Issue



Characterizing resists and films with VUV spectroscopic ellipsometry


07/01/2001







Pierre Boher, Patrick Evrard, Jean Philippe Piel, Christophe Defranoux, Jean Louis Stehlé, SOPRA S.A., Bois Colombes, France

overview
Spectroscopic ellipsometry has been the technique of choice to characterize thin films and multilayers in semiconductor manufacturing. Extending this technology into the vacuum ultraviolet range, down to 140nm, however, for emerging 157nm optical lithography applications, requires an environment that avoids the absorbing effects of oxygen and water vapor below the 190nm wavelength; these must be reduced to the parts/million range. In addition, the optical setup for VUV spectroscopic ellipsometry must put the monochromator in the ellipsometer's polarizer arm to avoid photobleaching, and the ellipsometer must work in a rotating analyzer configuration to minimize parasitic polarizations.

Compared to reflectance and transmittance photometry methods, ellipsometry has several advantages. First, measurement is made on a ratio of two signals (Rp/Rs, the reflection coefficients of the two polarizations parallel and perpendicular to the incidence plane). This measurement is less dependent on source fluctuations so the accuracy is generally better than a photometric one. Also, there is no need for a reference sample since the measurement is self-calibrated. Finally, two independent parameters — amplitude ratio and phase — are measured simultaneously compared to one measured parameter with reflectance or transmittance; this allows direct extraction of complex indices without using the Kramers-Kronig method. The measurement can be rapid thanks to the development of multichannel detectors [1].

Click here to enlarge image

Figure 1. Analysis of SE measurements at 157.6nm on a CaF2 substrate (insert). Squares indicate individual measured data points; lines show simulated results.

At 248nm and 193nm wavelengths, spectroscopic ellipsometry (SE) has been shown to be a very efficient method to characterize photoresists and antireflective coatings used in 248nm and 193nm deep-UV lithography applications [2]. Specifically, photoresist behavior versus exposure dose has been determined by this method [3].

SE at around 157nm wavelength is increasingly being required to support 157nm lithography, where resist layer thickness is generally less than what has been used conventionally. Thus, thickness and optical indices correlation becomes more difficult with photometry. Moreover, at this shorter wavelength, diffuse scattering from surfaces and contamination creates problems that favor ellipsometry over photometry. Another challenge associated with 157nm lithography involves finding suitable metrology methods for characterizing metal fluoride lens materials (i.e., CaF2, MgF2, SiOF, and LiF).


Figure 2. Extracted optical indices of a 157nm photoresist layer with uncertainties multiplied by 10 so they show up.
Click here to enlarge image

Fortunately, applications that require SE at or near the 157nm wavelength can now be addressed with vacuum ultraviolet (VUV) SE, which is capable of using analytical wavelengths from 720nm down to 140nm. This feature addresses two major requirements for doing SE at157nm: The entire beam path must be free of oxygen and water vapor because of strong molecular absorption bands; and a standard ellipsometric optical path with a monochromator just before the detector, connected to the analyzer arm, cannot be used. These requirements have been met by enclosing the ellipsometer in a glove box with continuous water and oxygen purification and by using a double monochromator in the polarizer arm. In addition to ellipsometry, the system can make photometry measurements (reflectance and transmittance versus angle of incidence and wavelength) at fixed polarization state and scatterometric measurements [4].

Click here to enlarge image

Our work has compared VUV SE results, in some cases, to grazing x-ray reflectance (GXR) results. Measuring thin film thickness with GXR does not require a structural model [5-6] because the refractive index n is always close to 1 and the extinction coefficient k is negligible at the 1.54Å wavelength that GXR uses. Indeed, with multilayers, GXR provides precise information about the periodicity of stacks and interlayer structures [5].

Research has shown that combining VUV SE and GXR is a very powerful technique for characterizing different samples. For example, they have been used to examine epitaxial

semiconductors and metallic films — SiGe [7], AlGaAs/GaAs [8], SiC [9], and Ti/SiO2/Si [10]. Other research has looked at antireflective coatings and photoresists for 365nm and 248nm lithography [11-12]. In other work, we have shown that the extraction of the optical indices of the films is much easier when the thickness is precisely determined by GXR [13].

CaF2 substrates
In our current work, we have taken SE and photometric measurements of CaF2 substrates, for which transmittance, sensitivity to contamination, and deposition of antireflective coatings are crucial issues [14-15], and compared the results to data from the literature. Specifically, we made variable angle VUV SE measurements at 14 different wavelengths in the 150-300nm range, making these measurements from 50-70° incidence at each wavelength.


