The use of optical metrology techniques in process control for microelectronic manufacturing has become widespread. These techniques are fast and non-destructive, allowing a higher sampling rate than non-optical methods like scanning electron or atomic force microscopy. One drawback of most optical metrology tools is the requirement that special measurement structures be fabricated in the scribe line between chips. This poses significant limitations regarding the characterization of lithography processes that may be overcome via in-chip measurements. In this paper we present experimental results for an in-chip optical metrology technique that allows direct measurement of both critical dimensions and overlay displacement errors in the DRAM manufacturing process. This technique does not require special target structures and is performed on the actual semiconductor devices.
© 2009 OSA
Scanning Electron Microscopy (SEM) has been traditionally used in the semiconductor manufacturing process for memory devices to monitor critical dimensions (CDs). However, with the rapid shrinking of devices at every new technology node and relentless demand for higher yield in recent years, the limitations of CD-SEM, such as the charging effect that compromises repeatability and accuracy, restrained sampling rate, and small aerial coverage, have been recognized as a bottleneck for further yield enhancement in advanced memory fabrication facilities. Standing out among the alternative approaches that have been developed is optical critical dimension (OCD) metrology  using optical scatterometry with rigorous coupled wave analysis (RCWA) modeling. CD-SEM measurements are typically limited to the top surface layer and must be applied at every process step. However, OCD can eliminate this redundancy and characterize the entire target structure down to the silicon substrate in a single measurement, provided photons are able to penetrate the target structure. In addition, CD-SEM may also damage the material via reaction to high-energy electrons, potentially creating fatal defects at the measurement site, whereas OCD has no such effect. With high throughput and larger sample coverage compared to CD-SEM, OCD also enables measurements at or near the wafer edge, which has been identified as the one of the key elements to yield enhancement. As a result, OCD has become an established metrology technology over the last few years, and is widely used in most memory fabrication facilities around the world.
In this paper, we show a novel OCD technology based on spectroscopic Mueller Matrix ellipsometry that can determine both CD and overlay registration error. This technique is applied to patterned structures in memory fabrication, where cross-sectional misalignment is observed between different layers. The existing method of image-based overlay (IBO) typically uses special targets located outside the chip which are typically two orders of magnitude larger in line and space dimensions than the actual cell devices. IBO is problematic in that it has shown deviations from the actual device overlay error within the die (chip) . The shrinking of devices in the current and future technology nodes makes the measurement of overlay errors in scribe line IBO targets insufficient to represent the true overlay error in the actual device area, especially due to process effects like micro-loading . This is becoming a pressing yield issue for memory manufacturers.
A direct overlay measurement in the device may result in substantial savings in real estate within the wafer since there is no longer a need for special targets on the scribe line. More sophisticated sampling schemes also become possible as much of the device areas in the chip are available for measurement, whereas metrology in the scribe lines is limited to a per die sampling rate. This technique is particularly useful when applied at the after-develop inspection (ADI) step, as it allows manufacturers to rework their process without scrapping wafers. It is also expected that OCD technology will become an indispensible metrology technique in advanced technology nodes below 40 nm, where double-patterning and/or immersion lithography are often required.
2.1 Experimental Configuration
The apparatus used in this experiment, shown schematically in Fig. 1 , consists of a focused beam rotating compensator spectroscopic ellipsometer (SE) in the PCSA configuration, where P stands for the polarizer, C for the rotating compensator, S for the sample and A for the analyzer. With the light sources used in this SE (J. A. Woollam Co. Model M2000), the wavelengths available are in the 200-1000 nm range. Data collection is performed in the specular mode at 65° angle of incidence and the azimuth angle (between the plane of incidence and the grating axis) can be chosen for maximum sensitivity. The light intensity at the spectrometer is described by the first component of the Stokes vector S and follows the equation
In the above R(θ) and Mi are the corresponding rotation and Mueller matrices for each optical element . The angle θ describes the azimuths of P & A and also the fast axis angle of the compensator C. The compensator is rotating continuously and by the analysis of the signal from the spectrometer one can obtain the elements of the sample Mueller matrix. In this single rotating compensator configuration we get 12 (first 3 rows) out of the 16 Mueller elements, see Collins  for details on data reduction.
