Machine learning-based methods for probe microscopy

Abstract

A probe microscope system comprising: a probe having a cantilever mount; a cantilever extending from the cantilever mount to a free end; a probe tip supported by the free end of the cantilever; a drive system for driving the probe toward and away from a sample; a measurement system for acquiring probe data in a measurement cycle while the probe tip interacts with the sample, the measurement cycle comprising a first drive stage in which the probe is driven toward the sample, and a second drive stage in which the probe is driven away from the sample, the probe data being acquired by measuring the parameters of the probe; a machine learning model; and a learning module for learning the machine learning model by machine learning, inputting the probe data as training data, the training data comprising a dataset of multiple measurements of the parameters of the probe, each measurement in the dataset being acquired in the same measurement cycle.

Description

Technical Field

[0008] ,

[0001] The present invention relates to a method for measuring a test sample using a probe, a method for training a machine learning model by machine learning, and a probe microscope system.

Background Art

[0002] A scanning probe system for scanning a sample to obtain information about the sample surface is known. Typically, a scanning probe system has a probe that approaches the sample surface and contacts the sample surface to obtain measurement points in order to obtain information about the sample.

Summary of the Invention

[0003] A first aspect of the present invention provides a method for measuring a test sample using a probe as described in claim 1.

[0004] Optionally, the output of the trained machine learning model includes an image, profile, dimensions, performance metrics, or information regarding the probe.

[0005] A second aspect of the present invention provides a method for training a machine learning model by machine learning as described in claim 3.

[0006] Optionally, the method of the second aspect further includes supplementing the training of the machine learning model by inputting additional training data from a computer model into the machine learning model.

[0007] A further aspect of the present invention provides a method for measuring a test sample using a probe, the method comprising training a machine learning model by the method of the second aspect to generate a trained machine learning model, and then measuring the test sample by the method of the first aspect using the trained machine learning model.

[0008] Probe data is acquired during measurement cycles in which the probe tip interacts with the sample (which may be a training sample or a test sample).

[0009] The measurement cycle includes a first drive phase in which the probe is driven toward the test specimen, and a second drive phase in which the probe is driven toward the test specimen. Probe data is obtained by measuring one or more parameters of the probe, such as the probe height and / or the probe angle and / or the probe shape.

[0010] Probe data is input into a machine learning model to train the model or to obtain output from a trained model. The probe data input into the machine learning model includes a dataset of multiple measurements of the probe's parameters, with each measurement in the dataset being acquired in the same measurement cycle. This can be contrasted with conventional methods where only a single data point (the data point indicates the height of the sample and provides a single pixel in the height profile image) is acquired per measurement cycle when in contact with the sample surface.

[0011] For test specimens, the trained machine learning model generates output based on the probe data input to the trained machine learning model. Optionally, the output includes characteristics of the test specimen (e.g., image, profile, dimensions, or figure of performance). Optionally, the output includes information about the probe.

[0012] Any of the following features apply to all aspects of the present invention.

[0013] Optionally, the dataset input to the machine learning model is acquired in the first and / or second driving phases.

[0014] Optionally, the dataset to be input into the machine learning model is acquired in the first driving phase.

[0015] Optionally, the dataset to be input into the machine learning model is acquired in the second driving phase.

[0016] Optionally, the dataset to be input into the machine learning model is acquired in the first and second driving phases.

[0017] Optionally, a portion of the dataset input to the machine learning model is acquired when the probe tip is interacting with the sample, and a portion of the dataset is acquired when the probe tip is not interacting with the sample.

[0018] Optionally, some probe data is acquired when the probe tip is interacting with the sample, some probe data is acquired when the probe tip is not interacting with the sample, and only the probe data acquired during measurement cycles in which the probe tip is interacting with the sample is input into the machine learning model.

[0019] Optionally, the sample may have features including trenches, holes, wells, or other depressions, and the probe is driven to move in and out of these features during the measurement cycle.

[0020] Optionally, in the first driving phase, the probe is driven downward at a position adjacent to the side wall of the sample, and in the second driving phase, the probe is driven upward at a position adjacent to the side wall, and during one or both of the driving phases, the probe tip interacts with the side wall.

[0021] Optionally, the sample has a surface that intersects with a side wall at a corner, and the method includes the steps of acquiring surface probe data in a surface measurement cycle in which the probe tip interacts with the surface, the surface measurement cycle comprising an approach drive step in which the probe is driven toward the surface, and a subsequent retraction drive step in which the probe is driven toward the surface, and the surface probe data is acquired by measuring the parameters of the probe, and inputting the surface probe data into a machine learning model, the surface probe data input into the machine learning model comprising a surface dataset of multiple measurements of the probe parameters acquired in the same measurement cycle.