Figure 3. GXR analysis of an oxide-nitride gate dielectric as shown in data on the right (i.e., sample no. 3 in Table 1). Circles indicate individual measured data points; lines show simulated results.
Click here to enlarge image

Our experimental data and simulations for 157.6nm (Fig. 1) show a well-defined Brewster angle at 58° angle of incidence on the two ellipsometric parameters (i.e., minimum of tany and zero crossing value for cosΔ). The shape of the cosΔ curve around the Brewster angle is due to the angular aperture of the measurement beam, which is not negligible and must be taken into account during analysis.

In our work, each single wavelength measurement was fitted independently using a very simple model where the refractive index of the substrate was adjusted. Our modeling precisely determined the refractive index at each wavelength and the thickness of the top surface roughness (e.g., n = 1.600±0.001 at 157.6nm and 5.4nm, respectively). When we plotted refractive indices versus wavelength overlaid with optical indices from literature [16], we found very good agreement above the 190nm wavelength, only noting a deviation below 190nm.

As already observed by others, in this wavelength range, sample surface contamination is crucial. We think that the discrepancy between our results and the literature data, which is more pronounced below 190nm, was due to a thin layer of contamination on the CaF2. Indeed, we found an indication of contamination when we made transmittance measurements over the same wavelength range and compared these with expected results from a perfectly clean substrate.

Photoresists and antireflective coatings
Our work also involved using VUV SE to characterize different kinds of photoresist layers and antireflective coatings, comparing the results to data from a standard UV-VIS ellipsometer.

Photoresists have been intensively studied at 248nm and 193nm using SE [17]. Before similar work could be done at 157nm, we needed to ensure that the measurement itself did not bleach the sample. Using the 157.6nm wavelength with our VUV SE instrument, we measured ellipsometric parameters continuously during two hours at the same location on a wafer with freshly coated photoresist and found no deviation (s ±0.001 for tany and ±0.004 for cosΔ).

Seeking to determine the capability of VUV SE to accurately characterize the optical indices of 157nm photoresist, we first made VUV SE measurements at three different incidence angles in the wavelength range 137-640nm (i.e., 2-9eV because plots versus energy provide better clarity in the UV range). This test revealed that the quality of the measurement was good within the entire wavelength range.


Figure 4. GXR analysis of one oxide-nitride gate dielectric as shown in data on the right (i.e., sample no. 2 in Table 2). Dots indicate individual measured data points; lines show simulated results.
Click here to enlarge image

Using a standard procedure [12] to simulate n and k, we simulated first in the visible-IR region assuming that the photoresist is transparent in this region and then fit thickness and refractive index using the dispersion law model. We extracted optical indices wavelength by wavelength by fixing the thickness and fitting the incidence angle dependence. Results deduced from the fitting procedure for the 157nm photoresist used in our test are plotted in Fig. 2, along with the uncertainties of n and k multiplied by a factor of 10 so that they show better on the graph. This data shows that the uncertainties are >±0.01 everywhere in the wavelength region. We confirmed these results by analyzing another sample coated with the same photoresist at a very different thickness; the extracted optical indices were in complete agreement with those of the first sample within the deduced uncertainties.

In addition to photoresist layers, the method above can be used to characterize antireflective coatings (ARCs) and optimize lithography structures at the wavelength of interest.

Thin oxynitride gate dielectrics
When thin oxynitride gate dielectrics are used [18], structure reliability at high electric field depends on layer thickness and on the nitridation level of the SiOxNy layer [19-20]. In our work, we found that a combination of GXR and VUV SE provides a precise analysis of these films.

For example, analyzing very thin gate dielectrics with low nitrogen content using GXR results in total reflection at a very high grazing angle, but reflectance decreases rapidly as the grazing angle is increased (Fig. 3). We found that the angular position of the total reflection threshold is related to the mean density value of the sample. With these very thin layers, it becomes difficult to deduce valuable information. The optical contrast between SiO2 and crystalline silicon is very small [19].

Click here to enlarge image

The contrast between nitride and oxide is more important. It produces interference fringes that are detectable around 1.5-2°. The angular position of these fringes is characteristic of layer thickness and the amplitude of the fringes is related to index contrast between layers and their composition. For each measurement, we applied a two-layer model adjusting at the same time the different layer thickness, the x-ray indices, and the mean roughness of the stack. We neglected the thin bottom SiO2 layer because its optical index contrast is very close to the silicon substrate.

