Image of perfection: Bringing up chips at IBM

READ OTHER MAY ARTICLES:  Contents, May 2006

EAST FISHKILL, NY—Manufacturing high-performance microprocessors in deep-submicron technologies on 300-mm wafers requires the utmost in process optimization and design-rule formulation and adherence. To ensure their company's products measure up to expectations, engineers at the diagnostic lab of IBM's Micro Test and Development (MTD) organization employ a host of imaging tools that can peer inside semiconductor wafers or packaged parts. Pat McGinnis, electrical engineer at IBM Technology Collaborative Solutions, works with six colleagues in the diagnostics lab here to employ ATE systems to exercise devices under test and to devise the fixtures necessary to interface to the DUTs within the various image systems they employ.

"Our organization's primary responsibility is power-on and test bring-up for IBM processors for the IBM Systems & Technology Group," explained McGinnis. "The mission of our lab—the image-diagnostics lab—is to support not only the power-on of all internal [IBM] 300-mm products, but also all 300-mm process development and external customer diagnostics. To that end, we have performed full SOC test and image-based diagnostics on the latest dual-core IBM POWER6 processor, on the STI [Sony, Toshiba, IBM] Cell processor, on the XBOX 360 game chip, and on the Nintendo PowerPC game processors."

An IBM-developed processors' first step within the MTD organization is through the primary MTD test lab, headed by Franco Motika. There, ATE systems including Advantest 6671/6672/6682 platforms and Teradyne Ultra Flexes put new products through their paces, looking for product and process problems through a range of structural and functional tests. If electrical tests suggest that a closer look is warranted, the DUTs could head across the hallway to the domain of McGinnis and the image-based diagnostics team.

The lab, McGinnis said, was originally dedicated to isolating process-development failures. But with the advent of 300-mm technology, separate process and product diagnostic efforts became counterproductive. Combining the two simplifies technology transfer, he explained, and is increasingly important as process development and product design occur in parallel. McGinnis said the lab is capable of performing a range of tests, from inline tests of the kerf macros located between dies on a wafer to complete functional tests of a final semiconductor design.

McGinnis himself is in his tenth year at IBM. "I actually started in Boca Raton, FL, with a group called SEDAB, an acronym that stood for 'special equipment design and build,' where we were working with hard-disk-drive, microelectronics, and other IBM manufacturing organizations. On my last project there, I was in charge of building atomic force microscopes. That's where I cut my teeth in the microelectronics world, and that paved the way for me to come here six years ago."

Pat McGinnis poses with a Suss Microtec NC-1 noncontact probe system. McGinnis and his team collaborate with Suss engineers to develop sample-preparation techniques that provide noncontact probe access through SOI substrates.

Because of the range of technologies that his lab addresses, McGinnis said, "The demands on the engineers in this lab are pretty high, because you have to know what everybody else is doing. I always like to think that we know a little bit of everything but are masters of none, so we rely heavily on our counterparts to help us create test programs and understand the fab process. I deal with every community that's imaginable within IBM—the front-end-of-the-line integrators, the back-end-of-the-line process-development engineers, the designers, the test guys, you name it."

He also deals with IBM's customers: "IBM always takes care of its customers. If one of our partners is having problems isolating fails based on our process, even though it's been transferred to the partner's fab, we will still look at the partner's hardware with our custom toolset."

In addition to addressing both process and product diagnostics, McGinnis's lab occupies a unique niche between the electrical-test and physical-failure-analysis functions. "Electrical test—whether it's inline test, final wafer test, or final module test—can tell you a lot." As a result of problems detected though electrical test, he said, IBM's Fishkill-based physical failure analysis (PFA) lab will ultimately use tools such as scanning electron microscopes to identify actual physical defects. "But with devices getting smaller and smaller, it's important to be able to point the PFA team to a pretty tight area" where defects might appear.

Further, McGinnis pointed out, hard defects no longer predominate. He cited the increasing frequency of condition-sensitive failures that are difficult to detect with traditional approaches. "That's where we come in. We combine traditional PFA tools such as emission microscopes with the ability to exercise devices using an ATE system. We actively exercise parts, be they wafers or modules, and we develop the fixtures that let us observe the parts using our imaging tools as they are exercised. We custom design all of our fixturing to fit into these various image-based tools that we use.

