Large-scale sample inspection with nanoscale resolution has a wide range of applications, especially for nanofabricated semiconductor wafers. Atomic force microscopes can be a great tool for this purpose, but are limited by their imaging speed. This work utilizes parallel active cantilever arrays in AFMs to enable high-throughput and large-scale inspections.
An Atomic Force Microscope (AFM) is a powerful and versatile tool for nanoscale surface studies to capture 3D topography images of samples. However, due to their limited imaging throughput, AFMs have not been widely adopted for large-scale inspection purposes. Researchers have developed high-speed AFM systems to record dynamic process videos in chemical and biological reactions at tens of frames per second, at the cost of a small imaging area of up to several square micrometers. In contrast, inspecting large-scale nanofabricated structures, such as semiconductor wafers, requires nanoscale spatial resolution imaging of a static sample over hundreds of square centimeters with high productivity. Conventional AFMs use a single passive cantilever probe with an optical beam deflection system, which can only collect one pixel at a time during AFM imaging, resulting in low imaging throughput. This work utilizes an array of active cantilevers with embedded piezoresistive sensors and thermomechanical actuators, which allows simultaneous multi-cantilever operation in parallel operation for increased imaging throughput. When combined with large-range nano-positioners and proper control algorithms, each cantilever can be individually controlled to capture multiple AFM images. With data-driven post-processing algorithms, the images can be stitched together, and defect detection can be performed by comparing them to the desired geometry. This paper introduces principles of the custom AFM using the active cantilever arrays, followed by a discussion on practical experiment considerations for inspection applications. Selected example images of silicon calibration grating, highly-oriented pyrolytic graphite, and extreme ultraviolet lithography masks are captured using an array of four active cantilevers ("Quattro") with a 125 µm tip separation distance. With more engineering integration, this high-throughput, large-scale imaging tool can provide 3D metrological data for extreme ultraviolet (EUV) masks, chemical mechanical planarization (CMP) inspection, failure analysis, displays, thin-film step measurements, roughness measurement dies, and laser-engraved dry gas seal grooves.
Atomic force microscopes (AFMs) can capture 3D topography images with nanoscale spatial resolution. Researchers have extended the capability of AFMs to create sample property maps in mechanical, electrical, magnetic, optical, and thermal domains. In the meantime, improving imaging throughput has also been the focus of research to adapt AFMs to new experimental needs. There are primarily two application domains for high-throughput AFM imaging: the first category is high-speed imaging of a small area to capture dynamic changes in the sample due to biological or chemical reactions1,2; the second category is for high-spatial resolution, large-scale imaging of static samples during an inspection, which is discussed in detail in this work. With transistor size shrinking down to the nanoscale, the semiconductor industry urgently needs high-throughput AFMs to inspect wafer-scale nanofabricated devices with nanoscale spatial resolution3.
The characterization of nanofabricated devices on a wafer can be challenging due to the vast scale difference between wafer and transistor features. Large defects can be spotted with optical microscopes automatically4. In addition, scanning electron microscopes (SEMs) are widely used for inspection down to tens of nanometers in 2D5. For 3D information and higher resolution, the AFM is a more suitable tool if its throughput can be improved.
With limited imaging throughput, one approach is to image selected wafer areas where nanofabrication defects are more likely to happen6. This would require prior knowledge of the design and fabrication process. Alternatively, combining other modalities, such as an optical microscope or SEM with an AFM for overview and zoom, is possible7,8. A wide-range, high-precision positioning system is needed to properly align the coordinate system between the fabrication and characterization tools. Moreover, an automated AFM system to image various selected areas is necessary to realize this functionality.
As an alternative, researchers have investigated different ways to increase AFM scanning speed. Since enabling high-throughput AFMs is a systematic precision instrumentation challenge, researchers have investigated various methods, including using smaller AFM probes, redesigning high-bandwidth nano-positioners9,10,11,12 and driving electronics13, optimizing modes of operation, imaging control algorithms14,15,16,17, etc. With these efforts, the effective relative tip and sample speed can be increased to a maximum of around tens of millimeters per second for commercially available single-probe AFM systems.
