Jacques Grange and David Pays, Surface Technology Systems; and Denis Cattelan and Jean Philippe Vassilakis, Jobin Yvon, Horiba Group

After more than 20 years of development by a host of R&D laboratories around the world, microelectromechanical systems (MEMS) have evolved to support a maturing, commercially successful industry. MEMS have finally achieved their long-promised potential as a disruptive technology.

A key process used to fabricate many MEMS devices is deep reactive-ion etching (DRIE) of silicon, which enables the creation of vertical trenches or holes that are typically in the 10- to 750-µm range. These deep trenches define the 3-D microstructures comprising the MEMS device. A widely used DRIE method is the Bosch process, which was patented by Robert Bosch GmbH (Stuttgart, Germany) in 1994 and then licensed to Surface Technology Systems (STS) in Imperial Park, Newport, UK.¹

To improve production yields in MEMS fabrication, it is essential to etch 3-D MEMS structures to a consistent depth wafer after wafer, requiring a high level of control. However, after-the-event ex situ control of etch depth is expensive. To achieve adequate real-time depth control without the need for ex situ metrology, advanced in situ metrology, fed back into the processing software, is required to terminate the etch process at the correct depth and hence obtain the required feature profile. Depth monitoring becomes ever more challenging as the structures required for MEMS devices deepen and their aspect ratios increase.

This article discusses the TDM-200 polarimetric camera from Jobin Yvon, Horiba Group (Longjumeau, France), which performs in situ control of DRIE silicon trench etch depth in MEMS manufacturing. The article details tests that were performed by STS to characterize the camera, demonstrating its ability to improve process control in a range of applications.

In Situ DRIE Silicon Etch-Depth Monitoring

Before the emergence of the Bosch process, conventional interferometry with a parallel or focused beam could be used to reasonably monitor the aspect ratios, feature sizes, and depths achievable by mainstream silicon etch processes.² STS began to expand the capabilities of the Bosch technique by developing the ASE DRIE process.^3–5 Nevertheless, even for early MEMS applications, aspect ratios, feature depths, and exposed areas restricted the use of single-spot or parallel-beam laser interferometry. It was found that signal strength was reduced both by small exposed areas (e.g., insulation trenches) and the loss of reflected-beam phase caused by multiple reflections in the trenches.

Applications such as silicon waveguides benefited from the development of the white-light interferometer by Jobin Yvon and its qualification by STS. The flexibility of the filtered-white-light, focused-beam technique allowed monitoring down to a depth of 10 µm on trenches that were typically 5 µm wide. However, most deep-etch processes still could not be monitored.

Figure 1: Schematic diagram of the Twin-Spot interferometric camera.

In 1997, Jobin Yvon's thin-film division, then Sofie Instruments, launched a twin-spot interferometric camera, which resolved the main issue encountered with a single-spot laser system: the impossibility of getting a spot to straddle a deep trench wall, be reflected at the top and bottom of the trench, and then recombine to form a clean, reliable interferometric signal.

Figure 1 illustrates the configuration of the Twin-Spot interferometric camera. Using a Wollaston prism, a single laser beam is split into two phase-locked beams that are projected onto the wafer surface. With Jobin Yvon's established confocal arrangement, the two reflected beams can be imaged simultaneously, recombined through the same prism, and measured by the detector. This method proved successful in challenging depth-monitoring applications.

Because of its cyclic nature, the Bosch process is complicated. Essentially, it alternates between three steps:

• The passivation, or polymer deposition, step, which covers the entire trench and the mask with a thin C₄F₈-based fluoropolymer.

• The passivation removal step, which removes the polymer from the bottom of the trench.

• The silicon etch step, a low-bias SF₆ etch of the exposed silicon at the bottom of the trench that allows overall directional (usually vertical) progression of the etch step.

When large features are etched continuously, the twin-spot interferometer can be used to directly recover a normal, near-sinusoidal interferometric signal. However, in the case of the Bosch process, the near-sine wave corresponding to the silicon etch step is interrupted by the passivation and passivation-removal steps. In such circumstances, an external trigger for gating the signal at the analysis stage is not readily available.

