AFM imaging with measurement storage and real-time noise reduction

Abstract

A method for operating an atomic force microscope (AFM) in real-time using noise reduction algorithms during AFM data acquisition. Total variation and non-local means noise reduction are preferred. As described herein, real-time images are produced with minimized sensor noise that do not require post-image acquisition processing to account for noise.

Description

[Technical field]

[0001] The preferred embodiments relate to Atomic Force Microscopy (AFM), and in particular to a real-time method for removing AFM sensor noise while preserving AFM metrology data. [Background technology]

[0002] A scanning probe microscope, such as an atomic force microscope (AFM), is an instrument that employs a probe having a tip to interact with the surface of a sample with a force suitable for characterizing the surface down to the atomic level with the tip. Typically, the probe is introduced to the surface of the sample and by providing a relative scanning movement between the tip and the sample, surface property data can be obtained for specific regions of the sample and a corresponding map of the sample can be generated.

[0003] A typical AFM system is shown diagrammatically in FIG. 1. An AFM 10 uses a probe arrangement 12 including a probe 14 with a cantilever 15. A scanner 24 generates relative motion between the probe 14 and a sample 22 while the probe-sample interaction is measured. In this manner, images or other measurements of the sample can be obtained. The scanner 24 is typically composed of one or more actuators that generate motion in three orthogonal directions (XYZ). Often, the scanner 24 is a single integrated unit that includes one or more actuators that move the sample or probe in all three axes, for example, piezoelectric tube actuators. The scanner may also be an assembly of multiple separate actuators. Some AFMs separate the scanner into multiple components, for example, an XY scanner that moves the sample and a separate Z actuator that moves the probe. The apparatus can thus generate relative motion between the probe and the sample while measuring the topography or some other surface characteristic of the sample, as described, for example, in U.S. Patent No. 5,399,233 to Hansma et al., U.S. Patent No. 5,399,233 to Elings et al., and U.S. Patent No. 5,399,233 to Elings et al.

[0004] In a typical configuration, the probe 14 is often coupled to an oscillating actuator or drive 16 that is used to drive the probe 14 at or near the resonant frequency of the cantilever 15. Alternative arrangements measure the deflection, torsion, or other motion of the cantilever 15. The probe 14 is often a micromachined cantilever with an integrated tip 17.

[0005] Typically, an electronic signal from a signal source 18 is applied under the control of an SPM controller 20 to cause an actuator 16 (or alternatively a scanner 24) to vibrate the probe 14. The probe-sample interaction is typically controlled by feedback by the controller 20. In particular, the actuator 16 may be coupled to the scanner 24 and the probe 14, but may also be formed integrally with the cantilever 15 of the probe 14 as part of a self-actuated cantilever / probe.

[0006] As previously mentioned, sample characteristics are monitored by detecting changes in one or more characteristics of the vibration of the probe 14, often when a selected probe 14 is in contact with the sample 22 and vibrates. In this regard, a deflection detector is typically utilized to direct a beam to the back side of the probe 14, where it is then reflected towards a detector 26. As the beam moves across the detector 26, an appropriate signal is processed in block 28 to determine, for example, an RMS deflection, which is sent to the controller 20, which processes the signal to determine the vibration changes of the probe 14. In general, the controller 20 generates control signals to maintain a relatively constant interaction between the tip and sample (or deflection of the lever 15) to maintain a setpoint characteristic of the vibration of the probe 14. More specifically, the controller 20 can include a High Voltage Amplifier 34 and a PI Gain Control block 32 that compares a signal corresponding to the probe deflection due to the tip-sample interaction with a setpoint to a circuit 30 to condition the obtained error signal. For example, the controller 20 may set a fixed point value A to ensure a generally constant force between the tip and the sample. S A set point phase or frequency may be used alternatively.

[0007] A workstation 40 may also be provided within the controller 20 and / or in a separate controller or linked system, or in a stand-alone controller, to receive data obtained from the controller and manipulate data acquired during scanning to perform point selection, curve fitting, and distance determination.

