Simulation, Modeling and Design Automation for High-Performance Chip, Package and Systems

近年来，微系统已经变得无处不在，复杂和越来越便宜，电子设计自动化工具的可用性是这个数十亿美元行业成功的必要先决条件。与此同时，这种系统的复杂性呈指数级增长。一个典型的微芯片现在包含数十亿个晶体管和许多计算核心。此外，微系统通常包括微机电系统（MEMS），各种传感器和光学互连，以形成复杂的多物理场微系统。同时，这种复杂性和相应的性能提高是在不显著增加成本的情况下实现的。事实上，虽然最先进的CPU的成本仍然很高，但现在可以以几年前的一小部分成本获得更简单但功能相对强大的计算机。该行业的一些关键推动因素是电子设计自动化（EDA）工具。这些工具允许系统的设计流程，以及设计和管理高度复杂系统的能力，同时遵守越来越快的产品更新周期。在这项研究计划中，我们的目标是从仿真和设计自动化的角度来解决与高性能芯片，封装和系统的信号完整性和电源完整性相关的关键战略问题。与此同时，我们将通过开发可以轻松并行计算的算法来考虑现代多核计算架构。更具体地说，我们将集中在以下一般主题：1。模型降阶（莫尔）：模型降阶背后的一般思想是，具有许多自由度（范围达到数百万）的大型非常复杂的系统通常具有相对较少的主导模式。我们的目标是开发莫尔算法，将大问题简化为小问题，以便在保持可接受的精度的同时有效地模拟大问题。这项工作的一个重要重点是开发技术，以减少一个大的系统成许多较小的系统，可以并行解决，从而使使用并行计算。2.基于测量和模拟参数的黑盒建模：在我们研究的这一重点中，我们将重点关注自动模型生成。对于许多实际应用来说，建立基于物理的数学模型是非常困难的。我们将开发工具和算法，用于根据测量或模拟参数自动获得电源和信号互连结构的数学模型。3.机器学习方法：近年来，机器学习领域取得了相当大的进展。这一领域与基于神经网络的建模和复杂电气系统的代理建模有许多共同之处。我们计划开发受机器学习启发的算法和方法，用于信号和电源完整性仿真和优化。

In recent years, Microsystems have become ubiquitous, complex and increasingly inexpensive, and the availability of Electronic Design Automation tools was a necessary prerequisite for the success of this multibillion dollar industry. At the same time, the complexity of such systems has increased exponentially. A typical microchip now contains billions of transistors and many computing cores. Also, microsystems often include microelectromechanical systems (MEMS), various sensors, and optical interconnects to form complex multiphysics microsystems. At the same time, this complexity and corresponding increase in performance is being achieved without considerable increases in cost. Indeed, while the cost of the most advanced CPUs remains at a premium, it is now possible to obtain simpler yet still relatively powerful computers at a fraction of the cost of a few years ago. Some of the key enablers of this industry are Electronic Design Automation (EDA) tools. Such tools allow for systematic design flows, and the ability to design and manage highly complex systems while obeying increasingly fast product update cycles. In this research program, we aim to address critical strategic problems related to signal integrity and power integrity of high performance chip, package and systems from a simulation and design automation perspective. At the same time we will take into account modern multicore computing architectures by developing algorithms that can be readily computed in parallel. More specifically we will focus on the following general themes: 1. Model Order Reduction (MOR): The general idea behind model reduction is that large very complex systems with many degrees of freedoms (ranging into the millions) often have a relatively small number of dominant modes. Our goal is to develop MOR algorithms that reduce large problems into smaller ones, such that they can be efficiently simulated while maintaining acceptable accuracy. An important focus of this work is to develop techniques to reduce a large system into many smaller systems that can be solved in parallel thus enabling the use of parallel computing. 2. Blackbox modeling based on measured and simulated parameters: In this thrust of our research we will focus on automated model generation. For many practical applications it is very difficult to develop physicsbased mathematical models. We will develop tools and algorithms for automatically obtaining mathematical models of power and signal interconnect structures based on measured or simulated parameters. 3. Machine Learning Methodology: In recent years there have been considerable advances in the machine learning field. This area has much in common with neural networksbased modeling and surrogate modeling of complex electrical systems. We plan to develop machine learning inspired algorithms and methodologies for signal and power integrity simulation and optimization.