I study heat transport in advanced microelectronics through thermal characterization, electrothermal simulation, and the development of numerical and machine-learning models.
Research overview
Materials and Interfaces
Frequency-domain thermoreflectance
Thin films & interfaces
Infer depth-dependent thermal properties from thermoreflectance data.
Explore moreDevices and Circuits
FinFET arrays
Illustrative temperature fields
Predict transient temperatures and thermal crosstalk in FinFET arrays.
Explore morePackages and Systems
Processor · power conversion
Embedded microfluidic cooling
Evaluate power delivery and cooling through coupled electrothermal models.
Explore moreMaterials and Interfaces
Thermal properties in nanoscale films can differ substantially from bulk values, and frequency-domain thermoreflectance (FDTR) estimates these properties through inverse analysis of measured surface responses. I develop a physics-informed neural network surrogate for FDTR to estimate depth-dependent thermal properties. My collaborative work examines the relationship between grain orientation and thermal conductivity in aluminum nitride films, and quasi-ballistic heat transport in vertically aligned hexagonal boron nitride.
Devices and Circuits
Self-heating and thermal crosstalk affect the performance of fin field-effect transistor (FinFET) arrays, yet repeated transient simulations are computationally expensive. I develop deep operator network (DeepONet) models that learn from finite-volume simulations to map the current temperature field and power dissipation profile to the temperature field at the next time step. For a 3.5 ns transient response, the trained model achieved approximately 2,500-fold speedup over the finite-volume solver, with over 92% of nodal errors below 3 K. Collaborative device-to-circuit work examines how self-heating in gate-all-around nanosheet transistors with backside power delivery affects circuit performance.
Packages and Systems
Power conversion close to a processor shortens electrical paths but concentrates heat beneath the chip. I develop automated, self-consistent electrothermal models of distributed vertical power delivery architectures with substrate-embedded microfluidic cooling. These models capture the feedback between device temperatures and electrical losses to determine cooling requirements. For a 48-to-1 V architecture at a 1 kW load, the coupled model predicted about 15.8% higher system losses than the initial uncoupled estimate.
At the same total module power, changes in spatial power loading can alter local temperatures. I develop DeepONet models for temperature prediction and use the models to infer spatial power loading from temperature observations. The predictions also guide the selection of candidate operating states within component temperature limits.
I develop STREAM, a numerical solver for transient heat transfer with prescribed velocity fields, and rationally enriched Chebyshev trunk bases for DeepONet models of high-Péclet-number entrance transport.
Earlier work: turbine rim sealing
At Seoul National University, I constructed and validated a turbine test rig to study hot-gas ingestion and rim sealing. I developed a patented rotor-disk attachment, the “swirler,” to investigate how enhanced wheel-space swirl improves sealing performance.