We found that the agreement between simulation and measured data of three oxide-nitride gate dielectrics (up to 8.2eV or 151nm) is very good along the angular range. In fact, the uncertainty associated with the different thicknesses is quite small (<0.1nm; see the GXR results summarized in Table 1). The data tabulated in Table 1 show that uncertainty associated with GXR thickness measurement is good except for sample nos. 5 and 7, where the contrast between oxide and nitride layers is too small, due to the low relative concentration of N versus O, to give well-defined interference fringes. The mean roughness is small for all samples.

Click here to enlarge image

Figure 5. Experimental (squares) and simulated (lines) VUV SE spectra of sample no. 2 from Table 2. Nitrogen content and SiO2 bottom layer are only fitted.

We measured the same samples using VUV SE in the 150-850nm wavelength range (1.5-8.2eV) at 75° of incidence angle versus phase shift, i.e., cosΔ. This parameter is the most sensitive to any change of layer thickness and composition of the nitride-oxide gate dielectrics. Detectable differences can be found between the three samples as expected from GXR measurements.

We analyzed the VUV SE results using the following procedure: The thicknesses of the top SiO2 and SiOxNy layers were fixed to the GXR values. A bottom SiO2 layer is then added and its thickness is adjusted. The composition of the SiOxNy layer is also adjusted (assuming a Bruggmann mixture of the optical indices of SiO2 and Si3N4 from the database). Results from this analysis are summarized in Table 1. Notice that the uncertainty associated with the Si3N4 content is small (<5% except for the two samples with small optical contrast).

In Table 1, x-ray photo-emission spectroscopy (XPS) measurements of the N/O ratio of the same samples are also reported. There is very good agreement between these chemical results and the combined GXR-SE measurement of the SiOxNy composition. In fact, the agreement between the results is very good even for the two samples with less optical contrast (i.e., sample nos. 5 and 7) for which XPS confirms the low nitrogen content.


Figure 6. VUV SE experimental spectra of AlO-ZrO on silicon at three different incidence angles.
Click here to enlarge image

Our combined analysis is then validated a posteriori. However, a standard analysis-based VUV-SE measurement alone is not as coherent. The reason is the high degree of correlation between thickness and composition for these very thin layers and the occurrence of more than one layer on the silicon substrate. The situation is different, for example, for the SiO2 gate oxide layer where the optical index is known. Otherwise, for more complex nitride-oxide structures, the GXR information is extremely important for the SE analysis [21].

Using thicker gate dielectrics with high nitrogen content, we also measured each sample with GXR prior to VUV SE (Fig. 4). Compared to Fig. 3, Fig. 4 shows more interference fringes from the thicker film and these are better defined due to higher nitrogen content. This allows us to apply a more complex model than that used above: We applied a three-layer model adjusting at the same time the different layer thickness, the x-ray indices, and the mean roughness of the stack. The third bottom layer is included to take into account the inhomogeneity of the nitride layer. Again, we neglected the thin bottom SiO2 layer, since its optical index contrast is very close to the silicon substrate.

Figure 4 shows very good agreement between experimental results and simulation along the angular range. The uncertainty associated with the different thicknesses is quite small (Table 2).

We also analyzed a second sample with a less homogeneous nitride layer. Assuming that the top and bottom interfaces of this layer were gradual, we independently fitted their thicknesses with GXR.


Figure 7. Optical indices of various new dielectric materials measured at wavelengths from 145-700nm.
Click here to enlarge image

Again, we measured the same samples with VUV SE in the 137-620nm wavelength range (2-9eV) at three different incidence angles (e.g., Fig. 5). We analyzed SE results assuming the same layer thickness as for GXR. We also added a bottom SiO2 layer and, to take care of the native oxide, adjusted for its thickness. We also adjusted for the composition of the SiOxNy layer, assuming a Bruggmann mixture of the optical indices of SiO2 and Si3N4 from a SOPRA database. From the simulation data shown in Fig. 5 and tabulated in Table 2, we notice that the uncertainty associated with the Si3N4 content is small (i.e., <5% for the two samples).

New dielectric materials
Some of the new dielectric materials, such as AlO or ZrO films, show generally non-negligible absorption in the VUV range. The characterization of the optical properties is then more informative in this region because it is generally more sensitive to the layer imperfections.

