This experiment shows the high throughput capability of an Atomic Force Microscopy system utilizing high frequency (AC55, 1.6 MHz) cantilevers on Park Systems AFM.
The throughput is increased to an industrial level with these cantilevers while maintaining performance in image quality and tip lifetime by using Park Systems industrial pattern recognition and operating software Park XEA.
Optical microscopy methods are no longer able to resolve the features of nanoscale technology reliably (including the 14 nm and 7 nm architectures).
Atomic force microscopy (AFM) has the resolution to serve this market, yet, past methods performed slower than optical options, resulting in a lower overall throughput, a drawback resulting in low adoption of the technology. Full automation of the method is one solution to this issue.
Using optical pattern recognition, automatic data analysis, automatic tip exchange, wafer handling and recipe optimization has made AFM with high throughput software a reliable technique for the characterization of nanoscale devices.
Adding high frequency scanning also increases throughput. While the method is not new, this data shows application of the higher speed scanning techniques in combination with automation software, enabling industrial scaling of the method.
Combining the power of automation, software like Park Systems XEA, and AFM systems like Park NX20, Park NX-Wafer, Park NX20-300mm, or Park NX-3DM can result in higher throughput for topography measurements.
In order to characterize the rms roughness (Rq) of a clean 200 mm silicon wafer, a Park NX20 system equipped with Park XEA software and 200 mm sample chuck was employed. For a sidewall angle measurement, a 150 mm patterned wafer was utilized.
The aim was to demonstrate the performance of the system in this configuration for these common applications, over time. To characterize the bare silicon wafer, a recipe was created to image 17 sites with 1x1 µm images for peak to valley (RPV) variation and rms roughness (Rq), five mm apart, over the horizontal axis of the wafer.
In addition, the center of the wafer was measured. Site variation was selected to show real measurement processes by having the system lift, move to another site, and re-approach. While testing approach parameters, different set point, and scan parameters were optimized for extended tip lifetime.
Larger cantilever driven amplitudes were discovered to image at these higher speeds but at the cost of hard tapping of the surface, which decreased tip lifetime. So, as generally applied in current industrial applications, lower amplitudes are utilized.
The second recipe was developed and used for a die-based wafer to measure angles with high speed. In this instance, based on feature characteristics, the image was optimized for 3 Hz.
The targeted feature was measured at five sites for 80 repeats using dye-based pattern recognition, positioning, scan-based post positioning, and analysis. The image was 2x10 µm at 512x64 pixels.
The angle was calculated with a 50% offset and fit to 10% bounds above and below this offset. In general, the feature was approximately 1.4 µm in height with an angle of around 27 degrees, as seen in Figure 1.
Figure 1. Pattern recognition and positioning as defined by the operator, 40 µm reference scan (green) compared with the reference (gray), leading to the final image (taken at 3 Hz) with offset (blue) and angle measurement (red).
Recipe Development in XEA 5.0
The same principles and software were applied for a manually loaded wafer tool as for automatic systems, like the Park NX-Wafer and NX-3DM in order to produce an automatic measurement routine using Park XEA. Park XEA is designed to supply the operator with a clean canvas and numerous tools to create custom measurements.
This template enables creativity to be applied by the operator. In XEA, the automatic measurement is known as a recipe. Within the recipe, the image and analysis process is known as the method and is embedded in the recipe. Multiple methods can be implemented in a single recipe.
As samples possess varied features and different parameters are required to be measured, the Park XEA recipe is designed to break down a wafer into smaller portions to be handled efficiently by the method.
A recipe can be designed for a blank wafer or a die-based wafer. Its key purpose is to find the location for measurement in addition to loading background settings for the wafer size and tip to be utilized (automatically switched on equipped tools). The method will gather the AFM image and analyze it.
When using pattern recognition for a die-based wafer, the lower-left corner of the die is found by the operator and the die is mapped out. Park XEA then automatically verifies and calculates the die size.
Wafer alignment can be employed to establish possible angle offset of the wafer by pattern recognition on two dies closer to the edge of the wafer in the same channel.
This is a common method that is employed for manually mounted wafers which produce a separate coordinate system specifically for the wafer and enabling precise positioning for the technique.
For a blank wafer, the edges of the wafer are found with pattern recognition via a method known as circle alignment. The coordinates have to be assigned to the wafer on a larger scale as there are no repeatable features on the wafer. Finer alignment choices are available; yet, blank wafers are usually featureless.
For each recipe, at each point specified in the recipe, a technique is applied. This can be a topography scan, profile, roughness measurement, or whatever is required by the customer. In this instance, finer positioning is applied with choices like optical pattern recognition at a location or a reference scan and offset.
The actual image scan parameters are also applied. Settings for automatic image processing and analysis are finally applied for data output. So, the recipe tells the system where to go, the method(s) tells the system what images to take and how to analyze them.
This template enables a number of measurements to be repeated efficiently over multiple positions on the wafer.
