Improving mmWave Device, Subsystem Design and Modeling

February 11, 2020 Anritsu Company

February 11, 2020

Millimeter wave (mmWave) frequencies are now designed into more direct applications and device/subsystem modeling efforts require even higher frequency measurements. This poses a challenge, as limitations in conventional test equipment make conducting accurate measurements a problem. Engineers need a solution that can reduce design turns, lower cost of test, and give them greater confidence in their designs. New technology in vector network analyzers (VNA) addresses these issues and presents a solution to high-frequency design validation.      

A broadband VNA system is now available that covers low frequencies through 220 GHz and accompanying probes can help resolve some of these measurement complications. It eliminates such problems as stitching together data from multiple bands, power control, linearity, and data stability. The overall performance of the system improves measurement performance in a variety of categories.

High-frequency Broadband Measurement Challenges

A typical mmWave test setup may consist of a broadband system supporting up to 110 GHz or 125 GHz followed by measurements using banded waveguide modules. In mounting and demounting modules and probes from the probe station, there are several issues. Let’s discuss a few.

Multiple Touchdown Cycles. For broadband device model parameter extraction and IC performance verification tasks, product and specifications require the same device under test (DUT) to be measured at numerous touchdown cycles of the DUT contacts. 

Small DUT Contact Pads. Another issue is that the DUT contact pads suitable for sub-mmWave probing are much smaller since parasitic pad reactance increases with frequency. As small-sized pads can support only a few RF probe touchdown cycles, it can become impossible to measure the same DUT over the whole frequency range and all temperature points.  

Test Cell Downtime. Frequent system reconfiguration and changeovers are required for wide-frequency range measurements. This increases the test cell downtime and risk of damaging expensive system components, such as wafer probes and VNA extenders. All of these increase an already expensive cost of test.

Data integrity. Since different probes are being used in the separate bands and various measurement modules are used, the data is acquired under different circumstances. While the calibration will correct for many differences (though not linearity, noise, spectral purity, drift or repeatability differences), the uncertainties and distributions in the different bands will likely vary. This raises some question on how to handle the inevitable data steps such as those shown schematically in Figure 1. It is an example of data stitching anomalies along with a change in expected uncertainties at an equipment boundary. Since different hardware, probes and calibration elements are employed in the two bands, the measurement characteristics can change significantly.

Figure 1
Figure 1

Adventure Known as De-embedding

Once the data is collected, de-embedding to the desired reference plane can sometimes be an adventure. Particularly in more current BiCMOS and CMOS processes, there may be 5-9 metal layers involved. The desired transistor reference plane may be at the bottom layer while the probe pads are at the top. This leads to a “network to de-embed” consisting of many vias and transitions, significant insertion loss, and mismatch. Since such fabrication processes are increasingly of interest at mmWave frequencies, the stability and basic accuracy of the S-parameter data can be critical. 

RF Drive Level Considerations

Another aspect of the measurement process, RF drive level, can be important for bare transistors and certain amplifiers at higher frequencies. The drive power may need to be -40 dBm or lower and accurate control of that power may be important. Figure 2 is an example measurement of a 120 GHz LNA where initially drive level was not considered (labeled “nominal,” was not flat, and reached -15 dBm in places) along with a measurement where the power was leveled at -40 dBm. Not only was the gain highly compressed in the first measurement, but because of flatness issues, the amount of compression varied significantly with frequency.

Figure 2
Figure 2

Another consideration is the linearity of the instrument receiver, as correcting for receiver distortion can be difficult. Since many mmWave receivers use relatively high order harmonic mixing, management of the internal LO system is important to maintaining linearity at the test port.

Novel Approaches Improve Testing

One way to address some of the challenges is a broadband VNA with sufficient levels of integration and system control to allow for reasonable stability and relatively high linearity receivers. The ME7838G (figure 3) is composed of a base VectorStar™ VNA and mmWave modules. The modules handle receiving chores from 30 GHz - 226 GHz using broadband forward couplers and a high LO sampling system based on III-V nonlinear transmission lines.

The port-referred third order intermodulation product of these downconverters exceeds 30 dBm, which helps with the linearity requirements. The modules handle source multiplication above 54 GHz and use a series of four multiplexers to inject energy from the respective multipliers. Since progressively less power is available at the higher mmWave frequencies, the highest frequency multiplexer is last and has the tightest coupling. Leveling loops, for both RF and LO, extend out to the modules to help improve measurement stability.

The integration of the receivers, couplers and multipliers in a small space helps with thermal uniformity and stability, as does the close location of the couplers to the probe tip. The test port is a novel structure supporting a coaxial mode (0.6 mm outer conductor diameter) but uses a precision UG-387 flange instead of a threaded body to form the outer mate. The increased mating area of the flanged interface improves durability and significantly reduces axial forces and bending moments of the connecting device. Such an interconnect system allows a single on-wafer probe to make measurements from 70 kHz to 226 GHz.

Precision probes also contribute to well-matched probing for improved calibration and measurement results over a wide frequency. The 220 GHz MPI TITANTM Probes incorporate perfectly-matched 50 Ω MEMS contact tips to ensure optimum results over extreme frequency range. The design provides excellent visibility of the tip contacts during touchdowns. The result is consistent positioning of the RF probe on calibration standards and small DUT pads for better repeatability and reproducibility, even by inexperienced operators.

Figure 3
Figure 3

Stability is important to maximize the time between probe tip calibrations and to avoid adding data distortions that may complicate model fitting or other analyses. The measurement of a thru line over 18 hours at 25+/-3 C is shown in Figure 4. As might be expected with temperature variations affecting LO cables, there is some increase in drift with frequency since the LO multiplication factor increases along that axis. 

Figure 4
Figure 4

To learn more about the design of the new broadband VNA technology and how it improves mmWave testing, download this new Anritsu white paper.

Previous Flipbook
Generating an .s2p File for a 220 GHz Probe
Generating an .s2p File for a 220 GHz Probe

The VectorStar ME7838G is a broadband system providing coverage to 220 GHz.

Next Video
How to Get Started and Connected to an MS46122B
How to Get Started and Connected to an MS46122B