Lower RF uncertainty boosts phone yield
Optimising the accuracy and repeatability of your RF test system significantly reduces false failures and attendant costs in mobile phone manufacturing.
Todd Wendle, Agilent Technologies -- Test & Measurement World, 4/1/2001
Mobile phones have to operate to very close test limits to meet their performance requirements. In practice, making measurements to close test limits means that a significant portion of the measurement’s distribution curve can exceed test limits. This means that indication of failure can be a result of measurement uncertainty rather than a faulty product (see Figure 1).
The fundamental functionality of mobile phones depends on RF performance and, typically, several measurements in mobile RF testing have distributions close to test limits. Three such measurements are transmitter maximum power, receiver sensitivity, and either frame or bit error rates. Take, for example, the 23 to 30 dBm maximum power specification range of a Class II CDMA phone. Manufacturers set their maximum power test limit very close to the lower limit of 23 dBm, say at 23.50 dBm. If the power measurement is 23.50 dBm, the product passes and moves on to the next test. If the same phone—because of measurement uncertainty—gives a measurement of 23.49 dBm, it fails. In this case, the phone needs re-testing or possibly even reworking. No matter how you “handle” failed phones, they slow production rates and add cost to the phone. Pin down uncertainty The four primary contributors to mobile RF measurement uncertainty are the mobile itself, the test fixture, cabling and connectors, and test instrumentation. To examine the mobile phone’s contribution to measurement uncertainty is beyond the scope of this article. But, briefly, as phone manufacturers try to reduce manufacturing costs by using less costly components, the phone’s contribution to measurement uncertainty can become significant. While it’s possible to reduce the phone’s measurement uncertainty through redesign, the first step is to reduce the uncertainty of the other measurement uncertainty contributors. Once the system uncertainty is understood and under control, the test results will then more accurately reflect the true performance and measurement uncertainty of the phone. Fixturing is probably the least understood contributor of uncertainty in test systems. RF isolation and shielding are key contributing factors. Adding other tests, such as audio or keypad functionality, may compromise an optimally shielded RF chamber. For example, RF noise can dramatically affect error rate test results without adequate RF isolation and shielding in the fixture. Cabling and connectors are the most volatile uncertainty contributors to an RF measurement system. A typical production environment has many connections and several metres of cabling for transporting RF signals. In some cases, cabling flexes continuously and the main test connector to the phone can undergo thousands of operations. The cabling molecular structure continually changes and the connector interface slowly wears away. Both these effects change standing-wave-ratio in the RF measurement path and increase measurement uncertainty. Test instrumentation specifications are critical. Some instrument manufacturers have very stringent and conservative specification procedures. Others do not. Furthermore, the language used to describe specifications differs from manufacturer to manufacturer. Specifications can be maximum worst case, typical, or sometimes non-existent. Specifications may state absolute accuracy or just repeatability. Overall, though, uncertainty in test instrumentation has to be significantly less than uncertainty in the phone. You should rigorously factory test your RF instrumentation to validate and guarantee its accuracy and repeatability. False failures The combination of RF measurement uncertainty and test distributions near test limits creates the big problem of false failures, or no-fault-found products. A false-failure phone is one that subsequently passes on an independent test system and therefore produces no fault diagnosis. Handling false-failure phones takes extra resources and equipment. Depending on your production flow, false failures reduce overall production capacity.
