Metrology is an essential aspect of nanomanufacturing. Increasingly, nanoscale three-dimensional features are demanded for advanced applications, increasing the demands on metrology. New fabrication techniques such as roll-to-roll (R2R) processes are being developed for manufacturing large-area nanotechnology products such as wire-grid polarizers (WGP), metal-mesh grids, and metamaterials. Angular scatterometry has demonstrated noncontact, optical capabilities for characterizing WGP and photoresist structures with sub-100 nm dimensions. However, existing implementations are not applicable to real-time R2R nanofabrication due to both a requirement of rotating a small sample over a range of angles and measurement times that are incompatible with a moving R2R web. This work demonstrates a high-speed approach (2.5 kHz scanning) to angular scatterometry where the optical beam is scanned rather than the sample mount. The design uses a scanning mirror and high numerical aperture (NA) optics to vary the incident angle over a range from ∼29° to ∼59°. Approaches to increase the angular range are discussed. The scatterometry results are in good agreement with off-line scatterometry results for plane surface, 1D, and 2D patterned samples.
There is an increasing demand for nanoscale manufacturing. Semiconductor circuits are approaching 5-nm scales in volume manufacturing,1,2 and features such as FinFET channels are three dimensional, increasing the demands on fabrication and requiring additional nondestructive metrology techniques.3 Large-area optical devices such as wire-grid polarizers4 and metamaterial lenses5 are being developed. Nanoscale contact grids6 are another emerging nanoscale manufacturing need.
Integrated circuit manufacturing uses a relatively mature tool set, is based on a stable crystal substrate, and metrology is usually off-line with statistical process control. Off-line ellipsometric scatterometry7–10 tools are the dominant optical metrology modality for wafer-scale manufacturing.
Increasingly, there is interest in high volume roll-to-roll (R2R) manufacturing at the nanoscale for emerging large-area applications. R2R manufacturing has many more degrees of freedom (stretching/warping of the web, fluidics issues related to resist application, etc.) and a less mature tool set that demands high-speed, noncontact, real-time metrology solutions consistent with R2R web speeds.
Angular scatterometry, based on high brightness laser sources and scanning the optical system across a range of incident angles, is ideally suited to the high-speed, noncontact requirements posed by R2R nanomanufacturing. Angular scatterometry only requires knowledge of material optical properties at a single wavelength, usually well removed from UV spectral ranges that exhibit high variability based on stoichiometry and film organization (grain boundaries, crystallinity).
Off-line angular scatterometry, based on monitoring the reflectivity of a periodically nanopatterned sample as a function of the angle of incidence, has proven capabilities to characterize feature sizes at sub-100 nm pitch on nanoimprint lithography devices.11,12 Angular scatterometry uses a fixed wavelength coherent laser source that provides high signal/noise and small focal spots allowing small metrology targets.13 Although off-line angular scatterometry provides a large angular range (8°–80°), it cannot be implemented during real-time fabrication processes since it requires a sample coupon and is based on rotation stages with slow response relative to R2R web speeds. Limits of optical angular scatterometry14 and ellipsometric scatterometry15 have been established, demonstrating capabilities well beyond current R2R manufacturing needs. The evident need is to develop an implementation of angular scatterometry that is compatible with real-time measurements on an R2R manufacturing line.
In this paper, we demonstrate the high-speed, real-time, noncontact, nondestructive capabilities of in-line optical angular scatterometry by moving the optical beam rather than the sample stage to vary the incident angle using high-NA optics (parabolic mirrors) and a high-speed galvanometer (2.5-kHz) to provide the angular scan. In our initial demonstration, a total angular range (Δθ) of ∼30° with an initial angle (θi) of ∼29° and a final angle (θf) of ∼59° is achieved; optical systems to increase this angular range to ∼79° are identified, demonstrating that in-line angular scatterometry is capable of nanoscale, noncontact, nondestructive, real-time metrology for R2R nanofabrication systems at speeds compatible with web motion during fabrication.
