Wafer-to-wafer control using on-demand pattern metrology
01/01/2005
Yield in the sub-100nm ground-rule regime strongly depends on controlling the semiconductor patterning process. Three types of measurements govern pattern control: critical dimension (including profile), overlay, and film thickness. Individual tools typically are devoted to each measurement type. At present, the tools are gathered in “farms” downstream from the lithography and etch sectors. The current throughput of metrology tools limits product sampling to run-to-run control of the manufacturing process. A new metrology-tool configuration is proposed that combines the capabilities of separately realized optical methods: microscopy, scatterometry, and ellipsometry.
During technology development, high-resolution metrology - scanning electron microscopy (SEM), tunneling electron microscopy (TEM), and atomic force microscopy (AFM) - must quantify within-chip process capability. For much of process characterization and control, however, metrology can be applied to “target” sites comprised of test patterns in planar regions of the chip or kerf. Appropriately designed targets enable the use of fast, relatively inexpensive, and nondestructive optical techniques that provide high sensitivity and precision. As has been demonstrated for both critical dimension (CD) and overlay applications, the precision of optical metrology can, in fact, exceed that of higher-resolution tools. Compliance to dimensional specifications is maintained by the predetermined correlation of within-chip behavior to the optically sampled targets. This is the principle behind in-line utilization of optical microscopy for overlay, scatterometry for CD, and ellipsometry for film thickness.
The current drive to embed optical metrology in process equipment, commonly referred to as integrated metrology (IM), promises the shorter response times necessary for wafer-to-wafer control. Given the multiple tooling types, however, the IM approach greatly increases complexity. Setup, calibration, matching, recipe management, systems integration, and reliability issues accompany the tool and data proliferation, to the detriment of overall productivity.
Optical metrology convergence
As a necessary precursor to effective IM, we consider here the convergence of CD, overlay, and film-thickness optical measurement capability to a common target and sensing module dubbed MOXIE (Metrology Of eXtremely Irrational Exuberance) [1]. The distribution of MOXIE modules, whether standalone or integrated, offers pattern metrology “on demand” for wafer-to-wafer control, as illustrated in Fig. 1. The broad pale-green arrow indicates the in-line production wafer flow through lithography and etch sectors. The orange arrows indicate the pattern metrology flow from off-line to in-line applications.
 Figure 1. On-demand distribution of MOXIE pattern metrology modules for both in-line and off-line applications. |
Existing high-resolution tools play a continued off-line role in site-specific calibration, characterization, and diagnostics, but MOXIE becomes the workhorse spanning off-line to in-line characterization and control. The standalone off-line MOXIE supports recipe creation, calibration, and matching functions for the distributed in-line tools. Black arrows show the application of in-line MOXIE output to feedback and forward control among process tools. The diamonds associated with in-line MOXIE indicate where specifications need to be applied to disposition wafers or lots.
Dual-channel spectroscopy and imagery
The envisioned MOXIE configuration performs orthogonal spectra and image detection from illuminated grating targets via biaxial lenses in reflection (zero-order) and diffraction (first-order) channels (Fig. 2). The illumination is incident at an angle Θ along the direction of the grating period (x axis). The zero-order ray is reflected at an angle -Θ relative to the z axis, and the first-order rays are substantially parallel to the z direction. For broadband or multiwavelength illumination in the range λ0±Δλ, the first order is dispersed in the direction of the grating period by the target itself; whereas the zero-order ray must pass through a wavelength-dispersive optical element, shown as a transmission grating on the left side of Fig. 2. The x-direction numerical aperture (NA) of the diffraction and reflection channel lenses must collect the wavelength-dispersed rays and project them onto the detectors. Simulations of the first- and zero-order spectra are shown above the detectors on the left in Fig. 2. The detected signals are proportional to the intensity dependence on wavelength I1,0(λ). The relative scales are not identical (for the example shown, I0 is on average two orders-of-magnitude greater than that of I1).
 Figure 2. Dual-channel spectroscopy and imaging system. |
In the y direction on the right side of Fig. 2, the diffraction and reflection channel lenses magnify and image the diffracted energy along the y direction in the planes of the respective detectors. Imaging on the y axis allows simultaneous collection of multiple first-order spectra that represent different target regions. Where a pitch G separates the regions, the y-direction NA of the lens must resolve the pitch. Simulations of the wavelength-averaged first- and zero-order signals are shown above the detectors. The detected signals are proportional to the diffraction-efficiency dependence on the width of the target grating elements I1,0(W). The zero-order I0(W) signal is dominated by the background reflectivity of the substrate and film stack. The signal decrease in the regions of the pattern image are due to the loss of energy to the first-order I1(W). The zero-order imaging varies across the field of view due to the tilt in the image plane with respect to the reflected rays.
