Conure: Expanded Surrogate-Based Artwork Generator for RFCMOS Transformers

This paper presents an enhanced version of Conure, a surrogate-based artwork generator designed for RFCMOS transformers. The tool now includes a graphical interface for visual artwork generation, parameter sweeps, and integrated EM simulations. Conure supports the generation of feed-forward artificial neural network-based surrogate models with reduced computational overhead. Using 65nm CMOS technology, we developed six transformer designs with ratios ranging from 1: 1 to 4:3 and trained surrogate models for each design. Model accuracy over all the configurations showed an R2 score larger than 0.9627. Notably, the models could be trained with only 105 data sets while maintaining high prediction model accuracy which significantly reduces computational time and improves efficiency in transformer-based circuit design.