Small, high-performance ICs require wafer-level-RF measurements

Continuing increases in the speed of semiconductor devices combined with higher levels of integration, on-chip wireless functions, and the usage of mixed-signal-device designs are driving new requirements for wafer-level-RF measurement. Many device categories that system designers have thought of as purely digital now include on-chip wireless functions. Therefore, parameters such as unity-current-gain crossover frequency and maximum frequency of oscillation have become even more critical for large groups of design and test engineers. Meanwhile, the march to higher speed continues even for standard logic devices. Also, RF is becoming more necessary for isolating, measuring, and modeling some key parameters in advanced processes, especially those that use advanced dielectrics. A key advantage of wafer-level-RF measurements is the fact that making measurements at the wafer state reduces development time by eliminating the time that it takes to package a die before testing. Also, measuring the device in the prepackage state eliminates potential disturbances from the package itself. The upshot is that many engineers who once relied primarily on low-frequency or dc measurements are now adding RF to their test suites.

From DC to RF characterization

Engineers who are currently relying on dc characterization need to be aware of the significant changes they must make when they move to RF techniques. DC characterization measures currents or voltages using meters and sources and provides a simple response that is not phase-dependent. Occasional dc-test-instrument calibration is necessary. Service personnel often perform this calibration at regularly scheduled service intervals, so users do not give much thought to calibration issues.

RF characterization, on the other hand, measures RF phase and magnitude using S parameters (scattering parameters) and reflection ratios. The RF technique is analytically convenient and provides a complex response, and the calibration process is central to making accurate measurements. For example, unless you properly analyze and characterize reflection ratios for the test apparatus and cabling, you can't accurately extend the measurement-reference plane to a wafer device using probes and cables. RF techniques rely heavily on frequent calibration (Figure 1).

RF-measurement methods

With RF measurement, a number of practical issues can add challenges to achieving highly accurate results. Figure 2 illustrates the contrast between an ideal RF and the real-world environment. Figure 2a shows the RF stimulus. In the ideal system, the RF source stimulates the DUT (device under test). Meanwhile, the acquisition system samples the magnitude and phase of the incoming or incident signal, the reflected signal, and the transmitted signal. Of course, you must terminate the device output in impedance that matches that of the test system. In an ideal system, by looking at the phase and magnitude of the three signals-incident, reflected, and transmitted-you can easily make such measurements as gain, delay, and reflection coefficient. However, in a real-world environment, extending the measurement reference plane to a device on a wafer or substrate is challenging because, at high frequency, you can't easily define exactly where the measurement system ends and the DUT begins (Figure 2b).

Items such as cables, couplers, connectors, and probes combine to create a nonidealized test situation. You must even consider the nonideal nature of the network analyzer. You can think of the RF-measurement system-the total system with all its elements-as an RF instrument with an error-adapter network that models the system's nonideal components. Only by modeling the system's nonideal components can you correct for such errors as poor response or noisy data and assure yourself of repeatable measurements. Ultimately, the system accuracy depends on the model's accuracy.

So, although calibration is important, consistently obtaining correct data critically depends on the repeatability of the measurement system itself. Without repeatability, no amount of vector-error correction or calibration can correct for random errors, such as noise or changes in attenuation, electrical length, phase, or other environmental factors. Therefore, when you extend the measurement-reference plane to the wafer, you must use high-quality components to minimize systematic errors.

Accurate RF measurement

The key system elements you need to achieve accurate and repeatable RF measurements are precision RF-wafer probes, cabling, impedance standards, and disciplined calibration and validation. RF probes provide the vital link between the test system and the DUT. Although RF-wafer probes first emerged many years ago, high-quality probes' basic characteristics haven't changed much over time. To deliver repeatable S parameters, an RF-wafer probe should exhibit a low return loss and low insertion loss, along with low-and highly repeatable-contact resistance. Low signal-to-signal crosstalk is desirable for dual-line RF probes. Also, it is important to minimize parasitic coupling at the probe tip, which can cause inaccuracy. Precision RF probes are essential for measurement of such properties as gain, return loss, and VSWR (voltage-standing-wave ratio).

