A potentially significant on-wafer high-frequency measurement calibration error

High accuracy radio-frequency (RF) measurements typically require a calibration to remove the undesired effects of the measurement apparatus. The calibration consists of measuring some combination of known standards such as short, open, load, through, and delay. When measurements are performed on-wafer for silicon RF integrated circuits (RFICs), a two-step calibration/de-embedding technique is typically used. First, the measurement system is calibrated to a reference plane located at the probe tips through measurement of calibration standards fabricated on an impedance-standard substrate. Second, on-wafer de-embedding standards are measured in an attempt to shift the reference plane to the terminals of the device under test (DUT). While significant effort has gone into the development of improved on-wafer de-embedding schemes, discrepancies between actual and de-embedded data still exist. In this article, we first discuss a specific case (a spiral inductor on silicon) for which there was a significant discrepancy between measurement and analysis. The problem is found to be with the measurement. This problem is detailed, and a technique we call "synthetic calibration" is described that can be used with any electromagnetic (EM) analysis to quantify calibration error for any proposed set of calibration standards. Due to the high expense and time required for wafer fabrication, it is important to successfully complete such a calibration validation prior to tape-out.