"My lab has an array of image-based tools with a custom-designed fixture set that allows for full tester control of a device while image-based analysis is conducted. Our image-based tools include a Credence Emiscope III PICA [picosecond imaging circuit analysis] tool, a Hamamatsu Phemos 1000 emission microscope, and various semi-auto probe systems that can accommodate various probing hardware sets, including a Suss NC-1 noncontact prober."

The IBM/Sony/Toshiba jointly developed Cell microprocessor is one device that has made the trip through IBM’s image-based diagnostics lab. Essentially a supercomputer on a chip, the Cell features a multi-core design with high-speed communications capabilities that can deliver 10 times the performance of chips used in personal computers. Courtesy of International Business Machines Corporation. Unauthorized use not permitted.

McGinnis noted that the 82000 lacks the speed of newer testers, but he doesn't consider that a major drawback for the diagnostics work the lab does. "In a diagnostics mode, everything can pretty much be run at a lower speed, and you get the same behavior at the low speeds as at the high speeds. And within the last five years, everything that runs at high speeds has gone to some type of onboard clock, so line loss or the round-trip delay through any of your test interfaces isn't an issue. All you need to do is be able to synchronize your tester to your device."

He did note, however, that the 82000 lacks the vector memory necessary to handle test data for some modern ASICs and microprocessors. That's one of the reasons his lab acquired an Agilent 93000, which now serves as "the main tester platform utilized in conjunction with the image-based tools."

McGinnis explained that a key advantage of the 93000 was that when his lab acquired the system, it was one of the few SOC testers, if not the only one, that integrated the pin electronics into the test head, making it easy to wheel the necessary test resources to the appropriate imaging tool. A key feature was a sufficiently long tether that can carry power and cooling water to the test head as the engineers maneuver it throughout the lab.

Many labs "take the approach of marrying a test system directly with a single tool—a probe station, emission microscope, or waveform tool like PICA," said Rich Oldrey, whose duties include engineering hardware design and build, test, and image-based screening. "We didn't go down that path. We chose to set a whole bunch of tools up and put a tester on wheels."

Figure 1.  Rich Oldrey displays an Agilent 93000 load board, which interfaces 93000 pin electronics with daughter-card modules or probe rings that hold and provide electrical connections to devices under test within imaging tools.

McGinnis attributed the success of the fixture scheme (Figure 1) to Oldrey and Darrell Miles, who is responsible for equipment maintenance, DUT hardware design and build, and image-based diagnostics support in the lab. Oldrey for his part attributed the success to the 25 years of experience he and John Sylvestri have in identifying the problems and pitfalls that can plague fixture-design efforts.

Figure 2.  Darrell Miles developed a water-cooled lid assembly that forms a seal between a steel plate and the ceramic substrate of a device under test. The low z height of the assembly enables it to easily fit within various imaging systems.

With fixturing in place, the next step is to run the DUTs through the tester interface and perform image-based analysis. "Our workhorse in this lab for image-based analysis is the Hamamatsu Phemos 1000 emissions microscope, which also has an integrated laser scanning microscope," said McGinnis. "Emissions has become a very practical way of easily and quickly finding different types of fails. Of course, it is obviously very hard to look through up to 15 levels of metal, but with our fixturing, we can make contact through a cantilever assembly for full tester control, and then we can flip the whole assembly upside down so we can look at the part working through the backside silicon."

The approach requires a confocal microscope with sufficiently sensitive detectors as well as the appropriate fixturing. Typical emission microscopes, McGinnis said, have CCD detectors, but for the Phemos 1000, he specified an MCT (mercury cadmium telluride) detector, which lacks the resolution of a CCD but provides an order of magnitude improvement in quantum efficiency, not to mention broader spectral bandwidth. Overlaying the resultant emissions and laser scanning microscope (LSM) images, he explained, enables the isolation of various types of failures, or at least can rule out potential product or process problems.

Other tools in the diagnostics lab include an Emiscope III from Credence Systems, which builds on the PICA technology invented by IBM. Emiscope adds the ability to acquire timing information from the same devices that have provided emissions data. McGinnis says that the Emiscope can acquire a waveform in a matter of minutes, whereas five years ago, a PICA system would require hours to acquire the waveform. He cited as a unique feature of the Emiscope a water-cooling-spray assembly, developed by Credence, that keeps parts under test cool despite the loss of built-in heat-sinking that occurs when backside silicon is thinned to 150 to 200 nm to permit photon acquisition.

