Incorporating combined functional/structural knowledge in library of neural models

A knowledge based technique is presented for the development of large sets of neural models which can be trained to represent engineering component library. Through establishment and reuse of a set of base neural models, which incorporate the functional knowledge of basic characteristics common to all library components, and a set of structural knowledge hubs, the proposed method substantially improves the quality of neural models while reducing the cost of library development through reduced need of engineering data collection and shortened time of training. A practical example of a library of electrical transmission line models for the design of high-speed VLSI packages is developed confirming the effectiveness of the proposed technique