Utilizing True Non-Contact™ in a Technique
True Non-Contact™ mode is employed to keep the tip out of contact with the surface in order to extend tip lifetime. This is done by keeping the probe in the attractive region of the Lennard Jones potential.
The Van der Waals interaction between the sample and the tip in this region is enough to image the surface while keeping the tip from tapping the surface. This keeps the tip sharp across long scanning distances.
Park AFMs are designed to work in this region. However, with the high-frequency cantilevers, higher resonance is usually achieved by making the cantilever shorter. As the cantilever is not only shorter, but stiffer, this makes the region between True Non-Contact and intermittent tapping smaller, and so harder to optimize for.
Regardless of this, the images show that it is possible to push this imaging method and achieve a very long tip lifetime with the AC55 (see Figures 2 and 3). The advantage is better throughput with maintained resolution.
Standard imaging of this sample was optimized for 0.8 Hz with a standard AC160 probe, with AC55 optimization set for 4 Hz in this instance. For these images (the Park AFM employed is capable of scanning at rates of 10 Hz and above) the limitations for the scan rate include the scan size, probe used, and sample surface features (see Figure 2).
Scan time for these images (256 x 256 pixels) decreased from around five and a half minutes to around one minute at 4 Hz. Throughput increase is dramatic over thousands of scans.
Figure 2. Comparison of AC55 and AC160 image resolution on 200 mm Si sample: (A) AC55 at 4 Hz RPV 12.498 nm Rq 1.555 nm; (B) AC160TS at .8 Hz RPV 11.3 nm, Rq 1.431 nm.
Rq and RPV Analysis
Multiple recipes and tip parameters were adjusted and tested in order to push the capabilities. The following AC55 seen in Figure 3, acquired 26.1 cm of scanning while maintaining image quality and resolution. Numerous tips were run with the same recipe to test track Rq and RPV. Figure 4 shows one such data set over 10.8 cm of scanning.
Figure 3. Silicon Surface 1x1 µm image at 4Hz: (A) First Image; (B) 100th Image; (C) 500th Image; (D) 1000th Image.
In order to achieve 1000 scans, each run consisted of 17 1x1 µm images across the horizontal axis of the wafer 5 mm apart for six sites, 5 cm from the origin, then the origin itself. RQ and RPV remain consistent across the measurement.
This resulted in 1020 images and 26.1 cm of scanning (each image was 128 lines at 512-pixel resolution). The same recipe was performed with new AC55s over 10.8 cm to track Rq and RPV, yet, the recipe was only repeated 25 times in this instance.
Figure 4. Variation in Rq and RPV over 17 sites, 25 repeats, 10.8 cm scanned (128 lines at 512 pixels).
An amplitude vs. distance curve (A/D Curve) is a quick technique of characterizing tip degradation generally, due to scanning beyond image quality and artifacts. The tip is driven as for a non-contact or intermittent tapping image and slowly lowered to the surface.
The change in the direction or phase of the Van der Waals potential can demonstrate the switch between the attractive to a repulsive regime of the Lennard Jones potential, i.e. showing roughly where the tip begins to contact the sample (the phase switches from negative to positive in this instance, showing an alteration in the slope of the potential).
In comparison, as seen in Figure 5, the new AC55 has an interaction with the sample similar to the same AC55 after 26.1 cm of scanning. The curve exhibits a maintained tip sharpness although some slight blunting has likely happened.
Figure 5. Amplitude/Distance and Phase/Distance Curves before and after 26.1 cm of scanning.
In its current state, this tip could continue to image. A fully dulled and unusable tip will show not only a much earlier switch in the phase but also the interaction is less consistent and more erratic because of the larger surface area, as the surface area of a blunted tip is much larger or chipped (see Figure 6).
Figure 6. Amplitude/Distance and Phase/Distance Curves after tip blunting.
Angle Measurement Die Based Wafer
After 80 repeats with such a large structure, angle measurement varied little and image quality stayed consistent over the 400 images for the die-based wafer angle measurement.
Figure 7. Variation in Angle over 80 Repeats.
The results above show the capability of using an AFM system in combination with high-frequency probes and scanning techniques with industrial automation. These examples demonstrate imaging speeds which are three to four times quicker than standard probes while maintaining data quality.
Furthermore, it shows that a single AC55 cantilever maintains image resolution for 26.1 cm of scanning at 4 Hz and micron tall angled features that measure 400 times at 3 Hz with a Park AFM and Park XEA operating software.
The AFM technology, like the Park NX20 system employed here, supplies increased scanning throughput while maintaining reliable data at an industrial level of automation with these higher resonance cantilevers.
Produced from materials originally authored by Ben Schoenek, Kevin Ryang, Byong Kim and Keibock Lee from Park Systems Inc.
This information has been sourced, reviewed and adapted from materials provided by Park Systems.
For more information on this source, please visit Park Systems.