Converter Circuits to Machine Learning: Optimal Feature Selection

Design in electronic circuit was application directed, utilizing ML models (regression/classification/clustering.. etc) to obtain application target, hence minimizing human error/intervention/effort/cost. The concept of how to properly transfer the different physical structures/properties of electric circuits from the known physical domain to ML domain was introduced in previous work. Adding to that, this paper illustrates how this transformation can be performed, explores all the possibilities of different feature representations of electric circuit parameters when mapped as a graph with edges and nodes, and proposes an optimal feature representation irrespective of the connection or the number of components the circuit has, in a form that can be easily interpreted and processed by machine learning algorithms. Many ML based applications such as circuit synthesis, automatic circuit layout generation, AI circuit design automation and many other applications are applicable with proposed methodology. The optimal features are applied on seven circuits of different circuit order, exploring the impact of feature selection on the classifier accuracy. The paper proposes a feature representation that results in optimal classifier, accuracy of 97.37%.