Improved Imaging Metrology Needed for Advanced Lithography

		At a Glance
	To meet the upcoming nodes in the ITRS, k1 needs to be pushed well below 0.4. To reach this, efforts are to be made through the entire lithography process. This article presents current and future metrology needs for imaging optics. In addition, for these low-k1 factors, the interplay between imaging optics and structures becomes increasingly significant. This might require challenging optimization procedures.

The International Technology Roadmap for Semiconductors (ITRS)¹ states that the half pitch for a DRAM structure should be 130 nm in 2002 and 100 nm in 2005. Longer term, it should go to 35 nm in 2014. Major steps in linewidth reduction can be achieved by decreasing the wavelength (l) and increasing the numerical aperture (NA). The wavelength is being reduced from 248 to 193 to 157 nm. Meanwhile, projection lenses with ever-larger NAs allow the extended use of a particular wavelength technology. The effects on resolution of these steps is calculated by



Here, CD is the critical dimension and k₁ is the process parameter.

Aside from the wavelength and NA, the illuminator, the reticles used, the quality of the projection lens, and resist processes all determine the ultimate resolution. The use of off-axis illumination, optical proximity correction (OPC) and phase shifting masks (PSMs) — generally called resolution enhancement techniques (RETs) — are all accounted for by k₁.

Figure 1 shows, on the left, the NAs necessary to achieve a certain node with a given exposure wavelength. The right side of the graph shows the calculated process factor, k₁. Apparently, with k₁ getting smaller and smaller, changing the wavelength and NA alone is not sufficient to reach the ITRS nodes, so RETs are needed. This has put more stringent demands on the quality of the projection optics and the dynamics of the scanner, with subsequent needs to measure and verify this quality. The measurements can be used to investigate contributions of the different modules to parameters specified on the wafer and to optimize the lithographic step. In this article we will discuss the metrology needs for the imaging optics, taking the ITRS values for the DRAM structures. The discussion is valid for high-NA 248, 193 and 157 nm systems. We will also expand the discussion to extreme UV lithography (EUVL).

ASML's 248 nm step-and-scan lithography systems incorporate high-NA Starlith lenses (shown here) from optics partner Carl Zeiss.

Off-axis illumination

Imaging is only viable as long as a part of the zeroth and first diffraction orders are captured in the pupil of the lens. At higher and higher resolutions, these orders are pushed to the edge of the pupil.

To continue to capture as much intensity as possible in the first diffraction orders, the illumination mode is adapted such that light is mainly at the edge of the diffraction orders. This is achieved with off-axis illumination. Figure 2 shows several examples of such modes.

1. Wavelength, numerical aperture and k1 all need to change to meet the upcoming nodes in the ITRS.

Where capturing the first order determines resolution, the overlap between the first and zeroth order determines depth of focus (DOF). All illumination modes shown are capable of printing 150 nm lines and spaces (1:1). The excellent overlap between zeroth and first order obtained with dipole illumination will yield superior DOF.

The particular off-axis illumination node affects the normalized image log-slope (NILS), which is a value for contrast, normalized to the linewidth. The NILS determines, for example, the mask error factor and exposure latitude. For k₁ values below 0.53, off-axis illumination yields higher NILS than conventional illumination. Below k₁=0.36 annular is preferred over quasar illumination.

2. Off-axis illumination pupil fillings calculated for 248 nm, NA=0.7 and 50% duty cycle at a pitch of 300 nm. From left to right, annular, quasar and dipole illumination are shown. Zeroth order is in blue, and first orders are red and green. For the dipole illumination, there is almost total overlap between zeroth and first orders.

The use of extreme off-axis illumination — i.e. very small poles with minimal overlap between zeroth and first diffraction orders — makes the contrast of the image sensitive to pole balances and pupil positioning and asymmetries. Because only a small part of the pupil is used, the aberration sensitivities will also increase, an issue that is related to the projection optics.

Projection optics

The quality of a lithographic lens (actually the quality of any lens) is often expressed by the RMS values of the wavefront (W) in the pupil of the lens.² ASML's lens supplier, Carl Zeiss, is improving lens technology such that the maximum RMS levels are below 0.025 l for advanced scanners. Whereas this is a sufficient metric to rank lenses overall, it does not give sufficient information on the quality of the actual imaging. For this, the wavefront is generally written as a sum of Zernike polynomials, W(r , Q ). These polynomials

describe the phase across the pupil plane, in polar coordinates r and Q .³ The quality of the lithographic lens can be presented by giving the Zernike coefficients, Z_n, with the common practice that only the first 37 terms are used.

