The last decade has seen sudden change and dramatic growth
in many electronics industries including
transportation, artificial intelligence, cryptocurrency,
robotics, high-performance computing, etc. Power
converters are critical for all of these applications, but
they are becoming a bottleneck to system
performance. Ideally, power electronics are invisible,
simply providing the exact current and voltage
required
for a system, with perfect efficiency and without undue
cost or size. In practice, they are
often expensive, inefficient, or larger than the system
they are servicing. In order to enable next-generation
technologies in these crucial industries, we need to
realize a step-change in power converter
performance.
High-Performance Converter Design
Much of my doctoral research has focused on improving the design of high-performance inverters to facilitate the advancement of electric aircraft. I’ve explored various limitations in
converter modeling, component selection, layout, and packaging. This work culminated in the design of a 10
kW, 14-level FCML inverter which achieved a power density >10x higher than the existing state of the art.
I’ve since conducted more detailed optimizations and have developed a new prototype which is projected
to attain another 3x reduction in size and greater than 1 MW/L power density. I also developed a novel
converter design for direct 400V-to-1V operation which could replace 3 stages of conventional power
conversion, thus enabling enhanced performance for next-generation data centers.
Thermal Management Techniques
The density and power consumption of modern graphics processing units (GPUs) and central
processing units (CPUs) is growing rapidly. Power converters are also reaching
unprecedented heat fluxes, and becoming constrained by thermal limitations. As a result, next-generation
electronics will require advanced liquid cooling solutions.
Along with the Energy Transport Research Lab (ETRL) at the University of Illinois Urbana-Champaign, I
helped explore novel dielectric coatings enabling immersion cooling with water.
Existing options for characterizing the performance of liquid coolers are unable to meet targets for
cost and configurability. It is risky, time-consuming, and expensive to test directly with end-use electronics,
while resistive heaters do not provide a realistic substitute. We have developed a novel thermal test vehicle
(TTV) architecture based on an array of power transistors. The devices can be individually
controlled to provide customizable heat generation and transistor junction temperature may be estimated
using measured on-resistance. Overall, this enables a more flexible, scalable, and affordable approach
for evaluating cooling solutions.
Design Process Improvement
There are countless power converter topologies, each with different components, loss mechanisms,
control techniques, etc., and many more are invented each year. Every application is unique, while some
have malleable parameters. As a result, it is difficult for a power electronics designer to understand which
topology is best-suited for a given situation. My goal is to develop techniques which simplify and clarify
the design process.
Existing methods for the estimation of passive component volume in a power converter are not
suitable for the analysis of transformers, omit certain scaling trends of inductors, and ignore passive
component losses. I've presented an improved approach which utilizes more precise methods to model
the magnetic components and incorporate losses. Although this theoretical analysis is useful, converter performance often
depends on many practical constraints such as cost, component availability, thermal limits, etc. As a result, a
more bespoke design optimization is necessary even after the topology is selected. Therefore, I've also developed a
detailed loss model and multi-objective optimization framework to refine the subtler aspects of the design.
