Benefits derived from integrated loadlock pumping
06/01/2003
Overview
Within a wafer fab, a high percentage of systems use a vacuum or controlled atmosphere for deposition and other process steps. The passing of wafers into and out of these systems via loadlocks becomes a significant portion of overall wafer handling. An integrated approach to loadlock design can do a lot to improve fab efficiency and contamination control.
The speed of loadlock venting and pumpdown directly impacts fab throughput, and the efficiency of contamination control affects wafer yield. An integrated loadlock subsystem that controls pumpdown and vent cycles can minimize cycle time and avoid problems such as turbulence that stirs up particles, and adiabatic cooling that produces condensation of acid-laden moisture droplets [1–3].
A key factor for loadlocks is that pumping them to below 10-7torr from atmospheric pressure is paced by the very slow desorption of water vapor. The amount of water adsorbed on the interior surfaces of a vacuum chamber from atmospheric air at 40–50% relative humidity can be equivalent to tens of monolayers or more.
Pumpdown is also slowed by inevitable virtual leaks — pockets of air trapped within regions having a low-conductance path to the main volume. This is why the exposure of a high-vacuum chamber or ultra-clean process chamber to atmospheric air is limited to essential maintenance procedures.
 Figure 1. A typical loadlock pumpdown and venting sequence. |
null
With a typical loadlock pumpdown and vent sequence (Fig. 1), the first part of the pump cycle is usually accomplished with a dry mechanical pump. A two-stage valve is often used to maintain laminar flow in the pumping line to reduce potential particle transport. If the required transfer pressure is <1mtorr, such as for physical vapor deposition, the pumping is transferred to a high-vacuum turbomolecular or cryogenic pump. The proper crossover pressure must be detected accurately for efficient pumpdown. When the required transfer pressure is reached, the inner loadlock valve is opened and wafers are moved to a transfer chamber or directly to a process chamber. Then the loadlock valve is closed and the loadlock is vented with dry nitrogen to atmospheric pressure in preparation for the next load of wafers.
Loadlock pressure transducer functions
During pumpdown, a loadlock pressure sensor transducer must perform several tasks: accurately monitor loadlock pressure from atmosphere to ~0.1mtorr; send a signal to control the operation of the two-stage valve; sense and signal crossover pressure if a high-vacuum pump is used; and sense final transfer pressure and provide a setpoint signal to the loadlock transfer valve.
During loadlock chamber venting, a loadlock transducer should be able to sense the pressure difference between the loadlock chamber and the external atmospheric pressure. The differential pressure is more important than absolute pressure, since the objective is to vent the loadlock to a slight positive pressure with respect to the ambient atmospheric pressure prior to opening the main loadlock door. However, even at sea level, the atmospheric pressure is rarely the standard 760torr. The mean atmospheric pressure depends on elevation and varies with local weather conditions. (For example, the mean atmospheric pressure over the US for a 31-day period in August 2000 ranged from 705–1080 millibar with a variation of ±20–30 millibar.)
If the loadlock is vented to a pressure significantly lower than the external atmospheric pressure, the loadlock door cannot be opened. At a pressure only slightly lower than the external pressure, the door could open, but inrushing gas could bring in moisture and particles and stir up particles within the loadlock. During venting, the pressure sensor should also be able to monitor venting speed and provide setpoints to the vent valve to control the vent cycle.
A programmed vent cycle
Particle contamination can often be reduced if vacuum system pumpdown and venting are programmed to reduce turbulence during the initial portions of pump and vent cycles [4–6]. For example, when venting a loadlock chamber, a two-step process can be used: a rapid vent to a crossover pressure followed by slow venting to atmospheric pressure. The crossover pressure should be high enough so that during the subsequent slow vent cycle turbulence will not stir up particles and condensables will not precipitate. An analogous situation occurs during pumpdown from atmospheric pressure; a two-stage pump cycle is often used to avoid condensation and turbulence.
Loadlock pressure is usually measured with a convection-enhanced Pirani gauge with a built-in crossover setpoint [7]. If the setpoint for the crossover from slow to rapid venting is in a region where the gauge has low accuracy, typically from 10–100torr for a conventional convection-enhanced Pirani sensor, the slow vent time can be longer than necessary.
In one case, the crossover pressure was specified at 80torr. A test with a capacitance manometer showed that the loadlock was actually slow venting to 120torr, which added two minutes to the process time. The Pirani gauge was replaced with a capacitance manometer sensor with a 100torr full-scale range. The crossover pressure was accurately sensed and two minutes were saved per cycle, resulting in a significant increase in throughput. A capacitance sensor is insensitive to gas species, an advantage if vent gases other than air or nitrogen are used.
Integrated two-stage vent valves
Soft-start isolation valves are available that simplify and automate two-stage pumping and venting. Some versions are actually two independent valves, a high-conductance main valve and an auxiliary, low-conductance bypass valve [7]. A better solution is a valve that integrates a bypass into the main valve. The unit has a small, pneumatically operated bypass poppet valve built into the center of the sealing face of a pneumatically actuated, bellows-sealed, high-conductance valve. Rather than using a long, small-dia. tube to restrict conductance, which can easily clog, the conductance-limiting element has four small adjustable orifices (Fig. 2) that take up less space than a tubular bypass.