The in-line optical angular scatterometry design demonstrated in this paper allows metrology of a moving web without interfering with the fabrication process. The scan speed is compatible with a projected web speed of 10 cm/s. Figure 1 presents the initial system design. The fundamental idea of scatterometry is to collect the zeroth order reflection from a periodically nanopatterned sample as a function of incident angle by focusing a polarized laser beam at a focal spot, large compared to the pattern periodicity, on the moving web. Our system uses a 405 nm collimated diode laser. After collimation, the laser beam profile is elliptical (1.8 × 3.5 mm2) and is focused on a small mirror mounted on an oscillating galvanometer (2.5 kHz) to vary the direction of the beam. The beam reflected from the galvanometer mirror is collimated by a second lens and refocused onto the moving web using a high-NA parabolic mirror. The upper and lower beam paths in Fig. 1 represent the limits of the angular scan. The number of resolvable angles is given by the ratio of the beam NA to the overall NA of the optical system. The effective NA of the beam is 8.67 × 10−3 (angular resolution of the optical system). The NA of the collimating lens is 0.25, which accommodates about 30 points across the scan range; no correction for the nonlinear time dependence of the mirror rotation was required over this small range, but corrections will be necessary if the angular range is increased. The 90° off-axis parabolic mirrors have a focal length of 10.16 cm, which gives a total incident angle scanning range at the sample of ∼30° with an initial angle (θi) of ∼29° and a final angle (θf) of ∼59°. The galvanometer oscillation at 2.5 kHz gives us a scan period of about ∼400 μs. For an ∼220 × 200 μm2 focal spot on the web moving at 10 cm/s, this allows averaging of about 5–10 reflectance measurements before the web moves a distance comparable to the focal spot. The collection optics is a mirror image of the illumination optics, with the galvanometer mirror, the sample surface, and the detector surface all being conjugate foci. Figure 3(a) presents the actual experimental setup mounted on the optical table; it is similar to Fig. 1 with the elimination of the two folding mirrors just above the web since we are using a small sample at present.
B. High-speed detection
In order to detect the signal at 2.5 kHz speed, a blue enhanced photoconductive Si photodiode (active area of 0.81 × 0.53 cm2) was operated under reverse bias to obtain the rise time required to resolve the reflectivity trace from the 2.5-kHz frequency galvanometer. Operating the detector at reverse bias with a 300 Ω load gave an experimentally measured rise time of 190 ns, sufficient to monitor the ∼200 μs reflectivity pulse.
Due to the large number of optical parts in the in-line setup, we need to calibrate the system. Our current lenses have a transmittance range of ∼92%–98%, and the polarizer has a transmittance of ∼15%–22%. Future designs will include a half-wave plate to maximize the intensity from the laser source. The moving galvanometer is coated with a layer of aluminum and a dielectric layer to increase its reflection; however, the reflectance is dependent on the incident angle varying from ∼80% to 91% over the angular range.
The parabolic mirrors also have a non-negligible angle-dependent reflectance variation. Because of all these optical effects, it is necessary to calibrate the system using off-line angular scatterometry, which has fewer optical components all of which are fixed relative to the beam; only the sample is rotated. The calibration constant was calculated using a well-defined flat surface. The surface used is a silicon wafer cleaned with HF to remove all native oxide contamination and coated with a 4.9 nm thick protective layer of Al2O3 (characterized by ellipsometry). Figure 2 presents the experimental curve obtained from the in-line setup compared to the off-line curve. Since the differences between the two curves are significant, the in-line experimental signal was multiplied by 21. For the off-line scatterometer, a sub-100 nm period WGP was used to demonstrate its capabilities.13 Clearly, at the higher angles, the angular dependences of the optics impact the measurement,
Figure 3 shows both off-line and in-line systems with incident beams and reflected beams indicated. The calibration coefficient varied nonlinearly with angle due to all of the angular dependent reflections/transmissions. Figure 4 presents the calibration coefficient as a function of incident angle and the mathematical fit that used to calibrate the experimental results from the in-line optical angular scatterometry.
