Statistical analysis of HBM channel performance
High-bandwidth memory (HBM) is a high-performance memory interface to achieve high bandwidth, low power dissipation, and advanced packaging, which uses more than a thousand channels. Thus, large power noise could occur due to the switching of thousands of I/O drivers, and HBM channel performance could be seriously affected. The time-varying nature of the power noise makes the HBM channel non-linear and the conventional statistical methods cannot accurately account for this non-linearity. In this talk, a statistical eye-diagram estimation method is presented to design and evaluate the HBM channels including both signal and power nets, where the power distribution network is modeled using lumped elements and I/O driver is assumed as an inverter using the 180nm MOS technology. The proposed statistical method is based on a multiple edge response (MER) approach, and the Data Bus Inversion (DBI) coding is implemented in the proposed method.
Chulsoon Hwang, Missouri University of Science and Technology
Chulsoon Hwang is with EMC laboratory at Missouri S&T (Formerly University of Missouri-Rolla) as an assistant professor. He received his PhD degree at KAIST in 2012 and worked at Samsung Electronics as a senior engineer from 2012 to 2015. He has authored or co-authored 50+ IEEE journal/conference papers and received a best paper award at AP-EMC 2017, DesignCon 2018 and an outstanding young scientist award at AP-EMC 2018. His research area includes RF desense, signal/power integrity, and electromagnetic interference.