Saptarshi Joshi · Ph.D. candidate · University of Illinois Urbana-Champaign
Thermal management of AI hardware, from the chip to the data center.
I work on liquid cooling for high-power electronics: manifold microchannel cold plates for next-generation GPU packages, additively manufactured cold plates, dynamic system-level models of liquid-cooled exascale AI data centers, and thermal management of Li-ion batteries. The aim is to remove more heat with less energy, so that AI, power electronics and electrified transport can keep scaling.
Background: a live heat-diffusion solve around a GPU package. Move the cursor to add heat.
Research
Manifold microchannel cold plates for high-power GPU packages
Cold plate design for next-generation GPU packages dissipating 10 kW and more, targeting thermal resistances near 3 K/kW.
Thermal-hydraulic design · Design optimization · Reduced-order and data-driven modeling · Fabrication and testing
Dynamic system modeling of liquid-cooled AI data centers
Transient models of direct-to-chip liquid-cooled data centers, from a single rack to the facility, aimed at water-free heat rejection with dry coolers and a total usage effectiveness near 1.01.
System-level modeling · Transient analysis · Energy efficiency · Heat rejection
Additively manufactured cold plates for GPUs
Monolithic copper cold plates with internal geometry that conventional machining cannot produce.
Design for additive manufacturing · Computational fluid dynamics · Experimental characterization
Li-ion battery thermal management
Two-phase immersion cooling and internal cooling channels for fast-charging cells.
Two-phase immersion cooling · Experimental setup design · Electro-thermal modeling
Other research
My M.S. thesis on cryogenic multilayer insulation, and undergraduate work in heat transfer and fluid mechanics.
Cryogenic and vacuum experiments · Uncertainty analysis · Experimental fluid mechanics · Machine learning
Selected publications
- 2026High-heat-flux liquid cooling of electronics with manifold microchannels: Thermal-hydraulic optimization and experimental validation
International Journal of Heat and Mass Transfer 268, 128924
Pareto-based design optimization of a copper manifold microchannel cooler for a six-device GaN board. Against the baseline design: 70–98 % lower pressure drop and 19–26 % lower thermal resistance, with heat fluxes up to 208 W/cm² and CFD within 3 % of experiment.
- 2026Metal additive manufacturing enables monolithic copper cold plate development with sinusoidal microchannels for direct-to-chip cooling
Advances in Heat Transfer (in press)
A copper cold plate printed as a single part, with wavy microchannels that cannot be machined. Thermal resistance of 0.113–0.137 K/W at 100–600 W, internal geometry within 5 % of design by X-ray CT, and 0.150 K/W junction to coolant on an NVIDIA RTX A4000 at full power.
- 2025Internal flow cooling of cylindrical lithium-ion batteries
Applied Thermal Engineering 281, 128458
Coolant routed through the core of a cylindrical cell to remove heat from the inside during fast discharge.
Currently
- Fourth-year Ph.D. candidate in the Energy Transport Research Laboratory with Prof. Nenad Miljkovic; expected graduation by May 2028.
- Presented our manifold microchannel work for GaN power electronics at ASME InterPACK 2026 in San Diego.
- Preparing a manuscript on dynamic system modeling of direct-to-chip liquid-cooled AI data centers.
- Open to internships for summer and fall 2027, and to full-time industry roles from 2028. My interests are in data center, semiconductor and electric vehicle thermal management. Get in touch.