In the reliabiliy lab at Luminus Devices, reliability manager Scot Solimine (left) runs accelerated life tests on the company's LEDs. Staff engineer Mike Brown analyzes any failure modes and suggests corrective actions. Photo by Adrien Bisson.
Billerica, MA—When lighting designer Kevin Adams designed the lighting for Green Day's "American Idiot" musical on Broadway, he wanted to illuminate the stage with white light. He found the light he wanted in stage lights that contain LEDs manufactured by Luminus Devices (Ref. 1).
Luminus Devices manufactures LEDs that appear not only in stage lighting, but also in portable projectors, retail stores, homes, and even street lights. They've also appeared in projection TVs.
Started in 2002, the Billerica-based Luminus produces LEDs as large as 12 mm2 that produce blue, red, green, and, of course, white light at its foundry in nearby Woburn, MA. Such a large size presents test issues not present in typical 1-mm2 LEDs. Assembly, packaging, test, and engineering take place in Billerica, where engineers design and test the LEDs under conditions that exceed specifications. Those tests assure users of consistent brightness, color, and reliability.
Luminus delivers LEDs packaged on either ceramic substrates or on metal-core PCBs (printed-circuit boards). When an LED is assembled onto metal-core PCBs (Figure 1), Luminus places either a glass window (shown) or a dome-shaped lens (not shown) over the device. The assembly includes a power connector that looks similar to those used in automotive fuses. The devices need heat sinks and large pins because they can operate at up to 30 A of forward current. The devices dissipate as much as 150 W, which produces temperatures as high as 150°C at the device junctions.
|FIGURE 1. These red, green, and blue LEDs can operate at up to 30 A of forward current. Courtesy of Luminus Devices.|
All that power produces light that's too bright for a person to look at directly. In fact, the LEDs are so bright that you can easily see them in sunlight. Indeed, engineers were running a life test with LEDs shining through a window on the day of my visit. See the video of pulsing light that I took while standing outside the building.
Injecting high current
Testing begins when wafers arrive from the fab. A wafer-probing system injects up to 10 A of current for 0.2 ms to 2 ms from a Keithley Instruments SMU (source-measure unit) into each device. The SMU also measures forward-operating voltage and reverse-leakage current. "We would like to synchronize two SMUs to produce 20-A current pulses," said test-engineering manager Michael Joffe. "Pumping 20 A through an LED will provide better prediction of a wafer's quality."
LEDs can generate as much heat as a professional-grade soldering iron, according to test-engineering manager Michael Joffee (right). Test engineer Aaron Plaisted commented that Luminus tests LEDs at current up to 36 A.
Photo by Adrien Bisson.
During a test, a spectrometer detects an LED's shade of color, including the dominant wavelength-the one that most influences how the human eye perceives color. "We do full spectral analysis on the LED," said Joffe. VP of engineering Arvind Baliga added, "We screen RGB [red-green-blue] LEDs at the wafer level based on forward voltage, reverse current, dominant wavelength, and brightness. White LED die are binned by color point after screening for forward voltage, reverse current, and brightness."
Baliga explained that the screened LEDs are then tested for uniformity, which lets engineers detect possible defects; a big LED can have occasional processing variations across its surface. When engineers detect defects, they look back through the fabrication process and locate the root cause.
After dice are sorted, they are assembled onto substrates or boards. Products for sale go to production test, but new designs first go through engineering evaluation. A team of six test engineers and technicians develop and maintain test stations for both engineering and for production. They also evaluate new designs.
The test engineers measure LED packages for forward voltage, reverse current, brightness, and color. They also measure a device's temperature. They can't, however, measure an LED's junction temperature directly. They measure temperature at the most accessible point and calculate junction temperature based on the thermal properties of PCBs, heat sinks, and the devices themselves. Temperature and forward current affect an LED's light intensity and color.
The engineers test LEDs using both pulsed and direct current. Testing LEDs at 30 A is more difficult than testing at, say, 1 A, particularly with pulsed current, because the shape of the current waveform affects LED performance. For example, overshoot will change the average intensity of the LED in the period when the LED is on.
Testing large LEDs
Dedicated test equipment isn't always available to engineers developing new technologies such as the 12-mm2 LEDs. "You can't buy COTS [commercial off-the-shelf] test equipment for our high-current devices," said Baliga. Thus, the test engineers at Luminus have to design, assemble, program, debug, and calibrate testers in house. Figure 2 diagrams a test system used in production and in engineering.
FIGURE 2. A test system consists of an integrating sphere that captures light from an LED under test. A custom driver circuit provides pulsed current up to 30 A.
