Pressure measurement and control in loadlocks
10/01/1997
Stephen P. Hansen, Kathryn Whitenack, MKS Instruments Inc., Andover, Massachusetts
A process tool's loadlock serves to isolate the high vacuum/low contamination environment of the process chamber from the atmospheric pressure environment of the cleanroom. To optimize cycle time and minimize yield limiting factors such as particulate formation and redistribution, accurate pressure management techniques are just as critical in the loadlock as they are in the process chamber. The proper monitoring and control of the loadlock pump and vent sequences demands several types of vacuum gauges to cover the needs of wide pressure range monitoring and high precision control functions.
The loadlock is the gateway to the modern, high-throughput process tool. Its primary purpose is to isolate the inner regions of the system from the general environment of the cleanroom workplace. The loadlock may be connected directly to a process chamber, or, in more complex cluster tools, there may be several intermediate staging and cleaning chambers before the wafers reach the process area.
We think of cleanrooms as low contamination environments. The cleanroom air, however, is dirty and wet when compared to the vacuum process environment. Airborne particulates are present, as well as water vapor in the form of humidity. The proper operation of the loadlock requires well controlled pumpdown and venting sequences to minimize particle transport and the precipitation of condensables on the wafer surfaces.
This article focuses on the function of the loadlock and a number of problems and solutions associated with proper pressure control in the loadlock cycle. By using absolute and gauge capacitance based pressure switches, turbulence, condensation, and particle count can be reduced. Benefits include reduced defects, increased yield and optimized cycle times. Much of this discussion is also applicable to single-chamber tools.
Loadlock cycle
The loadlock shuttles between the atmospheric pressure environment of the cleanroom and vacuum. The ultimate vacuum required at the end of the pumping cycle depends upon several parameters. On one hand, the pressure must be low enough so that particles can no longer be transported by air currents. On the other hand, the pressure should not be so low that it takes a long time to reach. A pressure in the range of a few tens of millitorr usually satisfies these requirements, and provides a good match to the pressure regimes found in the next stage of the tool.
Starting the pumpdown with a clean chamber helps avoid contamination effects. The influx of particles and condensables from the cleanroom environment may be partially avoided by providing a positive pressure dry nitrogen purge while the chamber door is open.
Getting from atmosphere to the required level of vacuum is fairly straightforward in terms of pumping technology. However, we have to go through the low (also called rough) vacuum regime between atmosphere and a few torr. In this region the gas is in viscous flow and has a fluid-like behavior. A number of difficulties can arise if the pumpdown is not conducted properly. Thus the excursion has to take into account a couple of practical realities:
- The first problem is turbulence. While cycle time would be minimized by a rapid pumpdown to the base pressure of the loadlock, such rapid pumping will stir up any particles that reside within the chamber and redistribute them, quite possibly on the wafers. In the viscous flow regime, if the rate of pumping exceeds a certain value, the flow will become turbulent. Slowing down the pumping process by reducing the effective pumping speed will minimize this effect.
- The second problem is adiabatic cooling. Adiabatic effects will be familiar to anyone who has observed that a bicycle pump heats up when used vigorously (heating due to rapid compression of the gas in the pump) or the condensation on a barbecue grill's propane tank (cooling due to rapid expansion of the propane as it leaves the cylinder). An adiabatic process is one in which no heat enters or leaves the system during a change in conditions. In a vacuum system, when the chamber is pumped, the gas expands rapidly. This will cause the gas within the chamber to cool to temperatures sufficiently low to condense low-boiling-point gases such as the water vapor that entered the loadlock when the door was opened. The net result is another source of contamination.
If the pumpdown is conducted more slowly, the process changes from one that is adiabatic to one that is isothermal, where there is no change in temperature. In this case, the pressure change is slow enough to permit an ongoing equalization of temperature with the walls of the chamber. Like the problem of turbulence, adiabatic cooling is problematic only in the low-vacuum region.
This process of starting the pumping cycle at a slow speed is often referred to as soft starting or soft pumping. The key to specifying the soft pump cycle is to answer the questions "How slow is slow?" and "For how long should the soft pump last?" An examination of a typical pumpdown sequence for a loadlocked system will illustrate the situation.
 Figure 1. Loadlock and loadlock pressure profile. |
Figure 1 shows the pumpdown sequence for a typical loadlocked cluster tool. The cycle for the loadlock begins with the soft-pump (A) to avoid the above-noted problems of condensation and turbulence. A common way of achieving soft pumping is by the use of a low conductance bypass line in parallel with the main isolation valve. The figure shows an adjustable needle valve in series with a pneumatic isolation valve. Fast pumping (B) commences at the point where condensation and turbulence will not have a strong effect, usually in the range of about 10 torr or slightly higher. At the crossover pressure (10 torr), any particles in the chamber cannot be transported by "floating," as occurs with dust at higher pressures. This second phase of pumping quickly brings the chamber well down into the medium vacuum range, typically below 100 millitorr. Once the pressure in the loadlock has reached the required low value (C), the slit valve connecting to the buffer chamber opens. The steady-state pressure in the buffer chamber (D) rises but quickly declines into the high vacuum range after the slit valve closes (E). The pressure at this crossover is low enough that the gas has lost any of its fluid-like behavior and the chamber-to-chamber transfer takes place with no damaging effects.
