Research Initiation Award: Multidisciplinary Design Optimization Development and Application in Electronic Package Design

在这项研究中，将充分开发两种新的多学科设计优化通用算法，并将其应用于电子封装优化。在一种算法中，将利用在子系统优化期间同时产生的局部二阶设计灵敏度来产生二阶系统灵敏度。并行的子系统优化和局部灵敏度计算将并行实施，以实现更高的效率。与以前的算法方案相比，该算法提供了一种用于量化子系统交互的计算成本低廉的过程。第二种算法将使用神经网络来建立系统设计空间的近似。使用神经网络来表示设计空间，将多学科设计优化策略从以前的算法提案中使用的点式设计技术转移到了一起。神经网络将使用并行子空间优化过程中积累的数据进行训练。这些算法将被开发为通用的设计优化工具，应用于各种复杂的耦合多学科系统。电子封装设计是一个可以通过应用多学科设计优化来实现立竿见影的效益的领域。在多学科设计优化中，子系统之间的复杂耦合被形式化地量化，并在并行设计过程中被利用。//

In this research two new general purpose algorithms for multidisciplinary design optimization will be fully developed and implemented in application to electronic package optimization. In one algorithm second order system sensitivities will be generated using local second order design sensitivities generated concurrently during subsystem optimizations. The concurrent subsystem optimizations and local sensitivity calculations will be implemented in parallel to achieve greater efficiencies. The algorithm provides a computationally inexpensive procedure for quantifying subsystem interactions as compared to previous algorithm proposals. The second algorithm will use neural networks to build approximations of the system design space. The use of neural networks to represent the design space moves the multidisciplinary design optimization strategy away from the point design techniques used in previous algorithm proposals. The neural network will be trained using data that is accumulated during concurrent subspace optimizations. The algorithms will be developed as general purpose design optimization tools for application to a variety of complex coupled multidisiplinary systems. Electronic packaging design is an area where immediate benefits can be realized through the application of multidisciplinary design optimization. In multidisciplinary design optimization the complex couplings between subsystems are formally quantified and exploited in a concurrent design process. //