[0008] AFMs may be designed to operate in a variety of modes, including contact and vibration modes. Operation is accomplished by moving the sample or the probe assembly up and down perpendicular to the surface of the sample in response to deflections of the cantilever of the probe assembly as it is scanned across the surface. Scanning typically occurs in an "xy" plane that is at least parallel to the surface of the sample, and vertical movement occurs in the "z" direction perpendicular to the xy plane. The term "generally parallel" is used because many samples have out-of-plane roughness, curvature, and tilt. Data associated with this vertical movement may thus be stored and used to construct an image of the sample surface that corresponds to measured sample characteristics, for example, surface topography. In a mode of AFM operation known as TappingMode® AFM (TappingMode® is a trademark of the applicant), the tip oscillates at or near the resonant frequency of the cantilever relative to the probe. A feedback loop attempts to minimize the "tracking force", i.e., the force resulting from the tip / sample interaction, and to keep the amplitude of the oscillation constant.

[0009] An alternative feedback device holds the phase or frequency of oscillation constant. In contact mode, such feedback signals are collected, stored, and used with data to characterize the sample. Note that "SPM" and abbreviations for specific types of SPM may be used herein to refer to microscope devices or related technologies, such as atomic force microscopes. In an improvement of the ubiquitous TappingMode® called Peak Force Tapping® (PFT), discussed in U.S. Pat. No. 5,399,433, U.S. Pat. No. 5,499,433 and U.S. Pat. No. 5,523,637, the feedback is expressly incorporated herein by reference and is based on the force (termed the probe-sample interaction force) measured at each oscillation period.

[0010] Regardless of the mode of operation, AFMs can resolve down to the atomic level on a variety of insulating and conducting surfaces in air, liquid, or vacuum by using a piezoelectric scanner, an optical lever deflection detector, and a tiny cantilever fabricated using photolithography techniques. Due to their resolution and versatility, AFMs are important measurement tools in a variety of fields ranging from semiconductor manufacturing to biological research.

[0011] In this regard, AFMs may be used in automated applications involving high precision manufacturing processes such as semiconductor manufacturing. AFMs have proven useful in the semiconductor space because they can provide high resolution measurements of nanoscale surface features (e.g., topography). However, AFM data is typically hampered by AFM system noise, which includes a sensor system such as the optical beam bounce array mentioned above that is used to collect metrology data.

[0012] Other denoising techniques such as Non-local means (NLM) (https: / / en.wikipedia.org / wiki / Non-local_means) and Total variation (TV) (https: / / en.wikipedia.org / wiki / Total_variation_denoising) have been applied to AFM image data to resolve sensor noise. In this case, manual image denoising is performed after image acquisition. The limitations of image data post-processing include additional time required to finally display an image with minimized sensor noise and the inability to adjust scanning parameters during data acquisition. Manual image post-processing cannot automate data acquisition and processing, which is required in a high-speed FAB (e.g., semiconductor manufacturing) environment.

[0013] As a result, the AFM field, including automated AFMs, has needed a solution to address and compensate for system noise, such as sensor noise. Improved data acquisition and display times are needed, and it is preferable to minimize post-image acquisition processing. Such a system would enable AFMs to produce clean, real-time images of a variety of sample surfaces with high resolution. [Prior art documents] [Patent documents]

[0014] [Patent Document 1] U.S. Reissue Patent No. 34489 [Patent Document 2] U.S. Patent No. 5,266,801 [Patent Document 3] U.S. Patent No. 5,412,980 [Patent Document 4] U.S. Patent No. 8,739,309 [Patent Document 5] U.S. Patent No. 9,322,842 [Patent Document 6] U.S. Pat. No. 9,588,136 Summary of the Invention [Problem to be solved by the invention]

[0015] Eliminate image noise and overcome the shortcomings of current AFM systems. [Means for solving the problem]