Figure 2 shows an example in which a phone has to fail twice before being moving to a diagnostic area. There are now two possibilities. First, the phone is a true failure or, second, the phone is a false failure because of measurement uncertainty. In either case, the phone travels to some type of rework area for diagnosis. In this area, a test system with a quite different set of cabling, RF connection, fixture, and instrumentation will repeat the test. This, in itself, only adds further to measurement uncertainty. For example, if the phone only marginally failed in the production test system, it has a high probability of passing in the rework RF test system. At that point, if it passes, it becomes a false failure and returns to the main production line for retest. In practice, there is a continuously varying quantity of phones moving into and out of the rework area, which is hard to manage. The quantity of phones is directly dependent on the main production line yield. The number of phones needing rework is a small percentage of the total production. Therefore, small changes in yield create relatively huge swings in the quantity of phones needing rework. Even a single percentage point change, up or down, in the production yield can severely impact the rework area. Eliminating even a portion of that variability improves the ability of that rework process to adequately handle the inevitable swings in the number of phones needing rework. A significant portion of that variability is made up of no-fault-found phones. The box “False Failures—Phew!” shows how expensive false failures can be.False-failure baseline Although you can reduce false failures by reducing measurement uncertainty, there are three steps to take before you can evaluate any measurement uncertainty reduction project. All three concern determining your existing RF test system measurement uncertainty, and documenting and costing the false-failure rework process. In detail you need to: • Determine the measurement uncertainty of the fixture, the connectors and cabling, and the test instrumentation. (Don’t forget to include the RF rework diagnostic system.) • Graphically represent the false-failure rework process, which differs from a truly failed phone. Determine labour costs and any equipment overhead costs in order to establish a cost per false failure. • Determine any opportunity costs associated with false failures by studying the false-failure rework process, line capacity, and per phone profit.You can’t evaluate any reduction project without doing this preliminary work. There are many different reduction methods, and without having a baseline it’s too easy to waste effort. Evaluating uncertainty With baseline work complete, it’s time to evaluate measurement uncertainty reduction methods. Using a simulation model eases the evaluation of false-failure costs. By simply plugging in different variables, the simulation model predicts the number of false-failure phones. Using this number along with some of the rework costing numbers from the baseline study, you can then calculate the cost of false-failure phones for the variable set inputted into the model. The following steps outline one evaluation method: • Determine the distribution mean and standard deviation for any measurement that has a distribution close to a test line limit. • Determine the existing false-failure rate. • Calculate your current false-failure cost using the existing false-failure rate and the rework data from the benchmarking data set. (A spreadsheet works well.) • Do a quick assessment of the current fixturing, connections, cabling, and test instrumentation. What do you know about each? What data are missing? Don’t forget to include calibration procedures and any RF test system maintenance. What’s different between the RF test systems? What’s the same? • List potential changes that you can make that may reduce measurement uncertainty. (See the box “Reducing RF Measurement Uncertainty”). • Using some type of model, determine the new false-failure cost with the potential changes in place. • Create a plan to implement the changes.As you can see, reducing measurement uncertainty is an on-going task that includes maintenance issues, new phone technologies and components, new services, new measurement techniques, and evolving manufacturing methodologies. Your work in keeping the RF test system as accurate and repeatable as possible will be a never-ending but rewarding experience. TME Todd Wendle is a product marketing engineer in the Services and Support Business Unit with Agilent Technologies. He is a specialist in service solutions for the wireless communication industry. | |||||||||||||||||||||||||||||||
| False failures—phew! No-fault-found phones are expensive as you can see from this example of a hypothetical TDMA phone. The false-failure rate calculation uses a simulation model that accepts the defined variables to model the results for the particular test for 100,000 phones. You then multiply the false-failure rate by the cost to move a false-failure phone through a typical rework process. Figures A, B, and C correspond to the three tests in the table below. The total false-failure rate is equal to 1.3%, or 13 phones per 1000. The rate is calculated by multiplying the probabilities that the phone will not be a no-fault-found unit and subtracting from 1—i.e., 1 - (0.996 x 0.997 x 0.994). This method assumes that: • single production line volume capacity is 3000 phones per day, Therefore, uncertainty cost is 518 Euros per day per line. With four lines and 30 production days/month, the uncertainty costs equate to a monthly cost of around 62k Euros, or a yearly cost of 746k Euros.
| |||||||||||||||||||||||||||||||
Lower RF uncertainty — what’s it worth?
Using the flowchart in Figure 2 (above), let’s step through an example of a hypothetical uncertainty reduction exercise centered on a single measurement. For the purposes of this example, the following assumptions apply: measurement is TDMA maximum digital power; benchmarked standard deviation for measurement uncertainty was measured with valid methodologies; rework labour-rate is 30.00 Euros; and the example ignores opportunity costs. There are two ways to calculate total uncertainty standard deviation. The worst-case way is to just add them together. A more realistic way is use the square root of the sum of the squares, which accepts that when several random variables are independent, the variance of the sum is the sum of variances. In this case, the value would be 0.71 dB as opposed to 1.1 dB. For this example we will stick with worst case. Now let’s fast forward through a measurement uncertainty reduction process (see the table below). The variable set for the measurement stays the same since no work was done to improve the phone’s performance for this test. All connectors were replaced with a higher precision connector, flexible cabling from the test rack to the fixture was replaced with semi-rigid cabling and the RF connections were tightened to the correct torque. New test instrumentation gives a power measurement repeatability specification of 0.05 dB. The projected saving is 569 Euros for every 1000 phones for this measurement alone. Based on 30 days per month, four production lines each with a 3000 phones capacity, the monthly projected savings are 205k Euros. Also notice that an additional 12,600 phones would pass this test the first time instead of becoming false failures. If you include opportunity in this example, the cost savings increase dramatically.
| |||||||||||||||||||||||||||||||
| Reducing RF measurement uncertainty Here are some useful ideas for reducing RF measurement uncertainty: |
