Hot-LEGO: Architect Microfluidic Cooling Equipped 3DICs with Pre-RTL Thermal Simulation
Runxi Wang, Jun-Han Han, Mircea Stan, Xinfei Guo
TL;DR
This work tackles the challenge of thermal-aware design for microfluidic-cooled 3DICs in the early design phase, where traditional tools struggle with speed or microfluidic modeling. It introduces Hot-LEGO, a Pre-RTL thermal simulation workflow built on and extending existing simulators (including Gem5, CACTI, McPAT, and HotSpot) to model performance, power, and convection-based cooling across vertical die stacks. The framework enables fine-grained, architecture-level exploration of cooling-backed stack configurations and facilitates rapid co-design between cooling, microarchitecture, and packaging. Preliminary results with SPLASH-2 and PARSEC demonstrate cooling-induced temperature reductions and shifts in hotspot distributions, highlighting the potential to guide early architectural decisions. Overall, Hot-LEGO aims to speed up the co-design cycle for microfluidic cooling in 3DICs and provide a portable methodology for exhaustive exploration of microarchitecture options.
Abstract
Microfluidic cooling has been recognized as one of the most promising solutions to achieve efficient thermal management for three-dimensional integrated circuits (3DICs). It enables more opportunities to architect 3DICs with different die configurations. It becomes increasingly important to perform thermal analysis in the early design phases to validate the architectural design decisions. This is even more critical for microfluidic cooling equipped 3DICs as the embedded cooling structures greatly influence the performance, power, and reliability of the stacked system. We exploited the existing architectural simulators and developed a Pre-register-transfer-level (Pre-RTL) thermal simulation methodology named Hot-LEGO that integrates these tools with their latest features such as support for microfluidic cooling and 3DIC stacking configurations. This methodology differs from existing ones by looking into the design granularity at a much finer level which enables the exploration of unique architecture combinations across the vertical stack. Though architectural-level simulators are not designed for signoff-calibre, it offers speed and agility which are imperative for early design space exploration. We claim that this ongoing work will speed up the co-design cycle of microfluidic cooling and offer a portable methodology for architects to perform exhaustive search for the optimal microarchitecture solutions in 3DICs.
