Machine Learning-Assisted On-chip Quadrilateral Interleaved Transformer Automatic Synthesis

The design of millimeter-wave on-chip passive transformers, which is an important component of radio-frequency (RF) circuits, has always depended on heavily computational full-wave electromagnetic (EM) simulations and specific design experience. A machine learning-assisted on-chip transformer automatic synthesis (OTAS) algorithm is proposed. This algorithm is applied to the quadrilateral transformer synthesis and is verified by designing a symmetrical quadrilateral interleaved transformer at 50 GHz to maximize available gain. OTAS can greatly reduce full-wave EM simulations while improving the performance of on-chip transformers.