Advanced Process Control for Dielectric CMP
By G. Dishon, D. Eylon, M. Finarov, R. Kipper, and A. Shulman, Nova Measuring Instruments, Ltd. P.O.Box 266, Rehovoth, Israel
Expanding use of chemical mechanical planarization (CMP) in advanced semiconductor manufacturing has motivated efforts to enhance the productivity and reliability of both the process and the equipment. Several approaches to advanced process control and advanced equipment control (APC/AEC) are being developed to improve process uniformity and overall equipment efficiency (OEE). NovaScan 210/420 integrated thickness monitoring systems (ITM) support a simple and reliable control method based on dynamic process control algorithms. Experimental results, applying closed loop control (CLC) of the polish time, show significant improvement in wafer-to-wafer uniformity.
CMP Process Monitoring Requirements
CMP is sensitive to changes in internal system variables and consumables as well as external parameters such as product types. A very large number of variables affect the removal rate within each wafer and between wafers, with conflicting variables forcing processing trade-offs. Machine-controlled variables are either fixed for a given configuration (e.g. platen and carrier rotation rates and directions, carrier configuration, insert material, and temperature) or adjusted during the process cycle (e.g. time, back-pressure, and pad conditioning technique and cycles). Process and materials dependent variables describe conditions on the incoming wafer (e.g. polishing step, material type and composition, pattern density and topography, wafer bow and warp, and stresses) and inputs to the process (e.g. slurry composition and flow, or pad material and structure).
In general, the CMP removal rate and uniformity vary with time during each wafer cycle and from cycle to cycle. This dynamic behavior imposes specific requirements on thickness measurement capabilities.
Monitoring choices for CMP
Three process monitoring approaches are used for thickness control in CMP: stand alone (SA) thickness measurement systems, end-point detection (EPD) devices, and on-line integrated thickness monitoring (ITM) systems [1].
Stand alone film thickness metrology
SA systems are the most commonly used due to their ability to provide full and accurate measurement parameters and ability to map product wafers. Performance specifications include fundamental measurement capabilities, spatial resolution, multi-layer determination, pattern recognition for product wafers, high throughput, ease-of-use, reliability, and up-time. Several high performance systems exist in the market.
For CMP processing, however, SA metrology is not the best solution. The dynamic nature of the process, the substantial variation with product type and materials, and the relatively low MTBF of the polishing system all require real time response.
End point detection systems
End-point detection (EPD) [2-4] based on in-situ sensors is being used in other processes such as plasma etching, and is adequate for CMP process control. The sensors may be classified into those that detect the end-point and those that continuously detect the thickness and removal rate [2]. EPD does not eliminate the need for frequent post-polish sample measurements on a SA metrology system, however.
In-situ sensors (ISS) are positioned in the interior of the processing area; measurements can not be carried out on actual micron-level size test sites. The data, averaged over a relatively large area, is thus less informative than data obtained by a SA tool. The ISS does not provide mapping capability or characterize local planarization, yet detailed uniformity information is often as important as the measured thickness. Interpretation of ISS signals is complex since the sensor is also affected by such irregular environment characteristics as electrical noise, slurry, mechanical movement, and temperature variations.
Integrated Thickness Monitoring (ITM) Systems
Integrated metrology combines the performance of a SA tool with the real-time response of an EPD. NovaScan ITM systems have metrology, pattern recognition and mapping capabilities comparable to SA measurement tools. The systems are very small, fitting inside the process equipment, and deliver very high throughput for wafer measurement while the next wafer is being polished. The wafer is measured in water while stationary. Several new technologies support these capabilities:
- Patented opto-mechanical scanning system allows measurement of the entire 8" wafer in a minimal footprint;
- Patented auto-focus system allows dynamic focusing on the wafer surface;
- Patented pattern recognition method allows global and local optical alignment on arbitrarily oriented wafers;
- Proprietary algorithms calculate film thickness from spectral data measured in water;
- Optical design allows a high signal-to-noise ratio in the visible and extended near IR spectral range with a small light spot, as well as wafer pattern imaging with varying optical contrast.
Although the measurement is carried out after the process is completed, not in real time, this limitation is not significant. Continuous and tight process control maintain the needed tolerances. The in-water measurement avoids drying before cleaning and enables rework, while locating the measurement system outside the actual processing area improves the metrology capabilities.
Process control
Advanced techniques to improve uniformity and increase overall equipment efficiency (OEE) [5] are based on measured data from processed wafers. Especially for CMP, a relatively young manufacturing technology, process control relies on phenomenology rather than on physical or chemical modeling. Algorithms used for process control include PID regulator, run-to-run control, neural networks, and design of experiments (DOE) [6-8].
ITM uses a simple and reliable control method based on self-adjusting control algorithms (SACA). The ITM concept dynamically adjusts both the control algorithm and the process during an actual production run (Fig. 1). In contrast, the DOE approach requires significant time and test wafers to calibrate the control algorithm. The SACA algorithm significantly reduces calibration time and effort.
ITM requires reliable communication links between the polisher, the measurement system, the factory host and a piggyback controller. Experimental results applying automated control to the polish time alone still show a significant increase in wafer-to-wafer uniformity.