The signal can be stitched to reconstitute a signal characteristic of the silicon etch alone. In fact, the passivation-removal and etch steps usually combine to form what is commonly referred to as the etch step. The triggering signal used in this case is the start of the etch step. A delay approximating the passivation removal time (which is not usually constant throughout the process) is introduced to enable the signal gating needed at the analysis stage. Calibration using an external metrology system is then required to establish the optimum delay time for specific process conditions.

Figure 2: Typical complex signal obtained using the Bosch process.

Early ASE processes typically operated with slow switching times and a relatively low etch rate. Etch time was typically 13 seconds, passivation time was 7 seconds, and the silicon etch rate was ~3–8 µm/min. These processes usually created a relatively shallow etch depth of ~100 µm. As illustrated in Figure 2, the process generated a very clear signal. Gating and stitching were simple, resulting in reliable, accurate depth targeting.

As etch depths and rates increased, new challenges arose. Because deeper features invariably have rougher bottoms, the signal reflected by the trench bottom became so faint that it lost its phase definition. Faster etch rates required higher-performance etch tools with laser beams that must traverse a very bright plasma region. Rapidly changing, intense plasma brightness was superimposed on the signal, generating noise that eventually drowned out a faint signal.

Ever-deeper features, ever-faster etch rates, and ever-smoother trench sidewalls stretched the twin-spot laser interferometer technology to its limit. The requirement for smoother trench sidewalls led to switching times that were so fast that less than a complete interference period per etch cycle was obtained, thereby rendering reliable signal recognition and stitching nearly impossible. Moreover, the now-widespread use of parameter ramping—the continuous adjustment of process parameters—rendered accurate gating of the signal extremely difficult, since the time taken to remove the polymer was varied throughout the process.

As a result, the twin-spot laser interferometer, when used in conjunction with high-performance ASE processes, typically performed adequately only down to an etch depth of 30 to 50 µm. In the case of processes with fast switching times and smooth sidewalls, the signal simply could not be interpreted.

Polarimetric Camera

To address the limitations of the twin-spot laser interferometer, the TDM-200 polarimetric camera was developed. The primary aim of the system is to measure the depth of a trench as it is being etched. To do so, two beams interfere with each other. One beam strikes the bottom of the investigated feature, while the other strikes the mask surface, serving as a height reference. The system measures the difference between the optical paths taken by the two beams.

Figure 3: Schematic diagram of the polarimetric camera.

The instrument setup is shown in the schematic diagram in Figure 3. A laser diode, combined with a polarizer, emits a polarized light beam that is split into two perpendicular, linearly polarized beams by a Wollaston prism. The two beams impact the sample at two different heights and are reflected back toward the prism, where they recombine. Because the reflection on the sample surface is at normal incidence, the respective polarizations of the two beams remain linear. What changes is the respective intensity of the beams and their relative phase. The information carried by the phase shift is valuable because it is closely related to the optical path difference. The phase shift is determined using polarimetry.⁶ The polarization analysis is performed using the combination of a 3000-rpm rotating quarter-wave plate and a linear polarizer. The intensity of the light that hits the detector after it has gone through the entire optical chain (including the two wafer surfaces) is measured as a function of time. An encoder is attached to the rotating wave plate and the intensity measurement is synchronized. The signal is then analyzed using a Fourier analyzer.

Signal Analysis

Signal analysis begins with a study of the polarization state of the beam leaving the Wollaston prism after the two reflected beams have recombined. The intensity (I) as a function of time (t) is derived by the equation shown in Figure 4, where r₁ and r₂ represent the reflectivity of both sample sites (mask and etched feature), Δ is the phase shift between them, A is the azimuth of the polarizer, and ω is the wave plate rotation speed.

Figure 4: Equation used to derive the intensity of the beam leaving the Wollaston prism after the two reflected beams have recombined.

A Fourier analysis performed on the signal provides access to the fundamental h₀ and the harmonics h_2s, h_2c, h_4s, and h_4c, from which Δ is obtained. Although Δ appears in h₀, its extraction from harmonics 2 and 4 is straightforward, as shown in the following ratio:

The measured Δ is the instantaneous phase shift between the two perpendicularly polarized beams reflecting on the sample. Finally, the phase must be unwrapped, and the corresponding relative depth (d) between the two surfaces is obtained by the following equation (the factor 2 means that the light travels twice the distance between the surfaces):

Advantages of the Polarimetric Method

While the twin-spot camera used only the intensity of the interfering beams to perform real-time measurement of trench depth during the etch process, the polarimetric camera exploits the resulting polarization state to extract the same information in a far more elegant manner and with far more precision. Indeed, the phase-shift measurement does not require optical fringe analysis, and its evaluation from the ratio above is independent of the continuous intensity resulting from the modulation of polarization. Hence, the evaluation is independent of the influence of source intensity drift, the window transmission factor, the evolution of the reflectivity ratio between mask and etched feature, and bright plasma emission.

