News/Tradeshows > Archives

Precision Positioning
Vertical integration powers precision positioning.

Aerotech's linear motor plano air bearing system shown with packaged servo amplifiers, high resolution encoder multipliers and UNIDEX 500 PC-bus based motion controller.

An Interview with Michael Formica, Aerotech Inc.

Lasers & Optronics: Within the broad optoelectronics industry, people often point to seminal events as milestones in the development of the technology. For example, the discovery of a particular type of laser may have opened new application areas. Are there such identifiable milestones within the motion control world in the recent past?

Mike Formica: Our industry serves a broad range of markets. We sell to the semiconductor industry, the laser  machining industry, the medical industry, the automotive industry, the packaging industry, imaging industry, data storage industry and so on. The figure above shows a linear motor plano air bearing system with packaged servo amplifiers, high resolution encoder multipliers and a PC-bus based motion controller. So, it is hard to say that any one milestone affected us directly. One industry may have advanced in one way in one year, and another industry may have advanced in another way the next year.

Having said that, over the past decade or so, most advances in precision positioning have been driven by the requirements of the electronics industry, principally by the semiconductor fabrication industry and to a lesser extent by the automated assembly requirements of electronic packagers — board stuffing, if you will. Many of the other industries we serve, in particular the medical industry, have benefited from these advances. But certainly, the semiconductor and computer industries are pushing this technology today, particularly on the high end.

The Semiconductor Industry Association (not to be confused with the Semiconductor Equipment Manufacturers Institute) produces what they call their "Semiconductor Industry Roadmap" every couple of years. The first one was published in 1992, the second in 1994. The most recent one was published in 1997. The 1997 roadmap is a 250-page document that identifies the changes in technology that need to occur for the semiconductor industry to get to the next generation of wafers. The roadmap tends to be a future-oriented document and as such is very good at describing the next generation wafer size and density plans of the semiconductor industry.

In the 1994 report, the benchmarks for a wafer generally revolved around feature sizes of about 0.35 m with a forecast of 0.25 m for 1998 and 0.18 m by 2001. The latest report, which came out in 1997, adjusted those forecasts to better reflect the state of the industry. The 0.18 m feature size was re-targeted for 1999, while 2001 was expected to see a feature size of 0.15 m. On a somewhat longer time scale, obviously with less predictive accuracy, they suggest that the semiconductor industry will reach feature sizes of 0.05 m by 2012. The semiconductor industry is extremely competitive and many companies have announced their own plans that exceed the targets of the SIA.

We, like various other suppliers to the semiconductor industry, view these roadmaps as telling us where the semiconductor industry believes it will have to go to reach its next generation of products. We then ask ourselves, 'What will we need to do to be able to support that level of performance?'

L&O: It's always nice for one of your principal customer groups to do your market research for you, isn't it?

M.F.: It certainly helps. However, because the semiconductor industry changes very rapidly and requirements vary depending on the specific customer, we still invest a tremendous amount of time researching the industry trends. The roadmaps certainly do tell us where we need to concentrate for the next couple of years. Historically, these roadmaps have tended to undershoot the target, leading to the frequent revisions. But that merely reflects a highly competitive environment.

L&O: In recent years, there has been a trending away from ball screws toward linear motors and a similar trending away from mechanical bearings to air bearings. How much have those trends been driven by the semiconductor industry?

M.F.: Very much so, although not in any sort of benchmarked manner. We cannot say, for example, that as soon as feature sizes went from 0.35 m to 0.25 m, everything changed. Historically, it has been much more a case of gradual, incremental change.

If you are dealing with 1 m or even 0.5 m moves, a ball screw is usually adequate. It gives good performance and is repeatable enough. But when you get down to tenths of microns, you start to feel the impact of friction, wear, vibration and backlash in the ball screw, not to mention speed and acceleration considerations. As feature sizes shrink, the impact of those factors becomes much more important.

Nor is this limited to the semiconductor industry. Five or ten years ago, electronic circuitry was almost all through-hole componentry. I could pick up a resistor and place it very easily. Now it is almost all surface-mount technology, with components placed by specialized machines to levels of accuracy unobtainable a decade ago.

For high-precision movements, the linear motor offers a great advantage over a ball screw. The linear motor is a totally friction-free device, so there are no wear or friction issues. Among other things, that allows faster motion, and thus more throughput, whereas a ball screw will heat up and change the stage's performance characteristics as you push throughput. As a ball screw moves, the friction generates heat that causes the screw to expand. When you are working in tenths of microns, this is not acceptable.

Conversely, the linear motor is not a panacea for every application, either. There are a lot of applications in which ball screws make perfect sense. It's not as if one technology were totally obsoleting the other.

L&O: Is there a similar analogy with mechanical and air bearings?

