Once a printed circuit board (PCB) design is complete and the required preproduction models built and tested, producing the assembly in volume should almost be automatic. Right? No way. While it’s true you can achieve a high test pass rate, it’s by no means automatic. There are too many process variables involved.
If you assume that a PCB layout has followed reasonable rules for clearances and component footprints, what can go wrong in production? One of the largest groups of problems concerns some aspect of soldering. A dirty lead or pad may result in no joint being formed or one with insufficient solder. Conversely, too much solder contributes to bridging shorts between component leads.
In addition to soldering problems, components could be incorrectly placed, or they may be the wrong components to start with. A comprehensive board test program is required to find bad assemblies and provide feedback to improve the production process.
One Size Doesn’t Fit All
A great deal of attention has been focused on the large, complex PCBs that have been developed for computer and telecommunications applications. The circuitry on these boards may be designed to operate at very high speeds, which can restrict the use and placement of test pads. Components are tightly spaced on high-speed PCBs because even short interconnection wiring can degrade signals.
Nevertheless, there are things that can be done to improve the situation. “Where time has been taken for the design and test engineers to work together to improve testability, higher test yields will result,” said Dave Tucker, the operations manager at Test Technology Associates in Lewisville, TX. “For example, part of the problem has to do with probe distribution. Many probes will be concentrated under high-density ball grid arrays (BGAs) unless the test and design engineers work together to spread out the test points. This gives you a better chance of making contact each time and puts much less stress on the PCB and BGAs.”
Bed-of-nails fixtures can be designed with more than 7,000 probes, but as the number of probes becomes large, false failures caused by faulty pin-to-board contact increase significantly. Some companies work to combine complementary test methods to avoid the need for very high pin-counts (Figure 1, see the November 2001 issue of Evaluation Engineering).
One approach by Agilent Technologies, Agilent AwareTest xi, uses X-ray inspection to reduce fixture pin-count. For example, if a PCB has several BGAs, X-ray inspection can confirm that all the balls are making contact without any shorts and that the solder fillets are adequate. The AwareTest xi software determines that fewer test pins are required in that area.
Another approach uses a specialized design-for-test (DFT) consultant. “Many companies aren’t large enough to have a test engineering staff,” Mr. Tucker continued, “so they hire an outside consultant to help with probe placement and testability. I’ve seen this approach work very well for telecom startup companies.”
At the other end of the board size/complexity spectrum, many limited-function PCBs are produced by the thousands, and each one has less than 100 nodes. For these boards, testers capable of handling 400 to 800 nodes are appropriate. Small PCBs often are built as a panel of four, eight, or more separate assemblies and tested in that form prior to singulation.
The types of ATE test capabilities also vary. Manufacturing defects analyzers (MDAs) test passive components and do not apply power to the board assembly. In-circuit testers (ICTs) deal with each part in an assembly and include powered IC along with passive tests. Functional testers determine that the assembly performs as designed when subjected to a range of expected input signals.
In practice, the three categories, MDA, ICT, and functional, really only form two groups: MDAs and ICTs are similar and often talked about interchangeably while functional test has a distinct and different purpose.
As the name implies, MDAs are intended to find manufacturing faults such as missing or wrong components and solder bridges. Passive components and groups or clusters of components can be tested by MDAs.
The impedance between two points is measured and compared against the expected value. Some machines require programming for each component while others can learn the required information by testing a known-good board.
Guarding is used to electrically isolate the component(s) being measured. ICs may be tested with a capacitive technique such as Agilent Technologies’ TestJet method or GenRad’s Opens Xpress to ensure that all leads are connected. Otherwise, MDAs assume the ICs on a PCB are good.
Figure 2 (see the November 2001 issue of Evaluation Engineering) shows a test fixture using the Opens Xpress technology. When the fixture is closed, the square metal areas in the lid are positioned over large ICs on the PCB being tested.
ICTs can do everything that an MDA does, but they also apply power to PCBs and test ICs individually. Backdriving ensures that an input to a device can be forced high or low even though it may be connected to logic outputs: sufficient current is supplied to force the output(s) to follow the test signal. Although some ATE manufacturers provide the means to monitor and limit backdrive time and current, it remains a controversial practice.
As device geometry becomes smaller, the risk of causing permanent but undetectable damage by backdriving has increased. One area especially sensitive to product reliability is the emergency siren and light business. Whelen Engineering, headquartered in Chester, CT, designs and manufactures strobe emergency lighting, siren amplifiers, and PA systems for emergency vehicles including fire, police, ambulance, and tow vehicles. A typical board is shown being tested in Figure 3, see the November 2001 issue of Evaluation Engineering.
Chuck Everett, the company’s chief test engineer, explained, “If you listen to the people who advocate not backdriving, they compare it to ESD. It’s comparable because you don’t know what kind of latent faults you may be putting into a product that might not show up for six months or a year.”
Have overstresses caused damage that isn’t readily apparent at the time you ship a product? “I have alternative ways of testing our products,” he concluded, “so I’d rather not take the chance of backdriving if I don’t have to.”
The CheckSum MDAs favored by Mr. Everett also include some functional test capability, with isolation modules in the MDA that limit voltage during powered test. Automation of the functional test on the same fixture immediately following the MDA component verification provides tremendous time savings and is used wherever viable—basically on boards operating at 12 V or less.
In contrast to the emergency services business, consumer product manufacturing emphasizes high throughput and low cost. Nevertheless, high test coverage and quality still must be achieved. Javier Merino, a principal engineer at Honeywell, uses MDAs to test smart gas-valve controllers and controls for furnaces and air conditioners.
