DPT (en almindelig omskiftningstest). Udtrækningen kombinerer to optimeringsmetoder (PSO og PS), og målfunktionen bruger WT (en tids-frekvensmetode) til at identificere, hvordan frekvensindholdet udvikler sig over tid. For at vurdere, hvor godt simuleringerne passer, anvendes Euklidisk afstand som mål. Resultaterne sammenlignes med en fabriksreference, som viste fejl i oversving og koblingsenergi. Simuleringerne viser, at den foreslåede algoritme giver en markant bedre tilpasning af oversving, EMI (elektromagnetisk interferens) og koblingsenergier end referencemodellen. Modellen stemmer godt overens med eksperimentelle data, forbedrer EMI-tilpasning under 30 MHz og forbedrer beskrivelsen af koblingsenergi ved tænding og slukning.
[This apstract has been rewritten with the help of AI based on the project's original abstract]