M.F.: In a broad sense it is, again, friction versus non-friction. For what you might call low-precision movements, mechanical bearings work very well. But as you shift to high-precision movements, mechanical bearings suffer many of the same shortcomings of ball screws, for identical reasons. For high-precision movements, air bearings offer advantages similar to those offered by linear motors.

Beyond the friction issues, an air bearing has inherently greater performance in terms of pitch, roll, yaw, straightness, and flatness. A typical air bearing will have straightness and flatness specs of well under a micron over large travels (300 mm), and much better numbers over smaller distances. This type of performance is extremely difficult with a mechanical bearing and not nearly as repeatable from system to system.

One of the great advantages of both air bearings and linear motors is that they maintain their performance over time. Since there are no contacting elements, there is no wear; the system performs the same regardless of whether the system is 10 days or 10 years old. The combination of greater performance and near-zero maintenance makes a linear motor air bearing solution quite attractive when compared to a ball screw and mechanical bearing.

Figures 1 and 2 show the performance of two nearly identical stages. One has a ball screw and the other a linear motor, but they are otherwise identical — same feedback device, same bearings, same cross section and so on. The two screen shots show the performance difference between a ball screw and a linear motor. The difference primarily comes from the effects of friction on the recirculating elements and the friction on the ball screw itself.



Figure 1. Velocity regulation with typical ball screw driven stage.

Figure 2. Velocity regulation with linear motor drive.

Now imagine taking the same stage and replacing the mechanical bearings with air bearings. If we then plotted a third graph, it would show a similar amount of improvement over Figure 2.

L&O: We've covered trends in accuracy and resolution pretty well. What other things are customers asking for?

M.F.: Those are the first two things most potential customers mention, and they are certainly key parameters. But then the typical customer says, 'Oh, by the way, it has to be twice as fast as the one I bought from you last time. And, by the way, it needs to last twice as long. Oh, by the way, it also has to be smaller than the last one. And it has to be cheaper, too.' Put that list together, and the challenges mount up quickly.

Don't forget that semiconductor fab lines today cost around a billion dollars. Minimizing downtime is crucial. Downtime isn't measured in terms of ten or twenty dollars per hour; it's more like tens or hundreds of thousands of dollars per hour. They want to set things up, get them running, and keep them running. That's another advantage of linear motors and air bearings — they are virtually zero-maintenance devices, as long as you don't do something foolish with them.

L&O: How do we get from where we are today to points further up the performance curve in the future?

M.F.: It will be a combination of several things. For example, linear motor technology has existed since the turn of the century, and the modern linear motor has been in use for more than 25 years. But it has only been in the recent past that they became widely used in motion control applications, because they provided a capability that other technologies did not.

Consider all the advances in electronics over the years. Many of them came about in part through the automation of the manufacturing process. That made electronic devices better products than their immediate predecessors. In turn, that has allowed us to take advantage of better controllers, allowing us to make better motion control systems. There is a nice feedback mechanism in play here. Controllers and computers that were not available to us ten years ago, at any price, are available to us now at reasonable prices. That allows us to make better positioning products.

If you want to move a 300-mm wafer from point A to point B, there are only so many methods available. The interplay of all the components in the system is where the advances will come from. It's basically a case of picking up a ten percent improvement here, and another ten percent over there, and so on. For example, the technology of linear motors hasn't changed very much, but the way you match motors to amplifiers has. The way the motion controller commands the amplifiers and the way it takes feedback information from the sensors are both key areas where significant advancements have recently occurred.

Five or six years ago, for example, controllers were able to handle data rates of 250 kHz or so, maybe up to 1 MHz. Now we can handle 32 MHz input rates. This enables faster moves and higher throughput rates, implying potential resolution of 8 nm at velocities of 250 mm/s. This was unheard of two years ago, or even last year. There won't be any sort of paradigm shift; rather, we will be taking advantage of existing technologies to make all the constituent parts a little bit better — better throughput on the controllers, better mapping algorithms, and so on. Microprocessors now have more memory and are faster, so we can data-map more points.

L&O: What exactly do you mean by mapping in this context?

M.F.: Error mapping involves taking a stage and running it versus a laser interferometer to get an exact measure of its performance. Then you can use that data to compensate for the inherent errors electronically. When you did that five years ago, using a 4 to 8 MHz chip, you could only do so much. Now with 80 or 100 MHz DSP chips, you can log a lot more data points and mathematically manipulate them more intensely. Today we can perform two-dimensional or even three-dimensional mapping. This capability didn't exist three or four years ago, at any price.

L&O: Are these all application-specific integrated circuits you are describing, as opposed to the garden-variety Pentium of today?