We have characterized three kinds of this class of dielectric films, specifically AlO, AlF and AlO-ZrO, using 145-700nm wavelength variable-angle VUV SE. With measurement of an AlO-ZrO film, for example, we found that it was completely transparent above 250nm, giving very nice interference fringes in this region (Fig. 6). For this data, we made the simulations measurement in the same way as we did for photoresist layers discussed earlier. The accuracy on the indices is very good, especially in the VUV region due to absorption.

Figure 7 summarizes the optical indices that we gathered for various films. As expected, the absorption varies rapidly below 250nm.

Conclusion
We have presented data from a new VUV SE metrology capability with a purged environment for measuring optical properties of substrates and thin films in the range 145-630nm, presenting data from experimental results on CaF2 substrate, thin oxynitride gate dielectrics, and photoresist layers. The results obtained with this new technique agree with data published in the literature and also with independent measurements derived from using SE with GXR, the latter, particularly, in measuring thin oxynitride gate dielectrics. It may be possible to add GXR capability alongside VUV SE within a purged environment. If so, we think that x-ray information will be helpful in this wavelength range due to its thinner width and great sensitivity to roughness and contamination.

Acknowledgment
The authors wish to acknowledge Sophie Bourtault's contribution to the article.

References

  1. P. Boher, J.L. Stehle, Materials. Science and Engineering, B37, 116, 1996.
  2. P. Boher, et al., SPIE, Vol. 2729, 1996.
  3. P. Boher, et al., SPIE, Vol. 3678, p. 126, 1999.
  4. The VUV SE capability is commercially available as the SOPRA Purged UV Spectroscopic Ellipsometer. GXR capability is commercially available as the SOPRA Gonio-Ellipso-Spectro-Photometer.
  5. P. Boher, P. Houdy, C. Schiller, J. Appl. Physics, 68, p. 6133, 1990.
  6. F. Bridou, B. Pardo, J. of X-Ray Science and Techn., 4, pp. 200-216, 1994.
  7. P. Boher, J.P. Piel, J.L. Stehlé, J. of Crystal Growth, 157, pp. 73-79, 1995.
  8. P. Boher, J.L. Stehlé, Presented at the MRS Fall Meeting, Boston, Nov. 27-Dec. 1, 1995.
  9. P. Boher, J.L. Stehlé, L. Hennet, Thin Solid Films, 294, pp. 37-42, 1997.
  10. P. Boher, et al., Thin Solid Films, 319, pp. 67-72, 1998.
  11. P. Boher, et al., SPIE, Vol. 2726, pp. 608-620, 1996.
  12. P. Boher, et al., SPIE's International Symposium on Microlithography, 1997.
  13. P. Boher, J.L. Stehlé, L. Hennet, Mat. Res. Soc. Symp. Proc., Vol. 446, pp. 369-376, 1997.
  14. T.M. Bloomstein, et al., J. Vac. Sci. Technol., B16, 3154, 1998.
  15. T.M. Bloomstein, V. Liberman, M. Rothschild, SPIE, Vol. 3676, 342, 1999.
  16. D.F. Bezuidenhout, in Handbook of Optical Constants of Solids II, ed. Palik, p. 815, Academic Press, 1991.
  17. P. Boher, et al., SPIE, Vol. 3677, 845, 1999.
  18. K. Kobayashi, J. Electrochem. Soc., Vol. 139, No. 5, p. 1693, 1992.
  19. L.K. Han, IEEE Electron Device Letters, Vol. 16, No. 8, p. 348, 1995.
  20. G.Q. Lo, IEEE Electron Device Letters, Vol. 13, No. 7, 1992.
  21. P. Boher, J.P. Piel, J.L. Stehlé, MRS Fall Meeting, Nov. 29-Dec. 3, 1999.

Pierre Boher received his engineering degree from Ecole Centrale de Paris and his PhD from Paris VI University. He is R&D manager at SOPRA S.A., 26 rue Pierre Joigneaux, 92270 Bois Colombes, France; ph 33/1 47 81 09 49, fax 33/1 42 42 29 34, e-mail [email protected].

Patrick Evrard graduated as a physicist from engineering school. At SOPRA, he is in charge of high-resolution spectrometer manufacturing and the design of spectroscopic systems.

Jean Philippe Piel received his PhD in solid state physics. At SOPRA, he manages production and develops new laboratory instruments.

Christophe Defranoux received his MSc in optics and materials sciences. He works in SOPRA's application department.

Jean Louis Stehlé is director of SOPRA's analysis department and has initiated most of the company's activities in spectroscopic ellipsometry.