III. RESULTS AND DISCUSSIONS
All our experimental results from the in-line system are adjusted by using the calibration factor shown in Fig. 4. All flat surface experimental results were compared to off-line angular scatterometry, ellipsometry, and simulation. For the patterned structures, in-line and off-line results were compared. A thin film calculation of our planar structures was used to compare with experimental results, and the fitting was accomplished by searching the simulation results for the lowest Root Mean Square Error (RMSE) to the experiments shown in Eq. (2),
A. Silicon wafer with native oxide
The first sample analyzed was a silicon substrate with a thin layer of native oxide. The results of the in-line setup were compared to off-line and fitted to the simulations to determine native oxide thin layer estimated at 1.36 nm thick by ellipsometry. Figure 5 presents the TE results. Both in-line (0.74 nm) and off-line (0.56 nm) estimates of the oxide thickness were below 1 nm. The RMSEs measured were similar: 1.116 × 10−2 for off-line and 1.2420 × 10−2 for in-line. These variations could be due to nonuniformity of the native oxide layer since no effort was made to measure the same spot in both experiments.
Figure 6 similarly shows TM reflection as a function of incident angle compared to ellipsometry and simulation. RMSE reflects a larger difference between the RMSE of the in-line and off-line; however, the thicknesses of the oxide layer for the off-line and on-line experimental results was close to each other during the fitting process. The off-line fit gave a native oxide thickness of 0.50 nm, compared to 0.66 nm for the in-line measurement. The RMSEs measured were 1.076 × 10−2 for off-line and 4.821 × 10−2 for in-line. Ellipsometry results show a thickness of 1.36 nm. Similar to the TE results, the scatterometry results were below 1 nm for the oxide layer.
B. Silicon wafer with an ∼100 nm SiO2 overlayer
This was a sample provided from J. A. Woollam with a nominal thermal oxide layer of ∼115.8 nm. The thickness was determined using ellipsometry and compared with off-line and in-line fitted values from the sample. Figures 7 and 8 present the TE and TM curves with the ellipsometry results as well as the fitted RMSEs of both in-line and off-line measurements. For TE and TM measurements, ellipsometry characterized the oxide layer at 113.98 nm; meanwhile, in-line was fitted at 114.0 nm and off-line was fitted at 113.90 nm. The fitted RMSE for the TE and TM after calibration was accurate while the fitting process was significantly closer than the previous sample. For TE, the RMSEs measured were 3.908 × 10−3 for off-line and 3.129 × 10−3 for in-line. For TM, the RMSEs measured were 6.393 × 10−3 for off-line and 3.896 × 10−3 for in-line. A thicker layer of oxide defines a more accurate index of refraction on the ellipsometry tools allowing a fit better the curve. These results show the capabilities of high-speed thin layer calculation with our current in-line system design. For the TM polarization measurement, similar to Fig. 7, results were obtained as shown in Fig. 8.
C. 1D undercut nanoscale Al lines on silicon
The third sample used on both systems was fabricated using interferometric lithography. The silicon sample was cleaned using acetone, methanol, and isopropyl alcohol, followed by DI water. The sample was spin coated with antireflection coating (BARC) ICON 16 at 3000 rpm for 30 s, which resulted in a thickness of about ∼150 nm. Photoresist NR7 500 was then spin coated at a speed of 3000 rpm giving a thickness of about ∼500 nm. Interferometric lithography was used to make a periodic 1D pattern of lines and spaces with a pitch of 500 nm. The sample was developed in MF-321 developer for 20 s. An ∼85 nm thick layer of aluminum was deposited on the photoresist patterned sample at a chamber pressure of 2 × 10−6 Torr, followed by a metal lift-off process. The resulting structure was dry etched using oxide etching to remove the residual layer of BARC (icon 16). The etching time was 1 min and 30 s. The roughing pressure, RF power, inductively-coupled plasma (ICP) power, and gas flow rate were 15 mT, 100 W, 300 W, and 21 sccm, respectively. The wall width measured was about 340 nm. The BARC layer underneath the metal layer was undercut leading to the mushroom shape. The lateral width of the BARC layer underneath the metal layer measured was about 140 nm. Figure 9 presents a 45° view SEM of the 1D fabricated sample.
