Research
Heat is the limit. I work on the limit.
AI accelerators are heading past 10 kW per package, racks past a megawatt, and batteries toward ten-minute charges. Each of those is a thermal problem before it is anything else. My work moves between the chip and the building: a cold plate that has to be designed with machine learning because the design space is too large, a rack that has to be modeled in time because the load never sits still, a printed copper part that only exists because of additive manufacturing, and a battery that stays cool because it is boiling.
Manifold cold plates for 10 kW-class AI GPU packages
Next-generation AI accelerators will dissipate 10 kW or more from a package about 80 mm square. I design manifold cold plates for that regime and, more importantly, the methodology that gets there with tens of simulations rather than thousands.
Conjugate CFD · PyFluent automation · Reduced-order modeling · Gaussian processes · Active learning · Additive manufacturing · Calorimeter-bar testing
Dynamic system modeling of liquid-cooled AI data centers
AI racks now draw 100–350 kW and every one is liquid cooled, yet nearly every model of that cooling is steady state while real loads swing by the second and weather moves the heat-rejection limit by the hour. I built a component-resolved transient model of a whole rack to answer the operational questions.
Simscape Fluids · Control design · Stiff DAE solvers · System identification · HPC batch campaigns · Energy metrics (TUE, PUE) · Reliability modeling
Additively manufactured cold plates for GPUs
Metal additive manufacturing lets a cold plate be printed as one piece with internal geometry no machine tool can cut. With 3D Systems and IQ Evolution I design, print, simulate and test such plates for GPU-class heat loads.
Design for additive manufacturing · Laser powder-bed fusion · Conjugate CFD · GPU-level testing · TPMS lattices
Li-ion battery thermal management
Fast charging heats lithium-ion cells; heat shortens their life and can trigger fires. I built and run a sealed two-phase immersion rig and lead modeling of coolant channels placed inside large prismatic EV cells.
Two-phase immersion · Rig design and commissioning · Battery cycling · Electro-thermal modeling · COMSOL and Fluent ECM · CT-validated geometry
Where this goes
Component-level cooling and system-level architecture are one problem.
The cold plate sets what the rack can do; the rack's dynamics set what the cold plate must tolerate. I want to keep working at that seam, in industry, on data-center thermal management, semiconductor packaging thermal, or EV thermal systems, where the next factor of two in power density is waiting on the thermal engineer.