Electronic Design Flow Improvement with Machine Learning Tools
The design of modern printed circuit boards (PCBs) is a challenging task and requires the compliance with a variety of specications. To reduce the risk of a poor design and getting the desired functionality, the design processs is accompanied by many electromagnetic simulations. In combination with the time requirement of an individual simulaion this results in a large eort which needs a lot of computational resources. Because future designs will have a higher complexity and larger integration level, decreasing the time requirement for an individual simulation is an important task to solve. Without improving the simulation mechanism the design ow will require more time which results in additional costs for the development. This project aims to improve the design ow by providing a more ecient design tool and process. In the area of machine learning algorithms some promising ideas are found. Publications in recent years provide methods to increase the eciency of optimization processes – for example generic algorithms for the placement of decoupling capacitors. With articial neural networks rst results are achieved by investigating the impedance of the power delivery network under the in uence of decoupling capacitors. Future work shall increase the applicability of articial neural networks to a wider range of simulation tasks. Therefore dierent aspects have to be investigated. The focus is on printed circuit boards which are commonly used in many electronic devices. The functionality and capabilities of printed circuit boards is well understood. One of the most important aspects within this project is to provide not only a faster simulation ow but to ensure the consistency of simulation results with existing tools and methods.