An integrating sphere is the heart of the tester. Its detector converts light from the LED into a voltage that a National Instruments data-acquisition card can digitize. That lets the tester measure light intensity. An optical fiber attached to the sphere transfers light to a spectrometer that measures the LED's optical spectrum and dominant wavelength.
The size of the sphere affects the resolution and accuracy of the light measurements. Test stations for production and engineering use 12-in. diameter integrating spheres. An R&D test station uses a 20-in. sphere, while stations that sort devices use 4-in. spheres.
Test stations also need fixtures, and Luminus engineers must design fixtures that cool the LEDs' 150 W of heat. "That's as much heat as in a professional-grade soldering iron," said Joffe. Furthermore, the LEDs must be tested under consistent current and temperature conditions. The test fixtures are water cooled, and the engineers have designed mechanical fixtures that provide consistent contact with heat sinks.
Engineering test stations must be flexible enough to test an LED regardless of its mechanical configuration. Engineers need to test devices mounted on boards with heat sinks or on devices in die form. The Luminus engineers designed an engineering test station that lets an experienced test operator run automated tests of 100 LEDs. A two-axis stage moves the LED under test into position under the integrating sphere. The test operator controls the stage to correctly position each LED under the sphere.
Test engineer Aaron Plaisted is one of three test engineers who build test stations. In the engineering lab, test stations need flexibility to accommodate new device packages. A custom PCB holds the LEDs and routes current to LEDs under test. A TDK-Lambda DC power supply provides current to an LED under test. The test engineers designed drive circuits controlled by pulses from a National Instruments multifunction data-acquisition card that route the current to an LED. "We test LEDs at current up to 36 A," said Plaisted. Precision 0.3-Ω, 10-W power resistors (Figure 3) in series with the driver circuits let engineers measure the current flowing into the LED under test with a data-acquisition card.
FIGURE 3. Power resistors in
series with LEDs produce a voltage used to measure LED current.
Because HB (high-brightness) LEDs are relatively new, their fabrication processes are not as mature as those of most semiconductors. Process engineers are still learning how to optimize production processes, and they rely on data from test engineers, reliability engineers, and failure-analysis engineers.
Finding where it fails
You probably expect incandescent bulbs to fail, but not LEDs. While LEDs can last for years, they still need reliability testing. LEDs can lose brightness over time, so even if an LED is still producing light, a user may perceive it as having failed if the light output diminishes. That's where reliability manager Scot Solimine and staff engineer Mike Brown come in.
Solimine spends much of his time in the reliability lab, which houses banks of LEDs under test (Figure 4). (Covers over the LED banks diffuse the bright light so it won't hurt anyone's eyes.) Solimine runs reliability tests on lots of about 100 devices assembled from at least three wafers. He typically tests about 20 devices for a particular condition such as accelerated aging, temperature, vibration, shock, or humidity.
FIGURE 4. Banks of LEDs in the reliability lab may run for 6000 hr or longer.
"Customers think LEDs should last for 25,000 hr and retain 30% of their brightness," he said. But 25,000 hr would make for a long test, so Solimine tests LEDs for 6000 hr, after which time an LED should retain at least 92% of its brightness. Figure 5 shows a typical plot of brightness over time.
FIGURE 5. An LED's brightness degrades slightly over time. A device is deemed acceptable if it retains 92% of its brightness after 6000 hr of use. Courtesy of Lumnius Devices.
The Department of Energy and the Environmental Protection Agency have developed a new test standard called LM-80, which specifies the methodology to be used in determining LED lifetime and reliability. Lighting manufacturers must show evidence of conformance to the LM-80 standard in order to receive Energy Star approval (Ref. 2).
Solimine assesses reliability using accelerated life testing. That lets him achieve 10 years' worth of life-testing data in a few months. He may have to sign off on a design after just 2000 hr of testing, but the customer may expect 60,000 to 70,000 hr of life from a device.
Accelerated testing on HB LEDs involves driving them harder than most customers do. For example, a customer may drive a 30-A LED with a 25% or 50% duty cycle at 360 Hz, so Solimine will test parts with continuous DC current. He also raises the ambient temperature to see how it affects brightness and LED life.
Combinations of junction temperature and current let him develop reliability models. In some tests, Solimine will raise device junction temperatures to 220°C just to see how an LED responds. Most tests use junction temperatures between 150°C and 180°C.
"If you run at temperatures that are too high, you could uncover failure modes that customers will never see," said Solimine. "We usually try to test at junction temperatures about 20% over published specifications." The same applies to current, which is why he tests LEDs rated for 30 A at 36 A. He may test smaller LEDs (3 A to 9 A) at twice their rated forward current.
