Ayelet Dag and Vladimir M. Rubinstein, Tevet Process Control Technologies; and Yitzhak Gilboa and Steven Hedayati, Cypress Semiconductor

Controlling oxide-layer thickness following chemical-mechanical polishing (CMP) is critical to die yield and device reliability. In addition to experiencing normal process variations, oxide thickness in shallow-trench isolation (STI) structures may vary greatly depending on trench width and trench etch uniformity. Conventional methods for measuring oxide-layer thickness involve pointing a small (<50-µm) light spot at a large, nonpatterned wafer area, such as a large pad or scribe line. In order to point the beam at the nonpatterned area, a vision system and pattern-recognition algorithms are used in conjunction with an accurate mechanical stage to move the wafer to the target measurement coordinates.

However, the need to isolate the measurement site from its surrounding films by accurately positioning a small spot of light confines oxide-thickness measurements to relatively wide (>50-µm) trenches. In the case of STI structures, oxide and nitride thicknesses in narrow (0.2–1.0-µm) trenches and dense patterns do not always correlate with thicknesses measured in wide trenches. Furthermore, thickness problems appearing in narrow trenches on the die may not appear in the pad areas monitored by conventional instrumentation. Therefore, the performance of accurate process control requires a technique that can measure film thicknesses within the die and on dense structures.

This article discusses an in situ thickness measurement system (IsTMS) from Tevet Process Control Technologies (Yokneam Moshava, Israel) which uses a large-spot-size broadband Fourier-transform reflectometry (LSBFTR) method for measuring silicon nitride and silicon oxide film thickness in STI applications. The IsTMS integrated metrology module was the subject of experiments in which it was compared with stand-alone and analytical tools.

STI Architecture

STI architecture is formed during wafer processing as part of the effort to electrically isolate adjacent transistors. Forming a dielectric isolation layer between the source/drain parts of two neighboring transistors prevents current leakage (crosstalk) between them. The STI process is used to manufacture most advanced integrated circuits at technology nodes below 0.25 µm.

A typical STI process flow includes several steps.¹

• Pad oxide growth. A thin (100-Å) layer of silicon oxide is grown using thermal oxidation furnaces. The silicon oxide serves as an intermediate layer between the silicon substrate and the silicon nitride that is deposited on top of the oxide.

• Silicon nitride deposition. A 1500-Å layer of of silicon nitride is deposited on top of the pad oxide layer using chemical vapor deposition (CVD) or low-pressure CVD. The silicon nitride forms an etch-stop layer and acts as a hard mask for the trench etch steps.

• Trench layer lithography. Following nitride deposition, a lithography process is implemented to form a photoresist mask for the trench etch steps.

* Trench etch. A nitride dry-etch step is followed by a silicon etch step to create deep (5000-Å) trenches within the silicon.

• Trench fill. After the removal of the photoresist mask, CVD is used to fill the trenches with silicon oxide. The silicon oxide layer forms a thick cover over the nitride area between the trenches.

• CMP oxide removal. CMP removes the oxide and stops after all oxide above the nitride has been removed.

• Nitride and pad oxide strip. The nitride is removed, leaving the trenches filled with oxide and a clear silicon area between them for transistor formation.

Figure 1: Schematic drawing of an STI structure before nitride strip.

Figure 1 shows a cross section of a typical STI structure. The silicon substrate after trench etch is shown in gray. The silicon trench usually has sloped walls as a result of the etching process. The silicon oxide remaining after CMP is shown in blue. Although the schematic indicates that the oxide surface is flat, the surface of the oxide remaining after CMP may exhibit thinning at the center of the trench—a phenomenon called dishing. The figure also shows the nitride layer after CMP.

Measuring Film Thickness

Film thickness measurements using classical spectroscopic reflectometry are based on the interference between light beams reflected from the top surface of the film and light reflected from the bottom surface. The intensity of the reflected light, as a function of wavelength, depends on the film thickness D and the complex index of refraction N of each medium through which the light passes. The reflectance from a single homogeneous film layer, for an illumination at normal incidence, is described in the equation

where

N_i = (n_i – jk_i),

n_i = the refractive index of layer i,
k_i = the extinction coefficient of layer i.²

Figure 2: Schematic drawing of light reflected from a single film layer with thickness D1 and complex index of refraction N1. The indexes 0 and 2 relate to ambient and substrate, respectively.

Figure 2 illustrates the reflection from a single layer with thickness D₁ and complex index of refraction N₁.

When light illuminates a large wafer area where adjacent films with different optical properties and thicknesses are present, the resulting reflectance is the sum of the reflectance from each of the films. Figure 3 presents a simplified example of a reflectance spectrum (excluding dispersion) expected from two different adjacent films as a function of wave number (λ^–1).

