While co-calibrated ports do not have a minimum length for their reference planes, short reference planes for boxwall ports, both orthogonal and diagonal, can cause problems with your de-embedded results. If the reference plane for a boxwall port is very short relative to the substrate thickness or the width of the transmission line, em may generate poor de-embedded results. This is due to one or both of the reasons discussed below.
As shown above, the port is too close to the device under test (DUT). There are fringing fields associated with the port and separate fringing fields associated with the DUT. If the port and DUT are too close, the fringing fields interact. The de-embedding algorithm (which is virtually identical to algorithms used in de-embedding measured data) is based on circuit theory and cannot handle fringing field interaction. See reference [68] in Sonnet References for a detailed description of the problem.
The other possible problem is that the calibration standard is too short: The calibration standard must be long enough such that the currents on the line(s) behave like standard transmission line(s). Since the calibration standard length for a boxwall port is based on the reference plane length, a short reference plane can result in inaccurate results. For more details about this problem, see Modifying Calibration Standard Lengths.
There is no precise rule as to how long a boxwall port’s reference plane must be made in order to prevent the above effects from corrupting the de-embedded results. The required length is dependent upon the circuit geometry and the nature of the analysis. However, we recommend that you use reference planes equal to or greater than several times the distance to the nearest ground metal. This is sufficient for most types of analyses.