cv
This is a description of the page. You can modify it in '_pages/cv.md'. You can also change or remove the top pdf download button.
Basics
| Name | Leo Han |
| Label | PhD Student |
| lxh4@cornell.edu | |
| Url | https://leoxhan.com |
| Summary | Third-year Ph.D. student at Cornell Tech focused on advancing the efficiency and sustainability of data centers across the full computing stack: from physical hardware to resource scheduling and software optimization. Research addresses critical challenges in sustainable cloud computing, including the fair attribution of carbon footprints to individual cloud applications and uncertainty quantification in embodied carbon estimates for computing hardware. |
Work
-
2023.09 - Present Graduate Student Researcher
Cornell Tech
Conducting research in sustainable cloud computing and carbon footprint attribution.
- Developed Fair-CO2, a game-theoretic framework for equitable and efficient attribution of data center carbon footprints to individual users, incorporating colocation interference effects and the impact of dynamic demand on hardware provisioning (ISCA'25).
- Enhanced hardware carbon modeling tools through probabilistic methods that quantify data and model uncertainties in embodied carbon estimates (CarbonClarity, ICCAD'25).
- Designing optimized agent-to-service interfaces for cloud software services that serve AI agent users, focusing on system-level resource efficiency and performance.
-
2023.05 - 2023.08 Undergraduate Student Researcher
University of Toronto
FPGA accelerator design for Bayesian inference algorithms.
- Designed FPGA-based accelerator for residual belief propagation, a Bayesian inference algorithm, leveraging task-level speculative parallelism to achieve scalable performance across many cores.
-
2021.05 - 2022.08 GPU Power Delivery Intern (Professional Experience Year)
Intel
GPU power management and performance optimization.
- Led end-to-end design of a tool for precision testing of GPU compliance to PCIe specifications for input voltage.
- Led enablement of key platform-wide power management and hardware power telemetry features on desktop and datacenter GPUs and accelerators which reduced power excursions by over 95%.
- Improved gaming performance (frames-per-second) by 40% while enabling stricter compliance to power limits on GPUs by tuning power management control loops.
-
2019.05 - 2019.08 Undergraduate Student Researcher
National University of Singapore
Energy harvesting research with dielectric elastomer generators.
- Optimized design of dielectric elastomer generator (DEG) to increase energy harvesting conversion efficiency by 14 times while decreasing prototype size by 4 times.
- Modularized DEG prototype to quickly and easily change between different biasing voltages and different dielectric elastomer capacitances, significantly reducing time required for experimental trials.
Volunteer
-
2019.11 - 2022.08 Toronto, ON
Team Leader
University of Toronto Solar Racing Design Team
Led a team of over 40 dedicated undergraduate and graduate students to design and build a solar-powered race vehicle for the 2023 World Solar Challenge, a 3000 km endurance race in Australia.
- Coordinated technical and business teams to meet project milestones and resource needs.
- Procured over $190,000 in funding and sponsorships through grants and industry sponsorships.
- Co-led design and manufacturing efforts with chief engineer and sub-system leads.
-
2018.09 - 2019.10 Toronto, ON
Electrical and Fabrication Team Member
University of Toronto Solar Racing Design Team
Team member contributing to electrical systems and structural design for solar race vehicle.
- Assembled and validated solar race vehicle's electrical systems for the 2019 World Solar Challenge.
- Designed, simulated, and built major composite structural components for solar race car.
Education
-
2023.09 - Present New York, NY
Ph.D.
Cornell Tech
Electrical and Computer Engineering
- Machine Learning Systems
- Algorithmic Game Theory
- FPGA Architecture
- ASIC Design
-
2018.09 - 2023.04 Toronto, ON
B.A.Sc.