[0016] The preferred embodiment overcomes the shortcomings of current AFM systems that use post-imaging processing to remove noise from an image by providing an AFM and corresponding method for removing noise from an AFM image in real-time as the AFM image is acquired. Real-time processing performs data processing within a short time frame during AFM image acquisition, providing near-instantaneous output. A baseline noise image is automatically acquired prior to AFM operation / image acquisition, and then a specified number of lines are imaged and denoised using an initial noise measurement. A block of scan lines is selected (e.g., 5 lines) and denoised in real-time to generate a "clean" version of at least one of the lines (the middle lines). The middle lines of each of the series of line blocks form the final image. Noise removal may be performed using known algorithms such as the NLM and TV algorithms previously described. These steps are repeated until the sample or region of interest of the sample is imaged. Thus, the method and system generate an AFM sample image in which the sensor / system noise is substantially removed, while preserving the measurements.

[0017] According to a preferred embodiment of an atomic force microscopy (AFM) method, the method includes the steps of engaging a probe of the AFM with a sample surface and oscillating the probe in an AFM operating mode. The method then includes measuring a system noise to generate a noise image. A relative scanning motion between the probe and the sample is then provided, and a deflection of the probe due to the providing step is measured. A probe-sample separation is controlled according to the AFM operating mode for a selected number of scan lines to generate a sample image for a selected number of scan lines. The method then includes de-noising the sample image using the system noise. The deflection for the next line of the scanning motion is measured to generate a new sample image, and the new sample image is de-noised in real time using the system noise.

[0018] In an additional aspect of this preferred embodiment, the selected number is at least two, and more preferably, the selected number is at least five, and the denoised sample image is intermediate the selected number.

[0019] In a further aspect of this embodiment, the step of removing noise includes using a non-local means denoising algorithm, a total variation denoising algorithm, or other image denoising algorithm.

[0020] According to another aspect of this embodiment, the scanning motion is a raster scan and the method includes, after the step of removing noise, repeating all steps including the step of removing noise until the region of interest of the sample is imaged.

[0021] According to a further aspect of this embodiment, the method optionally includes applying one of a low pass kernel and a deconvolution kernel to the denoised image. According to yet another preferred embodiment, an atomic force microscope (AFM) includes a scanner that provides relative scanning motion between a probe of the AFM and a sample, a detector that measures deflection of the probe in response to probe-sample interaction during AFM operation (the deflection being indicative of a sample property and stored with an image), and a controller that executes a denoising algorithm to denoise images of a selected number of scan lines in real time during AFM operation.

[0022] In another aspect of this embodiment, the selected number is at least two scanlines, and more preferably, the selected number is at least five scanlines, and the denoised image is midway between the at least five scanlines.

[0023] According to an additional aspect of this embodiment, the controller applies one of a low pass kernel and a deconvolution kernel to the denoised image to minimize offset effects between the trace and retrace AFM imaging operations.

[0024] These and other features and advantages of the present invention will become apparent to those skilled in the art from the following detailed description and the accompanying drawings. It should be understood, however, that the detailed description and specific examples, while indicating preferred embodiments of the present invention, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the present invention without departing from the spirit thereof, and the present invention includes all such modifications.