To further improve the imaging throughput, adding multiple probes to operate in parallel is a natural solution. However, the optical beam deflection (OBD) system utilized for cantilever deflection sensing is relatively bulky, which makes the addition of multiple probes relatively challenging. Individual cantilever deflection control can also be difficult to realize.
To overcome this limitation, embedded sensing, and actuation principles without bulky external components, are preferred. As detailed in previously published reports18,19, deflection sensing with piezoresistive, piezoelectric, and optomechanical principles can be considered embedded sensing, with the former two being more mature and easier to implement. For embedded actuation, thermomechanical with electrical heating or piezoelectric principles can both be utilized. Although piezoelectric principles can operate in a wider temperature range down to cryogenic environments, they can only support tapping mode AFM operations, since static deflection cannot be measured due to the charge leakage and static actuation suffering from hysteresis and creep. In prior work, active cantilever probe arrays using a piezoresistive sensor and the piezoelectric sensor have been developed for large-range imaging20,21, but have not been further scaled up for large-scale imaging or commercialized. In this work, the combination of piezoresistive sensing and thermomechanical actuation are selected as embedded transducers with static deflection control capability.
In this work, a novel "Quattro"22 parallel active cantilever array is used as the probe23 for simultaneous imaging using active cantilevers. To measure the cantilever deflection, piezoresistive sensors in a Wheatstone bridge configuration19 are nanofabricated at the base of each micro-cantilever to measure the internal stress, which is linearly proportional to the cantilever tip deflection. This compact embedded sensor can also achieve sub-nanometer resolution as the conventional OBD sensor. The governing equation of the Wheatstone bridge voltage output Uout in response to the applied force F or cantilever deflection z is shown in Equation 119 for a cantilever with length L, width W, and thickness H, piezoresistive sensor coefficient PR, and effective elastic modulus of the cantilever E bridge supply voltage Ub.
(1)
As dynamic tapping/non-contact mode operation is preferred for noninvasive imaging to avoid disturbing the sample, a thermomechanical actuator made of serpentine shape aluminum wires is used to heat up the bimorph cantilever made with aluminum/magnesium alloy24, silicon, and silicon oxide materials. At the microscopic scale, the time constant of thermal processes is much smaller, and the cantilever resonance at tens to hundreds of kilohertz can be excited by driving the heater with an electrical signal. The cantilever free end deflection zh controlled by the heater temperature ΔT relative ambience is shown in Equation 219for cantilever length L with a constant K, depending on the bimorph material thermo coefficient of expansion and geometric thickness and area. It should be noted that the ΔT is proportional to the heater power P, which is equal to the square of the applied voltage V divided by its resistance R.
(2)
As an added benefit, static deflection can also be controlled in addition to resonance excitation. This can be a particularly helpful capability to regulate the probe-sample interaction of each cantilever individually. Moreover, multiple cantilevers on the same base chip can be excited individually with the embedded thermomechanical actuator, which is impossible in conventional resonance excitation with piezo-generated acoustic waves.
Combining piezoresistive sensing and thermomechanical actuation, the active cantilever probe has enabled a wide range of applications, including collocated AF microscopy in SE microscopy, imaging in opaque liquid, and scanning probe lithography, with more details available in review25. For high-throughput inspection purposes, the active cantilever array is created with a representative AFM implementation example involving four parallel cantilevers, as shown in Figure 1. In the future, an industrial-scale system will be developed using eight parallel active cantilevers and tens of positioners28. To illustrate the scale using an example, with an in-plane spatial resolution of 100 nm, imaging an area of 100 mm by 100 mm would result in over 106 scan lines and 1012 pixels. With a scanning speed of 50 mm/s per cantilever, this would require a total of over 555.6 h of scanning (23+ days) for a single cantilever, which is too long to be practically useful. Using the active cantilever array technology with tens of positioners, the required imaging time can be reduced by around two orders of magnitude to 5-10 h (less than half a day) without making any compromise to the resolution, which is a reasonable time scale for industrial inspection purpose.