A tool like the Emiscope can show where edges are occurring, but it can't discern voltage levels. For that, McGinnis's team makes use of probing systems. McGinnis's atomic force microscopy experience led him to implement atomic force probing in the lab. In one approach, he enclosed a Cascade probing station within an acoustically isolated chamber. That approach, he said, enables him to make resistive contact with 100-nm targets without having to worry about acoustic noise in the lab.

Of course, resistive probing imposes loading on the circuit under test, and to get around that shortcoming, McGinnis has worked extensively with noncontact probing systems. He elaborated, "There are very small distinct electrical forces operating in the attractive van der Waals force range, and if you take advantage of maintaining device functionality through a synchronized tester interface, you can measure the dampening effects caused by these forces as they interact with your probe tip. You have to fly above the surface at a very stable height, and you have to be able to calibrate the mechanical dampening of the tip in order to detect the induced vibrations that are there. Then, the idea is that you digitize the signal and rely on synchronization and post processing so that what you end up with is a signal that not only gives you pretty good resolution for timing information but also a nonvolatile measurement of the peak-to-peak voltage, which is something that a light-based or an emission-based system can't do."

Initial efforts at noncontact probing yielded a system that could be made to work, he said, "but you practically needed four PhDs to run the thing. It wasn't a repeatable product. One of the big problems was synchronization through the tester interface."

That led to a collaboration between McGinnis and a team at Suss Microtec, lead by Dan Ouellette, noncontact probing manager at Suss. IBM provided test-device samples, and the Suss team provided the engineering to develop the NC-1, which Suss introduced as a commercially available tool last year (Refs. 1 and 2). Said McGinnis, "Dan and his cohorts have put a lot of time and effort into the system to make it repeatable and easy to use." The teams continue to collaborate on issues such as sample-preparation techniques that provide noncontact probe access through silicon-on-insulator (SOI) substrates.

Of course, the newest tools aren't always the best. In addition to the venerable HP 82000, McGinnis pointed out that the lab also has an old Zeiss LSM. "I call it the Bentley of the LSMs. It's old, it's ugly, but the darn thing works just as good as any of the other LSMs that are out there." And Miles, he said, has continued to enhance the tool over the last five years.

The continual enhancement of older tools helps the lab control costs. Said McGinnis, "You just can't keep throwing millions of dollars for another piece of equipment." Enhancement of older tools is part of that scheme; another is encouraging vendors to provide multiple functions within one piece of equipment. "Hamamatsu in particular has been a really good vendor for us to work with," given its willingness to integrate an emissions scope and LSM into one piece of equipment.

With the tools they have available, how does McGinnis's team know where to start? "Actually, in most instances, inline test data or wafer final test data gives us a good idea, because we know our own processes. We don't always get it right, but for the most part, I'd say 70% of the time, if I know I have gate leakage and I know I don't have a shorting mechanism, I just take the part over to the emissions microscope, and that will show me the devices that are leaking. If I know from a parametric sweep that I have a dead short, I can expect that OBIRCH [optical beam induced resistance change] is the best way to go. Tools like the noncontact probe system are used most often for design marginalities. The initial data tells us pretty much what tool to use."

Looking to the future of imaging diagnostics, McGinnis noted that spectrometry shows promise for SOI technology. Many researchers, he said, find they can quantify emissions wavelengths, and they can learn a lot about what's happening at the device level—for example, whether a transistor is saturated or near breakdown. "Unfortunately," he added, "I think microelectronics in general is just moving faster than the development of the detectors, so it's a constant game of catch-up." By the time such an imaging tool is perfected, he said, "It's probably going to be applicable to 90-nm technology after we've reached 45."

Nevertheless, McGinnis remains prepared to adapt. "One thing I've learned in my career and that I try to share with some of the newer guys is that you can't rest on your laurels. In this industry, what you did five years ago isn't applicable to what you are doing today. You have to reteach yourself, you have to retrain yourself constantly. And quite honestly, this is the greatest environment in the world to learn new technology. Our scientists at the Watson Research Lab, and here in Fishkill, and in Poughkeepsie, Burlington, Austin, and Rochester, MN, are always coming up with new ideas, and we are always coming up with new solutions to test what they are coming up with. So it's always interesting, and there's always something new."