Reproduce “Fine Grain 3D Integration for Microarchitecture Design Through Cube Packing Exploration” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yongxiang Liu, Yuchun Ma, Eren Kurshan, Glenn Reinman, Jason Cong

Researchers from UCLA and Tsinghua University developed a fine-grain 3D integration methodology for microarchitecture design, treating it as a cube packing problem with an architecture-driven co-optimization framework. This framework allows individual logical blocks to span multiple silicon layers, yielding an average 36% performance improvement and 11% lower average temperature compared to 2D single-layer designs, while effectively managing thermal challenges through automated thermal via insertion.

Key insights

  • Fine-grain 3D integration, by allowing logical blocks to span multiple layers, significantly reduces intra-block wire latency, which is a major contributor to overall system performance and power improvements.
  • The optimal 3D implementation for an architectural block varies (e.g., PP for area, BF for delay in some cases), necessitating a dynamic block selection mechanism within a holistic co-optimization framework.
  • Integrated thermal management, including thermal-aware floorplanning and automated thermal via insertion, is crucial for the practical viability of multi-layer 3D designs to mitigate increased power density and maintain acceptable operating temperatures.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.