Thermal Design Space Exploration of 3D Die Stacked Multi-core Processors Using Geospatial-Based Predictive Models

This paper presents novel 2D geospatial-based predictive models for exploring the complex thermal spatial behavior of three-dimensional (3D) die stacked multi-core processors at the early design stage. Unlike other analytical techniques, our predictive models can forecast the location, size and temperature of thermal hotspots. We evaluate the efficiency of using the models for predicting within-die and cross-dies thermal spatial characteristics of 3D multi-core architectures with widely varied design choices (e.g. microarchitecture, floor-plan and packaging). Our results show the models achieve high accuracy while maintaining low complexity and computation overhead.