I hear from a lot of engineers who have trouble getting their simulations and measurements to agree, so if you are having trouble, you are not alone. One of the most common problem areas is with the measurement of the Multilayer Ceramic Capacitor (MLCC) decoupling capacitors, and with good reason. Ceramic decoupling capacitors are sensitive to AC and DC bias. The frequency dependent Equivalent Series Resistance and very small parasitic inductance (ESL) also come into play. As if these details are not enough of a headache already, the required measurements are different for use in EM simulators, like Keysight ADS, and SPICE simulators, which do not include PCB impacts. This article will focus on calibration and de-embedding component measurements to help improve matching. Previous articles cover DC and AC bias effects.

What Accuracy Do We Need to Measure?

Typical MLCC decoupling capacitors are in the range of 1 uF to 22 uF and sizes from 0201-0805. The ESL of these capacitors range from roughly 100 pH to 300 pH. Wide geometry, X2Y, and interdigit capacitors are lower.

This means measuring the ESL term to an arbitrarily defined accuracy of 10% requires a measurement accuracy of 7 pH to 30 pH depending on the geometry. To achieve this accuracy repeatedly, we usually need to mount the capacitor on a PCB with RF connectors for attachment to the vector network analyzer (VNA).Figure 1. Most manufacturers include ESL data for their capacitors. These examples from the Murata Simsurfing™ software database show an 0402 capacitor and a wide geometry 0204 capacitor with ESL of approximately 200 pH and 70 pH, respectively.

What is Calibration with De-Embedding?

A traditional VNA calibration is performed by connecting the VNA and interconnecting coaxial cables to a calibrator. This could be a mechanical calibrator, or a faster and more accurate automatic electronic calibration module, as shown in Figure 2.

Figure 2. Mechanical calibrator (left) and automatic calibration module (right). Images courtesy of Copper Mountain Technologies.

The calibration process computes magnitude and phase correction terms for SHORT, OPEN, LOAD, and THROUGH (or RECIPROCAL), at each frequency, to force the measurement result to match predetermined calibrator values. The predetermined values are generally referred to as the “cal kit.” Lower cost mechanical calibrators include a cal kit based on polynomial equations. The equations are like a datasheet typical value and will vary slightly from calibrator to calibrator. More expensive devices, like the automatic calibration module, include a data-based cal kit, which uses actual s-parameter measurements, making it much more accurate, though also more expensive. After calibration, the VNA can accurately measure devices inserted between the two connectors. Interestingly, the one error this doesn’t correct is the low frequency ground loop error associated with the 2-port shunt through method. Achieving accurate low frequency measurements requires the ground loop isolator to be near-ideal.

Our goal is to measure the capacitor. Connecting the PCB between the two connectors won’t result in an accurate measurement of the capacitor because the PCB is also included in the measurement. The PCB introduces three errors: insertion loss, time delay, and return loss. Accurate measurement of the capacitor alone requires removal, or “de-embedding” of the PCB artifacts from the measurement. If the PCB is designed using a precision RF dielectric, with a tightly controlled 50 Ω path, including the PCB and the connectors, the insertion loss and the return loss become much less significant. The focus is then on the time delay, which results in phase error. Since this phase determines both the ESL of the capacitor and the resistive damping of the capacitor, the time accuracy is critical. We can demonstrate this using a simple simulation model representing the 50 Ω  path, the electrical delay and a simplified capacitor model. 

A simple simulation of the PCB is shown in Figure 3, with the component mounted in the center of the test PCB. The capacitor can be moved left or right, using the K parameter, which has little impact on the result. Only the total path length is critical. The simplified capacitor model is used so that we know the exact, and fixed, value of each parasitic.

Figure 3. Simplified model of the MLCC capacitor mounted on a PCB with 50 Ω impedance line connections (port extensions).

Simulating the PCB and mounted capacitor with a 0.01 ps delay essentially eliminates the PCB and results in the correct ESL of 200pH, as seen in Figure 4.

