Emulation-based transient thermal modeling of 2D/3D Systems-On-Chip with active cooling

摘要

New tendencies envisage 2D/3D Multi-Processor System-On-Chip (MPSoC) as a promising solution for the consumer electronics market. MPSoCs are complex to design, as they must execute multiple applications (games, video), while meeting additional design constraints (energy consumption, time-to-market, etc.). Moreover, the rise of temperature in the die for MPSoCs, especially for forthcoming 3D chips, can seriously affect their final performance and reliability. In this context, transient thermal modeling is a key challenge to study the accelerated thermal problems of MPSoC designs, as well as to validate the benefits of active cooling techniques (e.g., liquid cooling), combined with other state-of-the-art methods (e.g., dynamic frequency and voltage scaling), as a solution to overcome run-time thermal runaway. In this paper, I present a novel approach for fast transient thermal modeling and analysis of 2D/3D MPSoCs with active cooling, which relies on the exploitation of combined hardware-software emulation and linear thermal models for liquid flow. The proposed framework uses FPGA emulation as the key element to model the hardware components of 2D/3D MPSoC platforms at multi-megahertz speeds, while running real-life software multimedia applications. This framework automatically extracts detailed system statistics that are used as input to a scalable software thermal library, using different ordinary differential equation solvers, running in a host computer. This library calculates at run-time the temperature of on-chip components, based on the collected statistics from the emulated system and the final floorplan of the 2D/3D MPSoC. This approach creates a close-loop thermal emulation system that allows MPSoC designers to validate different hardware- and software-based thermal management approaches, including liquid cooling injection control, under transient and dynamic thermal maps. The experimental results with 2D/3D MPSoCs illustrate speed-ups of more than three ord-ners of magnitude compared to cycle-accurate MPSoC thermal simulators, at the same time as preserving the accuracy of the estimated temperature within 3% of traditional approaches using finite-element simulations for 3D stacks and liquid cooling.