Thermal Analysis and Design Guideline of Massive Orthogonal Stacking Assembly of IC (MOSAIC) Cube with Bump Connection Enabling SoC-DRAM Direct Stacking
Authors:
- 27 May 2025
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1595-1600
Abstract
The rapid growth of large language models demands unprecedented memory capacity and bandwidth, driving the need for innovative memory architectures. While HBM technology continues to advance, huge thermal resistance accompanied by more stacked memory chips and assembly costs impede further scaling. We propose Massive Orthogonal Stacking Assembly of IC (MOSAIC), a novel approach that arranges DRAM dies vertically like books on a bookshelf, enabling increase in the number of chip stacking without increase of thermal resistance. Through comprehensive thermal simulations, we demonstrate MOSAIC Cube's superior heat dissipation capabilities compared to conventional HBM-alike structures, enabling significantly higher chip stacking while maintaining thermal stability. Our study establishes design guidelines for optimal implementation, addressing critical parameters including chip height, memory thickness, and power density considerations. The architecture achieves substantially higher I/O connectivity while eliminating TSV-related constraints, demonstrating MOSAIC's potential as a viable solution for next-generation memory systems in high-performance computing and AI applications.Keywords
This publication has 11 references indexed in Scilit:
- Thermal and mechanical simulations of 3D packages with custom high bandwidth memory (HBM)2024
- Present and Future, Challenges of High Bandwith Memory (HBM)2024
- 13.4 A 48GB 16-High 1280GB/s HBM3E DRAM with All-Around Power TSV and a 6-Phase RDQS Scheme for TSV Area Optimization2024
- NVDRAM: A 32Gb Dual Layer 3D Stacked Non-volatile Ferroelectric Memory with Near-DRAM Performance for Demanding AI Workloads2023
- A 192-Gb 12-High 896-GB/s HBM3 DRAM With a TSV Auto-Calibration Scheme and Machine-Learning-Based Layout OptimizationIEEE Journal of Solid-State Circuits, 2022
- A study of micro-scale solder bump geometric shapes using minimizing energy approach for different solder materialsAin Shams Engineering Journal, 2022
- 3D V-Cache: the Implementation of a Hybrid-Bonded 64MB Stacked Cache for a 7nm x86-64 CPU2022
- Temperature-Aware DRAM Cache Management—Relaxing Thermal Constraints in 3-D SystemsIEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2019
- A Novel 3D DRAM Memory Cube Architecture for Space Applications2018
- HBM: Memory solution for bandwidth-hungry processors2014