The benefit of dual-channel detection is that first order is sensitive to the pattern attributes and insensitive to the properties of the unpatterned films, whereas, in the absence of a pattern, zero order is sensitive only to the films. By the simultaneous but separate detection of first- and zero-order, MOXIE enables optimum signal-to-noise (S/N) in the measurement of both the target pattern and the unpatterned films. The former is necessary for robust CD and overlay metrology, while the latter is necessary for robust film-thickness metrology.
For pattern measurement, “signal” is the component of diffracted intensity that contains pattern information. Pattern-unrelated variations that affect the diffracted intensity comprise “process noise.” These include variations in the thickness and (n, k) properties of the underlying films. Zero-order diffraction is swamped by background reflection from the unpatterned regions of the target. Thus, specular scatterometry requires detailed a priori knowledge of the film properties and laborious computation to extract the pattern-dependent signal from the background. On the other hand, first-order diffraction, like darkfield microscopy, is background-free - no signal exists in the absence of a pattern. Thus, the advantages of using first-order detection include reduced sensitivity to process noise, increased sensitivity to pattern variation, and more robust pattern recognition. The net result is improved S/N and ease of setup.
 |
To estimate the S/N improvement, we compute the scalar diffraction (polarization and profile effects are ignored) of 300-700nm plane-wave illumination from a grating of 100nm nominal width elements at a 1000nm period, defined in 300nm-thick positive resist with a refractive index of 1.73 (typical of ArF resists). The pattern is on a simple film stack consisting of a 500nm nominal thickness oxide film with a refractive index of 1.46 on a silicon wafer having a complex refractive index of 3.5+0.35i. The effects of varying oxide thickness over a ±6% ±30nm range at a fixed grating-element width are shown in Fig. 3, where the black lines indicate the nominal thickness and the gold lines show the effect of varying the thickness in 10nm increments. The diffraction efficiency for both orders is plotted on a logarithmic scale vs. wavelength. The zero-order intensity is dominated by the background reflection off the substrate. Averaged over wavelength, the pattern signal, which must be extracted from the background, is a fraction of the substrate reflection. In contrast, the first-order pattern signal is background-free, but its average intensity is another factor of 50 or so less than the zero-order signal. While the zero-order extracted signal is stronger in the absence of process noise, it exhibits much greater sensitivity to process noise. The relative sensitivities are shown in the first row of Table 1, where noise is represented by the range of variation divided by the signal (the extracted signal in the first-order case) averaged over wavelength. In this example, first-order detection has dramatically lower (more than two orders-of-magnitude) susceptibility to process noise.
 Figure 3. Relative diffraction efficiency and process noise susceptibility of zero- and first-order. |
On the other hand, as shown in Fig. 4, the first-order signal shows much higher sensitivity to the pattern changes we are trying to measure. We simulate the effects of the dose variation on the linewidth using an aerial image model for a 193nm wavelength exposure system with NA = 0.7 and a partial coherence of 0.6. Over a ±10% dose range, the change in the zero-order intensity is barely discernable compared to the first. The improvement in dose sensitivity is quantified to be ~30× in the second row of Table 1. The net result of reducing susceptibility to noise and increasing sensitivity to signal is an overall S/N increase by a factor >3500! This improvement can be realized if the illumination is sufficient to enable low-noise detection of the first order.
 Figure 4. Diffraction sensitivity to ±10% dose variation. |
For the sake of brevity, only the response of first-order diffraction to average CD changes induced by dose variation has been considered. Similar results apply to any changes to the grating element profile (sidewall angle, top loss, top rounding, bottom footing, etc.) induced by either dose or focus. The isolation of first-order intensity variation from the underlying substrate enables straightforward simulation of these responses without detailed knowledge of the underlying film properties and rapid empirical characterization of the response through a focus-exposure matrix. Modeling and signal library approaches similar to those employed in conventional scatterometry are applicable. In the MOXIE case, however, effective dose and defocus (EDD) can be determined directly from the first-order spectrum without the intermediate step of CD measurement.
Differential grating targets
As shown in Fig. 2, the angle of illumination is set relative to the target period so that the center wavelength of the illumination band is first-order diffracted normal to the wafer surface. To capture both positive and negative first-orders on targets at different orientations, the collection optic must rotate in synchrony with the illumination direction and target orientation. Positive and negative first-order signals are necessary to resolve phase and amplitude for overlay measurement [2]. Specific grating targets are suitable for measuring CD and overlay. CD targets are gratings of fixed period but variable linewidth W. A discrete differential CD target is shown in Fig. 5a. Overlay targets are interleaved gratings on different layers (A, B) of fixed period but variable relative position Δ.