Experimental
We tested the SACA approach in a real production mode on two different CMP tools—Strasbaugh's 6DS-SP and IPEC/Planar's Westech 372M. The computer running the control algorithms was connected to the NovaScan ITM system through a standard SECS-II interface. The experiments were intended to test the feasibility of the control algorithm; no communication was established between the ITM system and the CMP tool at this stage. The NovaScan 210 ITM system measured the mean thickness at 5 to 11 points on each processed wafer.
The operator manually fed the polisher parameters and the incoming initial oxide thickness to the SACA controller. After it suggested process parameters, based on previous process parameters and real-time measurement data, the operator manually entered the suggested polishing time for each spindle before each wafer polish. The experiment compared the mean oxide thickness of controlled and uncontrolled processes using different lots of the same product, polished on the same CMP tool.
Two major control methods are used in production CMP processes. One approach deliberately sets a relatively short polish time, so that the post-polish thickness will be close to the upper specification limit. This method prevents or minimizes wafer scrap due to overpolish, but more wafers have to be reworked. For this experiment, we processed several production lots of interlevel dielectric (ILD) products using the Strasbaugh 6DS-SP Planarizer with and without the SACA controller. The SACA reduced the number of reworked wafers to zero. After the operator set a "safe" polish time for the first wafer of each lot, the control algorithm quickly corrected and stabilized the process. The post-polish mean variation dropped from 7.5% to 4.3% standard deviation when using the control algorithm.
A second approach is more suitable for relatively stable production with a slowly decreasing removal rate. The polish time for the first wafer in the lot is set to be slightly longer than it should be to reach the target thickness, close to the lower specification limit. As the lot is processed, the removal rate gradually drops and the post-polish thickness gradually increases. The goal is to keep the last wafer in the lot near the high specification limit. This procedure keeps the polish time constant within the lot, but risks wafer scrap due to within-wafer (WIW) non-uniformity even though the mean is still in specification. With the SACA algorithm, the controller maintained the post-polish mean within ±300 Å from the target value.
The same process behavior was observed in a shallow trench isolation (STI) process running on the 372M Westech polisher. In this experiment the polish time for the first wafer was too high, and the first wafer in the lot was scrapped. Still, the control algorithm quickly corrected the polisher parameters and stabilized the process, bringing it well within the control limits.
These experiments assume that the incoming thickness of wafers within the lot is very stable. The control algorithm calculations use a single value for the pre-polish thickness. This assumption is not always correct. Large variations in the incoming thicknesses will modulate the controlled post-polish mean thickness. For more accurate control, the system must measure each incoming wafer thickness and use this data as a pre-polish input to the process control algorithm.
Because the calculation of process parameters is based on previously polished and measured wafers, uncertainty exists for the first wafer(s) of each lot. This problem can be solved by using the information from previously processed lots to calculate process parameters for the first wafer of the next lot. The difference in pattern density, materials, and so on between products must be considered.
Software Architecture and Configuration
SACA control based on ITM measurements requires communication channels between the ITM system, the CMP polisher, the fab host computer, and the controller. Communication between the ITM system and the fab host supports statistical process control (SPC) and recipe management. Communication between the ITM system/fab host and the CMP polisher provides the automatic process control.
Different communication schemes are possible. All these schemes must provide data collection on the fab host computer, centralized recipe management for both CMP tool and ITM system, and real-time process control. The NovaScan system uses the SECS-II protocol.
Conclusion
A fast, reliable ITM system allows dynamic, self-adjusting process control and thus can significantly improve CMP equipment efficiency. This in turn improves the COO and yield of the CMP process steps.
This report contains materials from 2 papers that were originally published at the 2nd and 3rd International CMP Planarization Conference (CMP-MIC) in February 1997 and February 1998, Santa Clara, California. Based on an article which first appeared in the April 1998 issue of European Semiconductor magazine.
References
1 G. Dishon et al, "On-line Integrated metrology for CMP Processing," 1996 CMP-MIC Proceedings, pp. 273-276.
(Back to article)
2 G. Dishon et al, "In-situ Detection of Film Thickness Removal During CMP of Oxide and Metal layers," 1996 CMP-MIC Proceedings, pp. 256-262.
(Back to article)
3 Robert Tolles et al, "Performance of the Applied Materials CMP System for Interlevel Dielectric Polishing ProcessUsing SinStep and Multi-step Process Sequence," 1996 CMP-MIC Proceedings, pp. 201-208
(Back to article)
4 A. Hu et al, "Real Time Control of Chemical Mechanical Planarization (CMP) Process," 1996 CMP-MIC Proceedings, pp. 235-240
(Back to article)
6 Hung-Wen Chiou et al, "PID Run to Run Control of CMP Removal Rate", 1997 CMP-MIC Conference, 1997 ISMIC - 200p/97/0375
( Back to article)
7 Albert Hu, et al, "Application of Run by Run Controller to the Chemical-Mechanical Planarization Process", IEEE Proceedings of the 16th International Electronics Manufacturing Technology Symposium, September 1994.
( Back to article)
8 James Moyne, et al, Sematech AEC/APC Workshop IX, Lake Tahoe, 9/1997.
( Back to article)