[0025] Preferred embodiments of the present invention are illustrated in the accompanying drawings, in which like reference characters refer to like parts throughout. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 is a schematic diagram of a prior art atomic force microscope (AFM). [Diagram 2] FIG. 1 is a simplified block diagram of the real-time noise reduction process for AFM image data. [Diagram 3] FIG. 1 is a block diagram of an AFM system for acquiring AFM data and performing real-time noise removal. [Figure 4] 1 is a flow chart illustrating the real-time denoising method of the preferred embodiment using non-local means or total variation filtering. [Figure 5A] 1 is a series of AFM images using a real-time denoising method of a preferred embodiment using total variation (TV) filtering and a deconvolution or low-pass kernel to remove the offset between the trace-retrace lines. [Figure 5B] 1 is a series of AFM images using a real-time denoising method of a preferred embodiment using total variation (TV) filtering and a deconvolution or low-pass kernel to remove the offset between the trace-retrace lines. [Figure 5C] 1 is a series of AFM images using a real-time denoising method of a preferred embodiment using total variation (TV) filtering and a deconvolution or low-pass kernel to remove the offset between the trace-retrace lines. [Figure 5D] 1 is a series of AFM images using a real-time denoising method of a preferred embodiment using total variation (TV) filtering and a deconvolution or low-pass kernel to remove the offset between the trace-retrace lines. [Figure 5E] 1 is a series of AFM images using a real-time denoising method of a preferred embodiment using total variation (TV) filtering and a deconvolution or low-pass kernel to remove the offset between the trace-retrace lines. [Figure 6A] 13 is a series of AFM images using the real-time denoising method of the preferred embodiment using non-local means (NLM) as well as deconvolution or low-pass kernels to remove offsets between trace-retrace lines. [Figure 6B]13 is a series of AFM images using the real-time denoising method of the preferred embodiment using non-local means (NLM) as well as deconvolution or low-pass kernels to remove offsets between trace-retrace lines. [Figure 6C] 13 is a series of AFM images using the real-time denoising method of the preferred embodiment using non-local means (NLM) as well as deconvolution or low-pass kernels to remove offsets between trace-retrace lines. [Figure 6D] 13 is a series of AFM images using the real-time denoising method of the preferred embodiment using non-local means (NLM) as well as deconvolution or low-pass kernels to remove offsets between trace-retrace lines. [Figure 6E] 13 is a series of AFM images using the real-time denoising method of the preferred embodiment using non-local means (NLM) as well as deconvolution or low-pass kernels to remove offsets between trace-retrace lines. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] The preferred embodiment relates to a denoising access scheme for Atomic Force Microscopy (AFM) that allows real-time improvement of high-resolution AFM data acquisition. The method described herein uses known denoising techniques in high-speed data processing architectures to account for sensor noise that is always present during AFM sample imaging and preserves measurements.

[0028] Referring now to FIG. 2, a simplified diagram of the process 100 of the preferred embodiment is illustrated. A sample 102 with a feature of interest imaged moving from left to right is displayed as step 1. This is displayed as a raw AFM image including anomalies such as sensor noise 104 that tend to reduce resolution. In step 2, the probe begins an AFM scan relative to the sample. A selected number of scan lines 106 corresponding to a moving block / buffer over the section of interest (e.g., 1000 nm×1000 nm scan area) is scanned and the data is sent to a processing block for real-time image denoising using known algorithms. The image buffer may be as low as two lines, but five lines are preferred. Five lines provides good denoising quality with sufficient line density while retaining sufficient throughput to suit the user's imaging needs. Two such preferred algorithms 108 include Total Variation (TV) filtering and Non-local Means (NLM). Non-local means denoising works best when scanning surfaces that contain self-similarity, such as periodic or textured backgrounds, rough surfaces, etc. Total variation denoising works best when scanning surfaces that contain flat regions with abrupt changes in height. Total variation denoising can handle even lower signal-to-noise ratios, but so-called staircase artifacts may occur.

[0029] The amplitude and spectral content of the AFM sensor noise is measured at the center of the scanned area of ​​interest and used with every image block / buffer and the corresponding noise removal algorithm. This noise may be periodic, random, or a combination of both. The noise profile removed from the original image is displayed in image 110 shown in FIG. 2. After noise removal, the middle of the rolling buffer of 5 lines is added to the displayed image as the entire sample section is scanned to generate an AFM image 112 with the measurements removed with the measured amount of sensor noise removed. The noise removal parameters (normalization parameter lambda for TV, and filtering parameter h for NLM) are selected and activated, and in this case are automatically adjusted for every 5 line data block by matching the measured noise amplitude to the difference amplitude between the image blocks before and after noise removal. In this manner, the amount of noise removed from the AFM image 102 is the same as the measured amount of sensor noise. The sample image 112 is much "cleaner" than the raw AFM image data 102, and features of interest can be identified with much higher resolution.