To capture large-area, high-resolution images, the nano-positioning system is also upgraded. For imaging wafer-scale large samples, scanning the probe instead of the sample is preferred, in order to reduce the size of the objects being moved. With the separation distance between active cantilevers at 125 µm, the scanner covers an area slightly larger than this range so that images from each cantilever can be stitched together during post-processing. Upon completion of a scan, the coarse positioner automatically repositions the probe to a new adjacent area to continue the imaging process. While the embedded thermomechanical actuatorregulates the deflection of each cantilever, the averaged deflection of all parallel cantilevers is regulated with another proportional-integral-derivative (PID) controller to assist the cantilevers during topography tracking. The scanner controller also ensures that the bending of each cantilever does not exceed a maximum threshold value, which may cause other probes to lose contact with the surface if the topography variation is too large.
The level of topography variation that can be tracked for cantilevers on the same base chip should it be limited, since the static deflection control range of the cantilever is on the order of tens of microns. For semiconductor wafers, the sample topography variations are typically on the sub-micrometer scale, so they should not be much of a big problem. However, with the addition of more cantilevers, the sample plane tilt with respect to the line of cantilevers can become a problem. In practice, eight parallel cantilevers with spacings close to 1 mm would still allow 1° of tilt angle, while adding more cantilevers can make the tilting control more difficult to realize. Therefore, using multiple groups of eight-cantilever probes placed on separated probe scanners is an ongoing effort to fully realize the potential of the parallel active cantilever probe principle.
After data collection, a post-processing operation is necessary to retrieve the desired information. The process generally involves removing scanning artifacts, stitching adjacent images to form an overall panorama, and optionally identifying the structure defects by comparing them to the desired geometry using suitable algorithms26. It is worth noting that the amount of data accumulated can be enormous for a large range of images, and data-driven learning algorithms are also being developed for more efficient processing27.
This article illustrates the general process of acquiring high-resolution AFM images using the parallel active cantilever array integrated into a custom AFM system. Detailed implementation of the system is available in22,28,29,30, and it is being commercialized with the model number listed in Table of Materials. All four cantilevers were operated in tapping mode excited by the embedded thermal-mechanical actuator. Representative results on calibration samples, nanofabrication masks, and highly oriented pyrolytic graphite (HOPG) samples (see Table of Materials) are provided to illustrate the effectiveness of this new AFM tool for large-area inspection.
1. Sample preparation for large-scale inspection
2. AFM instrument calibration and experiment setup
3. Topography imaging and parameter tuning
4. Post-processing and image analysis
To demonstrate the effectiveness of AFM large-range imaging using parallel active cantilevers for topography imaging, the stitched images of a calibration grating, taken by four cantilevers operated in parallel, are shown in Figure 2. The silicon wafer calibration structure has 45 µm long features with a height of 14 nm. Each cantilever covers an area of 125 µm by 125 µm, which gives a stitched panoramic image of 500 µm by 125 µm. The scanning speed was set to 10 lines per second at 1,028 pixels per line and channel in the amplitude modulation mode, so it takes less than 2 min to complete this large-area scan.
The merging of the images taken by each cantilever is performed by aligning the features on the edge of the adjacent images. With an actual imaging size larger than the cantilever separation, the merging is performed by correlating the features on the edges. It is worth noting that some vertical offset between each image in the in-plane Y-axis direction is also visible. This can happen due to the slight mismatch of the scanning axis with respect to the line of the four active cantilever arrays. However, the correlation method can be difficult to apply for boundaries without significant topography variation. Therefore, using correlation-based matching with prior offset knowledge to create panoramic images is the preferred method, compared to direct stitching using relative position offset to deal with these imperfect conditions of the instrument.
To verify the spatial resolution of the active cantilever array, high-resolution images of HOPG are taken, as shown in Figure 3, with a small in-plane image range of 5 µm by 5 µm and 1028 by 1028 pixels. HOPG samples are particularly suitable for resolution verification since the inter-plane spacing of graphite is around 0.335 nm31,32. Sub-nanometer out-of-plane resolution and in-plane resolution at several nanometers are demonstrated. As the separation distance between each cantilever at 125 µm is much larger than the 5 µm imaging area, these four images cannot be directly stitched, but the orientation trend of the imaged features between adjacent images aligns well with each other.