In addition, the polarimetric camera has a theoretical accuracy two orders of magnitude greater than that of the twin-spot camera (0.5 versus 80 nm). A comparison between the interferometric signal intensity obtained using a twin-spot camera and the phase signal (Δ) obtained using a polarimetric camera is presented in Figure 5.

Figure 5: Comparison between (a) the interferometric signal intensity obtained using the twin-spot camera, and (b) the phase signal (Δ ) obtained using the polarimetric camera.

In the case of the Bosch process, the camera can deal with any etch/deposition cycle without a process trigger or process-specific data-treatment optimization, since a special procedure to access the phase shift is not required.

Testing the Polarimetric Camera

The polarimetric camera was evaluated on a Multiplex^Pro ASE^HRM DRIE tool. The tool's decoupled, very intense plasma source processes the wafer by diffusing ions and radicals into the main chamber below. The source was chosen for the work discussed in this article to demonstrate that the polarimetric camera is insensitive to noise emanating from the plasma light. In fact, the laser beams traverse the brightest part of the plasma region, as shown in the schematic diagram in Figure 6.

Figure 6: Schematic diagram of the experimental setup.

In the figure, the camera is shown mounted on a swivel arm, which decoupled it mechanically from the top of the source. This setup ensured that thermal distortion of the source chamber would not perturb the experiment through either lateral drifting of the laser spots or a change of the beams' angle of incidence on the features of interest. The normality of the beam and its positional accuracy are key to performing accurate measurements.

Test Wafers. The samples used to carry out the experiment were silicon wafer pieces mounted at the centers of resist-coated silicon carriers. Cooling grease was used to achieve optimum heat transfer through the carrier to the wafer susceptor. Although wafer pieces with different feature sizes were used, most features were in the size range of 500 µm to 1 mm. The mask material used on the wafer pieces was either photoresist or silicon oxide (SiO₂). The carriers were electrostatically clamped during processing, and the wafer pieces were kept at temperatures typically ≤80°C using helium backside cooling.

Optical Setup. Before the process begins, the laser spots must be set up so that one spot is directed onto a masked area and the other onto an adjacent open area. The laser spots are approximately 50 µm wide and 250 µm apart. It is relatively easy to position the spots as required for feature sizes >200 µm. The tip and tilt of the camera are critical and must be adjusted so that the beams are directed onto the feature of interest with a normal incidence.

Process Variables. To test the camera's capabilities, a wide range of processes were performed using various process parameters. To obtain a baseline, a conventional etch process with a low etch rate and relatively long switching times was performed. That process did not challenge the camera's ability to monitor the etch depth.

After the baseline had been established, the etch rate was increased dramatically to test the camera's ability to monitor very fast phase changes without losing track of etch depth. Then switching times were decreased substantially to monitor the depth at which the twin-spot interferometer failed, and parameter ramping was introduced to test the camera's ability to monitor etch-rate changes. Next, the instrument's robustness was tested when processing was interrupted. The etch depth was also increased to determine the unit's limitations. Finally, the camera was tested for its ability to produce a clean signal when the size of the feature of interest approached the laser spot's size.

During the experiments, data acquisition was started manually during the gas-stabilization phase of the process (typically 10 seconds before the plasma was struck). When the process ended and the plasma was extinguished, data acquisition was stopped manually.

Ex Situ Metrology. The camera's measurements were confirmed using either a scanning electron microscope (SEM) from Leo Electron (now the Nano Technology Systems Division of Carl Zeiss NTS in Oberkochen, Germany) or, for the deepest etch, a white-light interferometric profilometer from Zygo (Middlefield, CT).

