Education
Stanford University
Sept. 2024 – Current
PhD Student, Electrical Engineering Department
- Advisor: Dr. Philip Levis
- Full fellowship from Samsung Electronics
The University of Texas at Austin
December 2012
Bachelor of Business Administration, High Honors · Bachelor of Arts, Economics, Highest Honors · Minor in Mathematics
Industry
Samsung Electronics Memory Division, Software Development Team
Senior Software Engineer, Associate Architect; Hwaseong, Korea
March 2023 – August 2024
- Developed firmware for NVM Express enterprise solid-state drives supplied to Oracle, Dell EMC, IBM, and HP
- Served as associate architect in defining new modules for metadata management and open time optimization
- Focus: metadata management, open time optimization, NAND flash exception management, power-loss protection
Software Engineer; Hwaseong, Korea
January 2016 – February 2023
- Graduated an intensive 16-month training program designed to provide comprehensive knowledge equivalent to an undergraduate degree in Electrical and Computer Engineering and received the best trainee award
- Developed firmware for NVM Express and SAS enterprise solid-state drives supplied to Dell EMC, IBM, and HP
Awards
CES Innovation Award
Jan 2022
World’s first PCIe Gen5 enterprise SSD (Samsung PM1743)
- Keeping the whole metadata cache in DRAM forces a DRAM budget near one thousandth of device capacity, which caps how large a drive can get.
- Built a partial DRAM scheme using NAND flash as the final cache stage, segmented for hit rate, with read and write caches split so write latency hides entirely.
- Added a dirty-cache program scheme that writes back only dirty entries rather than every cache line.
- Let drives ship past 32TB on less DRAM.
Presidential Award, Republic of Korea
Sept 2019
World’s first 32TB dual-port SAS enterprise SSD (Samsung PM1643)
- Sized metadata blocks against lifetime program and erase cycles at three drive writes per day over five years, returning the surplus to user capacity and raising throughput by 24%.
- Split metadata work into atomic units, cutting metadata-induced latency by 30%.