Comparative Study of Thermal Dissipation in Increasing DRAM Layers of HBM Using 3D FEA Simulations

• Published in: Journal of semiconductor technology and science Vol. 25; no. 2; pp. 142 - 147
• Main Authors: Song, Jeong-Hun, Yoon, SangWon
• Format: Journal Article
• Language: English
• Published: 대한전자공학회 01.04.2025
• Subjects: 전기공학
• ISSN: 2233-4866; 1598-1657; 1598-1657; 2233-4866
• DOI: 10.5573/JSTS.2025.25.2.142

Large language models (LLM) and generative artificial intelligence (AI) require extensive data processing and fast data transfer between components, increasing interest in high bandwidth memory (HBM). The high-speed data processing capability of HBM drives the need for next-generation HBM with additional dynamic random access memory (DRAM) layers. However, this increased stacking leads to more severe thermal issues, along with higher power consumption, potentially limiting HBM performance. This study explores these thermal challenges through 3D finite element analysis (FEA) simulations of simplified HBM models incorporating non-conductive film (NCF) layers. Three models with 4, 8, and 12 DRAM layers were simulated and compared. The results show that the maximum simulated temperature reaches 80◦C, close to the maximum allowable DRAM temperature, and approaches 110◦C, exceeding the recommended operational temperature for HBM. Therefore, this study highlights the potential thermal limitations of highly stacked HBM configurations. KCI Citation Count: 0