Lee Jung-hyun, a graduate student at the Korea Advanced Institute of Science & Technology (KAIST) TERA Lab, received the Best Student Paper — Honorable Mention at the 2026 Institute of Electrical and Electronic Engineers International Symposium on Electromagnetic Compatibility, Signal & Power Integrity, the semiconductor research group said Thursday.

During the symposium held from Aug. 3-7 in Dallas, Texas, Lee received the award for his paper, “Physics-Aware Tensor Learning for Data-Efficient Multi-Port Power Distribution Network Analysis.”

The paper proposed a new tensor-based method for efficiently storing and analyzing massive amounts of power distribution network (PDN) data generated in advanced semiconductor packages such as high-bandwidth memory (HBM) and chiplet systems.

Using a mathematical technique called Tucker decomposition to extract representative patterns from the data, the research team achieved a compression ratio of about 695 to 1 In a test involving a 256-port PDN, while maintaining the targeted reconstruction accuracy.

“I became interested in tensors while studying how artificial intelligence and supercomputers process high-dimensional data, and applied an idea that emerged from that process to signal integrity and power integrity problems in this research,” Lee said.

“Going forward, I hope to verify the accuracy and interpretability of tensor-based representations in more realistic semiconductor package structures.”

The symposium is one of the world’s leading international conferences on electromagnetic compatibility, signal and power integrity. Held annually since 1959, it brings together about 1,500 to 2,000 engineers and researchers from industry, academia and government institutions.

KAIST TERA Lab is a research group focusing on high-speed semiconductor packaging, signal and power integrity and next-generation memory architectures including HBM and high-bandwidth flash. The group is led by professor Kim Joung-ho, who is known for introducing the fundamental concept, design and architecture of HBM.