In situ chemical cleaning for improved ion implanter utilization
03/01/2008
EXECUTIVE OVERVIEW
The overall utilization efficiency of production ion implanters is strongly influenced by the build up of solid deposits inside the vacuum system. Atmel and ATMI have been jointly testing an in situ cleaning process that involves introducing a powerful fluorinating agent (XeF2) at regular intervals into the ion implanter. With use of in situ cleaning, we have observed doubling of ion source life, general improvement in overall cleanliness of the ion source region, and reduction in potential implant species cross-contamination.
The most common cause for ion source end-of-life failure is buildup of solid deposits inside the ion source arc chamber and on the high voltage electrodes in the neighborhood of the source.
Undesirable deposits inside and around the ion source lead to source lifetimes that are not only shorter than they should be, but are also widely variable depending on the implant process. Frequent, unpredictable interruptions of production in order to change the implanter ion source are difficult and expensive to manage in a production environment.
Chemical in situ cleaning
Atmel and ATMI have been jointly testing an in situ cleaning process, which involves introducing a powerful fluorinating agent (XeF2) at regular intervals into the ion implanter. The reagent is delivered from a standard cylinder, which fits in the gas box of any implanter. The cleaning vapor is introduced into the ion source twice daily for 10-15 minutes.
Tests were carried out at Atmel’s Colorado Springs manufacturing facility, using both a medium currrent and a high current implanter.
Results
The table shows a compilation of source lifetime data from the medium current implanter before and after introduction of the in situ chemical cleaning process. For these data, the source was running a dopant mix that included PH3 and AsH3, but not BF3. Prior to cleaning, average source life under these conditions was about 250 ±90 hrs., with source life limited by two common failure modes. The predominant failure mechanism was excessive leakage from the “suppressor” voltage supply.
 |
We believe that excessive suppressor leakage is caused by buildup of a conductive film on one or more of the suppressor insulators. The second most common source failure mechanism prior to introduction of in situ cleaning was shorting of components inside the ion source arc chamber caused by flakes of deposited material. The in situ chemical cleaning process was able to minimize both these failure mechanisms and extend source life by removing some or all of the deposits. After regular twice-daily cleaning was introduced, source lifetime on the production tool approximately doubled. In all cases, the failure mode of the sources run with in situ cleaning was an open filament condition, which represents the best possible life of the ion source.
We believe that the cleaning action of XeF2 occurs both within and outside the source arc chamber. Some of the cleaning vapor reacts with deposits inside the arc chamber, and some unreacted cleaning vapor diffuses out of the source arc chamber to reach the suppressor insulators, where it reacts with deposited film at ambient temperatures and in the absence of plasma. XeF2 is known to react under these conditions with P and As, both of which are likely constituents of the deposited material.
 Figure 1. Average suppression leakage current by wafer lot. |
Further evidence of the effect of XeF2 on the suppressor leakage current is shown in Fig. 1. Here the leakage current is plotted for the production medium current tool before and after introduction of in situ cleaning. Each data point represents the average suppressor current over the time taken to implant one wafer lot, and points have been plotted over the lifetime of several ion sources. Note that with XeF2 cleaning, leakage current never reached the upper control limit of 1.5mA, which would trigger unscheduled source maintenance.
Cross-contamination
At Atmel, we have studied the well-documented case of P contamination in a BF2 implant. The yield consequences of this contamination are so serious that some fabs will avoid scheduling phosphorus and boron implants on the same tool, resulting in loss of flexibility in scheduling the implant workload.
To study the effect of the XeF2 cleaning agent on this memory effect, we operated the high current implanter for about 200 hrs. in simulated production with a P+ ion beam from PH3 dopant gas. We then switched to BF3 gas and immediately implanted a bare silicon monitor wafer with a high dose (5 × 1015 ions/cm2) of BF2+. During this implant, the resolving aperture of the implanter’s analyzing magnet was opened more than usual to ensure that the contamination effect would be large enough to measure conveniently using SIMS analysis.
The cleaning effects of BF3, argon and XeF2 were compared by running each of the three gases and then periodically monitoring the amount of remaining contamination by implanting monitor wafers with BF2+ and using SIMS to measure the amount of P co-implanted with the BF2. Figure 2a shows a typical SIMS spectrum of implanted phosphorus. The peak in the phosphorus spectrum corresponds to the implanted depth of PF+ ions extracted from the ion source, while the dose corresponds to a contamination level of about 3% PF in BF2.
 Figure 2. a) Typical SIMS spectrum of implanted phosphorus and b) plot of the contamination level as a function of cleaning time with either BF3 or XeF2. |
Figure 2b is a plot of the contamination level as a function of cleaning time with either BF3 or XeF2. The plot is normalized to the contamination level immediately after changing from PH3 to BF3. Running BF3 plasma had little effect on the PF contamination even after 2 hrs. A similar result (not shown) was obtained with argon plasma. On the other hand, after only 15 minutes of in situ cleaning with XeF2, the PF contamination was reduced by a factor of two, and after 30 minutes, by almost a factor of five.
Improvement of overall fab efficiency
Prior to implementing in situ cleaning, Atmel’s medium current tool set averaged 3.3 source changes per tool per month, with the average source change and subsequent qualification tests taking ~5 hrs. per event. This translates to nearly 200 hrs. per tool per year of lost production time. Using in situ cleaning effectively doubled the source life, which adds 100 hrs. of additional production time to each medium current tool. In addition, we are saving test wafers and the time on fabrication and metrology tools needed for post processing the qualification wafers for up to 20 qualifications per year for each medium current implanter.
Acknowledgments
The authors would like to acknowledge the help of the following Atmel process engineering technicians who diligently ran the cleaning recipes: Stephanie Campbell, Jimmy Darrington, John Bonneville, Craig Hawley, and Jessica Rawson. We also thank Sheldon Bates, James Grim, and Neal Verzwyvelt for their work in collecting and processing test data.
James Dunn received his BS in business management from U. of Phoenix and his AS in electronics technology from Daytona Beach Community College. He is an equipment engineering section manager at Atmel Corp., 1150 East Cheyenne Mountain Blvd., Colorado Springs, CO 80906 USA; ph 719/540-1273, e-mail [email protected].