Engineers characterizing semiconductors and high speed components face increased challenges in broadband millimeter wave (mm-wave) testing. Developing accurate models often involves measurements at frequencies that range from near DC to 100+GHz. These measurements are typically performed using a vector network analyzer (VNA). In order to ensure accuracy, VNA re-calibration may be required as frequently as every hour. This consumes valuable test time and, in production, reduces throughput. While these limitations have always been a consideration, they are now becoming a major issue, due to the preponderance of mm-wave frequency applications. These include E-Band LOS links, 94 GHz imaging radar and 77 GHz long/medium range automotive radar as well as other emerging applications above 110 GHz.
Device Characterization Issues
Figure 1 Traditional broadband VNA system using waveguide frequency extension modules.
For adequate characterization of on-wafer devices, the greatest challenge is to obtain high quality S-parameter data over a sufficiently broad range of frequencies. The upper end of the measured frequency range must, at least, cover the operating frequency of the target application, although higher frequency coverage is required to properly represent the device model. For example, higher sweep frequencies provide more accurate predictive results when performing circuit simulation of MMIC designs. Measurements at low frequencies are important to identify the potential for low frequency oscillation when S-parameter models are used in simulations and also to understand how low frequency bias networks may cause memory effects at higher frequencies. Many model element extractions, such as the components of FET output admittance, rely on asymptotic frequency data, so the higher and lower in frequency that one can obtain accurate data, the better.
Making S-parameter measurements over a wide range of frequencies, potentially spanning tens of kHz to in excess of 110 GHz, has traditionally required both low and high frequency network analyzers. Apart from the issues of time and cost, this also leads to concatenation issues at the boundary between the low and high frequency measurements. A measurement is essentially an estimate with an associated uncertainty; hence, measurements at the high end of the low frequency system are unlikely to agree with those at the low end of the high frequency VNA.
Quality broadband mm-wave measurements have also been challenging due to limitations in calibration and measurement stability typically caused by high frequency multiplexing schemes, physically large and inhomogeneous measurement structures and complex receiver chains. Multiplexers are necessary to permit coaxial measurement up to 67 GHz to be combined with waveguide band measurements from 67 to 110 GHz in a single 1 mm test port connector. The loss through the MUX combiner and any coaxial cables between the multiplexer and the probe tip degrade directivity of the couplers in the waveguide path and typically has resulted in raw negative directivity. An example of this system is shown in Figure 1.
Figure 2 Measurement error due to directivity.
Directivity is a measure of the ratio of reflection from a short circuit to the level of the leakage signal through the coupler. The reflectometer measures the signal reflected by the load plus the leakage. After calibration, changes in phase between the leakage and the measured signal can result in directivity measurement uncertainty (see Figure 2). These phase changes can occur due to temperature and physical cable movement.
Addressing these issues has required technology changes for S-parameter measurement frequency extension modules above 70 GHz, architectural innovations to enable very low frequency measurements and innovations in coaxial connector design to achieve a capability significantly exceeding 110 GHz.
Technology for High frequency Sampling
The mainstays of microwave VNA receivers have been samplers based on step recovery diodes (SRD) or harmonic mixers. Harmonic samplers tend to have better conversion efficiency when compared to harmonic mixers, especially at higher multiplication numbers. A sampler based on nonlinear transmission line (NLTL) technology provides an even more beneficial alternative for generating high frequency harmonics much more efficiently. The SRD transfer function tends to drop off by 50 GHz, whereas NLTL technology offers a higher comb frequency with less drop-off in higher frequency performance (see Figure 3). Excellent dynamic range at 70 GHz and beyond without the need for large amplification is achieved with the promise of scalability to hundreds of GHz.
Figure 3 NLTL-based sampler vs. SRD transfer coefficients.
Anritsu’s engineers in Morgan Hill, working together with Anritsu’s fabrication facility in Japan, have successfully overcome many challenges, both process-related and physical (such as issues related to surface waves), to produce repeatable devices based on NLTL technology that achieve much greater dynamic range at frequencies of 110 GHz and beyond.
Architectural Solutions
An S-parameter measurement requires a combination of sources, directional devices to separate forward and reverse signals, and receivers. NLTL-based multipliers and samplers provide the enabling source and receiver technology, while the key to providing good directivity is eliminating the need for the MUX combiner, inherent in the waveguide frequency extension modules.
Figure 4 NLTL-based broadband frequency extension module.
Figure 5 Miniature 70 kHz to 110 GHz frequency extension modules.
Figure 4 shows the architecture of a frequency extension module using GaAs NLTL devices integrated with couplers on a coplanar waveguide substrate. Everything is built around a through-line. Measurements below 30 GHz are made in the parent VNA, while the directional devices for measurements at 30 GHz and beyond are located within the frequency extension module, with no intervening loss due to a MUX combiner. This provides high directivity and results in more stable measurements. Using this system, improved stability over the 70 kHz to 110 GHz frequency range has been demonstrated with S21 deviations of less than 0.1 dB in amplitude and 0.5° in phase over a 24 hour time period. This allows for longer intervals between calibration and improved measurement accuracy. The use of the NLTL technology enables a VNA system dynamic range of up to 109 dB at 110 GHz. The product of this technology and architecture is shown in Figure 5. As can be seen, the extremely small size of the modules means that, in many cases, the probes can be directly connected to the modules further enhancing stability.
Figure 6 VNA based on a hybrid bridge architecture.
While couplers provide great performance at higher frequencies, a coupler designed to operate to 70 GHz will fall off in performance at 500 MHz and below. This issue is solved in hybrid bridge architecture, as shown in Figure 6. Above 2.5 GHz, couplers are used, whereas below 2.5 GHz, RF bridges used as the directional devices provide a flatter raw conversion performance all the way down to 70 kHz (see Figure 7). The complete VNA system, as embodied in the Anritsu ME7838A (70 kHz to 110 GHz, operational to 125 GHz), shown in Figure 8.
Figure 7 Low frequency performance extended with a hybrid bridge architecture.
Figure 8 ME7838A in bench-top coaxial two-port measurement configuration.
Connector Technology
Making broadband measurements at a single test port requires a coaxial connector for operation at low frequency. W1 (1 mm) connectors permit measurements to be made up to 110 GHz with limited operational capability to 125 GHz. For traceable measurements beyond 110 GHz, smaller connectors are needed. Anritsu has already demonstrated a 70 kHz to 145 GHz broadband VNA system featuring NLTL-based frequency extension modules with a newly developed 0.8 mm coaxial connector.
Conclusion
A new VNA architecture provides better test data and more accurate information to improve and refine device models. The ability to combine stable, accurate high frequency and low frequency S-parameter data in a single frequency sweep eliminates the need for data concatenation and its associated errors. Improved stability allows more time spent measuring, because less time is needed for recalibrating. Reduced measurement uncertainty provides engineers with greater confidence in their mm-wave designs.
Steve Reyes, product marketing manager, network analyzers, for Anritsu. Steve has 20+ years of sales and marketing experience of wireless communication equipment to semiconductor, component and equipment manufacturers.