[0030] As part of the autotuning, the image buffer size can be rescaled to remove outliers before total variation or non-local means processing. A singular value is defined as a data point that deviates from the mean by more than a specified standard deviation (sigma). The singular values ​​are restored once the denoising process is complete. 6-sigma is the preferred accuracy threshold; however, thresholds of 1-sigma or higher can be used.

[0031] Optionally, a deconvolution kernel of selected size (e.g. 3×1) (filtering by a weighted average of pixels adjacent to the pixel in question, showing how the filtered value of the pixel depends on the neighboring pixels) can be used with the total variation algorithm, while a low-pass kernel of size 3×1 can be used with the non-local means algorithm to efficiently remove the apparent trace-retrace offset, which arises due to high scanning speeds and parachute-like scanning artifacts. However, larger kernels can also be used.

[0032] The sensor noise amplitude scaled by a factor of 2 / 3 is a good target for the total variation to eliminate the residual RMS error, but other scaling factors smaller or larger than 1 can be used effectively.

[0033] For NLM denoising, a study window of size 5x5 provides useful results in the AFM space, but other NLM study window sizes can be used depending on the image resolution.

[0034] A scanning probe microscope apparatus 150 (e.g., AFM) according to a preferred embodiment is shown in FIG. 3. In this embodiment, a probe 152 having a tip 154 ​​is held by a probe holder (not shown) supported by a piezoelectric tube scanner 156. The scanner 156 is a "Z" or vertical scanner that responds to sample characteristics in a closed-loop control system to position the tip 154 ​​relative to the sample 158 during AFM imaging. The tube scanner 156 is coupled to an XY scanner 160, preferably a piezoelectric tube, that is used to raster the probe tip 162 relative to the surface of the sample during operation of the AFM. A mechanical Z-stage 162 is used to provide large movements in the Z direction between the tip 154 ​​and the sample 158, for example to align the tip 154 ​​and the sample 158 in preparation for AFM image acquisition. The sample 158 is mounted on an XY stage 164 which primarily provides general XY motion for positioning the probe 152 over a region of interest on the sample 158. An XY stage controller 166 controls the stage 164 to position the probe / sample over the region of interest. However, the stage 164 may also be configured to provide relative scanning motion (e.g., raster) between the tip 154 ​​and the sample 158 at a selected scanning speed. The controller 166 is also responsive to an AFM controller 176 for positioning the image scan over the region of interest. The controllers 166, 174 are implemented by a computer 180.

[0035] In operation, after the tip 154 ​​is coupled to the sample 158, rapid scanning of the sample begins using the XY scanner 160 in an AFM mode (e.g., PFT mode) as previously described. The interaction of the tip 154 ​​with the surface of the sample 158 causes a deflection of the probe 152, which is measured by an optical beam-bounce deflection detection apparatus 168. The apparatus 168 includes a laser 170 that directs a beam "L" from the rear facet of the cantilever 155 towards a photodetector 172 that transmits the deflection signal for rapid processing, e.g., to a DSP 176 of an AFM controller 174.

[0036] The AFM controller 174 continuously determines a control signal depending on the AFM operation mode and sends it to the piezo tube 156 to maintain the Z position of the probe 152 with respect to the sample 158, more specifically the deflection of the probe at a feedback setpoint. The controller 174 also performs real-time processing of the sensor noise using TV or NLM algorithms. This AFM control is described in more detail with reference to FIG. 4.

[0037] Referring to FIG. 4, a method 200 for real-time denoising of AFM image data is illustrated. In block 202, the AFM tip is aligned with the sample surface. An initial measurement of the raw system noise (i.e., sensor noise) is taken in block 204 and saved as output noise 206. For a noise measurement scan, the scan area is very small (e.g., 1 nm×1 nm) and the number of scan lines is small (e.g., 10). This noise is generally consistent and stable over time, so this measurement is useful for denoising the final acquired AFM image. Then, in block 208, the AFM scans, for example, the first three lines of the sample area of ​​interest and collects AFM metrology data. The method 100 denoises (e.g., using a TV or NLM denoising algorithm) in block 210, and a denoised "line 1" 212 of the AFM scan is captured and displayed as part of the final image.

