Improving production efficiency withfaster vacuum loadlocks
12/01/2002
overview
This industry's quest for increasing production efficiency means that not even subsystems are exempt from scrutiny. For example, equipment designers are analyzing all details associated with vacuum loadlocks and their effect on fab cycle time. A thorough analysis has shown that entrance loadlock speed is improved by minimizing chamber and foreline volumes without sacrificing pump speed, and exit locks are faster and cleaner using a novel gas inlet.
The International Technology Roadmap for Semiconductors (ITRS) outlines the need for a reduction in factory cycle time from 1.4 to 1.1 days by 2010 — a 27% improvement. The ITRS suggests that tool throughput is one of the main contributors to factory cycle time. The logic is that improvements in throughput without increasing tool cost will reduce effective cost of tool ownership.
In semiconductor manufacturing, a wafer is moved from atmosphere to vacuum and back to atmosphere 30–50 times, making these transitions through vacuum loadlocks. Thus, loadlock performance may have a significant effect on a tool's throughput, particularly if a wafer's process time is relatively short.
null
 Figure 2. An optimized loadlock pumpdown. |
null
A current trend avoids outgassing contamination from cassettes or wafers [1] by adopting smaller loadlocks that move only a few or even a single wafer at any given time. A typical cycle — wafer insertion at atmosphere to vacuum robot pick up — for such a small loadlock is 30–60 sec. Large portions of this cycle time involve pumpdown and vent times that can be 10–30 sec each.
To understand the impact of loadlock cycle time on throughput, we used computer modeling to track wafer flow through a cluster tool. The modeled results shown in Fig. 1 assumed the cluster tool had four process modules operating in parallel and running the same process. These four modules are serviced by a single vacuum robot and an atmospheric robot for moving wafers into the entry loadlock and out of the exit loadlock.
We analyzed tool throughput as a function of loadlock pumpdown and vent time, robot pick-and-place time, and wafer-processing time. We found that for all operating conditions, a faster pumpdown time for loadlocks improves tool throughput. The effect of these improvements becomes greater as processing time is reduced. For short processing times, loadlocks limit tool throughput (i.e., any further reduction in processing time does not yield appreciable cycle-time improvement). Our conclusion here was that as the industry moves to shorter cycle times, faster loadlocks are needed to reduce the total tool cycle time.
Design and operation
Reducing pumpdown time for an entrance lock requires the optimization of the loadlock chamber and its vacuum connection and vacuum pump. The conventional approach of attaching larger pumps to an existing chamber may well be counterproductive, because the volume that needs to be pumped increases with any increased diameter of the vacuum connection.
Proper design optimization should address four key variables: chamber volume, vacuum system foreline and pump volume, chamber purge flow rate, and the outgassing behavior of a wafer. With proper optimization, the pumpdown time of a single-wafer entrance lock can be <3 sec to a pressure <100mtorr (Fig. 2).
For the exit lock, vent cycle time deserves equal consideration. A cycle 2 sec (Fig. 3) can be achieved with a single gas inlet without overstressing the wafer.
 Figure 3. An optimized loadlock vent cycle. |
null
To achieve desired cycle times, chamber volume must be minimized. For a single-wafer chamber, we accomplished this by designing a novel internal wafer-lifting mechanism that eliminates the need for vertical robot movement in the chamber and allows a chamber volume of <1.5 liters. This loadlock design can accommodate existing robot configurations and is compliant with dimensional Semi standards for the robot end-effector exclusion volume. The lifting mechanism uses a vacuum wafer chuck to detect and secure the wafer during rapid evacuation and fill cycles.
The vacuum system for our loadlock uses a single onboard dry pump mounted directly below the loadlock chamber, thus keeping the foreline length at a minimum. In the pumpdown cycle, the pump runs full speed with no throttling. The foreline diameter is the same pump connection diameter.
 Figure 4. Loadlock purge efficiency during wafer transport. |
null
The operational sequence of the loadlock has four steps (Table 1): wafer insertion, pumpdown including gate valve operation, wafer removal, and chamber venting including gate valve operation. These steps take approximately 5.7, 3.4, 5.7, and 3.6 sec, respectively. The total cycle time is ~18.4 sec from atmosphere to vacuum (i.e., one way).
While a similar pumpdown time can theoretically be achieved with a conventional loadlock volume of 6–10 liters and a large pump-set, volume minimization avoids large flow rates associated with the rapid pressure changes of large volumes causing particle disturbance by turbulent flow patterns.
