Time-Domain Simulation Based Multi-Domain Optimization of Gallium Nitride-Based DC/DC Converters utilizing Spice Models

Abstract:
Usually electrical engineers are relying on simplified analytical or simulative dimensioning rules in the design of DC/DC converters and then on repetitive redesign iterations. However, this leads to suboptimal designs, because of the huge solution space spanned by all available design parameters, ranging from the circuit architecture to choice of component technologies and sizing. The objective of this work is the development of an automatic multi-domain optimization utilizing the simulation program LTspice interfaced with Matlab to enable rapid prototyping and to provide a comprehensive optimization method. The underlying database of off-the-shelf inductors and capacitors with gravitational, volumetric, temperature and electric characteristics of actual components allows an extraction of a Pareto front for various performance parameters, like power density over converter efficiency. To validate the proposed simulation, a GaN-based 48V to 24V (up to 5 A) converter is analytically described, then simulatively optimized and compared with experimental results. Because of the vast parameter permutations of a power electronics system, a simple parameter sweep requires lots of simulation steps. Therefore, a Genetic Algorithm is used to shorten simulation time by 33 %. With this optimized converter, an approximately 1.5 times higher power density with only 5% of efficiency sacrifice could be achieved, providing a peak efficiency of 97.2%, at 120 W output power.