In the absence of depolarization, as it is the case with our samples, the sample-light interaction is described by the Jones reflection matrix
The Jones matrix depends on the angle of incidence, azimuth, wavelength as well as structural details of the sample. The 4 × 4 Mueller matrix, used to describe our sample-experimental apparatus arrangement, is related to the above 2 × 2 matrix by
Where the matrix T is given by
2.2 Misalignment or overlay error detection from the sample Mueller matrix
To perform an OCD measurement, we compare the experimental Mueller matrix elements to simulated data using the RCWA formulation for 3D objects (2D periodic array) . The model is adjusted by means of nonlinear regression until an acceptable match is found. The Mueller elements are sensitive to the profile details of the structures as well as any asymmetries present, e.g. any misalignment between objects at different levels in the structure. The Physics behind the use of Mueller SE for the detection of misalignment relies on the key fact that the Jones cross-polarization reflectance coefficients (rps and rsp) are anti-symmetric for symmetric structures and this condition is violated in the presence of structural asymmetry. For symmetric gratings the specular, or 0th order, cross reflection coefficients in the conical mount, are known to be anti-symmetric, i.e. rsp = −rps . When the grating symmetry is broken this relationship is violated and we have rsp ≠ −rps, which can be exploited for misalignment control, or overlay metrology. This property of the Jones matrix for symmetric structures translates for the Mueller elements available from our SE into:
In the regime of small overlay errors there should be a linear relationship between the (anti-) symmetry breaking term and overlay displacement δx, i.e.
3. Samples & Experimental Description
3.1 Recess Channel Array Transistor (RCAT) structure
To prove the principle of simultaneous in-chip CD and overlay measurement using the scatterometry and modeling method, we use the DRAM device with recess channel array transistor (RCAT) structure at the litho patterning step, after the shallow trench isolation (STI) process. Figure 2 shows a transmission electron microscope (TEM) image of the cross section of this structure. It consists of a shallow trench isolation (STI) array at the bottom, photo-resist lines at the top, and several thin film layers in between. In Fig. 3 we show the models together with an aerial TEM image of the RCAT structure. Figure 3(a) is a 3D model of the bottom STI structure. The silicon islands are arranged in a non-orthogonal lattice with pitches of 286 and 249 nm and a lattice angle of 54.95° [Fig. 3(c)]. Each silicon island has a CD ratio around 10 between the long and short axes and is rotated by an angle of 26° with respect to the horizontal axis. The space surrounding the silicon islands is filled with silicon dioxide, and then a silicon oxynitride layer of 35 nm and an anti-reflective coating layer of 30 nm are deposited. The resist lines developed on top of the STI structure have a pitch of 143 nm and a height of around 100 nm. The measurement is done at the after-development-and-inspection (ADI) step as shown in Fig. 3(b). Figure 3(c) shows the top down view of the RCAT-ADI structure with the resist lines aligned perfectly with respect to the center of the STI islands. When there is an overlay error, the resist lines are shifted relative to the islands. The effect of such shift is shown in Fig. 3(d), a SEM image of an after-etch-inspection (AEI) RCAT structure with certain overlay error. The overlay error makes the two ends of the elongated STI island non-equal, therefore breaking the symmetry of the structure. This breaking of symmetry will be exploited in the next section for the overlay measurement with MM-SE.
3.2 Overlay response
We first study the spectral response of MM-SE measurement to overlay error. Because of the complex 3D lattice of this RCAT structure, it is possible to take MM-SE data along one of the several high symmetry azimuth angles. A sensitivity study of the SE plane of incidence azimuth angle reveals that the spectral response to overlay error depends strongly on the measurement azimuth and the best sensitivity is achieved at 26° azimuth angle with respect to the top grating lines and orthogonal to the long axis of the STI island (A detailed description of sensitivity studies for azimuth angle optimization can be found in ). The MM SE measurement, therefore, was carried out at 26° azimuth angle.
The structures described in the previous section were fabricated on a 300mm silicon wafer and a 0.3 ppm radial linear overlay displacement expansion was programmed at the lithography step, i.e. as one moves from edge to edge across a diameter of the wafer the misalignment between top and bottom structures varies from −45 to + 45 nm in a linear fashion. Figure 4 shows the combination of two measured MM SE elements, M13 + M31, across the diameter of the wafer orthogonal to the resist lines. The spectra have well formed oscillations across the wavelength range. The oscillation with highest amplitude is between 245 and 265 nm.