Test Results

Fast Etch Rate. The first test was performed to evaluate how well the polarimetric camera monitors a silicon etch process under high-etch-rate conditions. The trace shown in Figure 7 was acquired for an etch rate of 15.9 µm/min. The wafer pieces were coated with a resist mask and the process was run for 22.5 minutes. At the end of the acquisition period, the camera recorded a depth of 357.5 µm, while the SEM recorded a depth of 366.2 µm.

The difference between the measurements is attributable to the fact that the camera and the SEM did not necessarily monitor the same features. Etch-rate uniformity was found to be poor on the mounted wafer pieces, primarily because of edge effects. Nevertheless, the agreement between the in situ and ex situ measurements was still within 3%.

Figures 7a and 7b show measurements acquired from the camera during the etch and deposition cycles and a corresponding SEM image, respectively. The flat section of the signal illustrated in the inset in Figure 7a corresponds to the deposition step, during which the depth did not vary greatly because the polymer film thickness was negligible. The sloped section seen in the inset represents the etch step, during which the depth varied over time. The slope corresponds directly to the instantaneous etch rate: the steeper it is, the faster the rate. Similarly, the slope of the trace at any point in time represents the average etch rate resulting from the entire Bosch cycle.

The camera measures depth regardless of how fast (or slow) the etch rate is. In this case, the instantaneous etch rate was on the order of 33 µm/min. In fact, the camera's optics, electronics, and software are designed to monitor instantaneous etch rates up to 100 µm/min.

Fast Cycle Times. In the second experiment, parameter ramping and fast switching times were applied (etch and deposition steps were ~2 seconds). A wafer piece was masked with SiO₂ and the process time was 22 minutes. At 4 µm/min, the etch rate was much lower than in the first experiment. The camera recorded a depth of 100.2 µm, while the SEM recorded a depth 104.5 µm. The error was within measurement accuracy.

The first and second experiments demonstrated that the camera is not affected by the mask material. Either an oxide or a photoresist mask can be used as long as the reflected signal from the mask is strong enough.

Figures 8a and 8b show measurements acquired from the camera during a fast etch process and a corresponding SEM image, respectively. The etch depth (red curve) shown in Figure 8a is not linear because it is dependent on the etch rate (blue curve). Even though the etch rate varied during the process, as expected, and the deposition and etch cycle times were very short, the camera monitored the etch depth accurately.

Process Interruption. In the third experiment, the process conditions were identical to those used in the fast-cycle-time experiment, except that a photoresist mask was used in place of an SIO₂ mask. Approximately two-thirds of the way through the experiment, the etch process was stopped for about one minute. The result of the interruption appears in Figure 9 as a step, where the etch depth did not change. When the process resumed, the camera continued to accurately monitor the depth as if the process had not stopped, leading to results very similar to those shown in Figure 8. At an etch rate of 4.44 µm/min, the camera recorded a depth of 110.9 µm, while the SEM recorded a depth of 108.8 µm.

Figure 9: Polarimetric camera data showing a process that was interrupted and then resumed.

Deep Etch. In the fourth experiment, a smooth etch process was run for nearly two hours. As in the third experiment, a photoresist mask was used. Some process parameters were ramped during certain phases of the etch process. The aim of the experiment was to let the process run until the camera failed to monitor the etch depth.

To achieve a very accurate ex situ measurement, the etch depth was measured using a Zygo interferometric profilometer. A SEM micrograph of the feature probed is presented in Figure 10. The discrepancy between the depth recorded by the camera and that recorded by the profilometer was just over 1%, well within the measurement error that can be expected from two different optical instruments probing two somewhat different areas of an etched feature. At an average etch rate of 4.3 µm/min, the camera recorded a depth of 503 µm, while the profilometer recorded a depth of 497 µm.

Figure 10: SEM micrograph of the deepest trench that could be monitored using the interferometric profilometer.

Although the aim of this experiment was to determine the greatest depth that can be measured by the polarimetric camera, that goal could not be achieved because of roughness at the bottom of the trench. In this case, the peak-to-valley value of the roughness was measured to be ~1.2 µm. When that value reaches approximately twice the laser wavelength (670 nm), the intensity and phase of the beam reflected off the bottom of the trench cause the lock onto the phase shift to be lost, rendering further monitoring of the etch depth impossible. As the roughness increases, the energy ratio deteriorates (decreases), as illustrated in Figure 11. Monitoring the energy ratio provides a good indication of how accurate the measurement will be.