3. Predicted and measured L1-L5 values, for several field points on four different scanners, show a good correlation between model and experimental values.

The impact of aberrations, for the lithographer and chip manufacturer as for the tool manufacturer, can best be illustrated by an example. Figure 3 shows the measured and predicted linewidth variation of a five-bar structure, L₁-L₅. The prediction is made from lithographic simulations, using a calibrated full resist model and the full aberration set of the lenses used in the experiment.

Clearly, the model is capable of predicting lithographic performance. Looking further into the predictive model, it turns out that the coma terms primarily determine the linewidth variation. Based on such correlations a lithographic specification can now be translated into a lens aberration specification. If we, for example, take the maximum allowed linewidth variation for L₁-L₅ to be 15 nm, one can derive a maximum allowed phase difference due to coma.

Aberration sensitivity increases linearly with decreasing resolution.⁴ Knowledge of the aberration levels of a lens becomes increasingly important. More important is to apply this knowledge to improve the actual imaging performance of the scanner for the layers to be printed.

Aberration measurements

Modern lithographic lenses are qualified during manufacturing using a phase measuring interferometer (PMI). For monitoring purposes and optimization, the PMI data is insufficient and there is need to measure the wavefront in situ on the scanner.

The wavefront in the pupil is measurable by sampling the pupil. This can be done a) by selecting specific angles at which the light enters the lens, b) by imaging various test structures with different diffraction patterns, or c) by changing the NA and the partial coherence (s). The first technique is being used by Litel in its In-Situ Interferometer.⁵ The last technique uses the flexibility of modern scanners, which have the ability to automatically set the NA and s. Hence, for this technique, in principle, no additional hardware is required — only a scanner job to run the exposures.

Based on this idea, we will present three methods. The first two methods, FAMIS and DAMIS, are simple and straightforward, based on measurements routinely made on a scanner, but only giving particular aberrations. The third method, ARTEMIS,⁶ is capable of providing the full set of Z_n. The methods all measure changes of the image at multiple illumination settings (hence "MIS" in their names).

The changes are proportional to the aberration levels of the lens and the proportionality factors are dependent on NA/s. These factors can be calculated with a lithographic simulator. To obtain the individual Zernike coefficients, the equation to solve (using least square techniques) is then

 

where  is the vector containing the 37 Zernike coefficients, W the matrix with the sensitivities and C the vector containing the measured effects at the illumination settings. The assumption that a linear sensitivity model can be used has been verified with simulations and, more importantly, by comparing the predicted lithographic result from Eq. 3 with experimental results.

FAMIS and DAMIS measure the focal plane and distortion, respectively, at multiple illumination settings. FAMIS measures the Zernike coefficients Z₉, Z₁₆, Z₁₂ and Z₂₁. These terms are a cause for iso-dense bias and CD differences for horizontal and vertical structures.

Focus is measured using our standard method, FOCAL.⁷ The standard ASML distortion measurement technique uses 8 µm lines. For distortion of relatively large structures, only the sensitivities for the coma terms have a significant value. Hence, DAMIS yields Z₇, Z₁₄ from x distortion and Z₈, Z₁₅ from y distortion. These aberrations induce overlay errors and left-right asymmetries.

The total exposure, measurement and analysis time, all on a scanner, is three hours for FAMIS and 30 minutes for DAMIS. The current reproducibility is 2 nm.

ARTEMIS is based on Dirksen's Aberration Ring Test (DART).⁸ It uses a circular phase object, with a 180° phase shift, which prints as a ring in resist. Aberrations induce a shape change in the ring, which depends on the focus position. The inner and outer contours of the ring are written as a Fourier series, and the order of the Fourier harmonics corresponds to the angular order of the Zernike polynomials.

By measuring at multiple illumination modes, the Fourier components can be separated into the full set of Zernike coefficients. The exposure, measurement and analysis time is about eight hours. The actual measurements are done on a SEM and take about six hours. Current reproducibility is 2 nm.

Correlation with performance

Aberration measurements should not be done just to marvel at the optical quality of the projection optics. These measurements are only of value if the obtained information helps to achieve better imaging. For this, the correlation with lithographic performance must first be established. Then system parameters need to be found to counteract the influence of the aberrations. We will give a few examples of this strategy.

4. Aerial image shape in presence of coma (left). Change in curvature of the aerial image as a function of wavelength shift. This allows control over the amount of linear coma by the wavelength.