 Figure 2. Bypass pumpdown curves for various bypass conductance settings pumping a 9-liter volume; pumpdown time to 1torr can be varied from ~30 sec to more than 3 min. |
null
The second portion of the vent cycle can be very rapid (<15 sec), so rapid pressure sensing is required. A small pressure drop occurs during rapid backfilling after the gas supply is shut off and a negative setpoint may be required to compensate for signal delay.
Miniaturization of pressure sensors
A conventional convection Pirani pressure gauge has three limitations that prevent efficient monitoring of loadlock pressure. The first is slow response time. The second is relative insensitivity from 10–100torr, which is the range where crossover setpoints are often required. The third limitation is the inability to sense differential pressure — the difference between the loadlock pressure and the actual external atmospheric pressure. Two new pressure transducers, combined in one sensor body, overcome these limitations.
Miniaturizing the Pirani sensor improves response time and sensitivity and eliminates the flat zone between 10–100torr and the dependence on sensor orientation that are typical of convection Pirani gauges. The microminiature Pirani sensor (e.g., MicroPirani) is constructed using IC fabrication processes. A thin nickel film replaces the conventional Pirani hot filament. The small size and low heat capacity reduces the response time. The narrow gap between the heat source (the nickel film) and the heat sink (a silicon base and cover) improves the sensitivity in the critical 10–100torr range. The low-pressure range is increased by on-chip ambient temperature compensation. The result is a sensor with a pressure range from 10-5 to 1000torr with ±5% accuracy, a response time of 150 msec, and an operating temperature range of 0–40°C.
The conventional diaphragm differential pressure sensor can be miniaturized using MEMS fabrication. This sensor measures the difference between the pressure being sensed and a reference pressure, in this case ambient (atmospheric) pressure. The differential piezoresistive sensor (piezo sensor) has polysilicon piezoresistors deposited on a thin silicon diaphragm; one side is at atmospheric pressure and the other at the pressure being sensed. The polysilicon resistivity change is proportional to the diaphragm's strain (deflection). As with the MicroPirani sensor, precise temperature compensation can be built in. The sensor output is linear with differential pressure within ±30torr of the external atmospheric pressure (Fig. 3).
 Figure 3. The output of a piezo sensor is linear with pressure; note that the range extends 30torr above and below atmospheric pressure. |
null
The differential piezo sensor is better suited for loadlock applications than a convection Pirani gauge because it measures the difference between the local atmospheric pressure and chamber pressure. This is what is really needed for loadlock control, which requires a low differential pressure when the door is opened to reduce mixing with ambient air and contamination of the wafers. In comparison to the convection gauge, which can drift over time, the piezo reading at atmosphere is stable and is usually within 0.25torr of true atmospheric pressure (see "Pressure sensor response times when tracking loadlock pressure"). The differential pressure range is ±30torr with ±1torr accuracy. The response time is 10 msec, faster than a conventional Pirani gauge or a capacitance manometer.
Space is at a premium in production process tools. A MicroPirani and the piezo sensor can be combined in one ~1.5 in.-dia. package. Thus, one loadlock sensor device replaces two separate components and measures pressures from 10-5torr to atmosphere. Outputs can be provided for setpoints for the control of two-stage pumpdown crossover and venting differential pressure.
Youfan Gu, Ole Wenzel, HPS Products, MKS Instruments, Boulder, Colorado
Acknowledgments
Baratron is a registered trademark and MicroPirani is a trademark of MKS Instruments Inc.
References
1. B.Y.H. Liu, Semiconductor International, pp. 75–80, March 1994.
2. J. Zhao et al., Solid State Technology, pp. 85–89, Sept. 1990.
3. P. Borden, Semiconductor International, p. 190, May 1988.
4. J.F. O'Hanlon, J. Vac. Sci. Technol., A 5(4), p. 2067, Nov./Dec. 1987.
5. M.H. Hablanian, Research & Development, 31(4), pp. 81–86, April 1989.
6. G. Strasser, M. Bader, Microcontamination, pp. 45–48, 100–101, May 1990.
7. S.P. Hansen, et al., Solid State Technology, 40(10), pp.151–158, Oct. 1997.
Youfan Gu is manager of research at HPS Products, MKS Instruments, 5330 Sterling Dr., Boulder, CO 80301; ph 303/449-9861, fax 303/449-2003, e-mail [email protected].
Ole Wenzel is plant manager at HPS Products, MKS Instruments, Denmark Branch.
Pressure sensor response times when tracking loadlock pressure
The response of the MicroPirani, a differential piezo gauge, a conventional convection Pirani transducer, and a 1torr full-scale Baratron capacitance manometer were compared by mounting these sensors on the same vacuum chamber and performing fast pumpdown and venting cycles. The figure at left shows the response of the vacuum sensors during a rapid (15 sec) pumpdown. The MicroPirani takes about 22 sec longer to read 10-3torr than the Baratron. However, the convection gauge takes more than two minutes longer than the MicroPirani. The figure at right compares sensor outputs during fast venting. The ambient atmospheric pressure (at ~5400-ft elevation) was ~630torr. The piezo sensor had the fastest response time. The convection Pirani had the slowest; the reading continued to drift upward due to temperature or electronic effects.
 |
null
 |
null
Various pressure sensor responses during rapid pumpdown cycle (left) and for fast venting to an external atmospheric pressure of 630torr (right). 1torr full scale.