Figure 11: Energy ratio of reflected beams. A loss of signal occurred because of roughness at the bottom of the trench.

Conclusion

An evaluation demonstrated the adequacy, robustness, reliability, and repeatability of an in situ DRIE silicon trench depth-monitoring method involving the use of a polarimetric camera on a DRIE system. The method resulted in improved control of DRIE silicon trench etching in MEMS manufacturing applications. It was found that the camera comfortably monitored etch depths at instantaneous silicon etch rates as high as 33 µm/min. It was robust even at very fast process switching times. In addition, the camera was shown to be adequate for monitoring features with either photoresist or silicon dioxide masks, the materials most commonly used in the industry. The monitoring method was successful even when the process was interrupted and resumed. Finally, the camera can monitor trench depths down to through-wafer etch levels of 500 µm.

The insensitivity of the new method to changes in the reflectance of probed surfaces overcame most of the limitations associated with an interferometric approach. The most accurate control tests showed an agreement between in situ and ex situ measurements on the order of 1%. However, the experiments discussed in this article did not aim specifically at proving the accuracy of the camera, which will require further work.

The method is limited by the laser wavelength and spot size. In these experiments, the signal was lost when the roughness of the etched feature surfaces reached a level of approximately 1.2 µm peak to valley. Moreover, because of the nature of the laser spot size, the width of the endpoint feature must be larger than approximately 200 µm.

In conclusion, many MEMS applications can benefit from improved process control and the higher yields resulting from the combination of a high-performance silicon etch tool and real-time monitoring of the etch process.

References

1. F Laermer and A Schilp, Patent DE4241045 (U.S. Pat. No. 5,501,893), 1994.

2. J Canteloup, R Brückner, and T Moore, "Interferometry and Imaging," European Semiconductor (March 1995): 14–17.

3. J Hopkins et al., "The Benefits of Process Parameter Ramping during the Plasma Processing of High Aspect Ratio Silicon Structures," in Proceedings of the MRS Fall Meeting 1998, vol. 546 (Warrendale, PA: Materials Research Society, 1999), 63–68.

4. SA McAuley et al., "Silicon Micromachining Using a High-Density Plasma Source," Journal of Physics D: Applied Physics 34, no. 18 (2001): 2769–2774.

5. LA Donohue et al., "Developments in Si and SiO₂ Etching for MEMS-Based Optical Applications," in Proceedings of SPIE, Vol. 5347: Micromachining Technology for Micro-Optics and Nano-Optics II (Bellingham, WA: SPIE, 2003), 44–53.

6. RMA Azzam and NM Bashara, Ellipsometry and Polarized Light (Amsterdam, NY: North Holland Pub., 1977).

---

Jacques Grange is systems research manager in the R&D group at Surface Technology Systems (STS) in Imperial Park, Newport, UK. Since joining the company in 1994 as a process engineer, he has contributed to the early development of the ASE process and endpoint detection products. He received degrees in material physics and integrated electronic devices from the National Institute of Applied Sciences of Lyon in 1990 and 1991, respectively. (Grange can be reached at +44 1633 652400 or jacques_grange@stsystems.co.uk.)

David Pays is an R&D engineer at STS, where he is responsible for the development of endpoint detection systems. Before joining the company in 2002, he received a degree in physics, optics, and electronics from the National Higher School of Physics in Marseilles, France. (Pays can be reached at +44 1633 652400 or david_pays@stsystems.co.uk.)

Denis Cattelan is an R&D optical engineer in ellipsometry and process control in the thin-film division of Jobin Yvon (Chilly-Mazarin, France), where he has been since 2000. He received a degree in physical and instrumental optics from the Ecole Supérieure d'Optique in Orsay, France. (Cattelan can be reached at +33 1645 41300 or denis.cattelan@jobinyvon.fr.)

Jean Philippe Vassilakis heads the software team in the thin-film division of Jobin Yvon. A software developer for real-time control since 1983, he cofounded Sofie Instruments, which became part of Jobin Yvon in 1983. Vassilakis graduated from the Orsay Technical Institute in 1982 and received an MS in software engineering from the Centre National des Arts et Metiers in Paris in 1990. (Vassilakis can be reached at +33 1645 41300 or jpvassilakis@jobinyvon.fr.)