[0022] Optionally, a portion of the surface dataset is acquired during surface measurement cycles in which the probe tip interacts with the surface, while a portion of the surface probe dataset is acquired during surface measurement cycles in which the probe tip does not interact with the surface.

[0023] Optionally, the sample has an upper surface intersecting a side wall at a convex corner and a lower surface intersecting a side wall at a concave corner, and the method is to acquire upper surface probe data in an upper surface measurement cycle in which the probe tip interacts with the upper surface, the upper surface measurement cycle comprising an approach drive step in which the probe is driven toward the upper surface and a subsequent retraction drive step in which the probe is driven toward the upper surface, and the upper surface probe data is acquired by measuring the parameters of the probe, and the upper surface probe data is to be fed into a trained machine learning model The steps include: inputting data, wherein the top probe data input to the machine learning model includes a top dataset of multiple measurements of probe parameters acquired in the same measurement cycle; acquiring bottom probe data in a bottom measurement cycle in which the probe tip interacts with the bottom surface, the bottom measurement cycle including an approach drive phase in which the probe is driven toward the bottom surface, and a subsequent retraction drive phase in which the probe is driven toward the bottom surface, the bottom probe data being acquired by measuring the parameters of the probe; and inputting the bottom probe data into a trained machine learning model, wherein the bottom probe data input to the machine learning model includes a bottom dataset of multiple measurements of probe parameters acquired in the same measurement cycle.

[0024] Optionally, a portion of the top surface dataset is acquired during top surface measurement cycles when the probe tip is interacting with the top surface, a portion of the top surface dataset is acquired during top surface measurement cycles when the probe tip is not interacting with the top surface, a portion of the bottom surface probe dataset is acquired during bottom surface measurement cycles when the probe tip is interacting with the bottom surface, and a portion of the bottom surface probe dataset is acquired during bottom surface measurement cycles when the probe tip is not interacting with the bottom surface.

[0025] Optionally, the probe parameters may include a height parameter indicating the probe's height, or an angle parameter indicating the probe's angle.

[0026] Optionally, the probe can be driven toward the sample by moving the cantilever mount toward the sample, or moved away from the sample by moving the cantilever mount away from the sample.

[0027] Optionally, the probe is driven towards the sample by a deformation of the cantilever shape (e.g., by bending the cantilever), and the probe is driven away from the sample by reversing the deformation of the cantilever shape (e.g., by straightening the bend of the cantilever).

[0028] Optionally, the data set input into the machine learning model includes 10 or more measurement values obtained in the same measurement cycle, or 100 or more measurement values obtained in the same measurement cycle, or 1000 or more measurement values obtained in the same measurement cycle.

[0029] Optionally, the data set input into the machine learning model includes a plurality of measurement values of the probe parameters obtained in the first driving stage and / or a plurality of measurement values of the probe parameters obtained in the second driving stage.

[0030] Optionally, the method comprises the steps of scanning a sample with a probe, performing a plurality of measurement cycles while the probe tip is interacting with the sample, each measurement cycle including the step of obtaining probe data by measuring the parameters of the probe, each measurement cycle including a first driving stage in which the probe is driven towards the sample and a subsequent second driving stage in which the probe is driven away from the sample, and inputting the probe data from the measurement cycles into a machine learning model, the probe data input into the machine learning model including a global data set including a plurality of data sets of a plurality of measurement values of the probe parameters, each measurement value of each data set being obtained in the same measurement cycle.

[0031] Optionally, the probe parameters include a height parameter indicating the height of the probe. The height parameter may represent the height of the cantilever base in the cantilever mount, the height of the cantilever's free end, or the height of any other part of the probe. Optionally, the height parameter is read by interferometry.

[0032] Optionally, the probe parameters include an angle parameter indicating the probe angle. The angle parameter may indicate the angle of the free end of the cantilever or the angle of any other part of the probe. Optionally, the angle parameter is read by an optical lever. Optionally, the angle parameters include a bending angle parameter indicating the bending angle of the probe, or a torsion angle parameter indicating the twisting angle of the probe.

[0033] Optionally, probe data is obtained by measuring two or more parameters of the probe, and the probe data input to the machine learning model includes multiple measurements of each parameter of the probe obtained during the same measurement cycle. In one example, the two or more parameters could include a height parameter indicating the height of the probe and one or more angle parameters indicating the angle of the probe. In another example, the two or more parameters could include a bending angle parameter indicating the bending angle of the probe and a twist angle parameter indicating the twist angle of the probe.

[0034] Optionally, the method further includes the steps of detecting an interaction between the probe tip and the sample, and triggering a second drive step based on the detection.

