Neural Network-Based Thermal Simulation of Integrated Circuits on GPUs

Abstract

Abstract-With the rising challenges in heat removal in integrated circuits (ICs), the development of thermal-aware computing architectures and run-time management systems has become indispensable to the continuation of IC design scaling. These thermal-aware design technologies of the future strongly depend on the availability of efficient and accurate means for thermal modeling and analysis. These thermal models must have not only the sufficient accuracy to capture the complex mechanisms that regulate thermal diffusion in ICs, but also a level of abstraction that allows for their fast execution for design space exploration. In this paper, we propose an innovative thermal modeling approach for full-chips that can handle the scalability problem of transient heat flow simulation in large 2-D/3-D multiprocessor ICs. This is achieved by parallelizing the computation-intensive task of transient temperature tracking using neural networks and exploiting the computational power of massively parallel graphics processing units. Our results show up to 35× run-time speedup compared to state-of-the-art IC thermal simulation tools while keeping the error lower than 1°C. Speedups scale with the size of the 3-D multiprocessor ICs and our proposed method serves as a valuable design space exploration tool.

Enabling HPC for QoS-sensitive Applications: The MANGO Approach

“…MANGO will extend the experience acquired in the latest research on advanced compact modeling for liquid-cooling monitoring [21] to explore the time constants of thermal and energy control knobs to develop next-generation cooling technologies for HPC systems. In particular, we will explore the use of a novel passive thermosyphon (gravity-driven) cooling technology that will attempt to include multiple parallel heat sources at multiple elevations to eliminate energy consumption.…”

The MANGO FET-HPC Project: An Overview

“…MANGO will extend the experience acquired in the latest research on advanced compact modeling for liquid-cooling monitoring [15] to explore the time constants of thermal and energy control knobs to develop next-generation cooling technologies for HPC systems. In particular, we will explore the use of a novel passive thermosyphon (gravity-driven) cooling technology that will attempt to include multiple parallel heat sources at multiple elevations to eliminate energy consumption.…”

Accelerating thermal simulations of 3D ICs with liquid cooling using neural networks

“…A NNbased thermal simulator for conventional air-cooled 2D/3D ICs was proposed in [9]. The authors proposed to train a neural network to reproduce the dependence in Eq.…”

“…The accuracy of the NNbased simulator depends upon the training algorithm, the quality of inputs, the approximations involved in the construction of the NN and the error tolerance imposed on the training process. Furthermore, the NN-based simulator proposed in [9] exploits the predominantly vertical heat flow that characterizes a conventional air-cooled IC as shown in Figure 2. This implies that there is little lateral spreading of heat, meaning temperatures of nodes in a given layer of IC that are far apart do not affect each other significantly.…”

“…Once trained, the properties of large-scale parallel operations and low data transfer overhead of this NN-based simulator make it an appropriate candidate for implementation on GPUs. Recently, a NN-based simulator for 3D ICs has been presented by the authors of [9]. However, they tackle only the thermal modeling issue of conventional aircooled ICs, without considering liquid cooling.…”