Pressure Measurement and Control Keep Pace with Processing
Alexander E. Braun, Associate Editor -- Semiconductor International, 10/1/1998
C ompanies supplying pressure measurement and control (PM&C) equipment for the semiconductor industry's vacuum- and gas-related needs are attentive to their OEM customers' needs, which are, in turn, driven by those of end-users in the fab. PM&C suppliers must not only be able to provide OEMs the needed technology and materials but also anticipate future needs in order to perform the necessary R&D that will enable them to meet these needs. Companies supplying pressure measurement and control (PM&C) equipment for the semiconductor industry's vacuum- and gas-related needs are attentive to their OEM customers' needs, which are, in turn, driven by those of end-users in the fab. PM&C suppliers must not only be able to provide OEMs the needed technology and materials but also anticipate future needs in order to do the necessary R&D that will enable them to meet these needs.Pressures Decline
The migration to lower process pressures is one of the trends challenging PM&C equipment manufacturers. Conventional plasma CVD processes are becoming high-density plasma CVD processes, which means operating pressures are shifting from tens or hundreds of millitorr down to 2 or 1 millitorr levels. One Torr is equivalent to 0.019372 lbs/in. 2 absolute (psia). Normal atmospheric pressure at sea level is standardized at 760 Torr, or 14.72 psia.
For the manufacturer of a capacitance manometer that measures a process pressure at 2 millitorr, it meansthe product must have the accuracy and dynamic range to give a reasonable reading. An accuracy of 2 millitorr, ±1 millitorr, is unacceptable; it would be ±50%. So the gauge's dynamic range must be at the same order of magnitude it operated at when a process took place at 100 millitorr.
"We must offer more and more lower full-scale range gauges," said Paul Blackborow, vice president of marketing for MKS Instruments Inc. (Andover, Mass.). "A few years ago the lowest range gauge we offered was a 1 Torr, full-scale gauge; then it became a 100 millitorr full-scale gauge; now we're offering gauges down to 20 millitorr full-scale, so that when a 2 millitorr process is measured it falls within the transducer's reasonable range."
 |
|
Fig. 1. A typical backside cooling system as would be installed on a plasma processing system (etch or PVD). A backside pressure of a few Torr assures optimum heat transfer without a significant pressure differential across the wafer. Under normal circumstances, the gas flow will be within certain limits. If the wafer does not seat properly on the chuck, the flow will rise as the pressure control system attempts to hold the pressure set point. This is detected by the MFM, signaling the system to interrupt the process. (Source: MKS Instruments) |
Temperatures Rise
Higher process temperatures is another major trend PM&C equipment manufacturers face. "In the old days it was sufficient to make one of our Baratron capacitance manometers operate at room temperature," Blackborow said. "When more stability became necessary, we heated the sensor inside to create a uniform temperature environment to produce stable readings" (Fig.1).
When chemistries that reacted on chamber surfaces began being used, byproduct condensation became an accuracy-impairing concern, making necessary instrumentation capable of operating at higher temperatures. The PM&C industry responded by designing a product capable of operating at 100°C internal temperature. This avoided condensation from either the reaction's reactants or its byproducts in things like metal etch or certain CVD processes.
With the advent of more exotic CVD processes for low- and high-K dielectrics chamber temperatures have continued rising to where there are now versions of these instruments operating at 200°C, capable of measuring processes without material condensation in the vacuum gauge, ensuring accurate readings.
Thus, every vacuum component from the gas inlets down to the pump must be heated. There are heated valves, heated pumping lines, flanges, throttle valves etc., to keep condensation from taking place in the vacuum lines. These heated process components are becoming common, while working temperatures continue rising, and temperature control accuracy in the region they cover becomes tighter.
Hanging It All Together
|
|
Fig. 2. The calibration of gauges for state-of-the-art applications needs the facilities offered by an advanced lab. Here, each gauge can be seen being individually calibrated at various pressure values and supplied with a memory module matched to its own calibration data. (Source: Granville-Phillips) |
There are various digital networking field bus protocols being adopted for process control: DeviceNet, LonWork and Profibus, with DeviceNet apparently leading.
Most new sensors must have digital outputs compatible to one of these protocols: Pirani gauges, capacitance manometers, throttle valves, hot cathode gauges, cold cathode gauges the whole range of PM&C components. An advantage of a process control field bus such as DeviceNet is that it not only provides information about pressure, but also data on the gauge's condition. In a capacitance manometer's case, for example, the gauge will show not only that the chamber is at 2 millitorr, but also that it is at the proper temperature and does not need zeroing that given week. This capability permits the development of flexible intelligent systems (Fig. 2).