M.F.: They are not based on Pentiums, but neither are they ASICs. They are DSPs and RISC processors for the most part. The Pentium in your desktop PC is, in effect, running a front end for you, with these DSPs and RISC processors doing the heavy calculations in the background. The motion controller does what motion controllers do (including mapping), but in certain cases you are also able to make some corrections on the front end.

At least, that's the general view. A lot of the machine vision companies do heavy number crunching on the front end with their desktop PC. For them it is critical that the controller takes care of the motion portion, so their processors can focus on their process. The dividing line between mapping on the controller versus the PC varies by industry.

Not only are we using a lot of DSPs and RISC processors in our controllers, but as programmable gate arrays have become more powerful, we have begun using them heavily, too.

L&O: Motion control companies speak of 'matching' amplifiers to controllers, for example, and also of 'optimizing' them. Those terms seem to be synonymous. Are they?

M.F.: The terms are quite similar, both pointing to a different trend in the motion control business. In the early 1990s, after the recession of 1991, a lot of the big companies went through heavy downsizing. Prior to that, most of these companies had large in-house engineering staffs. These staffs generally did what might be called systems integration. In terms of the motion control business, such a company might buy a stage from us, a controller from one of our competitors, and amplifiers and feedback circuitry from another competitor, and put the system together themselves. Often, they would go back-and-forth with the vendors two or three times, requiring changes to get the most out of the full system. In the end, the system would usually work, but it could require an enormous effort from all parties.

Those days of in-house engineering of production systems are mostly gone now. As many of these companies have downsized, their engineers have become more focused on their core business, and no longer have the resources nor the expertise in-house to do those things. Many companies simply do not do systems engineering any more. They out-source it, mostly back to us. This has forced motion control vendors to become much more systems oriented than we had been six or eight years ago.

With that said, 'matching,' in the customer's mind, pertains to getting a complete system from a single source that will do what it is supposed to do, right out of the box. 'Optimizing' pertains to getting the maximum performance out of a system.

Consider amplifiers, both linear and pulse-width modulation. Each has advantages in certain cases. Depending on the actual motor you are using and its characteristics, you may want to use one type of amplifier or the other. Or you may want to 'tune' the characteristics of the amplifier to the performance characteristics of the motor. Modern amplifiers boast the capacity to make many changes in the software and even some changes in the hardware. Ultimately, it is the difference between having something that does 110 percent and something that 'just does the job.

L&O: But it can't be all bad to be more of a systems house than you were some years ago. You get to keep a larger core of corporate knowledge and, as a vertically integrated systems vendor, you should enjoy higher margins than a strictly components house, shouldn't you?

M.F.: That is true, but it is expensive. It is a challenge to develop the proper mix. But after you sink a few million dollars into R&ampD in controllers and amplifiers and the other aspects of motion control, you ultimately get to a better solution.

Vertical integration is the pathway to improvement. We have a saying here, 'You are only as good as your weakest link.' By that we mean that you might have the best air-bearing system in the world, with the best motor in it. But if your amplifier underperforms, or your cabling scheme has not been implemented intelligently, your system will not perform well. It's as simple as that.

This approach has significant benefits for our customers, as they can rely on us for the continued development of more technically advanced motion control products. This does result in lower development costs, and shorter product development cycle times.

L&O: Do you have any case histories to relate, regarding the advantages of vertical integration?

M.F.: We had a situation not long ago where a machine-vision company was doing very high-speed machine-vision inspection. A year or so ago they developed a concept for a new inspection system and took the concept to a trade show, where it was well accepted. The image-capture portion of the system worked quite well.

Their company decided to buy bearings from one vendor, motors from another, and so on, thinking, "We'll do this ourselves. How tough can it be?"

Well, many months passed and they now have the answer to that question. It was a lot tougher than they thought. Ultimately they came to us and asked us to put a complete motion control system together for them, which we were able to put together and get running in a day or two. In hindsight, they wasted a year of development time when they should have been focusing on their core business.

Let's say someone buys parts from half a dozen vendors and does his own systems integration. When the system doesn't work to his satisfaction, whom does he blame? The motor company? The bearing company? The controller company? In many such situations the truth is that none of the vendor companies are to blame. Each component does exactly what it was designed to do. It's the interaction between the parts that proves to be critical.

As an extreme example, we once sold multiple components to a customer, who then did his own system integration. He ended up with a system that was almost entirely ours, with just the controller from another vendor. When the system didn't perform the way the customer wanted, we got the call. We would have loved to blame the other guy's controller, but it performed exactly to specification. When we saw what the customer was trying to do, it became apparent that he had mismatched our equipment, which was easy for us to fix.

That's the power of vertical integration in motion control applications. I'm sure you can find similar stories in a lot of other fields, too.

Over the past decade or so, most advances in precision positioning have been driven by the requirements of the electronics industry.

Vertical integration is the pathway to improvement.