Test your testers for R&R
Automated test systems need to operate within designed parameters so that they produce reliable and repeatable results. False failures mean good products don’t ship, and false passes can result in field problems. To keep a test system operating properly, you need to know how well it worked when it was put into operation or last calibrated, and you need to monitor the station for deviations.
The methods for testing a tester can vary depending on the tester’s use and specifications, but there are many common techniques. One involves the use of check standards, which could be components such as resistors, voltage sources, current sources, or frequency sources that may be built into a test system or attached when needed. You can also use known-good devices—often called “golden” devices—to verify a tester’s integrity, or you may want to rely on instrument self-test and loopback tests at regular intervals to keep testers running properly.
Set a baseline
Before putting a test system into service, you need to characterize it. Here’s where a calibrated check standard or golden UUT (unit under test) can help. Such components let you establish a set of baseline measurements for each instrument and for the entire system. The best check standards are calibrated using methods traceable to NIST (National Institute of Standards and Technology) or another national lab. A golden UUT lets you check all of the connections in your system using the same cables, connectors, and fixtures that the system will use every day.
When you characterize a tester, always record and plot your results. You’ll need them to compare against future measurements. Plots can reveal trends before the tester exceeds its limits. Over time, you’ll make thousands of measurements that let you develop a performance history of R&R (reliability and repeatability).
Larry Raymond, president of Intrinsic Quality, a test-system developer, explained how to establish a baseline set of measurements for an in-circuit tester. “Use a reference board, or better yet, a set of boards with component values you trust,” he said. Raymond recommends that you make enough measurements to establish a good statistical sample, typically hundreds to start. Then, you should plot a histogram so you can verify the tester’s repeatability.
Raymond explained that sometimes one of his customers will provide just one board for verifying a tester. When that happens, he requires hundreds of measurements to prove that the tester is repeatable.
If possible, you should measure the components on several boards. That will give you a better statistical sample. It will also give you a clue as to how the components’ values vary, which lets you establish the tester’s R&R.
If you have more than one sample UUT, you should look for inconsistencies in measurements. For example, you may find that a large set of capacitance measurements range between 120 µF and 132 µF. In such a case, Raymond suggests that you set the tester’s acceptable limits for the component limits to perhaps 110 µF and 140 µF. If measurements fall outside those limits, you should check the tester against a reference board. Center your test window around the mean of the distribution of the baseline measurements. If the tester is operating within specifications, you may need to change tester parameters such as test current to compensate for the errors.
Letting the tester test itself is a common technique that many engineers use. Marius Gheorghe, engineering manager at test-system integrator Ideal Aerosmith, explained that the company often provides loopback TUAs (test unit adapters) that connect signal sources to measuring instruments. Figure 1 diagrams a test system that uses a loopback TUA to connect its oscilloscope and arbitrary waveform generator.
Figure 1. a) A TUA makes the instruments available to the DUT, while (b) a loopback TUA routes signals from the waveform generator to the oscilloscope for system test. Courtesy of Ideas Aerosmith.
Figure 1a shows the normal operation where the waveform generator’s outputs and the oscilloscope’s inputs come to a mass-termination panel. A TUA that attaches to the panel during normal operation routes signals to a DUT (device under test). Diagnostic software in the system controls both instruments and verifies that the waveform generator’s signals are within limits.
Gheorghe suggests letting your loopback TUA make the connections between the signal sources and the measuring instruments and switching subsystems. “Don’t route signals used for system self-test entirely through the system,” he said. “Use an adapter. You’ll get more flexibility.”
Diagnostic panels such as the loopback TUA in Figure 1b may do more than just connect two instruments. They can hold test components such as resistors, voltage sources, current sources, or oscillators. They can also connect instruments to any test and calibration resources that may reside in the test system, such as DMMs (digital multimeters), oscilloscopes, and signal sources.
Test systems that consist mostly of card-based instruments may have just one stand-alone measuring instrument: an oscilloscope. Engineers often use the oscilloscope not only as part of the system, but also as a diagnostic tool. The oscilloscope can check signals as they pass through switches, cable, connectors, and mass-termination panels.
Mark Carlson spent 30 years as a test engineer with Texas Instruments. During that time, he developed testers for the company’s analog and digital ICs. In addition to using a loopback TUA, the testers he developed have a “calibration bus” that lets engineers check and calibrate power-supply outputs (Figure 2).
The calibration bus connects test equipment to resistors with 0.01% tolerance and to an ADC. The resistors provide a precision load for the power supplies, while the ADC measures power-supply voltages. Carlson explained that the ADC makes two voltage measurements on each power supply at the low and high ends of its range. A PC then calculates slope and offset (mX+b) to calibrate the power supply’s output. The results go into a look-up table. Whenever the system needs an output voltage, it refers to the look-up table before sending a command to the power supply. The TUA also holds a NIST-traceable voltage reference for the ADC, making its measurements credible and verifiable.
