Whiskers - tiny metal filaments that grow in data centers - can knock out servers. Are they a fluke or a crisis in waiting?
Still, data centers with old or inexpensive materials or equipment run the risk of whiskers forming, breaking off, getting into computers and short-circuiting them. Metal whiskers may cause unusual, sporadic problems, or they may cause a data center's power supplies to short out en masse. Since Network World first covered metal whiskers in 2004, new research and environmental legislation have changed how people approach the issue.
Explaining metal whiskers
The source of metal whiskers is steel that has been electroplated with zinc or tin to prevent rusting, according to Robert Sullivan, senior consultant at the Uptime Institute in Santa Fe, N.M. When manufacturers deposit zinc or tin through electroplating, he says, the process can introduce stresses that cause whiskering. Zinc- and tin-plated metal has been used in computers, server racks, floor tiles and the like.
Sullivan says that hot-dipped galvanizing — a process in which metal is dipped into molten zinc or tin — does not produce whiskers. He adds that if metal is electroplated, a powder coating process can be used to prevent whiskers from forming.
Electroplating involves immersing one metal, such as steel, in water, and running a current through the water to get another metal, such as tin or zinc, to bind to it. Hydrogen atoms from the water also can be deposited on the metal, and that can lead to what Sullivan calls “compressive stress.” To relieve some stress, molecules of the metal’s outer layer get pushed out in a process that Sullivan says is “like trying to empty the ocean with a kid’s sand pail.” Stress remains, however, so this process happens over and over, causing a tiny metal whisker to form. Sullivan says that zinc, tin, and cadmium all can produce whiskers through this process.
According to Sullivan, most manufacturers of metal tiles and the like use metals and processes that don’t cause whiskering, but old or inexpensive materials may pose a risk. He says that in one data center that was less than five years old, he found metal whiskers on an economy-grade metal bar that was used to support ceiling tiles. Only the bottom of the bar was powder-coated, and zinc whiskers formed on the other sides. Sullivan says metal whiskers generally take about two years to form, although he has seen them crop up in as little as six months.
Mixing lead into tin or zinc prevents metal whiskers from forming, and new, lead-free solder could introduce a new source of whiskers. “I think we’re just starting to see the tip of the iceberg on that,” says Rich Hill, president of data-center cleaning company Data Clean. Sullivan disagrees: “I don’t see that soldering is an exposure to the creation of either zinc or tin whiskers,” he says. He adds that lead restrictions should not cause whisker problems for other parts, either. “If you’re plating with a lead-zinc alloy . . . it won’t grow zinc whiskers,” he says, “but there are very few people that are doing that,” so restrictions on the use of lead should not pose a problem.
Who has metal whiskers?
Documentation of metal-whisker problems in data centers is hard to come by.
“Whiskering is something that people keep close to their vest,” Hill says. “You don’t want your clients to know you have [metal] whiskers.” He adds that this especially is true in the case of collocation facilities and other organizations whose reputations are built on data-center reliability. Hill said that although he didn’t know of any clients that would be willing to talk about metal whiskers, he has seen the problems the tiny filaments can cause. “We’ve heard things go ‘pop’; we’ve had clients that lost a hundred power supplies in a weekend” due to metal whiskers, he said.
NASA is no stranger to metal-whisker problems: the organization runs a Web site that covers metal-whisker research, and metal whiskers have caused failures in NASA equipment, including a flight control system. According to a 2006 NASA presentation, metal whiskers have caused equipment failures in satellites, telecommunications equipment, missile programs and nuclear power plants. A presentation by the U.S. Department of Energy Office of Environment, Safety and Health Evaluations says metal whiskers caused a nuclear power reactor to shut down in April 2005.
Layne Maly, director of communications for data-center user group AFCOM, found less evidence of problems with metal whiskers. He said in an e-mail that she had asked the majority of the group’s members whether they had encountered such problems: “The responses I’ve received back all say the same thing — ‘I have not had a problem with metal whiskers, and I don’t know anyone who has.’”
Other reasons for whiskers
Stress from manufacturing might not be the only factor in the growth of metal whiskers. Research on the effects of humidity, electrical current and other factors is contradictory and inconclusive, according to NASA’s metal-whisker Web site, and so the organization argues these factors should not be ruled out. In addition, NASA says that stress from sources other than manufacturing — for example, scratching or bending metal — also could cause whiskers to grow. For his part, Data Clean’s Hill says there are “no conclusions out there as to what causes whiskering.” The Uptime Institute’s Sullivan acknowledges that higher temperatures can speed the process, but he says that stress is the root cause of the phenomenon and that humidity and other factors have no effect on whiskers. He maintains that the metal-plating process is the primary source of whisker-inducing stress: “I have never seen a hot-dipped galvanized surface grow whiskers,” he says.
Not everyone agrees that galvanized metal is immune to metal whiskers, however. A 2002 white paper by the University of Maryland’s Center for Advanced Life Cycle Engineering (CALCE) says that “existing approaches are not sufficient to control tin whiskering in high-reliability systems.” A UMD presentation noted that at the end of 2004, only 33% of component suppliers surveyed were testing their lead-free parts for tin whiskers.
“Right now it’s a minimal problem, but it still exists,” Sullivan says about metal whiskering. He notes that the problem has “gone down exponentially” since it was discovered 12 or 13 years ago. Still, he says, it is remarkable that the issue still exists at all. “[Some] people [are] not paying attention, not doing due diligence in their data centers, and all of a sudden they’re finding [metal whiskers],” he says. According to Data Clean’s Hill, on the other hand, the metal whisker problem is “still going about the same level that it was a few years ago.”
By most accounts, metal whiskers pose a credible, if not critical, risk to data centers. Although information about them can be sparse and sometimes contradictory, that should not be a reason to ignore them. As one NASA presentation says, “The absence of evidence is not evidence of absence.”