Another advantage lies in preventative maintenance. It may be programmed, for example, that a certain manometer needs zeroing once every two weeks. If it requires zeroing three times in a week it is flagged to be checked. That level of control will enhance understanding of tool performance, based on the sensors' statistical history.
Future DeviceNet implementations will have some gauges interrogating other components about the vacuum, enabling them to zero themselves using peer-to-peer communication over the bus. A particular gauge will "know" it is supposed to check itself and will interrogate, for example, the hot cathode gauge as to whether it is at 10-7, and when it is, it will zero itself. If it is out of spec, it will signal the controller that there is a problem. The system will alert the operator when it is out of spec, rather than having the operator interrogate it. Some of these developments are still dependent, to a certain extent, upon how 300 mm progresses, but intelligent devices already exist. When everything is networked and controlled together, the complete system will be intelligent, and performance quality will take a quantum leap.
Michael Borenstein, vice president of engineering at Granville-Phillips (Boulder, Colo.), said he also sees DeviceNet as a technology prime mover, but that work needs to be done first. "With DeviceNet, the cost of interfacing vacuum gauges is moved upstream from the system to the component and may someday have the effect of reducing overall system cost," he said. "For this to happen, the industry-wide committee will first need to finalize standards, which will then remain stable. Even when this happens however, the most important attribute of a sensor is its ability to measure a process parameter reliably, repeatably and accurately."
Borenstein pointed out that adding intelligence to an unstable or unreliable vacuum gauge is like painting racing stripes on an old car and expecting it to run better.
Nevertheless, this kind of open architecture will give more plug-and-play capability, making it simpler for OEMs and end users to swap between vendors.
The Move to Integration
Once OEMs used catalogs to put systems together and bought flanges from one company, valves from another, heaters from a third and pressure switches from yet another. The chances of all components working optimally together from the beginning were slim. This is why the integration of PM&C devices into subsystems and systems is a major trend. OEMs have achieved good chamber environment control and now want to heat all the vacuum lines to better control that environment. Suppliers are being told to integrate components into subsystems from the chamber to the pump designed and optimized to work in concert.
There are two drivers to component integration. One is product integration - an integrated set of components usually works better and enhances performance. The second is a reduction in complexity and cost for the OEM. If the OEM can get the necessary components in one subsystem, it means one-stop shopping with only one part number and one item being bought. This simplifies purchasing and reduces inventory management, since the OEM deals with one item bought from one vendor instead of 10 bought from as many others.
Like its competitors, Pfeiffer Vacuum (Nashua, N.H.) is consolidating components into subsystems and systems. According to Suzanne Garacau, instruments product manager, the company has come out with what they call a "full range gauge," which combines two gauges a Pirani and cold cathode gauge in one housing. The company also offers a gauge based on the hot-filament principle, which is also two gauges in one, a Pirani and hot cathode ionization. Within two to three years, Pfeiffer expects to have three gauges in one by adding a capacitance manometer to the Pirani and hot cathode mix in one housing.
Conservatives In High-Tech Clothing
But within this move toward integration there is a paradox; new components must not be too different.
Steve Hisel, sales marketing manager for Leybold Inficon (East Syracuse, N.Y.), observes that like the rest of us, OEMs are not inclined to abandon the familiar for the new. "Even though processes are changing, we still use the same type of ionization gauges, capacitance manometers and thermal conductivity types sensor. That's OK; these gauges are tried-and-true methods of measurement and have been around for years. We keep improving them to meet new process requirements and manage to make them operate at slightly lower pressures and capacitance manometers to work at slightly higher temperatures."
However, Hisel sees the end of the road rapidly approaching. "These are, however, mature technologies, and we're getting closer to their performance limits," he said. "The problem is they're familiar mainstays, and our customers are reluctant to give them up, even for something better. Even small, subtle changes that make a component seem different are not easily accepted. It is understandable that they may be concerned about whether a new gauge may adversely affect a process or reduce process reliability, and this doesn't make them very willing to look at new technology, even though we're approaching the limits of what the older technology can do."
As a computer engineer put it, "A capacitance manometer may be the best choice for process control, because it is gas-independent." "However, there are limits to how thin a diaphragm can be made and to how much the gain on the electronics can be boosted to get an additional measurement decade before the noise level increases."
|
|
Fig. 3. Because of the complexity of matching each individual component to make the whole work accurately and reliably, some aspects of the manufacture of capacitance diaphragm gauges require the human touch. (Source: Leybold Inficon) |
Some major manufacturers firmly disagree with this gloomy assessment. As MKS' Blackborow puts it, "While it is true today's technology is based on initial work done years ago, it is undergoing continuous development and is in no sense out of date, the proof being OEM requirements continue being met to their satisfaction and that of their end-users. There are no perceived fundamental limits for current or near-future technology for the processes identified on the SIA Roadmap."
