110–170 GHz On-Chip Calibration Using Deep Neural Networks

For millimeter-wave (mmWave) integrated circuits, an on-chip calibration method with high accuracy is crucial. For accurate calibration, this paper proposes a thru-reflect-line neural network (TRL-NN), by incorporating machine learning techniques into the classical thru-reflect-line (TRL) calibration method. Compared with conventional TRL, the proposed TRL-NN demonstrates enhanced accuracy. The TRL-NN utilizes deep neural networks (DNNs) to predict <tex-math notation="LaTeX">S </tex-math> -parameters of the device under test (DUT). The proposed method also achieves high calibration accuracy even when non-standard calibrators are employed. For validation with comparison, 100 randomly generated DUTs are calibrated using both proposed TRL-NN and classical TRL methods. The proposed TRL-NN method is further experimentally demonstrated to accurately calibrate a transmission line covering the whole D-band of 110–170 GHz.