Although Luminus engineers test their LEDs under conditions that exceed published specifications, some customers exceed even those conditions. One customer pumped 40 A through the LEDs. They failed.
When Solimine tests LEDs under continuous current, he can use a single TDK-Lambda power supply to test as many as 40 devices connected in series because it has enough voltage to keep the LEDs forward biased (each LED drops about 4 V). In total, the lab has about 100 kW of electrical power for driving devices.
Solimine runs reliability tests under conditions that stress an LED more than most customers. But customers keep finding new LED applications. Solimine has added tests as new applications arise. For example, he runs one test using 30 A pulsed current for 0.2 s on and 1.2 s off, which is more like a thermal cycle compared to running them at 360 Hz, because at that frequency, the LED is on for much shorter periods. Other tests run at low current with high ambient temperatures, while others run at high current.
Not all of the current pumped through an LED produces light. An LED's circuit model is a resistor in parallel with a diode. Thus, a shunt current passes through the resistance, not through the diode. When running at low current, say 2 A, a shunt current of 500 mA can make a difference in reliability because 25% of the current doesn't illuminate the LED. Another customer, however, may run the same device at 30 A. There, a loss of 500 mA is less significant. An LED's shunt resistance increases over time, which fixes the leakage problem.
Failure modes for large devices are different from those for small devices. For example, a defective area that causes failure of a 1-mm2 LED may have little impact on the performance or reliability of a large Luminus LED, but larger defect areas increase the likelihood of failures. Figure 6 shows the difference between an acceptable device and a failed one.
| || |
|FIGURE 6. Uniformity tests reveal the light intensity across a device's surface. These images show (left) acceptable uniformity and (right) a failed device.|
Learning from failures
When failures occur, Solimine relies on staff engineer Mike Brown to analyze them. Brown analyzes failure modes both from internal tests and from field returns. He looks for repeat failure modes and makes suggestions on corrective actions.
Brown, who has worked at Luminus since 2003, explained, "When we started, we didn't understand the failure modes of the devices. Our wafers may have 70 different layers of doping in the layers, any one of which can go out of control. We've discovered and fixed numerous failure modes over the years and spent much time analyzing failures to learn about them."
As Brown learned more about failure modes, he could recommend better ways to test production parts. At one time, Luminus would burn-in 100% of its production products until engineers learned to better control the manufacturing process. That improved wafer yield, but still left failure modes caused by packaging. Thus, Brown and other Luminus engineers had to learn how packaging caused failures and how to fix them.
"As we've increased production, we've learned how to minimize failures in less time. We used to burn-in for 4 hr. Today, we can get reliable products with a 5-min stress test." Brown has uncovered packaging-related failure modes such as those caused by epoxy.
Baliga, the VP of engineering, emphasized that with each new product, Luminus engineers encounter fewer failures because they've learned how to minimize them. Still, new failure modes occur as customers try new applications. Brown noted that Luminus engineers learn a great deal from customers and how to fix new failure modes. "Each new application is an opportunity to learn about failure modes," he said. Some customers use the LEDs with low current for best efficiency while others want the highest power.
What is a failure? Customers have different expectations. To some customers, a loss in brightness constitutes a failure. The amount of acceptable brightness loss also differs among customers. One user has an end-of-life specification of a 50% drop in brightness. Another might consider an LED to have failed if it loses 10% of its brightness in the first year, which is why Luminus engineers specify an 8% dropoff in one year to be a failure mode. To others, only a complete loss of light is a failure.
Over time, Brown has developed a library of failure modes by intentionally overstressing the company's LEDs. Because of that experience, he can often tell the failure mode just by looking at the device, such as when that customer pumped 40 A through an LED. He also found one customer who was pumping too much current through a device without realizing it. The customer eliminated the problem by changing driver circuits.
Luminus engineers also try to emulate customer applications. The video of the pulsing light in the online version of this article shows one example. The pulsing pattern is unique to a particular customer, so engineers ran that pattern on an LED 24 hr a day to find where it might fail. Given the brightness of the LED, its light illuminates the parking lot at night during such tests. Fortunately, there are no homes across the street, so it doesn't bother anyone.
1. "LD Kevin Adams Uses VLX Wash for ‘American Idiot,'" Projection, Lights and Staging News, July 2010. www.plsn.com.
2. IESNA LM-80-08, "IES Approved Method: Measuring Lumen Maintenance of LED Light Sources," Illuminating Engineering Society, New York, NY. www.ies.org.