Figure 3: Simulated reflectance spectrum (excluding dispersion) as a function of wave number from two adjacent films with different optical properties and thicknesses.

Tevet's Fourier-transform spectrometric reflectometry method calculates the thickness of layers by initially applying a Fourier transform (FT) to the reflectance curve—reflectance versus wave number R(λ^–1)—and by converting the result into a spectrum, denoted by the equation S(ƒ) = FT[R(λ^–1)]. The relative frequencies ƒ_m of the peaks in the S(ƒ) spectrum are then determined, and the peaks associated with specified thickness values are isolated for further analysis using proprietary mathematical algorithms, which calculate the corresponding thicknesses accurately.

For example, applying the Fourier-transform technique and the fitting algorithms to a reflectance curve such as that shown in Figure 3 results in a spectrum with two peaks, representing the thicknesses of two films (see Figure 4). A change in a layer's thickness causes a shift of the peak relative frequency associated with that layer, allowing direct thickness monitoring from an analysis of the FT spectrum.

Figure 4: Simulated Fourier transform of reflectance curve from two films, represented by the two peaks.

When this technique is applied to typical IC devices, additional mathematical analysis is required for fine peak resolution, since film stacks and structures are complex. Each film may have different thicknesses when deposited on top of different underlying topographies. In such complex cases, a thickness decomposition method can further separate thicknesses. This method is based on high-resolution spectral analysis.³

System Design

The IsTMS integrated metrology module employs several sensors to measure multiple points across product wafers. Using parallel-beam optics, each sensor radiates a light spot onto the wafer and collects the reflected light, which is then transferred to a spectrophotometer for spectral analysis. Performing thickness measurements using a wide (20-mm-diam) spot size, which is similar to the size of the lithography field, simplifies the integration of the thickness-measurement module into the process tool by eliminating the need for pattern recognition, autofocusing, and wafer alignment.

The IsTMS can be operated in situ, or it can be seamlessly integrated into the process tool if the measured wafer can be viewed directly. Simultaneous measurements of multiple points across the wafer are accomplished using the parallel optical sensors. The unit's wide spot size and simultaneous measurement capacity enable it to measure multiple points across the wafer in less than 2 seconds (data acquisition takes less than 1 second, and data analysis takes another second). Because of its speed, the system has no negative impact on process tool throughput.

Because of its large-spot-size measurement capability, the instrument can perform on-product and in-die measurements. In STI applications, large-spot-size measurements provide thickness information from dense patterns and from materials of different thicknesses, such as oxide and nitride layers. The results obtained from performing large-spot-size measurements represent an average value of the specific film thicknesses under investigation across the die (i.e., oxide within trenches or nitride between trenches). Furthermore, the technique eliminates the sensitivity to micrononuniformities associated with measuring small, nonpatterned areas using small-spot-size optics. These features make the method suitable for performing process monitoring and statistical process control.

Experimental Results

A five-sensor IsTMS measurement module was used to perform post-CMP thickness measurements on 20 product wafers containing an STI structure. The STI structure consisted of trenches with widths varying between 0.22 and 1.0 µm. The thicknesses of both silicon oxide and silicon nitride films were measured simultaneously on the top, bottom, left, right, and center of each wafer. In order to compare the performance of the thickness-measurement module with other systems, measurements in all cases were taken from the center of each wafer.

After being measured using the IsTMS module, the wafers were remeasured using a stand-alone metrology system. Stand-alone measurement sites across the wafers were chosen so that the stand-alone tool's small light spot was positioned in the center of the area that had been previously illuminated by the much larger IsTMS light spot. Adhering to the stand-alone tool's standard measurement procedure, the investigators performed each measurement on a special test structure located in a nonpatterned area close to the die. Because the stand-alone tool uses different measurement schemes for different materials, separate recipes were used to measure oxide and nitride films.

Figure 5: Correlation between IsTMS and stand-alone measurements from STI oxide layer (normalized thickness).

Normalized measurement results from both the IsTMS measurement module and the stand-alone system were then compared. (All the results reproduced in this article are normalized.). Figures 5 and 6 show the correlation between measurements from the IsTMS module and the stand-alone tool for oxide and nitride films, respectively. Although the IsTMS measurements were performed on the die while the stand-alone measurements were performed on a test structure outside the die, correlation coefficients of R² = 0.88 for the oxide films and R² = 0.84 for the silicon nitride films were obtained. The good correlation between IsTMS and stand-alone measurements demonstrates that the IsTMS's sensitivity to thickness changes is at least as good as that of the stand-alone tool. The thickness difference between the samples investigated in this experiment was less than 1 Å.