This assessment seems to be reinforced by the coming of technology such as DeviceNet, which will enable already intelligent sensors to do better in systems. Field-bus compatibility allows a sensor to be used to its fullest extent, perform on-the-fly control and tailor it to the process. It will make it possible to use gas-dependent gauges. Many of these gauges will perform well under differing process conditions, because it will be possible to direct them to adapt. This capability should give today's mature technologies a new lease on life (Fig. 3).
For the immediate future, it appears OEMs will continue using current technology adjusted to meet process goals but will have to overcome their reluctance for the new.
Most OEMs follow a conservative "copy-exact" philosophy. Even when something better is available, if OEMs and end-users have a process that gives them today the results that they had yesterday, they want it exactly copied tomorrow so that the new start-up fab can achieve it. It does not matter whether the better technology is in the process area or the control system, continuity is the thing, and the new system must be an exact copy to ensure this.
This can turn evaluation processes into a Byzantine labyrinth. A PM&C supplier may have a new product under evaluation by a group within an OEM company only to have it turned down by another group after the first one approved it. It may not matter that the new device eliminates all the older one's disadvantages and is smaller, longer-lasting and lower in cost. The company is unwilling to change. As a PM&C sales engineer puts it,"We were told they could not use our product (which they admitted was better), because they buy from a single supplier to get volume discounts. Also, one of their major customers specifies the old device, because they have many of them in stock. The inertia is incredible."
Some PM&C component manufacturers hope copper will break through this technological logjam, when even higher temperatures and lower process pressures will become necessary.
Process gas delivery times are growing shorter. This is a result of the smaller volumes of gases being used, as well as the need for increased throughput. This presents gas delivery component suppliers with yet another set of challenges.
According to Andrew Dribnak, director of product marketing for the Veriflo division of Parker Hannifin Corp. (Richmond, Calif.), this has forced its engineers to develop a different way of evaluating a regulator's performance. "If an entire process cycle lasts five seconds, then the pressure and flow in the first second represents a large percentage of that cycle."
Traditional methods of evaluating the performance of a regulator use bourdon tube pressure gauges and rotameters. By the time the operator records the delivery pressure as the flow is varied, the regulator has already stabilized. The fluctuations that exist in that first second of the gas delivery often cannot be quantified with these methods.
|
|
Fig. 4. The production of pressure measurement and control components requires manufacturing facilities similar to those used in the production of semiconductor devices. This often means working in environments such as this Class 10 cleanroon. (Source: Veriflo Corp) |
Combining pressure transducers and mass flow controllers with a data-acquisition system capable of monitoring 10,000 outlet pressure data points per second enables Veriflo engineers to quantify fluctuations at the onset of flow. Because most semiconductor fabrication processes are performed at very low flow rates, efforts were concentrated there. "At the pressures and flows typical in these processes, the movements of the regulator poppet must be controlled in millionths of an inch," pointed out Dribnak. Better test methods have led to a better understanding of the device and regulator components' requirements. Component roundness and alignment become critical to providing stable pressure under these conditions (Fig. 4).
This kind of testing has other benefits as well, in that it allows products to be tested in ways that closely duplicate their final application.
Still, Dribnak does not foresee drastic changes in regulator technology. "Advancements in the next few years will not be in the form of new gas delivery technology, but rather from working with the other gas component suppliers to provide a 'systems' approach to overcome gas delivery hurdles."
Surprisingly, particularly in view of the delicacy and criticality of some of these processes, several major PM&C component manufacturers grumbled privately about insufficient communication between them and their OEM customers. By far the consensus appears to be that the data pipe must be scraped clean and widened.
"One of the issues we face is that because our industry is growing so fast there are many new, inadequately trained, people in it," said an engineer at an OEM company. "Much of the experience has either moved up the management ladder or been let go in favor of younger (and cheaper) engineers. Some of these guys wouldn't know a Pirani gauge if it bit them on the rear. Often the unsatisfactory results we get from a component are directly traceable to this - they just don't know how to use the stuff."
