Multiscale Simulation of Rarefied Gas Flow for Engineering Design

微处理器芯片存在于我们日常生活中与之互动的大多数设备中。从移动的设备、电视、汽车、冰箱、加油站泵、为网络和社交媒体基础设施提供动力的服务器-这个名单是无穷无尽的。根据摩尔定律，微处理器的功率大约每两年翻一番，这是通过使芯片的功能更小，每单位面积安装更多的晶体管，并推动整个消费电子市场每年价值超过1万亿英镑。为了继续满足推动摩尔定律的工业和社会需求，我们需要克服一些流体动力学建模挑战。下一代光刻机需要制造更小，更快的微处理器芯片和在操作期间过冷高性能芯片所需的新设备，可以通过理解和准确预测气体在微米/纳米尺度或类似真空条件下的行为来实现。在这些多尺度流动问题中，流体动力学通常是不直观的，并且我们通常用于建模和设计工程流动问题的所有方程，例如使用Navier-Stokes方程的飞机和船舶周围的流动，在这里不再有效，因为气体不再处于局部热力学平衡，这些经典方程在此基础上制定。直接模拟蒙特卡罗（DSMC）是模拟这些非平衡气体流动的最先进的软件。它是一种数值稳定性很高的随机粒子方法，可以求解三维几何中气体的分子性质。然而，因为它是一种粒子方法，它需要贪婪的计算成本来产生对工业重要的规模的工程解决方案。DSMC也表现不佳，如果这些流动是在低速，由于固有的热噪声的颗粒阻挡可测量的signals.In这个项目中，我们建议开发一种新的多尺度方法，一个DSMC与计算成本较低的模型，如计算流体动力学（CFD）中使用的。我们将使用替代建模和贝叶斯推理将DSMC和CFD求解器连接起来，从而在模拟效率和准确性方面产生重大变化。在我们的工业合作伙伴的大力支持下，我们将把该项目的成果转化为免费的开源计算求解器，并在英国的OpenFOAM软件中发布，并通过实验数据进行验证。该项目的工业重点将是处理器芯片制造，芯片热管理和电喷雾技术，但基本方法是一般的新方向，在其他研究和工业领域。

Microprocessors chips are in most devices we interact with in our daily lives. From mobile devices, TVs, cars, fridges, petrol station pumps, servers that power the web and social media infrastructure --- the list is endless. Microprocessors have been doubling in power roughly every two years following Moore's law, which has been enabled by making the features of the chips smaller, fitting more transistors per unit area and driving the entire consumer electronics market worth more than £1 trillion per year. In order to continue to satisfy the industrial and societal demand that drives Moore's law, there are some fluid dynamics modelling challenges that we need to overcome. The next-generation of photolithography machines that need to manufacture smaller, faster microprocessor chips and the new devices required to supercool the high-performance chips during operation can be enabled by understanding and predicting accurately how gases behave at the micro/nanoscales, or in vacuum-like conditions. In these multiscale flow problems, the fluid dynamics is often unintuitive and all equations we normally turn to for modelling and designing engineering flow problems, such as flow around aircraft and ships using Navier-Stokes equations, are no longer valid here, because the gas is no longer in local thermodynamic equilibrium, on which these classical equations are formulated. The direct simulation Monte Carlo (DSMC), is the state-of-the-art software for modelling these non-equilibrium gas flows. It is a stochastic particle method with large numerical stability and can resolve the molecular nature of gases in three dimensional geometries. However, because it is a particle method, it requires a voracious computational cost to produce engineering solutions of scales that matter to industry. DSMC also performs poorly if those flows are at low speed, due to the inherent thermal noise in the particles blocking the measurable signals.In this project, we propose developing a new multiscale method, one which combines DSMC with computationally cheaper models such as those used in Computational Fluid Dynamics (CFD). We will produce a step change in simulation efficiency and accuracy by connecting DSMC and CFD solvers using surrogate modelling and Bayesian inference.With strong backing from our industrial partners, we will turn the outcome of this project into a free open-source computational solver released in the UK's OpenFOAM software that is validated with experimental data. The industrial focus of this project will be on processor-chip manufacturing, chip thermal management and electrospray technologies, but the underlying method is general to new directions in other research and industrial areas.