Loopback tests and instrument calibration let you verify that a signal path is working properly, but what if the measurements seem wrong? You must troubleshoot the system. Here, reference devices can help you isolate errors.
For example, if your system has a DMM that measures power-supply voltage and the DMM has a self-test, then run the test. Try checking the meter with a known voltage that you’ve checked on another meter. Check the power supply’s output. Is it what you expected? If it is, you likely have a problem with a cable, connector, or switch.
Test adapters that use golden UUTs may be far more complex than those that simply route signals between instruments. The board in Figure 3 contains relays, an FPGA, and other devices that test engineer Todd Grey of Maxim Integrated Products uses to test digital ICs.
The FPGA works in combination with the relays under software control to connect a DUT to a power supply and to the appropriate pull-up components for the DUT’s communications bus. The pull-ups can be either resistors or FETs, depending on pull-up requirements for the DUT.
How often to check?
Once you’ve established a baseline for a tester’s performance, you should periodically check it. You can use a loopback setup, a reference component, or a golden UUT to conduct periodic checks. As a start, you can take advantage of self-test features built into your test equipment. You can run self-tests before every shift, every day, or every week, or you can run them at longer intervals, depending on how much you can trust the test equipment’s stability.
Grey uses the test board in Figure 3 at least once a week. He performs his tests using the tester’s built-in diagnostics, a set of golden devices, and load boards. He will run complete characterization tests on the golden devices and compare the results against previous results. These tests provide greater test coverage than production tests. He often uses a separate DMM and oscilloscope when checking the tester.
Checking your instruments frequently can help you catch errors caused by signal-path degradation or by an instrument drifting out of calibration, especially if you use that instrument to test an important product specification. David Buxton, senior test engineer at Tektronix, noted that “The last thing you want is to have a calibration lab report that one or more specifications, for an item of test equipment, was received out of tolerance. When that happens, an investigation should follow, which can lead to a recall notice for a free recalibration of the instrument.”
Figure 4. Test data from six test stations shows that readings from station 5 are near the high limit. Courtesy of Tektronix.
Buxton noted that you can apply SPC (statistical process control) analysis to the test data to look for trends. If the reference component is built into the system, then you can program it to automatically run a check and log the results.
Checking production test data can provide clues about a tester’s health. Look for trends in the measurements on production parts. A trend could indicate that an instrument is drifting out of its tolerance or that something else is wrong. Figure 4 gives an example of measurements made on five test stations, while Figure 5 shows data from a single tester.
Figure 5. Test data shows a sudden measurement dropoff, but not from all measurements, which could indicate intermittent performance. Courtesy of Tektronix.
In Figure 4, the data from station 5 indicates a measurement problem with the station. For example, if you notice that a power supply’s voltage is dropping, it could indicate an increased resistance in a switch, connector, pin, or relay. Test pins get soft, and they may not make good contact over time. In addition, poor test-clamp alignment may also cause measurement errors.
Figure 5 illustrates a problem from a single test station. The plot highlights a drop in a measured value, and the measurements are erratic. This could indicate an intermittent instrument problem or a signal-channel problem that will require troubleshooting.
If you encounter situations like those depicted in Figures 4 and 5, you can start by checking the test equipment against a known reference. You can also run the same measurement on a different instrument or tester. If you see significant differences in a measurement from one tester to another, you know to suspect the tester.
You should try to minimize the number of obstacles in the suspected measurement path. Running a signal from a signal generator through an amplifier, an attenuator, and a test head to a spectrum analyzer means that any of them could cause a measurement error. You will likely have to check each component.
Signal paths that carry RF signals such as radar or serial data streams of 10-Gbps Ethernet have more possibilities for signal errors than low-frequency signals because of losses and reflections. Chris Scholz, field applications engineer at LeCroy, recommends that you use a TDR (time-domain reflectometer) to characterize signal paths and that you periodically rerun TDR measurements, looking for changes in results that could indicate a change in impedance.
TDRs are usually an option on high-bandwidth sampling oscilloscopes. Their wideband analog front ends let them measure reflections on repetitive pulses with rise time of tens of picoseconds (Ref. 1). You can use TDR measurements to calculate S-parameters that characterize a signal path in the frequency domain. You can also use a VNA (vector network analyzer) if you have one.
Knowing the characteristics of your signal path, you can “de-embed” or compensate for channel losses in your measurements. If you shut down your tester for maintenance say, every six months, then that’s the time to run a TDR measurement. You need to make measurements with a TDR or VNA for every test fixture that carries high-frequency signals.