Several methods exist to determine the proper pumpdown profile. The dimensionless Reynolds number is a measure of turbulence and can be calculated from the fluid velocity at the outlet of the chamber. In general, a Reynolds number of less than 500 will assure no turbulence [1]. Another applicable method is the Z-factor analysis from which one can develop a pumpdown profile that avoids condensation effects [2, 3]. Since turbulence mobilizes particles, a particle monitor mounted on the outlet of the chamber will indicate the onset of turbulence.
 Figure 2. Time constant. |
Each phase of the pumpdown is in the form of a declining exponential curve. The time constant associated with the pumpdown curve is equal to the volume of the chamber divided by the effective pumping speed (that is, the speed at the outlet of the chamber). If, for example, the chamber has a volume of 50 liters and the effective speed is 5 liters/sec, the time constant will be 10 sec. On the pumpdown curve, the time constant (10 sec in our example) is the point at which the pressure has declined to about 37% of its initial value. Assuming the initial pressure was 1 atm (760 torr), one time constant would be the time to get to about 281 torr (Fig. 2). It takes a little more than 4 time constants to go from atmosphere to 10 torr.
The needle valve of Fig. 1 can be adjusted to give the proper pumpdown profile. A fixed orifice may also be used. Two-stage valves that integrate isolation and soft-pump bypass functions in one valve body are becoming more popular (Fig. 3).
 Figure 3. Photo of a two-stage valve incorporating isolation and soft-pump functions. (Photo courtesy of HPS Division, MKS Instruments Inc.) |
Establishing and monitoring the pumpdown is one thing. Controlling the profile with a high level of consistency is something else. The next section looks at the basic gauging requirements for a loadlock and then examines several deficiencies of the system. Capacitance-based pressure switch solutions to each of these problems will be described.
Basic loadlock gauging – the convection-enhanced Pirani gauge
The convection-enhanced Pirani (CEP) gauge is the type of gauge most commonly encountered on loadlocks. We will look here at its principles, characteristics, strengths, and weaknesses.
Convection Pirani gauges are used on loadlocks for one simple reason: they are uncomplicated, relatively inexpensive, and can cover the entire pressure range associated with loadlocks. A typical CEP will measure from 10+3 to 10-3 torr.
The CEP is an indirect gauge — it does not measure true pressure (defined as the force that the gas exerts on a specific area) but rather some characteristic of the gas that changes in a predictable way with pressure. In the case of the CEP, that parameter is the thermal conductivity of the gas.
The operating element of a Pirani sensor is a wire, held at a constant, elevated temperature, in the range of 120°C. Heat transfer from the wire is by molecular conduction: molecules collide with the filament and depart, carrying energy away. If the sensor is at a very low pressure, say about 10 millitorr, a certain amount of power has to be fed to the wire to maintain its temperature. As the pressure increases, the density of gas molecules around the wire increases proportionally. This greater density of molecules causes the heat to be conducted away more quickly. Since the wire wants to cool, the gauge's circuitry feeds more power to the wire to maintain the desired temperature. This mode of operation works most effectively from about 1 millitorr to several torr, whereupon the heat transfer characteristic will flatten as conduction becomes independent of pressure. As the pressure increases further, convection currents will start to circulate around the wire. These convection currents also transport heat from the wire in a repeatable fashion.
The exploitation of two distinct modes of operation, one associated with the medium vacuum regime and the other associated with viscous flow at higher pressures, gives the CEP its extended range of operation.
Controlling the loadlock cycle
In the loadlock pumpdown, the critical parameters are the general pressure vs. time profile, proper crossover pressure to fast pumping, and reaching a satisfactory base pressure. The accuracy and repeatability of the CEP are sufficient for the first and third items. The way that the CEP operates, however, does lead to a couple of problems when this gauge is used for control applications where some level of precision is required, especially during the transition from soft pumping.
In the region between about 10 and 100 torr, the sensor is passing between the molecular conduction and convection modes and loses sensitivity. This transition can be seen in the pressure vs. voltage characteristic plot of the CEP (Fig. 4).
 Figure 4. General form of pressure vs. characteristic output voltage in a convection-enhanced Pirani transducer. |
Also, if the pressure is changing while the gauge is in convection mode, a forced convection component is introduced as gas either moves into or out of the volume of the gauge. This additional component will increase cooling of the wire, causing the gauge to indicate a higher pressure than the actual pressure. Therefore, if the Pirani is used to control the crossover to fast pumping, the gauge is in the area of its lowest accuracy and repeatability. Consequences include crossing over too soon — turbulence/adiabatic cooling issues — or too late — lost cycle time (Fig. 5).
