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

用于高频电子电路产量驱动设计的先进统计建模和优化技术

• 批准号: RGPIN-2017-06420
• 负责人: Zhang, Qijun
• 金额: $ 2.7万
• 依托单位: Carleton University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=711577
• 关键词: Advanced; Statistical; Modeling; Optimization; Technologies

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.

本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中的功能性、复杂性、信号速度和带宽的增加，构建块组件和子系统的设计规范变得更加严格。这反过来又使得组件和子系统中不可避免的制造公差和工艺不确定性对整个系统性能的影响更加明显，影响生产良率并对设计提出挑战。作为设计过程的一部分，直接考虑参数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，是足够成熟的等效电路为基础的设计是不是有效的，今天的电磁（EM）/多物理为基础的设计，因为高昂的计算成本。一个新的统计建模和优化范式是必要的，使EM和多物理场的成品率驱动的设计。拟议的研究将应对这些挑战。 建立在我们最近在统计神经空间映射和认知驱动技术的RF/微波设计的进步之上，这项研究探索了基于EM/多物理场的统计建模和产量优化的新前沿。本研究将开发新的优化算法，利用快速参数EM模型与或没有粗糙的工程模型，大大减少了EM结构的成品率优化的计算费用。 本研究将结合空间映射与知识基神经网络模型，建立电磁结构的统一参数化建模算法。 将开发新的动态统计神经空间映射算法，用于非线性器件的统计建模，涵盖高频和低频（如捕获效应）响应的统计行为。 该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，开辟建模和设计的新前沿。提出了一类新的电磁场空间与多物理场空间映射的微波优化算法。 长期方向是统一的EM/多物理场为基础的方法，快速和准确的统计建模和下一代高频电子设计的产量优化。长期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。它有助于创造新的知识和高素质的技术人员在高频电子设计领域的培训。