For practical applications in semiconductor inspection, an EUV lithography mask for creating semiconductor features is imaged using the parallel active cantilever array. An overall stitched panoramic image with a 5 nm spatial resolution covering an area of 505 µm by 130 µm is shown in Figure 4. The height of the structure pattern is around 60 nm, with various areas of the circuit clearly shown in the image. At 10 lines per second, the 101,000 by 26,000 pixels are captured within around 40 min, which is significantly faster than conventional AFM systems.
Figure 1: Large-area AFM for large-area sample inspection example implementation using a single array of four active cantilever probes. (A) Large-area imaging of a silicon wafer sample using the custom AFM with probe scan configuration and a large sample stage. (B) Simplified schematic of the AFM system with an optical microscope zoomed-in view of the area of four active cantilevers wire-bonded to the SD card shape printed circuit board (PCB). (C) SEM image of tapping actuation for one of the four active cantilevers showing a maximum amplitude over 30 µm. (D) SEM image of the active cantilever array with a schematic showing the serpentine-shaped thermomechanical heater and piezoresistive stress sensors at the base of the cantilevers for deflection measurement. Scale bar = 50 µm. Please click here to view a larger version of this figure.
Figure 2: Panoramic merging of images taken by four active cantilevers simultaneously over a total width of 500 µm. Each cantilever scans over an area of 140 µm to create some overlap between the 125 µm separation of the cantilever tip in amplitude modulation dynamic tapping mode. The image is taken at 10 lines per second with a resolution of 1,028 pixels in each direction. The sample is a silicon test structure with 45 µm long lines at a height of 14 nm. The top four separate images taken by each cantilever are stitched to form the panorama image on the bottom. The figure is adapted from Ahmad et al.22. Please click here to view a larger version of this figure.
Figure 3: High-resolution AFM images of HOPG samples. The images are captured simultaneously with four cantilevers with a 3 µm by 3 µm area, captured at 10 lines per second with a resolution of 1,028 pixels in each direction. (A–D) Topography images captured in amplitude modulation dynamic tapping mode by cantilevers 1-4, respectively. Please click here to view a larger version of this figure.
Figure 4: Record of four 2D and 3D single EUV mask images captured simultaneously by the four AFM channels at 10 lines per second in amplitude modulation dynamic tapping mode. The imaging field of a single image is 130 µm x 130 µm. (A) Four 2D images. (B) Four 3D images. (C) Overall 3D stitched image with 500 µm by 500 µm obtained with four images of 125 µm, where 5 µm is the overlap between the single fields. The image iss 101,000 by 26,000 pixels with a 5 nm spatial resolution. Please click here to view a larger version of this figure.
As demonstrated in the representative results, an active cantilever array can be used to capture multiple images of a static sample in parallel. This scalable setup can significantly improve the imaging throughput of large-area samples, making it suitable for inspecting nanofabricated devices on semiconductor wafers. The technique is also not limited to man-made structures; as long as the topography variation within a group of active cantilevers is not too large for the cantilever array to handle, high-throughput imaging can be realized.
In addition to enabling high-throughput, large-area inspection, an active cantilever probe array offers several advantages in terms of imaging setup. First, there is no need to perform laser-cantilever alignment for probe installation. In terms of operation, this reduces the operator overhead. More importantly, the gain of the cantilever deflection sensor is fixed and does not change between experiments. Quantitative force and deflection measurements can be taken with these probes both in tapping mode and contact mode29,33,34. This also makes the imaging process more reliable, as drifting of the laser alignment for long-term imaging is longer be a problem. Second, the thermomechanical cantilever drive avoids the spurious structure resonance of the cantilever holder, which can become a problem during resonance sweep in conventional piezo acoustic actuation. The conventional resonance excitation technique uses a piezoelectric actuator placed on the base support chip of the AFM cantilever. Since the vibration generated is acoustically propagated through the entire base chip, cantilever resonance excitation may indeed interfere with each other. However, the thermomechanical actuation directly acts on the individual cantilever, and the base support chip remains stationary. As the mass of the base support chip is several orders of magnitude higher than the cantilever, the interference between the parallel active cantilevers is negligible. Third, the compact size of the active cantilever allows easier parallel integration for probe scan configuration. This means the sample can remain static, and multiple probe positioners can scan simultaneously at different speeds during imaging, which helps to maximize the effective utilization of each cantilever.