 Figure 5. Controlling crossover with the Pirani gauge. |
A second way of controlling the crossover is to time the crossover to the start of the pumpdown. This eliminates any gauge repeatability problems. However, should the pumpdown profile change due to variables such as pump performance, small seal leaks, changes in the gas load, variations in the soft-pump needle valve setting, or orifice size, etc., the chamber will not reach the correct pressure in the specified time. If the error is on the high side, turbulence and condensation due to adiabatic cooling are still a problem (Fig. 6).
 Figure 6. Constant time control of crossover. |
The solution to these problems is to add a gauge that has a high level of accuracy and repeatability at the crossover point. Absolute pressure switches based on capacitance manometer technology meet these requirements. The device incorporates two capacitance electrodes plated in a concentric bull's-eye arrangement on a ceramic disc that is positioned parallel to a thin, flat, tensioned Inconel* diaphragm. When the pressure on both sides of the diaphragm is equal, the diaphragm is flat and the bridge is balanced. As pressure changes, the diaphragm deflects toward or away from the ceramic disc. The center electrode capacitance changes more or less, respectively, than the outer electrode capacitance, causing the bridge to be balanced. The signal is then compared using a comparitor to the switch point. Figure 7 shows a photograph of a capacitance-based switch along with a diagram of the switch's sensor capsule and its position in the unit.
 Figure 7. Photo of capacitance-based switch with a cut-away illustration of the sensor capsule. (Photo courtesy of MKS Instruments Inc.) |
A typical switch will have an accuracy of 0.5% of full scale (FS). In our example application with the crossover at 10 torr, if a 10 torr FS switch were installed on the loadlock, the crossover error would be no more than 0.05 torr.
Controlling the vent cycle
After reloading a wafer cassette at the end of a cycle, the loadlock is returned to atmospheric pressure. Since we still want to avoid particle movement, just as we soft-pumped the chamber from atmosphere, dry nitrogen is introduced slowly ("soft vent") to avoid turbulence. The loadlock door opens when the chamber has returned to atmospheric pressure.
While the CEP on the loadlock will give a good representation of the speed of the venting cycle, it again fails us when we need to know that the chamber has reached the ambient pressure of the cleanroom. This is because the CEP is an "absolute" gauge — it is referenced to vacuum. Cleanroom pressures will vary according to location (a 13 torr drop per 1000-foot rise) and due to weather patterns. At a given location, atmospheric pressure changes constantly, sometime as much as 20 torr in a 24-hour period. Therefore, were we to rely on the Pirani gauge, the internal pressure of the loadlock would almost certainly be well over or under the ambient pressure. This could cause the door to fail to open or ingest room air (with all of its humidity and particulates) in the case of underpressure, or, in the case of overpressure, to expel whatever gases are in the loadlock into the cleanroom.
Installing an atmosphere referenced mechanical switch on the loadlock is one common way to address this problem. These devices consist of a diaphragm that is coupled to the cleanroom on one side (the reference side) and to the loadlock on the other. When the two pressures are equal, the diaphragm becomes flat and an attached position sensing electrical switch closes, sending a signal to the control system.
 Figure 8. Effects of at-atmosphere accuracy. |
The primary disadvantage of these mechanical switches is that they lack repeatability. Therefore, the chamber might still be over or under the desired equilibrium condition. Here, another type of capacitance manometer based switch (Fig. 8) is useful. The switch would be atmosphere-referenced and set to a specific setpoint within the range of the transducer. In a loadlock application, a 10-torr FS switch would be set to switch at a 0 torr differential. With an 0.5%-FS accuracy switch, the loadlock door would activate precisely when the setpoint is reached. The loadlock of Fig. 1 shows the types of gauges that are necessary for proper monitoring and control of the loadlock cycle.
Conclusion
Determining the profile of the overall loadlock cycle demands attention to a number of damaging effects that can contaminate the wafer. Proper gauging on the loadlock provides pressure profile monitoring as well as accurate and repeatable control at the critical operating points. From a manufacturing standpoint, we have reached an optimal point where contamination is minimized without compromising cycle time.
References
- "Sematech Guide for Contamination Control in the Design, Assembly, and Delivery of Semiconductor Manufacturing Equipment," SEMASPEC #92051107A-STD, available from Sematech.
- J. Zhao, "Thermodynamics and Particle Formation during Vacuum Pump-Down," PhD thesis, Univ. of Minnesota, 1990.
- B.Y.H. Liu, et al., "Particle Generation During Vacuum Pump Down," Institute of Environmental Sciences Conference Proceedings, 1991.
 |
Stephen P. Hansen received his BSEE from Northeastern University. He is responsible for development and deployment of MKS's educational programs and products and is the owner of two patents. Prior to joining MKS, he worked for 23 years in process development in semiconductor-related areas. Hansen is an AVS member and is active on the History and Education committees. He also edits the bimonthly History Column for the AVS Newsletter. MKS Instruments, fax 508/975-0093, e-mail [email protected].
 |
Kathryn Whitenack received a BS in chemistry from Bates College and an MBA from Boston University. She has been the product manager for Pressure/Vacuum Switch at MKS for one year. Prior to that, she worked for Johnson Matthey, most recently as Sales and Marketing Manager.
*Inconel is a registered trademark of Inco Alloys International