In some cases, it is desirable to perform post-processing steps such as desorption or cooling in the loadlock. Using nitrogen purging and repeated rapid pumpdowns, desorption can be performed in a mode often referred to as "cycle purging." Due to the high speed of the loadlock pumpdown, this method can be employed without a large sacrifice in cycle time. The fast loadlock has also been evaluated for wafer cooling. For high-temperature applications, wafers need to be cooled from as high as 1100°C down to 70°C before they are returned to the atmospheric FOUP. Wafer cooling in the loadlock is accomplished by heat conduction to a chuck and convection with a purge gas. Cooling can be achieved in 10–20 sec depending on incoming wafer temperature.
Contamination and particle control
Loadlocks must be specifically designed to minimize or eliminate the introduction of air and moisture when the atmospheric gate valve is open. Any moisture intrusion raises the possibility of condensation during pumpdown of the entrance loadlock and subsequent transport of moisture to the process chamber as wafers are moved. Moisture in the exit loadlock increases the potential for corrosion from the combination of process gas transport, wafer outgassing, and moisture adsorbed on the walls. In addition, a loadlock must not add any particles, scratches, or defects to a wafer, above the allowed budget.
Our fast loadlock design uses a nitrogen purging system to prevent the ingress of atmospheric contaminants into the chamber while the atmospheric gate valve is open. We evaluated various purge gas diffusers, locations, and flow conditions to minimize the intrusion of atmospheric gases. Using oxygen as an indicator, we used a residual gas analyzer (RGA) to measure atmospheric contamination in the chamber during nitrogen purge.
We found that when using a rapid vent sequence with a series of nozzles delivering nitrogen into the loadlock, oxygen concentration was high (~100,000ppm) at nine measured points around the perimeter and at the center of the chamber (these tests were done without wafer transfer). We reasoned that this was likely from air entrainment due to the turbulent characteristics of the purge flow. Alternatively, when we used a purge-flow diffuser located near the rear of the chamber, the oxygen concentration decreased to <0.1ppm at all measured locations. In addition, measurements taken 25mm from the door opening show an oxygen concentration of <1ppm. These results showed us that oxygen ingress into the loadlock can be minimized by proper design.
With further testing, we evaluated the purge efficiency during wafer transport (Fig. 4), taking RGA data near the front-center of the loadlock and 25mm from the door opening. The results indicate some perturbations in oxygen as wafers are moved. These were likely caused by a disruption in the laminar flow purge gas, which is strongly influenced by design details of the robot end effector.
We tested for front-side particles using pre-cleaned, bare, 300mm silicon wafers loaded on a robot simulator that provided automatic transport of wafers into and out of our loadlock chamber. Using one wafer out of the sealed carrier box as a control, we placed the remaining wafers on the simulator and cycled these through the chamber. Two wafers were cycled 50 times with no pumpdown. Three wafers were processed 50 times with a full cycle. Using a KLA-Tencor Surfscan 6420 for particle measurements, we found no added particles on any of the cycled wafers (Table 2).
Acknowledgments
The authors thank Frank Jansen and Chris Case for technical discussions, direction, and assistance in preparing this article. BOC Edwards is a trading name used by affiliate companies of The BOC Group plc. KLA-Tencor is a registered trademark of KLA-Tencor Corp.
Reference
1.R.L. Wise et al., Proceedings of the International Symposium for Semiconductor Manufacturing, San Francisco CA, pp. E1–E4, Oct. 6–8, 1997.John Dickinson received his ME in mechanical engineering and materials from Queen Mary & Westfield College, University of London. He is an engineering manager at BOC Edwards.
Daimhin P. Murphy received his BS in mechanical engineering from California Polytechnic State University, San Luis Obispo. He is a mechanical engineer at BOC Edwards.
Frederick Tapp has more than twelve years' experience in the semiconductor industry. He is a lead electronics applications engineer at BOC Edwards.
Peter Kimball received his ME from Northeastern University. He is a business development manager at BOC Edwards, 301 Ballardvale St., Wilmington, MA 01887; ph 800/848-9800, fax 978/658-7969, e-mail [email protected].
John Dickinson, Daimhin Murphy, BOC Edwards, Santa Clara, California Frederick Tapp, Peter Kimball, BOC Edwards, Wilmington, Massachusetts