A typical PCB may have 120 components and from 40 to 60 test nodes. The PCBs are manufactured and tested as a panel that contains up to eight separate assemblies. About 70% of the components are through-hole style, and the remainder is surface-mount. But the big challenge for Mr. Merino is the 20,000-PCB per-day volume.
Different brands and sizes of end-user equipment require slightly different controller boards. This means that one PCB design may be built in several ways by using different component values or inserting jumpers. The same test fixture is used for all variants of a particular design. Separate MDA test software files accommodate the changes. At present, the company is running seven different product lines through separate MDAs and fixtures.
“The biggest problem is the pins, the contacts. I don’t have problems with the test equipment,” Mr. Merino said. “I have a lot of problems with the pins. There is no choice other than the pogo-style test pins, but they soon wear out with our volume and in our environment. We have a dusty environment here in Tijuana,” he continued, “so we maintain a positive pressure inside the test rooms. Even so, the pins do become contaminated, and there is no way to clean them. As soon as the boards start to fail, we know that most likely it’s a pin problem, not a board problem.”
Xantrex Technology’s Distributed Power Division, formerly Trace Engineering, manufactures power supplies, AC-to-DC and DC-to-AC converters, and chargers used to supply backup and renewable power for homes and small businesses. A separate MDA now is used for each product line, although this change occurred only recently.
As little as a year ago, according to Sami Bouzeid, production engineer at Xantrex’s Arlington, WA, facility, “each of the stations that we had could run any type of board, plus we had 10 different kinds of PCBs with a dozen variants of each one. In addition to the 120 test files, we changed test fixtures too. Separate testers are much better. In an emergency, because the ATE is all the same, we can temporarily move PCBs to another machine.”
Because of the nature of the products Mr. Bouzeid tests, MDAs can’t achieve 100% test coverage. For example, it’s common design practice to bypass a very large electrolytic capacitor with a smaller ceramic one. This is done to extend the frequency range over which the combined capacitor has a low impedance. Unfortunately, there’s no way to test for the presence or correct value of the smaller capacitor.
“If you have a 100-µF electrolytic capacitor in parallel with a 0.1-µF capacitor, you must account for the tolerance of the larger one. If that is ±5%, or more typically ±10%, the 0.1-µF part can’t be distinguished,” Mr. Bouzeid continued. “So, you still have to do some amount of visual inspection. There’s no other way.”
Even separate large capacitors may complicate test. If a constant current is used to charge the capacitor so that its value can be determined by measuring the voltage after some time, you must program a delay into the tester. To maintain throughput, you want the time to be as short as possible, but because of the large tolerance often associated with large capacitors, you must delay for the maximum time. To avoid the problem entirely, Xantrex installs some capacitors by hand after the PCB assemblies have been tested.
Another problem involves a small resistance such as 27 W in parallel with a diode. If voltage is applied to reverse bias the diode, the resistance can be read. However, over 30 mA is required to create the necessary 0.6- or 0.7-V drop across the resistor to forward bias the diode, and a diode test may not provide sufficient current. Visual inspection also is used in this type of situation.
The last opportunity you have to find PCB problems before your customer reports them is at functional test. But this manufacturing stage does much more than simply test boards. Honeywell’s Mr. Merino said, “Some of the boards go into the gas valve, and they are tested as part of the unit. We test them in a simulated working environment.”
Whelen Engineering’s Mr. Everett highlighted other aspects of functional test. “High-voltage strobes use up to 600 V, and you’re not going to test that on an MDA. Some PCBs will be functionally tested in the actual product housing because it may provide heat sinking or have connectors for I/O cables. It doesn’t do any good to make sure the boards are working only to find out later that the I/O connectors were miswired.”
As many manufacturers have confirmed, functional test setups are very product-specific. In the case of Whelen Engineering’s products, the functional test software is based on National Instruments’ (NI) LabWindows/CVI. This programming environment was chosen for several reasons.
The drivers necessary to address the NI data acquisition boards that are part of the functional test hardware are bundled with LabWindows/CVI. Equally important advantages of LabWindows/CVI are the efficiency and maintainability Mr. Everett feels he has achieved.
In spite of the often-stated productivity improvements associated with graphical programming languages, Whelen Engineering chose LabWindows/CVI because it is based on the C language. Mr. Everett said, “The screen the test operator sees looks very similar to what he would see if I had written the test program using LabVIEW. The difference is that behind the graphical user interface the programming language is C, and that’s a big advantage.
“C programming has become a universal standard in the technical world. There still is a learning curve with LabWindows/CVI,” he continued, “but it’s very short. In fact, there have been cases when I haven’t been available, and some of our programmers who ordinarily work on product development were able to update the functionality of a test program with little difficulty. That wouldn’t have been possible if the test program hadn’t been written in C.”
Obtaining a high pass rate for board ATE requires up-front planning that starts at the design stage. For example, at Whelen Engineering, board designs are standardized to provide a minimum of 100-mil test-pin spacing, allowing the fixtures to be built using fairly robust large-diameter probes. In addition, board designers are required to provide access to all signal nodes on the bottom of the board to accommodate single-side test access. These conservative rules result in mechanically simple and robust fixtures that help eliminate false failures due to poor pin-to-board contact.
For large and complex boards, using a DFT consultant or a DFT optimization program such as Agilent AwareTest xi can avoid major testability problems in production. As Test Technology Associates’ Mr. Tucker put it, “Obviously, the key is to get to market quickly, but time must be taken for design and test engineers to build testability into the design. If not, we all pay for it down the line due to false failures and untested devices.”
Return to EE Home Page
Published by EE-Evaluation Engineering
All contents © 2001 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.