Blackborow concurs. "Customers like to think they know what they're doing, and they do, but they'd be better off also using the manufacturer's expertise. We want to be part of their design team. We have solid experience in vacuum measurement, and when they have a vacuum issue and try to solve it themselves they just reinvent the wheel. If we're brought in early in the cycle, we can offer a range of solutions, with all their tradeoffs, that they can choose from. That way, neither they nor we waste time."
So change comes slowly and in small increments in the PM&C industry.
From all accounts it would appear its technology is comfortably
well-ahead of the wave - there is more and better PM&C technology than
OEMs or end users seem to want or need. The consensus is that there will
be nothing radically different in this area for another two years, just
minor improvements on what is already here. 
Ultrahigh Purity Transducers - Keys to Accuracy
Richard Rosenblum Because errors or faults caused by incorrect pressure measurements can cause lost production or system downtime, the pressure transducer is a key component of gas distribution systems. Semiconductor manufacturers blame pressure transducers for shutdowns, due to their need for frequent recalibration and replacement. Technicians will suspect a faulty transducer even if other components of the gas distribution system are at fault. While there are many technologies used to make a pressure transducer, the ones covered here are applicable to the gas distribution system. Semiconductor Strain GaugeIn this type of device, four resistors comprising the strain elements on a Wheatstone bridge change with pressure. The bridge is driven with an excitation, and the pressure signal is sensed at the terminals. This arrangement is common to strain gauges. A mature architecture, its construction process is dependent on the quality and manual placement of four silicon strain elements to a stainless steel (or other) diaphragm with an epoxy mount. This element is considered drift prone, because the epoxy may cause pressure reading drift with time, temperature and pressure. Although the resistors' high output makes amplification easy, this is futile if there is a drift in the basic elements. Additional error sources include thermal gradients resulting from sudden gas pressure changes at the diaphragm surface (j-T effects), temperature coefficient of expansion mismatch between silicon bridge elements and the metal diaphragm, and any resistor instability. Capacitance GaugeCapacitance cells are an improvement on semiconductor strain gauges, because epoxy is eliminated. These usually measure low or vacuum pressures and tend to have large diaphragms and package sizes. The capacitive elements are excited with an alternating current. Distance changes between the two electrodes due to pressure produce a capacitance change. This is sensed as a frequency change and demodulated to produce a DC output. The pressure change is only several picofarads, which makes electronics more complex and sensitive to temperature changes. When diaphragm travel is extended to improve sensitivity, the transducer becomes more susceptible to linearity errors, requiring compensation. Often, two capacitors are used, one to measure diaphragm deflection due to pressure, the other changes due to temperature or other effects. Long-term electronic or mechanical instability effects cannot be compensated for, requiring periodic adjustment. Some capacitive transducers have mechanical coupling between the flow path diaphragm and the capacitive plates. This eliminates most off flow path volume but complicates temperature compensation, because process temperature is a distance from the internal case temperature by the coupling. The internal coupling can add to temperature-, hysteresis- and vibration-induced errors. Polysilicon Strain GaugeIn this Wheatstone bridge implementation, silicon is deposited directly onto a metal diaphragm using CVD. This eliminates the epoxy interface, which can improve device stability. There is still a mismatch between the temperature coefficient of expansion between the silicon and metal diaphragm, leading to large raw temperature errors as well as j-T effects. This type of device is sometimes called a thin film strain gauge. However, it is not the same as a metal thin film strain gauge. Bonded Foil Strain GaugeHere, four individual gauges are made of a metal foil, sandwiched between a thin laminate such as a polyimide insulator. Each gauge is epoxy-bonded to the optimal point on a stainless steel diaphragm arranged as a Wheatstone bridge. Like the silicon strain gauge, gauge placement and selection and application of the epoxy are key. The sensing element requires careful attention to the adhesive interface. The output is less than the silicon strain gauge, requiring amplification. Metal Thin Film Strain Gauge
This is a strain-sensitive Wheatstone bridge made out of a stable
metal film. Its nichrome resistors have very low temperature
coefficients, which can be further reduced by doping. Once used
for very accurate aircraft transducers, the technology has found
its way into the UHP gas distribution system. A metal diaphragm is
sputtered in a vacuum with a glass insulator followed by the
stable conductive layer. The result is a molecular bond that
minimizes long-term drift. Four resistors are then etched or
laser-defined to form a Wheatstone bridge pattern. There is an
excellent match between the temperature coefficient of expansion
between the metal diaphragm and the metal thin film resistors,
which keeps raw temperature errors and j-T effects very small.
Because the sensor is metal-based, it requires internal amplifiers
capable of handling smaller signals than those of a silicon-based
bridge. This is easily accomplished with some current generation
CMOS operational amplifiers. |