[0038] The AFM raster scan then continues at block 214 to image the fourth line of the sample area. At block 216, the AFM image data sent to the DSP is processed using a selected denoising algorithm to denoise the image based on lines 1-4. This operation produces a denoised line 2 (218) of the scan. If the image block / buffer is selected as 5 lines (preferred) (the block / buffer size may be 2 to 5 lines or more), the method 100 proceeds to block 220 to scan the Kth (e.g., 5th) line of the AFM data acquisition scan. At block 222, the AFM data / image is denoised based on lines K-4 to K. In a preferred embodiment, the denoised K-2 lines (the middle line of the 5) are captured and output 224 for display. The denoised data is added to the displayed image to correct for sensor noise. More specifically, the noise removal parameters are automatically adjusted for every 5-line data block by matching the measured noise amplitude (block 206) with the difference amplitude between the image block before and after noise removal, so that the amount of noise removed from the AFM image is nearly identical to the measured amount of sensor noise.

[0039] Next, method 100 queries whether all lines (N, the number of lines to be imaged) of the sample section to be imaged (i.e., the region of interest) have been scanned in block 226. If not (K < N), the variable K is incremented by 1 in block 228, and the corresponding line is scanned by AFM and controlled to block 220. As previously described, the data is output as an intermediate line with noise removed and new noise removed from the image at output 224. On the other hand, when the scanning is almost complete (K = N), method 100 removes noise from the image data based on lines N - 3 to N in block 230 to generate a noise-removed line N - 1 (232). To obtain the noise-removed line N, method 200 removes noise from the AFM image data based on lines N - 2 to N using the selected algorithm (TV or NLM). Using the noise-removed line N 236, a noise-removed sample image is generated while preserving the measurement (surface roughness, both-side depth of the sample shape, e.g., half-space line / trench, etc.), and method 200 ends at block 238. Optionally, a deconvolution (TV noise removal) or low-pass (NLM noise removal) kernel 240 (e.g., 3×1) may also be employed to efficiently remove the trace-re-trace offset (common AFM image anomaly). Corresponding exemplary images are illustrated in FIGS. 5 and 6, which will be described below.

[0040] Referring to FIG. 5A-5E, a series of images are shown that were generated based on the method described above utilizing Total Variation (TV) denoising. NIQE is the Naturalness Image Quality Evaluator (no-reference image quality score). The lower the value, the better the image quality. The AFM image is 512×512 pixels and the region of interest is 10 μm×10 μm. FIG. 5A is an AFM image collected without denoising. Artifacts that appear as "ghosting" due to system sensor noise can impair the discrimination of surface features and result in suboptimal image resolution. FIG. 5B shows an even smoother and cleaner image with improved surface feature resolution due to real-time denoising of sensor noise as previously described using TV denoising. Sensor noise is shown in FIG. 5C.

[0041] In FIG. 5D, an optional deconvolution kernel is used to remove trace-retrace data offsets between data collected in the trace direction (e.g., left to right) and retrace directions (e.g., right to left). The kernel operates to adjust the AFM data on a pixel-by-pixel basis. For example, adding a 3×1 kernel with weights of 0.25, 0.5, and 0.25 to the TV optimization can effectively average the average of all trace-retrace offsets between the current scan line and the lines before and after the current scan line. FIG. 5D is a real-time denoising result image using the kernel, and FIG. 5E shows the residual sensor noise and the removed trace-retrace error.

[0042] Figure 6A is an AFM image acquired without noise reduction. Similar to Figure 5A, the image contains artifacts due to system sensor noise, resulting in less than optimal resolution of surface features in the image. In this case, NLM noise reduction is used. Figure 6B shows improved resolution of surface features with real-time NLM reduction of sensor noise, which is shown in Figure 6C.