Of course, the probe needs to match the DUT layout. Most RF probes have two or three contact points at the probing end-either a two-contact GS (ground-signal) probe or three-contact GSG (ground-signal-ground) probe, with some dual or specialized probes providing even more contacts.

Figure 3 illustrates some of the parasitics that come into play when you use a probe to contact a wafer device. The yellow portion of the diagram shows the probe needles, and the blue section shows the DUT. The ground in the figure is the back surface of the wafer. As you can see, parasitics are factors with both the on-wafer probing pads and the probe needles.

The contact-pad size can affect the parasitic values and can cause differences even when you use a calibration standard. Through proper design and by properly selecting probes, you can minimize parasitics at the probe tip. The objective is to keep parasitics small and repeatable. Figure 4 compares some common RF-probe-tip approaches and contrasts GS pads with GSG and a microstrip configuration with shielding. Figure 4a shows a two-contact GS-probe configuration with some of the fringing fields on one side coupling to the wafer or to the wafer chuck. Because of the lack of field control, you should not use GS probes for frequencies higher than 10 GHz. Figure 4b shows a GSG probe. As you can see, by controlling the ground path directly through the probe tip, this three-contact configuration does a much better job of terminating the field lines. Some minor interaction occurs with the DUT, but the fields on both sides of the DUT are consistent, making the three-contact approach better for higher frequencies. Figure 4c illustrates an advanced configuration in which the probe tip uses a microstrip line with a coplanar transition at the probe needles. Microstrip transmission lines on the probe's thin-film tips confine fringing fields more tightly than do conventional flexible coplanar tips, and the resulting improved field confinement reduces unwanted coupling to nearby devices or other probe tips, thus increasing RF-measurement accuracy. Besides preventing the field lines from interacting with the DUT, microstrip minimizes crosstalk, which makes it possible to configure dense, fine-pitch multitip probes that simultaneously handle more test points and higher frequencies.

RF cabling

Whereas much of the focus is correctly on the probe interface, you should also consider the importance of proper RF-cabling design, configuration, and quality. Although cabling represents a relatively small percentage of the overall test-system cost, in an RF-test environment, skimping on cabling or assuming that just any cable will do the job never makes sense.

Cables should provide low attenuation, high flexibility, and highly repeatable electrical recovery following flexure. In addition, they should exhibit consistent phase stability over a range of temperatures, especially because you make many RF measurements while holding the target devices at elevated temperatures. Just a few degrees of temperature change can significantly affect the electrical length of a low-quality cable. To minimize undesired effects, you must use stable, carefully specified cables and keep them clean and properly torqued. Too often, cables have been lying around the lab for years and are of unknown quality. Dirt, nicks, and excessive torque often compromise the cable connector, even though the cable material itself may be good. Some users go to great pains to use low-cost cables, taping them down so that they don't pick up vibration from movement in the room and to protect them from drafts and vibration from air-conditioning units and the like. Despite such precautions, such systems still experience significant random calibration errors because, occasionally, someone bumps a cable with his elbow, and the cable can't recover quickly enough from the shock. A better approach is to invest in RF cabling that can withstand the demands of the test process and then to configure and maintain the cabling in a disciplined manner.

Standards

The final key element of an RF system is calibration, which is essential for highly accurate RF measurements. Calibration, or vector-error correction, involves careful, accurate RF measurement of known standard devices or impedance standards. By measuring the known standards, you can identify and remove undesired effects from subsequent measurements of the DUTs. Calibration standards typically include short, open, load, and through delay. The short is just that-with all probe tips contacting a common conductor. You typically measure the open using a probe-in-the-air method, with no connections for any of the probes. You measure load using a precision-trimmed, 50? load resistor. For the through delay, you use a short transmission line of known delay-usually 1 psec. Convenient off-wafer mounting of the standards is also important to allow easy integration of calibration into the measurement process-ideally, with readily available standards mounted as discrete independent substrates.