[0035] In a further aspect of the present invention, a probe microscope system is provided, comprising: a probe having a cantilever mount; a cantilever extending from the cantilever mount to a free end; and a probe tip supported by the free end of the cantilever; a drive system for driving the probe toward and away from a sample; and a measurement system for acquiring probe data in a measurement cycle between the probe tip interacting with a sample, the measurement cycle comprising a first drive stage in which the probe is driven toward the sample, and a second drive stage in which the probe is driven away from the sample, wherein the probe data is acquired by measuring the parameters of the probe; a machine learning model; and a module for inputting probe data into the machine learning model, wherein the probe data comprises a dataset of multiple measurements of the probe parameters acquired in the same measurement cycle.

[0036] Optionally, the module includes a learning module that inputs probe data into a machine learning model and trains the machine learning model through machine learning.

[0037] The drive system may have a linear actuator to move the cantilever mount. Alternatively, the drive system may have a thermal drive system that changes the shape of the cantilever by illuminating or heating the cantilever. [Brief explanation of the drawing]

[0038] Embodiments of the present invention will be described below with reference to the attached drawings.

[0039] [Figure 1] Figure 1 is a schematic diagram of a scanning probe system according to one embodiment of the present invention. [Figure 2] Figure 2 is a schematic diagram of the measuring device incorporated into the system shown in Figure 1. [Figure 3] Figure 3 shows a method according to one embodiment of the present invention. [Figure 4] Figure 4 is a schematic diagram showing a series of measurement cycles. [Figure 5] Figure 5 schematically shows the trajectory of the probe tip during two sidewall measurement cycles at different distances adjacent to the sidewall. [Figure 6] Figure 6 shows the sample and data set collected from the sample using the method according to one embodiment of the present invention. [Figure 7] Figure 7 shows how to train a machine learning model using machine learning. [Figure 8] Figure 8 shows a method for measuring a test sample using a probe. [Figure 9] Figure 9 shows two measurement cycles on a flat surface of a training sample. [Figure 10] Figure 10 shows the overall measurement cycle for the indentation features of the training sample. [Figure 11] Figure 11 shows the training sample and associated dataset. [Figure 12] Figure 12 shows the overall measurement cycle for the characteristics of the indentation in the test sample. [Modes for carrying out the invention]

[0040] A scanning probe microscope system according to one embodiment of the present invention is shown in Figure 1. This system comprises a piezoelectric driver 4 and a probe 1 having a cantilever 2 and a probe tip 3. The bottom of the piezoelectric driver 4 provides a cantilever mount, and the cantilever 2 extends from the cantilever mount from its proximal end or base to its distal free end. The probe tip 3 is supported by the free end of the cantilever 2.

[0041] The probe tip 3 has a conical or pyramidal structure that tapers from its base to a point at its distal end, which is the point of closest interaction with the sample 7 on the sample stage 11a. The sample has a sample surface that defines the sample surface axis, perpendicular to the sample surface and extending vertically in Figure 1. The cantilever 2 has a single beam with a rectangular profile extending from the cantilever mount 13. The cantilever 2 has a length of approximately 20 microns, a width of approximately 10 microns, and a thickness of approximately 200 nm.

[0042] In this example, the probe tip 3 is tapered up to a certain point, but in other embodiments, the probe tip 3 may be specially formed according to the measurement of the side wall. For example, the probe tip 3 may have a flared shape.

[0043] The cantilever 2 is a thermal bimorph structure composed of two (or more) materials with different thermal expansions, typically a silicon or silicon nitride base with a gold or aluminum coating. The coating extends along the entire length of the cantilever and covers the back side of the tip 3. An illumination system (in the form of a laser 30) under the control of a photothermal (PT) drive 33 is positioned to illuminate the coated upper side of the cantilever with an intensity-modulated radiation spot.

[0044] The cantilever 2 is composed of a monolithic structure with uniform thickness. For example, the monolithic structure is described in Albrecht T., Akamine, S., Carver, TE, Quate, CFJ, Microfabrication of cantilever styli for the atomic force microscope, Vac. Sci. Technol. As described in A 1990, 8, 3386 (hereinafter referred to as "Albrecht et al."), Si It can be formed by selectively etching a thin film of O2 or SiN4. The tip 3 may also be formed integrally with the cantilever, as described by Albrecht et al. They may be formed by an additional process such as electron beam deposition, or they may be formed separately and attached by adhesive or other attachment methods.

[0045] The wavelength of the working beam 32 output by the laser 30 is selected to ensure good absorption by the coating so that the cantilever 2 bends along its length to move the probe tip 3. In this example, the coating may be located on the opposite side of the sample so that it bends downward toward the sample when the cantilever 2 is heated, or alternatively, the coating may be located on the same side as the sample so that it bends away from the sample when the cantilever 2 is heated.