From Eq. (6), the amplitude of M13 + M31 directly corresponds to the asymmetry induced by the overlay error of RCAT ADI structure and the spectrum between 245 and 265 nm gives best sensitivity to overlay shift. Figure 5 shows the average spectral response to overaly shift within this wavelength range. The mean response is nearly linear within ± 15 nm of overlay error.
4. OCD (RCWA) Modeling and Overlay Results
An OCD model was built using NanoDiffractTM software to perform a non-linear regression of the model to best fit the MM SE spectra (Fig. 6 ). Besides overlay shift, the model also provides the profile of the STI island (x & y CD, height, SWA), profile of the photo-resist line (CD, height, SWA), and the thickness of the films between the two gratings. It is necessary to float all these parameters to track process variation and achieve good fitting quality. Figure 6 shows a typical set of measured and best-fit spectra for 11 MM SE elements (M11≡1 in our configuration) from the RCAT ADI structure. Note that the M13 (M23) and M31 (M32) spectra are nearly the reverse of each other. The change induced by the overlay shift is relatively small compared to the amplitude of the spectrum.
The overlay error, as well as other profile information, was obtained by fitting the above OCD model to the measured spectra. Figure 7 shows a plot of the correlations between the overlay and CD of the photoresist line (Fig. 7 insert) from the in-chip Mueller-SE measurement against reference techniques. The CD reference was taken by CD-SEM in the device area and the overlay reference was taken by IBO on the targets in the scribe line surrounding the device area. The photoresist CD has a linear correlation to the CD-SEM results with an R2 of 0.986 and the overlay has a linear correlation to IBO with an R2 of 0.987. These excellent correlations demonstrate the superior capability of MM-SE technique combined with OCD modeling for measuring simultaneously CD and in-chip overlay on complex DRAM structures. This additional capability not only extends the application of OCD techniques to overlay metrology, but also greatly improves the turn-around time and cost of ownership (COO) by reducing the metrology steps and equipment required.
5. Concluding remarks
By utilizing a broadband spectroscopic rotating compensator ellipsometer with Mueller matrix capability we demonstrate a novel overlay and critical dimension metrology technique for advanced semiconductor manufacturing. Exploiting fundamental symmetries in the physics of periodic structures and polarized light we proved that it is possible to measure overlay registration error in the device area of actual memory chips. Contrary to traditional imaging overlay metrology this technique does not require special targets placed outside the device area therefore allowing a better understanding of the impact of in-chip overlay and CD errors on production yield. The results of this paper can be extended and applied to high-volume manufacturing, not only for other DRAM structures, but also to flash memory and logic structures.
References and Links
1. C. J. Raymond, “Scatterometry for Semiconductor Metrology,” Handbook of silicon semiconductor metrology, A.C. Diebold, ed., (Academic press 2001), Chap. 18, p.477–514.
2. N. P. Smith, “Overlay metrology at the crossroads,” Proc. SPIE 6922, 0277–0286 (2008).
3. C. Hedlund, H.-O. Blom, and S. Berg, “Microloading effect in reactive ion etching,” J. Vac. Sci. Technol. A 12(4), 1962–1965 ( 1994). [CrossRef]
4. K. Rochford, “Polarization and polarimetry,” NIST publication, http://boulder.nist.gov/div815/81503_pubs/PPMDocs/Rochford-EPST-02.pdf
5. R. W. Collins, Handbook of Ellipsometry, H. G. Tompkins and E. A. Irene, eds., (William Andrew Publishing & Springer-Verlag, 2005), Chap. 7.3.3, p 546–566.
6. L. Li, “New formulation of the Fourier modal method for crossed surface relief gratings,” J. Opt. Soc. Am. A 14(10), 2758–2767 ( 1997). [CrossRef]
7. L. Li, “Symmetries of cross-polarization diffraction coefficients of gratings,” J. Opt. Soc. Am. A 17(5), 881–887 ( 2000). [CrossRef]
8. P. Vagos, J. Hu, Z. Liu, and S. Rabello, “Uncertainty and Sensitivity Analysis and its application in OCD measurements,” Proc. of SPIE 7272, 72721N–72721N–9 (2009)