For this sample, off-line and in-line experimental results were compared. Figures 10 and 11 present the TE and TM reflectivity versus incident angle curves for this structure. In-line experimental results were calibrated using the same calibration curve described previously. Both off-line and in-line show similar experimental trends, proving the capabilities of scanning at high-speed nanoscale periodic 1D structures.
D. 2D silicon nanotubes on a silicon wafer
The fourth sample measured was a nanopatterned wafer with a 2D array of hollow silicon tubes fabricated by nanoimprint lithography, atomic layer deposition (ALD) coating, and reactive-ion etching (RIE). The wafer is 100 mm in diameter, and the process flow is shown in Fig. 12. The pillars are at an ∼200 nm pitch. The SiO2 layer is ∼30 nm high. The resist layer is ∼35 nm high, and the silicon has been etched ∼75 nm deep. The outside diameter of the tubes are ∼135 nm, and the inside diameter of the tubes is ∼100 nm. The fabrication procedure consisted of nanoimprint lithography to define photoresist pillars on the silicon wafer. Then, low temperature ALD of SiO2 was used to form sidewalls and a cap on the photoresist. Reactive ion etching was used to remove the cap, and the SiO2 sidewalls acted as a mask to etch the silicon nanotubes. Figures 13(a) and 13(b) present side view diagram and SEM image of the 2D structure.
This sample represents a periodic complex 2D nanostructure. Figure 14 shows the measured reflectivity as a function of incident angle for TE polarization; a top view diagram of the structure is included. The slight variation between the off-line and in-line measurements is likely due to sample nonuniformity.
IV. ALTERNATIVE DESIGN
Two issues with the current design are (1) the limited angular range and (2) the proximity of the folding mirrors (Fig. 1) to the web. The necessary optomechanics for mounting and aligning the folding mirrors adds to the difficulty. Due to these challenges, an alternative design has been developed.
Because of the web-interference constraint, we have considered the use of 45° parabolic mirrors that allow bringing the beam from the side and increase the angular range due to the availability of shorter focal length mirrors. Also, the optical part mounts can be easily designed to avoid proximity to the moving web. Using 45° off-axis parabolic mirrors with a focal length of 11.9 cm allows a total angular range up to ∼50° with commercial off-the shelf optics. Figure 15 presents an alternative solution showing the beam incidence from the side of the parabolic mirror and avoiding interference problems with the web movement. The NA angular range can be further increased to ∼79° with a variation of the focal length and the off-axis cut using custom parabolic mirrors.
V. SUMMARY AND CONCLUSION
This work demonstrates high-speed, noncontact, nondestructive, in-line angular optical scatterometry for metrology at speeds compatible with R2R nanofabrication. Angular optical scatterometry is a powerful metrology tool adaptable to in-line capabilities due to a high brightness, single wavelength laser source and rapid optical scanning. Our in-line design has proven to scan with a 2.5-kHz galvanometer giving us a trace and retrace time around ∼400 μs. Higher speed galvanometers are available as necessary. A typical vertical displacement (vibration) of a web in operation is 0.5 mm. Our system has a focal depth of 8 mm, so the web displacement will not pose a problem. Measurement consistency with established off-line scatterometry system has been demonstrated for planar, 1D, and 2D nanoscale samples. Our design with a volumetric space requirement of <0.110 m3 and with no components extending into the plane of the web is directly applicable to R2R manufacturing. The use of 45° parabolic mirrors allows us to scan an angular range up to ∼50°, and the center angle can be adjusted to match the sensitive angular range of specific structures by varying the configuration. An in-line optical scatterometry system is under development for real-time monitoring in an R2R tool for metrology of WGP and metal-mesh grid fabrication using nanoimprint lithography. The in-line, noncontact, real-time, nondestructive monitoring capabilities have been demonstrated.
This work is based on the work supported primarily by the National Science Foundation (NSF) under Cooperative Agreement No. EEC-1160494. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.