FOCUSED LOOK: A SEMI task force tested more than 35 tools in the course of writing voltage sag standards.

New standards give chipmakers the power to maintain their yields

The power supply to your fab is like the air you breathe, says Gary Malhoit. You don't think about it until your supply is interrupted, you realize you're suffocating, and you decide, hey, you'd "better get some air." Malhoit, a Motorola engineer at the chipmaker's MOS 5 fab in Arizona, uses the analogy to draw attention to an important new set of SEMI standards on equipment voltage.

Fab management often doesn't pay much attention to its power supply until glitches occur, explains Malhoit. And, as any process or facilities engineer will tell you, such sags in voltage occur regularly. According to experts, each year the average fab experiences one dozen breaks in the plant's steady supply of electricity. They become a particularly acute pain when a batch of processed 8-in. wafers suffers, says Malhoit, who was on the task force that wrote SEMI F47 on voltage sag immunity. The document became official within the past six months.

The interruptions primarily have two effects on yields, the engineer asserts. "Initially, you have an immediate problem with all types of equipment and then other types of equipment may actually ride through the power blip." Unfortunately, back-up systems such as the cooling-water supply may go down, increasing the temperature to process tools and creating a damaging ripple effect. Reasons for the power dips vary. Causes include a squirrel in the switching gear, mishaps with grounding straps at the fab, trees touching power lines, sagging lines caused by summer heat, and August air-conditioning loads.

Michele Negley, program chair of the task force that coordinated the writing of the four documents, thinks the standards will have broad benefits for the semiconductor industry. "We believe they're going to save the industry a lot of money and help it gain a lot more control over the process. Imagine if it's one of those uncontrolled interruptions. A voltage snag could hit you at any time. It not only causes yield problems and a lot of downtime and materials scrap, but it also hurts your whole business cycle."

The goal of the standards drive was to make tools immune to voltage sags without the use of a pricey uninterruptible power supply, or UPS. Recently approved standard F47 aims to improve tool compatibility with the electrical environment. An accompanying test method, F42, defines how to characterize the susceptibility of equipment to voltage sags ranging from 3 to 60 cycles long. Additional guidelines for electric utilities suggest methods for measuring power quality performance, establish a root-cause analysis program, and offer service options. Voters gave the nod to this benchmark in February.

The guidelines also address "how to deal with other things in facilities," says Negley, director of energy solutions for New West Energy, an electric service provider for clients in the western United States. "For instance, if you lose compressed air to a tool then, even though the electrical system rides through [a dip]...if the equipment driving the compressed air goes down because of a voltage sag, you still have a production loss."

Writing the standards required a collaborative effort that drew on the expertise of several utility companies, the Electric Power Research Institute, major semiconductor manufacturers, and tool suppliers. The Power Standards Testing Lab, a California-based company that also helped with the standards, is offering four voltage sag tutorials on its Web site at http:// www.powerstandards.com.

To describe a typical voltage sag, Negley likens an electric grid to a net. "Visualize a big net over your head in a room. That net represents the electric grid. Walk toward a corner of the room, take a couple of fingers and bring the net down to the floor, or as far down as you can bring it. At that point where your fingers are pulling it down, that's where a short circuit occurs.

 

"Usually, not too far, say within a few inches on either side of that point is a circuit breaker. Within about a tenth of a second the breaker senses a break and it opens. It's almost like a pair of scissors. The net snaps, and you're holding a couple pieces of string. That's an outage."

Ten feet away from the break there may not be much impact, she continues. "But when you have a short circuit anywhere in an electric system you have an outage. At that point, for that tenth or twelfth or fifth of a second, you have a voltage snag...until the circuit breaker works."

The entire grid may experience 15 outages. "The short circuit that causes those 15 outages causes a voltage sag across the whole system to varying degrees. The closer you are to the outage, the worse it is. Because of that, fabs experience about a dozen process-threatening voltage sags throughout the year."

When voltage dips below 80% of normal, "then something in the process is probably going to burp," Negley points out. "If it gets down to 60% of normal then there's probably going to be a processwide upset that's going to take a while to dig out of. We've got a standard that says equipment should ride through a voltage sag down to 50% of normal for the first 0.2 second. Then it steps up higher and higher the longer the voltage sags, and the less ride-through you need to have in the equipment."