University of Toronto
Engineering Science
- Computer Architecture
- Operating Systems
- Computer Security
- Electronic Devices
Awards
- 2023.09.01
- 2023.09.01
Cornell Tech Digital Life Initiative (DLI) Doctoral Research Fellowship
Cornell Tech
Doctoral research fellowship award of $5,000 USD
- 2023.05.01
NSERC Undergraduate Summer Research Award
Natural Sciences and Engineering Research Council of Canada
Undergraduate summer research award of $7,500 CAD
- 2019.05.01
Engineering Science Research Opportunity Program – Global – Research Fellowship
University of Toronto
Global research fellowship award of $4,000 CAD
Publications
-
2025.01.01 Fair, Practical, and Efficient Carbon Accounting for LLM Serving
CarbonMetrics'25
Workshop paper on carbon accounting for LLM serving. Authors: Yueying Li, Leo Han, Edward Suh, Christina Delimitrou, Fiodar Kazhamiaka, Esha Choukse, Rodrigo Fonseca, Liangcheng Yu, Jonathan Mace, and Udit Gupta
-
2025.01.01 Metrics for Data Center Embodied Carbon
CarbonMetrics'25
Workshop paper on metrics for data center embodied carbon. Authors: Leo Han, Yueying Li, and Udit Gupta
-
2025.01.01 CarbonClarity: Understanding and Addressing Uncertainty in Embodied Carbon for Sustainable Computing
ICCAD'25
Probabilistic methods for quantifying uncertainties in embodied carbon estimates for computing hardware. Acceptance rate: 25%. Authors: Xuesi Chen, Leo Han, Anvita Bhagavathula, and Udit Gupta
-
2025.01.01 Fair-CO2: Fair Attribution for Cloud Carbon Emissions
ISCA'25
Game-theoretic framework for equitable and efficient attribution of data center carbon footprints to individual users. Acceptance rate: 23%. Authors: Leo Han, Jash Kakadia, Benjamin C. Lee, and Udit Gupta
-
2024.01.01 Understanding the Implications of Uncertainty in Embodied Carbon Models for Sustainable Computing
HotCarbon'24
Workshop paper on uncertainty in embodied carbon models. Authors: Anvita Bhagavathula, Leo Han, and Udit Gupta
-
2024.01.01 Towards Game-Theoretic Approaches to Attributing Carbon in Cloud Data Centers
HotCarbon'24
Workshop paper on game-theoretic approaches to carbon attribution. Authors: Leo Han, Jash Kakadia, Benjamin C. Lee, and Udit Gupta
Skills
| Programming Languages | |
| C | |
| C++ | |
| Python | |
| Go |
| Cloud Computing | |
| Docker | |
| gRPC | |
| Kubernetes |
| Machine Learning Systems | |
| PyTorch | |
| Nvidia Nsight Systems/Compute |
| Digital Design | |
| Verilog | |
| SystemVerilog | |
| Xilinx Vitis/Vivado |
| Laboratory | |
| Function generators | |
| Power supplies | |
| Oscilloscopes | |
| Micro-soldering |
Languages
| English | |
| Native speaker |
| Chinese (Mandarin) | |
| Native speaker |
Interests
| Research Interests | |
| Sustainable Cloud Computing | |
| Data Center Efficiency | |
| Carbon Footprint Attribution | |
| Hardware-Software Co-optimization | |
| Machine Learning Systems | |
| AI Agent Workloads | |
| Resource Scheduling |
Projects
- 2023.09 - 2025.01
Fair-CO2: Carbon Attribution Framework
Developed a game-theoretic framework for equitable and efficient attribution of data center carbon footprints to individual users, incorporating colocation interference effects and dynamic demand impacts.
- Published at ISCA'25 (23% acceptance rate)
- Game-theoretic approach to carbon accounting
- Addresses fairness in cloud resource attribution
- 2023.09 - 2025.01
CarbonClarity: Embodied Carbon Uncertainty Quantification
Enhanced hardware carbon modeling tools through probabilistic methods that quantify data and model uncertainties in embodied carbon estimates.
- Published at ICCAD'25 (25% acceptance rate)
- Probabilistic carbon modeling
- Uncertainty quantification framework
- 2024.01 - Present
AI Agent-Service Interface Optimization
Designing optimized agent-to-service interfaces for cloud software services that serve AI agent users, focusing on system-level resource efficiency and performance.
- Cross-stack co-optimization
- Resource efficiency for AI workloads
- Cloud service optimization
- 2023.05 - 2023.08
FPGA-based Bayesian Inference Accelerator
Designed FPGA-based accelerator for residual belief propagation, leveraging task-level speculative parallelism to achieve scalable performance across many cores.
- FPGA hardware acceleration
- Speculative parallelism
- Bayesian inference optimization