Figure 4 illustrates the effect of coma on the aerial image. It is clear that the image becomes severely asymmetric, and the larger the amount of coma, the larger the curvature of the aerial image. Hence coma is suspected to cause the L₁-L₂ variation. Sensitivity analyses show the effect to be 1.5 nm CD variation per 1 nm linear coma (Z7).

To remove the variation, a system parameter needs to be found to remove the linear coma. This parameter is the wavelength. Figure 4 also shows the curvature per wavelength shift. The relation is linear and passes zero. Thus a wavelength shift can be performed at which the linear coma term is zero.

5. Linear coma as a function of slit position before (strongly tilted curve) and after wavelength optimization.

Figure 5 shows the difference before and after the wavelength optimization. Shifting the wavelength reduced the coma significantly. The L₁-L₂ asymmetry was reduced from 50 nm to ~10 nm, verifying the calculated sensitivity.

A second example is on the optimization of a DRAM isolation layer. The brick wall structure is critical on the linewidth difference between top and bottom of the "bricks," C-D as shown in Figure 6. A simulation study with k₁=0.37 was done for this structure to calculate the aberration sensitivities for all Zernike coefficients. Especially coma and three wave contribute to C-D. With these sensitivities we could calculate the expected performance for a 248 nm, high-NA scanner. The aberration levels of the system were such that it classifies as a "golden" lens, i.e. RMS < 0.025 l.² The combination of very low-k₁ imaging, the coma and three wave sensitivities, and the aberration distribution along the slit result in a C-D value that varies strongly when the structure is printed horizontally. By rotating the mask, a better performance is reached.

Future needs

6. Calculated predicted C-D across the slit for an isolation layer (left). The prediction was made for k1=0.37 and a 248 nm, high-NA scanner of good quality, RMS < 0.025l. Vertical orientation of the mask gives better performance.

The capabilities of the techniques described above must be stretched with decreasing k₁ values. The underlying budget shows that there is room for improvement, but it remains a challenging task to achieve the improvements. The information from these measurements can be used to optimize the lithographic process in a production environment, which needs fast and accurate prediction models and a flexible production process.

For nodes below 70 nm, next-generation lithography (NGL) has to be in place. EUVL is a leading contender. An enormous reduction of the exposure wavelength to 13 nm allows the decrease in critical dimension even though the NA is expected to be smaller than 0.3. This gives a k₁ factor of about 1.

Even so, a well-corrected EUVL imaging system needs to meet the Marechal criterion |RMS| £ l/14. The tolerance on absolute RMS wavefront aberration will thus not change, making it again necessary to have adequate in situ checks and controls on imaging key parameters.

Joost Sytsma, who joined ASML in 1996, is departmental manager for the Imaging Systems Development group. He has a Ph.D. in chemical physics from Utrecht University (the Netherlands).

Hans van der Laan, at ASML since 1995, is a senior designer working on projection lenses and aerial image qualifications. He has a Ph.D. in applied physics from Leiden University (the Netherlands).

Marco Moers, at ASML since 1996, is also a senior designer working on projection lenses and aerial image qualifications. He has a Ph.D. in near-field optical microscopy from Twente University (the Netherlands).

Rob Willekers is project leader for Imaging Qualification & Correlation. He joined ASML in 1998 after working for Fokker Space. He has a Ph.D. in high Tc superconductivity from Delft University (the Netherlands).

---

REFERENCES

1. International Technology Roadmap for Semiconductors.
   
2. C. Progler, D. Wheeler, Optical Microlithography XI, 1998, p. 256.
   
3. M. Born, E. Wolf, Principles of Optics, (Sixth edition) Oxford: Pergamon Press, 1993, p. 465.
   
4. D.G. Flagello, J. Mulkens, C. Wagner, Optical Microlithography XIII, 2000, p. 172.
   
5. K. Rebitz, A. Smith, Microlithography World, Summer 1999, 10-26 (1994).
   
6. P. Dirksen, C. Juffermans, A. Engelen, P. de Bisschop, H. Muellerke, Optical Microlithography XIII, 2000, p. 9.
   
7. P. Dirksen, W. de Laat, H. Megens, Optical Microlithography VIII, 1995, p. 701.
   
8. P. Dirksen, C. Juffermans, R. Pellens, M. Maenhoudt, P. de Bisschop, Optical Microlithography XII, 1999, p. 77.
   

Acknowledgments

Thanks to all colleagues in the Projection Lenses group, Jan van Schoot, Jo Finders, Jan Mulkens, Donis Flagello, Kevin Cummings, Anton van Dijsseldonk, Hans Meiling and Christian Wagner of Carl Zeiss.

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

There are no other articles written by this author.

SPONSORED LINKS