Negley believes chipmakers using the standards should be able to avoid nearly all power glitches. "We determined that if the equipment rode through those types of voltage sags unaffected, then that eliminates 92% of all voltage sags that happen and it would not cause a process interruption." The interruption in voltage "can take two forms," Malhoit says. "It can go to zero, or it can go to intermittent levels and stay down there for five or six cycles. That's all it really takes to play havoc with pretty much all the tools in the fab."

Quantifying the costs to wafer processing is difficult, notes the Motorola engineer. "It depends on what you're making and where you're at in the process. If you've got a lot of wafers in diffusion where it's a batch process [your yields can take a hit.] If it's a single wafer in a stepper it's not as big a deal. Usually the lamps are lost, and those are $1500 a whack." Etch tools and CVD systems where six to seven wafers are processed at a time are other examples of susceptible tools.

Some chipmakers—Intel among them­have installed expensive equipment such as a dynamic voltage restorer (DVR) that eliminates the sags. The DVR uses the same principle as variable speed drives to create "long chains of cascading devices to get the voltages they need." The solution compensates for power dips. The downside of this approach is that the devices "cost too much unless your profit margins justify their use," Malhoit says.

The solution costs between $2 million and $10 million. And price is not the only consideration. "The bad thing about the DVR is you end up protecting the coffee pot and everything else," he adds. For a new facility it makes a lot more sense because you can separate out the critical loads when you design the plant. Otherwise, [with older plants] you're protecting the Coke machine."

"We actually spent a heck of a lot of time and money researching this," Negley says. "We wanted to test equipment and see at what level it could perform reasonably" through a voltage sag. The task force monitored 15 sites over two years and examined "what type of voltage sags happened out there on the power grid. Then we did a best fit [to determine] what you could design equipment to reasonably ride through...if you picked the right relays and so on."

The task force tested more than 35 tools for 4- to 12-in. processes. "We spent several hundred thousand dollars worth of testing on semiconductor equipment," Negley recalls. "When we did this testing we found it's not the rocket science part of the equipment that goes down; it's the basic building blocks like relays and undersized power supplies that are causing the tools to go down. Once we realized that we determined how best to find a reasonable standard."

Although the bulk of the work was done in the United States, the companies involved are global companies and most likely will implement the standards throughout their operations worldwide, Negley says. "My feeling is that this is a global standard and it will grow into those other areas. We didn't talk about numbers of cycles because, as you know, there are 60 cycles in the United States and 50 cycles abroad. Instead, we talked about fractions of seconds to keep all measurements international."

In addition to Malhoit, the task force had engineers from a number of chipmakers. Intel, AMD, Texas Instruments, and IBM have been adopting the standards, according to Negley. Eventually, even Motorola, which has shown less interest than its counterparts, will choose to use F47, she asserts, adding that the chipmaker "is not typical" and has decided to handle the issue differently. Malhoit thinks semiconductor manufacturers will adopt the guidelines more readily as 300-mm wafer production continues to climb. "The larger the wafer, the more sense it makes [to use F47]."

Fab managers also meet with utilities in order to iron out any problems. For example, Malhoit says that at a Motorola site where he worked in the past the management met monthly "to review incidents we had." The chipmaker developed a policy that "if the utility was going to be doing any switching in a substation within a certain number of points they were required to notify Motorola to let the managers know that." The message was "rebroadcast to specific managers responsible for wafer production."

Although this approach didn't necessarily prevent all process interruptions, the warnings of known changes in power supply allowed process engineers "to take all precautions at work from Time A to Time B when the window of vulnerability, so to speak, was greatest," Malhoit recalls. "Then managers knew when the mistake would happen."

Manufacturers often will take steps to institute safety measures in small ways "but when it comes to something major like power we don't have a lot of safeguards in place." The issue doesn't get the attention it deserves because of the "recency effect," says the Motorola engineer. Calling it "just common sense with a sophisticated label on it," Malhoit says the term is management-speak for "whatever most recently gets anybody's attention." When a glitch occurs there's a scramble to deal with it, but as time passes it draws less and less interest, he explains.

It's a pennywise-and-pound-foolish attitude that will change, believes Malhoit, who agrees that F47 and the accompanying guidelines will enjoy wide acceptance. Indeed, the response so far has been heartening for the task force chair. "The documents are being implemented," Negley says. "That's what brings joy to my heart after we put so much work into it."