Integrated process control using an in situ sensor for etch
04/01/2002
by Z. Sui, C. Frum, H. Shan, D. Lymberopoulos, Applied Materials Inc., Sunnyvale, CA
B.J. Su, Yi-Hong Chen, TSMC Fab 6, Taiwan, ROC
Overview
The migration to tighter geometries and more complex process sequence integration schemes requires having the ability to compensate for upstream deviations from target specifications. Doing so ensures that downstream process sequences operate on work-in-progress that is well within control. Because point-of-use visibility of work-in-progress quality has become of paramount concern in the industry's drive to reduce scrap and improve yield, controlling trench depth has assumed greater importance. An integrated rate monitor for etch-to-depth (dual damascene trench etch and spacer etch) applications has been developed for controlling this parameter [1].
In dual damascene applications, the trend is toward reducing the thickness of the middle stop layer, or, preferably, eliminating it, to improve the effective dielectric coefficient of the interconnect. In spacer etching, the use of thinner gate oxide makes its integrity and the avoidance of punch-through crucial for subsequent implant steps.
Using interferometry
Integrated interferometric tools employ light wave reflection and interference phenomena to determine etch progress characteristics (e.g., depth and etch rate). The interferometric signals (i.e., fringes) that are recorded at normal or off-vertical incidence as the dielectric film is etched can be used to trigger etch completion once a dialed-in target depth has been achieved (dual damascene trench etch) or when the targeted remaining thickness is met (spacer etch).
Traditionally, interferometric techniques have been used for measuring blanket film thickness [1-5]. However, patterned wafers produce interferometric signals that are difficult to deconvolve as multiple etch fronts contribute to the observed signal. The temporal normal reflected light intensity during etching contains frequency components introduced by etching the photoresist mask surface and the exposed dielectric surface. To extract the frequency component associated with etching dielectric films and amplify its signal-to-noise ratio, the interferometric tool employs two techniques.
One technique uses the polarization effect. When normal-incident unpolarized light is reflected from a dielectric surface, the reflectivity of the linear polarization component parallel to the direction of the trench differs from the one perpendicular to the trench. This difference is more significant when the trench width is smaller than the wavelength used for detection. At the same time, the reflectivity of both polarized components from the featureless area of photoresist is almost the same. Therefore, the ratio of these two signals can greatly reduce the interference signal from the featureless area of photoresist.
The second technique involves digital signal processing (DSP), which employs several real-time filters to average out the magnetic field and separate the undesired photoresist signal from the dielectric film signal. The real-time signal consists of three frequencies originating from the photoresist (ωPR), dielectric film (ωox), and the magnetic field modulation (ωmag), respectively. (The presence of a magnetic field modulation was inherent in the system used for the evaluation.) The signal attributed to the dielectric film is that remaining after filtering has first removed the contribution from the magnetic field and then the contribution from the photoresist.
The integrated rate monitor (iRM) uses both of the techniques described above to isolate the signal emanating from the dielectric and enhance the signal-to-noise ratio (SNR). (The reflected light gets cross-polarized to filter out contributions from the featureless photoresist; digital signal processing is further employed to enhance the SNR.) The monitor requires no external light source for its operation; instead, it uses the naturally emitted light from the plasma in the process chamber. The reflected interference signal from the wafer is collected in real-time by the endpoint module placed above the wafer on top of the chamber's lid.
Dual damascene trench etch
The reduction in the effective dielectric constant (keff) of the interconnect is the dominant force behind the proliferation of efforts to optimize process integration and the development of a growing array of low-k interlevel dielectric (ILD) materials. The keff can be approximated by the sum of dielectric coefficients of the films weighted by their respective thicknesses.
One means of reducing the keff is to select a low-k ILD material. To date, undoped silica glass (USG, k ≈4.2) and fluorinated silicate glass (FSG, k ≈3.0-3.7) have been used as ILDs. As technology moves beyond 180nm, however, other low-k (<3.0) dielectrics are being considered. Typically, a wide range of keff values is achieved as a result of compromises in integration schemes dictated by the need for manufacturability. The keff penalty resulting from such compromises can be as high as 20-25%.
Barrier films are also used in the dual damascene structure for several applications, including copper diffusion barrier, etch stop, and hardmask. The k values for these films range from 4.0-5.0, which is substantially lower than those for the traditional silicon nitride barrier films (k ≈6.5). At least two barrier/etch stop films are present for each ILD layer. A typical eight-metal-layer device will require 16 layers of barrier films. The k< value for these films can thus make a significant contribution to keff. Consequently, reducing the thickness of the barrier layer, or eliminating the etch stop layer, would reduce the keff and permit ILD materials to be extended to the next device generation with minimal need for new material/process-sequence combinations — at nominal risk.