Figure 4. Simulation of the capacitor impedance magnitude (blue) and the imaginary impedance (red), used to calculate the parasitic inductance (ESL; with no delay, the ESL is exact).

The Bode 100 and Bode 500 Calibrate Differently

The Bode 100 and Bode 500, popular VNAs from OMICRON Lab, calibrate 2-port impedance in a very different way than a traditional VNA. The traditional VNA calibrates in S-domain using eight measurements. These include SHORT, OPEN, and LOAD for each port and THROUGH (or RECIPROCAL) and ISOLATION for a total of eight calibrations. The ISOLATION calibration is used to correct the low frequency ground loop error of the SETUP. The typical VNA calibrator does not include the ISOLATION ports, so it can not correct the ground loop error.

Figure 5. Simulation of the capacitor impedance (blue) and the imaginary impedance magnitude (red), used to calculate the parasitic inductance (ESL); with a 1ps delay, the error is just above 10%.

De-embedding

Calibrating the Bode 100/500 for precise measurement, therefore, requires the de-embedding of the PCB as part of the calibration process. Combining the calibration and de-embedding into a single process greatly simplifies the procedure while also increasing the accuracy of the measurement.

A calibrator needs to be created to precisely match the design of the component PCB mount. These boards must perfectly match, particularly in time delay. This requires extreme control of the material dielectric constant, the physical size of the board, and the precision of the 50 Ω traces. 

Given these idealities, a Z-domain calibrator was created for 2-port OPEN, SHORT, and LOAD. The OPEN is simple, since it is just the PCB without any component mounted. The THROUGH is a very precise 50 Ω resistor. Since the frequency is above 100 MHz, and we hope to calibrate to 3 GHz, the resistor is a microwave, surface trimmed, frequency stable resistor. Ordinary resistors are trimmed differently and won’t be accurate above 100 MHz. 

The SHORT is the most challenging, requiring a near ideal short with the ground loop error included and without any component electrical length added. The electrical length, including the connectors, for the three calibrators and the measurement PCB must all precisely match within a total error of approximately 1 ps or approximately 0.15 mm including the effects of trace impedance error.

After performing the OPEN, SHORT, and LOAD calibrations, the component is mounted on a matching PCB for measurement. 

The Picotest UC10-2.92 Calibrator is universal, since it includes OPEN, SHORT, LOAD, THROUGH, and ISOLATION calibrators, and a cal-kit, for support of traditional VNAs and 2-port OPEN, SHORT, and LOAD for the Bode 100 and Bode 500 VNAs.

The universal calibrator and matching PCB mounts are shown in Figure 6.

Figure 6. The PICOTEST UC10-2.92 Universal Calibrator, SMA wrench and the matching component mount are shown side-by-side. 

Example Measurements

T

Figure 7. A 1.2 nH chip inductor was measured using a Copper Mountain Technologies S5085 VNA calibrated with an ACM2509 calibrator and the PCB is removed using automatic port extensions. The same mounted chip inductor is measured with the Bode 500 using the Picotest UC10-2.92 Universal Calibrator. The results agree within approximately 5 pH.

Conclusion

The traditional VNA calibration includes a cal kit to remove variations in the measurement setup, including the calibrator inaccuracies, and cable errors. The PCB mount is de-embedded using port extensions. This method does not work for the Bode 100 and Bode 500. 

The Bode 500 requires a precise “ideal” calibrator to both calibrate and de-embed the measurement using only three calibrations rather than the traditional eight. This simpler three-step calibration method also corrects the low-frequency ground loop error. The mounting PCB time delay must exactly match the calibrator time delay for accurate measurements.

Providing a cal kit for this universal calibrator allows it to be used with all traditional VNAs as well as the Bode 100 and Bode 500 VNAs, resulting in highly accurate measurements for improved modeling and simulation measurement correlations.