Figure 6: Correlation between IsTMS and stand-alone measurements from STI nitride layer (normalized thickness).

After optical thickness measurements were performed, two product wafers were randomly selected for scanning electron microscope (SEM) cross-section measurements. The wafers were cross-sectioned where the IsTMS measurements were performed. Three SEM cross-section measurements were made on each wafer.

Figure 7: SEM cross section of STI structure before silicon nitride strip that was compared with the IsTMS thickness measurement of the same structure.

IsTMS results from dense wafer structures were compared with the SEM cross-section results. An example of a SEM cross section is illustrated in Figure 7. Table I compares IsTMS and SEM cross-section measurements for two wafers. The left-hand table summarizes the measurements from wafer 1, while the right-hand table summarizes the measurements from wafer 2. Three SEM-based thickness measurements per wafer (W01-1, 2, 3 and W02-1, 2, 3, respectively), average SEM-based thickness values, and the difference between the IsTMS and SEM average values are shown. The agreement between the two methods was better than 2%. The discrepancy between them was smaller than the SEM resolution for thickness measurements.

Method Oxide Film Thickness Nitride Film Thickness IsTMS 0.99932 0.995009 SEM cross section W01-2 value W01-2 value W01-3 value Average value 0.98357 0.999325 1.017331 1.0 0.979414 0.979414 1.041797 1.0 IsTMS value – SEM value –0.006752 –0.004991

Method Oxide Film Thickness Nitride Film Thickness IsTMS 1.017225 0.997995 SEM cross section W01-2 value W01-2 value W01-3 value Average value 0.986965 1.01257 1.000931 1.0 1.002506 1.046366 0.952381 1.0 IsTMS value – SEM value 0.017225 –0.02005

Table I: Comparison between IsTMS and SEM cross-section measurements from STI structures of wafer 1 (top table) and wafer 2 (bottom table). The data represent normalized thicknesses.

Conclusion

In film-thickness experiments, adjacent silicon oxide and silicon nitride films in STI structures located on product wafers were measured using a large-spot-size broadband Fourier-transform reflectometry method. Comparison tests showed that there is good agreement between the film-thickness measurement results obtained from the IsTMS system and those obtained from a stand-alone tool and a SEM.

While the IsTMS method discussed in this article performs as well as conventional systems in STI process control applications, it has the added advantage of being an integrated system that can measure different films simultaneously in dense patterns on product wafers.

References

1. B Lee, Modeling of Chemical Mechanical Polishing for Shallow Trench Isolation. PhD diss., Massachusetts Institute of Technology, 2002; available from Internet: http://www-mtl.mit.edu/Metrology/PAPERS/Lee-PHD2002-Thesis.pdf.

2. M Born and E Wolf, Principles of Optics (Oxford, UK: Pergamon Press, 1980).

3. J Proakis and D Manolakis, Digital Signal Processing (Upper Saddle River, NJ: Prentice-Hall, 1996).

---

Ayelet Dag is an application manager at Tevet Process Control Technologies in Yokneam Moshava, Israel. She received MS degrees in physics and in quality assurance and reliability from Technion—Israel Institute of Technology. (Dag can be reached at +972 4 9591775 or ayelet@tevet-pct.com.)

Vladimir M. Rubinstein, PhD, is chief scientist at Tevet Process Control Technologies and a founder of the company. His main interests are in the fields of signal and image processing, simulation methods, precise optical measurements, and metrology. Rubinstein is a member of SPIE and the International Association of Science and Technology for Development. He received an MS in electrooptical engineering from the Moscow Institute of Electronics and Mathematics and a PhD in applied physics from the National Research Institute of Optic-Physical Measurements, also in Moscow. (Rubinstein can be reached at +972 4 9591775 or vrubin@tevet-pct.com.)

Yitzhak Gilboa is manager of technology development in the process development group of Cypress Semiconductor (San Jose), where he is responsible for developing fabrication methods in the areas of CMP, thin films, and diffusion. He received a BS in aerospace engineering and an MS in materials science from Technion—Israel Institute of Technology, and an MBA from San Jose State University in California. (Gilboa can be reached at 408/943-2719 or yeg@cypress.com.)

Steven Hedayati previously worked as a process development engineer at Cypress Semiconductor. He has extensive experience in the semiconductor industry in the areas of CMP and substrate engineering. In addition to his work at Cypress, he has worked at OnTrak and National Semiconductor. He received a BS and MS in materials science from the Technical University of Berlin, Germany. (Hedayati can be reached at 408/232-2556 or ssh@cypress.com.)