Visual overview
From the component to the system
The systems my research works on, followed from where the heat is generated to where it leaves the system: AI hardware, from the GPU package to the data center, and Li-ion batteries, from the cell to electric vehicles and battery storage at the data center. Each step links to the research behind it. The drawings are schematic.
AI hardware
01 · ~5 cm
Package
The heat starts in the GPU die and the memory stacks around it. Next-generation packages are expected to dissipate 10 kW or more, at heat fluxes well above 100 W/cm².
02 · ~10 cm
Cold plate
A liquid cold plate on the package carries the heat into the coolant. We design manifold microchannel and additively manufactured cold plates and test them on GPU hardware.
03 · ~2 m
Server and rack
In each server tray the cold plates are joined by coolant supply and return lines. A coolant distribution unit circulates the coolant through the rack and passes the heat to the facility water loop.
04 · ~100 m
Data center
Many racks share one facility water loop, and dry coolers reject their heat to the outdoor air; a large AI data center has thousands of racks. We model this whole chain in time, from the chip to the outdoor air, for liquid-cooled AI data centers.
Li-ion batteries
05 · ~10 cm
Cells
A lithium-ion cell generates its heat inside, and most quickly during fast charging and discharging. Cells perform best between about 15 and 40 °C, with only a few degrees of difference across a cell and between cells. We study two-phase immersion cooling of single cells and small packs, in which a dielectric fluid boils on the cells and holds their surface near the boiling temperature.
06 · ~5 m
Electric vehicle
In an electric vehicle the cells sit in the floor on liquid-cooled plates, and a coolant loop carries their heat to a chiller at the front. Fast charging is usually the largest heat load on the pack. Our work on cooling channels placed inside the cell is aimed at the large prismatic cells used in vehicles.
07 · ~10 m
Battery storage at the data center
Large AI data centers are adding battery energy storage on site: for backup power, to smooth the fast swings in power drawn by AI training, and to support the grid. Each container holds several megawatt-hours of cells and has its own liquid cooling unit. Here the two chains meet: we are developing models of battery energy storage to integrate with our data center facility models.
Why the energy matters
Across both chains the aim is the same: keep every part within its temperature limits with as little energy as possible. AI computing and electric vehicles are among the fastest-growing loads on the grid, and electric vehicles and battery storage are central to decarbonizing transport and power. At that scale, the energy spent on cooling adds up, and so does every saving.