 Figure 5. Differential a) CD, b) overlay, and c) combined target. |
A continuously differential overlay target is shown in Fig. 5b. For first-order detection, a signal is only generated in the presence of a target so that illumination does not have to be fully landed. Thus, overlay and CD targets may be combined and measured simultaneously within the illumination area, as shown in Fig. 5c. In addition to increasing throughput for multiple measurements and minimizing target area, simultaneous measurement of multiple targets is essential to in situ calibration and matching. Calibration to discrete or continuous offsets embedded in the target designs is performed with each measurement. In this manner, measurements on any given target by multiple MOXIE tools are matched to a common mask offset.
The CD and overlay grating targets are characterized by a known period P and an unknown width W or an unknown relative position Δ. The size of the period relative to the illumination wavelength dictates the angle Θ. The P >λ constraint ensuring detectable first-order diffraction serves also to ensure MOXIE ground-rule scalability independent of illumination wavelength.
In contrast to specular scatterometry, illumination is not required to sense subwavelength period structures. To utilize broadband or multiwavelength sources spanning the visible regime, a grating period in the neighborhood of 1µm or greater is required. While the lower limit on period is of little consequence to overlay targets, it implies a low duty-cycle grating (P/2W >>1) for CD targets comprised of sub-100nm features. Low duty-cycle targets do not limit the achievement of CD control.
The gratings can be comprised of process-compatible features - lines, trenches, contacts, etc. - even arranged at subwavelength periods if need be, provided the overall structure has a major period P >λ. Therefore, the target can mimic the dose and focus response of any appropriate set of features. In the profile measurement case, where top and bottom CD variation can be distinguished, the sensitivity of isolated features to dose and focus makes them the preferred basis for CD control [3]. More generally, however, dose and defocus effects are distinguishable as long as at least two pattern-dependent attributes having distinct response to dose and focus are determined with each measurement. The MOXIE diffraction channel provides the spectral response at two or more wavelengths. The first-order suppression of film stack sensitivity allows direct measurement of EDD without the explicit determination of top and bottom CD. EDD measurement is equivalent to the in-line CD measurement of within-chip or kerf features with CD response that has been precharacterized over the anticipated range of dose and focus variation by any available metrology, including both high-resolution and optical techniques [4]. MOXIE projects one-time off-line characterization to high-frequency in-line sampling on product.
 |
The imaging of the unpatterned region along the central x axis of the illuminated target isolates it from the adjacent patterned regions in the reflection channel of Fig. 2. As indicated in Fig. 5c, this region is used to measure film thickness in the reflection channel - by conventional ellipsometry - simultaneous with the measurement of CD and overlay in the diffraction channel. For 20 element gratings of 5µm length and 1µm period, differential targets to capture x- and y-oriented CD and overlay, allowing a 10µm separation between patterned areas for film thickness measurement, fit in a roughly 50µm2 area. Table 2 summarizes the MOXIE advantages relative to more conventional pattern-metrology approaches.
Conclusion
In configuring MOXIE, we have borrowed from the well-characterized capability of separately realized optical methods: microscopy, scatterometry, and ellipsometry. From microscopy we take separate imaging targets, but only in the dimension perpendicular to the grating period. From scatterometry we take spectral analysis for both overlay and CD, but suppress substrate interaction and gather multiple spectra in parallel. From ellipsometry we take the analysis of angled reflection through polarization for film thickness measurement. By combining these concepts in an appropriately designed optical module, we anticipate that they will perform as well as in their independent configurations. The whole, however, becomes greater than the sum of its parts.
First-order detection improves CD and overlay metrology S/N. The ability to combine targets increases the effective throughput, enforces a common sampling plan, and reduces target area. Differential targets enable in situ calibration and matching. Net benefits include a decrease in precision-to-tolerance, complexity, and setup and cycle times. MOXIE empowers semiconductor manufacturers to improve process control and capability with a scalable, cost-effective pattern-metrology instrument that can be made available to the process on demand.
References
- C.P. Ausschnitt, “A New Approach to Pattern Metrology,” Proc. SPIE, Vol. 5375, February 23, 2004.
- J. Bischoff, et al., “Light Diffraction-based Overlay Measurement,” Proc. SPIE, Vol. 4344, pp. 222-233, 2001.
- C.P. Ausschnitt, S.Y. Cheng, “Modeling for Profile-based Process Window Control,” Proc. SPIE, Vol. 5078, 2004.
- C.P. Ausschnitt, et al., “Process Window Metrology,” Proc. SPIE, Vol. 3998, pp. 158-166, 2000.
Christopher P. Ausschnitt received his BS from Princeton U., his MSEE from the U. of Pennsylvania, and his ScD from the Massachusetts Institute of Technology. He is a senior technical staff member in the IBM Microelectronics Division, 23 Wyman Rd., Lexington, MA 02420; e-mail [email protected].