 Figure 1. Typical interferometric trace with resultant structure. |
If the middle etch stop layer between the trench ILD and via ILD is eliminated, accurate control of the trench etch depth becomes imperative. The integrated rate monitor addresses this need in situ. It can be activated at the endpoint of the hardmask open step and monitored during the subsequent dielectric etch step. Compared to timed etches, the "pure" dielectric trace from the interferometric module (Fig. 1) permits much more precise monitoring and control. In fact, interferometric control has demonstrated repeatable etch depth results (1% 1σ, wafer-to-wafer) in the absence of an embedded etch stop layer. The combination of optical emission spectroscopy (OES) and interferometry ensures wafer-to-wafer repeatability by delivering consistent etch-to-depth results in the oxide layer regardless of variations in the thickness of the overlying hardmask layer on different wafers.
 Figure 2. Integrated rate monitoring compensates for etch rate variations. |
Figure 2 shows the value that interferometry offers to maintain process control when the etch process drifts due to etch rate variation. In this specific example, the RF power was deliberately varied within ±10% of the setpoint of 1000W while etching a simple stack of blanket SiO2 on Si. The RF power variation (900-1100W) naturally had an impact on the etch rate and etch depth for a timed process (59 sec, based on the setpoint of 1000W), as shown on the left side of Fig. 2. The same RF power variation was applied while the process was under interferometric control; although the wafers were subject to the same variation in etch rate, the etch depth was tightly controlled (right side of Fig. 2).
 Figure 3. Uniformity and repeatability wafer-to-wafer using integrated rate monitoring. |
Figure 3 shows the repeatability achieved in trench depth when using the interferometric sensor. In this case, the integrated measurement system was triggered by an OES as the etch front moved through the barrier anti-reflective coating (BARC) into the low-k film. The combined solution delivers repeatable low-k dielectric etch-to-depth results regardless of incoming BARC thickness variation.
Spacer etch
Reliable etch control is crucial as gate sidewall spacer thickness (100-500Å) and gate oxide thickness (25-50Å) are decreased to improve device performance. In this case, an interferometric sensor can be used to control a soft landing on the gate oxide.
 Figure 4. Typical optical emission and endpoint traces during spacer etch; the method can be used as a predictive endpoint for soft landing. |
By providing a timely trigger to switch from a process step characterized by a high etch rate and low selectivity to oxide to a step that exhibits low etch rate and high selectivity to oxide, an integrated interferometric monitor can minimize oxide loss as compared to the same step-change trigger based on optical emission. A delayed trigger of a single second can cause removal of ~8Å of gate oxide (Fig. 4).
A typical interferometric trace for spacer etch applications is shown in Fig. 5 and illustrates the consistent behavior of the trace regardless of differences in the thickness of the incoming nitride film. The nitride etch portion of the trace has the appearance of a fringe with a maximum and a minimum clearly visible. The positions of the maximum and minimum in the trace signal are related to constructive and destructive interference in the reflecting light beams and depend on the remaining nitride film thickness and its refractive index. It is important to observe that the locations of the two extremes are independent of the incoming nitride film thickness and occur at the same remaining film thickness.
The efficacy of using an integrated rate monitor while performing spacer etch to determine the thickness of the remaining film is based on the above observations. As the nitride layer is etched, the software can recognize the occurrence of the minimum in the real-time signal and identify a trigger point. Although trigger point time depends on the incoming nitride thickness, it will always occur at the same time interval before the nitride endpoint, providing that the etch rate remains the same from wafer to wafer.
 Figure 6. Plot of the remaining nitride thickness Vs. the landing time for 700Å SiN wafers. |
The remaining nitride film thickness has been measured optically using a commercially available metrology tool based on ellipsometry. As expected, it decreases in a linear fashion as the landing time increases. Etch depths predicted using interferometric signals and optically measured layer thickness correlate closely (Fig. 6).
Conclusion
The examples presented in this article demonstrate that the integrated rate monitor, using polarization and digital signal processing, enhances control of etch-to-depth processes and can also be implemented as a predictive endpoint in a wafer manufacturing environment for dual damascene trench etch and spacer etch applications.
References
1. R.M. Azzam, N.M. Bashara, Ellipsometry and Polarized Light, North Holland, Amsterdam, 1977.
2. M. Born, E. Wolf, Principles of Thin Film Optics, Pergamon, New York, 1965.
3. P.A. Heimann, R.J. Schultsz, J. Electrochem. Soc., 131, 881, 1984.
4. P.A. Heimann, J. Electrochem. Soc., 132, 2003, 1985.
5. R. Bruckner, J. Canteloup, J.P. Vassilakis, Solid State Technology, 262,
262, 1997.
Acknowledgments
iRM is a registered trademark of Applied Materials Inc.
For more information on the subjects mentioned in this article, please contact Dimitris Lymberopoulos, senior product marketing manager, in the Dielectric Etch Division of Applied Materials Inc., Sunnyvale, CA; ph 408/584-1679, email [email protected].