Regarding limitations, handling samples with large topography changes can be challenging due to the maximum deflection limit of each cantilever. Some special consideration needs to be taken during the sample preparation and installation. As the sample being handled is on the macroscopic scale, the tilt with respect to the scan plane should be minimized to ensure good tracking of the topography. Tilting of the surface larger than 1° with respect to the scanner stage can result in out-of-range cantilever deflection control that may cause damage to the probe. For nanofabricated structures on a semiconductor wafer, the flatness is typically guaranteed and no polishing is needed. This also avoids potential damage to the fine features to be imaged. The surface finish of conventional computer numerical control (CNC) machining on the micrometer level should be sufficient for the active cantilever array to handle. For generic samples, polishing may be required at the cost of altering surface features to be captured. A CNC machine is used to remove large unwanted protruding features. If large topography variation cannot be avoided, such as on a curved surface, using an array of two parallel active cantilevers with tilting control accommodates large topography variation. Multiple separated positioners would be needed for parallelization to further improve imaging throughput with more cantilever probes. Using nanofabrication techniques, it is also possible to fabricate a nanoscale nano-positioning system on the Z-axis to better address this problem in a more compact design35.
To fully realize the potential of the parallel cantilever array, especially for semiconductor inspection purposes, more engineering developments are in progress to commercialize the system. The goal is to integrate a probe with an array of eight active cantilevers into a three-axis piezo scanner and replicate tens of such structures with precise motion control for parallel imaging. With this setup, a 60 mm2 area with a 100 nm spatial resolution can be imaged within 30 min, which should be sufficient for many inspection applications. Using dynamic mode imaging in non-contact mode, the probe-sample force interaction is small at the cost of a slower imaging rate. As a trade-off, contact mode can increase the imaging speed significantly, but can increase the probe-sample interaction force and may result in sample damage or probe tip wear. To further ensure the longevity of these probes, diamond tips can also be used to significantly reduce the probe tip wear for long-term, continuous operation. To ensure good imaging performance, the imaging environment should be controlled to have low vibration and dust, to avoid particles landing on the sample during the imaging process.
In terms of software improvements, automated parameter tuning for a large number of controllers is being investigated. Adaptive scan speed and resolution adjustment are desirable for imaging samples with large property variations. Automated stitching of thousands of images and identifying defects from billions of pixels using machine learning-based algorithms can further help make this technique even more useful in research studies and industrial inspection.
The authors have nothing to disclose.
The authors Ivo W. Rangelow and Thomas Sattel would like to acknowledge the German Federal Ministry of Education and Research (BMBF) and the German Federal Ministry of Economics Affairs and Climate Action (BMWK) for supporting parts of the presented methods by funding the projects FKZ:13N16580 "Active Probes with diamond tip for quantum metrology and nanofabrication" within the research line KMU-innovativ: Photonics and Quantum Technologies and KK5007912DF1 "Conjungate Nano-Positioner-Scanner for fast and large metrological tasks in Atomic Force Microscopy" within the funding line Central Innovation Program for small and medium sized industries (ZIM). Part of the work reported here was funded by the European Union Seventh Framework Programme FP7/2007-2013 under Grant Agreement No. 318804 "Single Nanometre Manufacturing: Beyond CMOS." The authors Ivo W. Rangelow and Eberhard Manske gratefully acknowledge the support by the Deutsche Forschungsgemeinschaft (DFG) in the framework of Research Training Group "Tip- and laser-based 3D-Nanofabrication in extended macroscopic working areas" (GRK 2182) at the Technische Universität Ilmenau, Germany.
Active-Cantilever | nano analytik GmbH | AC-10-2012 | AFM Probe |
E-Beam | EBX-30, INC | 012323-15 | Mask patterning instrument |
Highly Oriented Pyrolytic Graphite – HOPG | TED PELLA, INC | 626-10 | AFM calibration sample |
Mask Sample | Nanda Technologies GmbH | Test substrate | EUV Mask Sample substrate |
NANO-COMPAS-PRO | nano analytik GmbH | 23-2016 | AFM Software |
nanoMetronom 20 | nano analytik GmbH | 1-343-2020 | AFM Instrument |