[0043] In Figure 6D, an optional low-pass kernel is used to remove trace-retrace data offsets between data collected in the trace direction (e.g., left to right) and retrace directions (e.g., right to left). The kernel operates to adjust the AFM data on a pixel-by-pixel basis. For example, a 3 × 1 kernel with weights of 0.25, 0.5, and 0.25 added after NLM processing can effectively average the average of all trace-retrace offsets between the current scan line and the lines before and after the current scan line. Figure 6D is a result image of real-time noise removal using the kernel, and Figure 6E shows the residual sensor noise and the removed trace-retrace error.

[0044] In addition to TV and NLM, various other denoising techniques can be applied to process each image block / buffer without affecting the measurement of the final image, including median filtering, Fourier domain denoising, Wavelet domain denoising, etc.

[0045] Preferred embodiments relate to methods and apparatus that provide high resolution AFM images that require no image post-processing and are substantially free of sensor noise. The noise reduction technique is fully automated and runs in real time as the image is acquired, so noise is not visible as the image is acquired. AFM sensor noise is suppressed and no additional image distortion occurs.

[0046] Although the best mode contemplated by the inventors for carrying out the present invention has been disclosed above, the practice of the present invention is not limited to the above-mentioned mode, and it is apparent that various additions, modifications, and rearrangements of the features of the present invention can be made without departing from the spirit and scope of the concept of the present invention.

Claims

1. 1. A method for an atomic force microscope (AFM), comprising: The method comprises: Aligning the AFM probe with a surface of a sample; vibrating the probe in an AFM operating mode; measuring system noise to generate a noise image; providing a relative scanning motion between the probe and the sample; measuring a deflection of the probe in response to the providing step and controlling a probe-sample separation in response to an AFM operating mode for a selected number of scan lines to generate a sample image for the selected number of scan lines; denoising the sample image in real time using the system noise; measuring the deflection for a next line of the scanning motion to generate a new sample image; denoising the new sample image in real time using the system noise; A method comprising:

2. The method of claim 1 , wherein the selected number is at least two.

3. the selected number is at least 5; The method of claim 2 , wherein the lines of the sample image that have been denoised are intermediate the selected number.

4. The method of claim 3 , wherein the denoising step comprises using a non-local means (NLM) denoising algorithm.

5. The method of claim 1 , wherein the step of removing noise comprises using a total variation (TV) denoising algorithm.

6. The method of claim 1 , wherein the scanning motion is a raster scan.

7. The method of claim 1 , further comprising, after the step of removing noise, repeating all of the steps including the step of removing noise until a region of interest of the sample is imaged.

8. The method of claim 1 , wherein the mode is one of a peak-force tapping (PFT) mode, a contact mode, and a tapping mode.

9. The method of claim 1 , further comprising applying one of a low-pass kernel and a deconvolution kernel to the denoised image.

10. An atomic force microscope (AFM), a probe of the AFM aligned with a surface of a sample and oscillated in an AFM operating mode; a scanner providing relative scanning motion between the probe and the sample; a detector that measures deflection of the probe in response to the relative scanning motion during AFM operation; A controller, controlling a probe-sample separation according to an AFM operation mode for a selected number of scan lines to generate a sample image for the selected number of scan lines; the controller executing a denoising algorithm to denoise images of the selected number of scan lines in real time during AFM operation using the measured system noise to generate a noise image; Including, The deflection is indicative of a sample property and is stored as an image; An atomic force microscope, wherein the detector measures the deflection for a next line of the scanning motion to generate a new sample image, and the controller is configured to use the system noise to remove noise from the new sample image in real time.

11. The atomic force microscope of claim 10 , wherein the selected number is at least two scan lines.

12. the selected number is at least five scan lines; 12. The atomic force microscope of claim 11, wherein the denoised image is median of the at least five scan lines.

13. The atomic force microscope of claim 10 , wherein the denoising algorithm is one of total variation (TV) denoising and non-local means (NLM) denoising.

14. 11. The atomic force microscope of claim 10, wherein the controller applies one of a low pass kernel and a deconvolution kernel to the denoised image to minimize offset effects during trace and retrace AFM imaging operations.

15. 13. The atomic force microscope of claim 12, wherein the AFM operates in one of a peak force tapping (PFT) mode, a contact mode, and a tapping mode.

Images