Advanced Statistical Modeling and Optimization Technologies for Yield-Driven Design of High-Frequency Electronic Circuits

这项研究的目标是开发下一代技术，用于无线和有线通信系统中高频电子元件和封装的统计建模和优化。随着无线和有线通信系统的功能、复杂性、信号速度和带宽的不断增加，对构建块组件和子系统的设计规范变得更加严格。这反过来又使部件和子系统中不可避免的制造公差和工艺不确定性的影响在整体系统性能中更加明显，从而影响生产产量并对设计构成挑战。作为设计过程的一部分，直接考虑参数中的不确定性的统计建模和产量优化变得重要。然而，传统的统计建模和成品率优化技术对于基于等效电路的设计来说已经足够成熟，但由于高昂的计算成本，对于今天的基于电磁(EM)/多物理的设计来说并不有效。为了实现基于EM和多物理的成品率驱动的设计，需要一种新的统计建模和优化范例。拟议的研究将解决这些挑战。*建立在我们在统计神经空间映射和认知驱动的射频/微波设计技术的最新进展的基础上，这项研究探索了基于EM/多物理的统计建模和产量优化的新前沿。这项研究将开发新的优化算法，利用快速的参数EM模型，无论是否有粗略的工程模型，都可以大大减少EM结构屈服优化的计算费用。该研究将结合空间映射和基于知识的神经网络模型，为EM结构创建统一的参数化建模算法。将开发新的动态统计神经空间映射算法，用于覆盖高频和低频(如陷阱效应)响应的统计行为的非线性器件的统计建模。这项研究还旨在通过将基于EM的统计设计扩展到基于多物理的统计设计来开辟建模和设计的新前沿。介绍了一类新的用于微波优化的电磁空间和多物理空间的映射优化算法。长期方向是基于统一的基于电磁/多物理的方法，为下一代高频电子设计快速准确的统计建模和成品率优化。长期的影响将是更快的设计周期，更低的设计成本，更好的设计质量和更高的制造产量。它有助于在高频电子设计领域创造新的知识和培训高素质的技术人员。

The objective of this research is to develop next generation technologies for statistical modeling and optimization of high-frequency electronic components and packages in wireless and wireline communication systems. With increasing functionality, complexity, signal speed and bandwidth in wireless and wireline communication systems, the design specifications for the building block components and subsystems become more stringent. This in turn makes the effects from unavoidable manufacturing tolerances and process uncertainties in components and subsystems more pronounced in the overall system performance, affecting production yield and posing challenges in design. Statistical modeling and yield optimization directly taking into account the uncertainties in parameters as part of the design process become important. However, conventional statistical modeling and yield optimization techniques that are mature enough for equivalent circuit based design are not effective for today's electromagnetic (EM)/multiphysics-based design because of the prohibitive computational cost. A new statistical modeling and optimization paradigm is necessary to enable EM and multiphysics based yield-driven design. The proposed research will address these challenges.*******Built on top of our recent advances in statistical neuro-space mapping and cognition driven technologies for RF/microwave design, this research explores new frontiers in EM/multiphysics based statistical modeling and yield optimization. This research will develop new optimization algorithms exploiting fast parametric EM model with or without coarse engineering models, dramatically cutting the computational expenses of yield optimization of EM structures. The research will create unified parametric modeling algorithms for EM structures combining space mapping and knowledge-based neural network models. New dynamic statistical neuro-space mapping algorithm will be developed for statistical modeling of nonlinear devices covering the statistical behavior of both high- and low-frequency (such as trapping effects) responses. The research also aims to open a new frontier in modeling and design by extending EM-based statistical design to multiphysics-based statistical design. A new class of space mapping optimization algorithms with mapping between EM space and multiphysics space for microwave optimization will be introduced.*******The long term direction is a unified EM/multiphysics based methodology for fast and accurate statistical modeling and yield optimization for next generation high-frequency electronic design. The long-term impact will be faster design cycle, lower design cost, better design quality and increased manufacturing yield. It contributes to creating new knowledge and training of highly qualified technical personnel in areas of high-frequency electronic design.***