Accurate calibration requires a combination of the impedance standards themselves, an off-wafer-mounting site for the standards, and software to manage the calibration and system validation. Many people are just starting to use RF in their test suites, and, with some of the new requirements, such as multiport and differential measurements, calibration has become more challenging, complex, tedious, and error-prone. Calibration-management software addresses these issues not only by automating the entire calibration process, but also by providing guidance for system setup, tracking of system elements, and help to users in understanding the process. The wizards and tutorials that are part of comprehensive calibration software can speed learning, more quickly get users up and running, and help with optimization of measurement sequences. For example, software can remember and track all of the probe dependencies for various calibration values, such as which electrical values apply to a particular probe. This information ensures that calibration errors can't occur simply because an operator enters a wrong value when setting up a system. If undetected, such a simple error can propagate a long string of incorrect values and incorrect calibrations, resulting in numerous incorrect measurements over an extended period. As with cabling, an investment in calibration software can greatly enhance the value of the entire RF system.

Many customers use only an open transmission line or open stub for calibration validation. In measuring an open stub, you can look at the reflection coefficient on the Smith chart. With increasing frequency, it should show an inward spiral and roll-off. If it's a good match and good calibration, the open stub should also show a linear-phase-transmission response.

Wafer-level errors

Three of the most common errors that affect wafer-level RF measurement are incorrect probe planarization, dirty connectors on cables or probes, and improper probe placement on standards. Planarization is the process of using the adjustment on the probe arm to make sure that all of the probe tips make contact with the device in a parallel plane. For example, with a GSG probe, it is critical that all probes come in uniform contact and have equal scrub areas on the DUT pads. One of the tools for making planarization easier, a contact substrate, provides a check area for making probe contact and examining the contact scrub marks. You can then make probe-planarity adjustments to ensure a good contact. If possible, you should use systems with dual off-wafer-mounting sites for both a contact substrate and an impedance-standard substrate. You should also provide proper cable-strain relief to make sure that the cables don't pull the probes out of planarity. In addition, calibration software is useful in that it offers optimization tools that make it possible to electrically determine the consistency and integrity of probe contacts.

The second frequently encountered error condition relates to material accumulation on cable connectors or probes. A simple discipline such as occasionally cleaning and swabbing the cable connectors with isopropyl alcohol can make a world of difference, as can regularly checking the connector fit with a torque wrench. The probe needles themselves also can pick up debris over time, especially when probing on aluminum. Get in the habit of always checking for properly cleaned probe needles to provide electrical stability and low contact resistance.

Visually inspect probe tips and first clean them with air. For standard needles, you can also use a soft brush, isopropyl alcohol, or both. Remember to brush away from the body of the probe. For some more advanced probes, such as those based on thin-film technology, it's more appropriate to use semisticky materials, such as a gel pack, which can cleanly attach itself to the debris and pull it off the probe tip.

Another common error is improper probe overtravel, or skate, especially when contacting impedance standards. "Skate" is the distance that the probe moves across a device or standard when you apply a downward force to make contact. The degree of skate electrically affects the measurement, whether it is on the DUT or on a calibration standard. Both excessive and insufficient skate can significantly affect the results by changing the inductance of the contact interface (Figure 5). To help control this effect, calibration-standard substrates typically incorporate alignment marks that allow users to adjust the amount of scrub to control overtravel and maintain electrical stability. Once again, proper use of software helps. By using software to automate the calibration, you can get consistent probe placement-not just on the scrub, but also in controlling x, y, and z placement. Research has repeatedly proved that calibrations under automatic software control are more repeatable than those using manual probe placement.

As devices scale down and add functions, it is clear that the need for RF-measurement techniques is increasing. Also, the development of new differential, multiport devices has driven a movement to RF disciplines. In addition, new differential, multiport devices and balanced measurements rely on RF probes with better crosstalk performance and for better software to prevent errors due to increased calibration complexity.

Although you need to prepare for a future that includes RF measurements, the good news is that the necessary tools and processes have for years been in place and undergoing refinement. The most important factors are understanding the key differences between RF and lower frequency measurements; making wise investment in a high-quality wafer-level test system; and developing solid disciplines for system setup, calibration, and maintenance.