[0046] The piezoelectric driver 4 expands and contracts in the Z direction in response to the piezoelectric drive signal 5 at the piezoelectric driver input. As will be further described below, the piezoelectric drive signal 5 causes the piezoelectric driver 4 to repeatedly move the probe toward and away from the sample 7 during a series of measurement cycles. The piezoelectric drive signal 5 is generated by a piezoelectric controller (not shown). Typically, the piezoelectric driver 4 is mechanically guided by a bend (not shown).

[0047] The measurement system 80 is positioned to detect the height and angle of the free end of the cantilever 2, which is directly opposite the probe tip 3. The measurement system 80 includes an interferometer to measure the height of the free end of the cantilever and a quadrant photodiode (QPD) to measure the angle of the free end of the cantilever. Figure 1 shows the measurement system 80 in schematic form only, and Figure 2 shows a more detailed diagram. Light 100 from the laser 101 is split into a sensing beam 103 and a reference beam 104 by a beam splitter 102. The reference beam 104 is directed to a appropriately positioned back reflector 120 and then returns to the beam splitter 102. The back reflector 120 is positioned to provide a fixed optical path length with respect to the vertical (Z) position of the sample 7. The beam splitter 102 has an energy-absorbing coating and splits both the incident beam 103 and the reference beam 104 to generate first and second interferograms with a relative phase shift of 90 degrees. The two interferograms are detected by the first photodetector 121 and the second photodetector 122, respectively.

[0048] Ideally, the outputs from photodetectors 121 and 122 are cosine and cosine signals with a 90-degree phase difference. Furthermore, they have no DC offset, have equal amplitude, and depend only on the cantilever position and the wavelength of laser 101. Known methods are used to monitor the outputs of photodetectors 121 and 122 while varying the optical path difference in order to determine and apply the errors resulting from the outputs of the two photodetectors being perfectly harmonic, equal amplitude, and orthogonal in phase. Similarly, the DC offset level is also corrected by methods known in the art of this application.

[0049] These outputs of the photodetector are suitable for use with reversible interference fringe counters and interference fringe division devices 123 in conventional interferometers, which can be provided as dedicated hardware, FPGA, DSP, or programmed computer. The phase-orthogonal interference fringe counter can measure the displacement of the cantilever position with an accuracy of λ / 8, i.e., down to 66 nm for 532 nm light.

[0050] Known interference fringe splitting techniques allow for improvements to sub-nanometer-scale accuracy based on the arctangent of the signal. In the above embodiment, the reference beam 104 is adjusted to have a fixed path length with respect to the Z position of the sample 7. Thus, it can be reflected from the surface of the stage 11a on which the sample 7 is placed, or from a back reflector linked to the position of the stage. The reference path length may be longer or shorter than the length of the path taken by the beam 103 reflected from the probe. Alternatively, the relationship between the reflector and the Z position of the sample does not need to be fixed. In such an embodiment, the reference beam may be reflected from a fixed point, which has a known (but variable) relationship with the Z position of the sample. Thus, the tip height is estimated from the path difference measured by the interferometer and the Z position of the sample relative to the fixed point.

[0051] An interferometer detector is an example of a homodyne system. The specific system described herein offers several advantages. By using two phase-orthogonal interferograms, it becomes possible to measure the displacement of the cantilever over multiple interference fringes and therefore over a large displacement range. Examples of interferometers based on these principles are described in U.S. Patent No. 6,678,056 and International Publication No. 2010 / 067129. Alternative interferometer systems capable of measuring changes in optical path length may be employed. A suitable homodyne polarization interferometer is described in European Patent No. 1,892,727, and a suitable heterodyne interferometer is also described. This is described in U.S. Patent No. 5,144,150.

[0052] Returning to Figure 1, the output of the interferometer is a height signal on the height detection line 20, which is input to a surface height calculator (not shown) and a surface detection unit (not shown). The surface detection unit is configured to generate a surface signal on the output line of the surface detector each cycle when it detects the interaction between the probe tip 3 and the sample 7.

[0053] Furthermore, the reflected beam is split into a first component 107 and a second component 110 by the beam splitter 106. The first component 107 is directed to a segmented quadrant photodiode 108 via lens 109, and the second component 110 is split by the beam splitter 102 and directed to photodiodes 121 and 122 to generate height signals on the output line 20. The photodiode 108 indicates the position of the first component 107 of the reflected beam on the photodiode 108 and generates angle data 124 that changes according to the tilt angle of the cantilever relative to the sensing beam 103.

[0054] The angle data 124 includes a deflection / bend signal that indicates the bending angle of the cantilever, i.e., the angle that changes as the cantilever bends along its length. Therefore, the deflection / bend signal indicates the bending angle of the cantilever. The deflection / bend signal can be determined by the difference between the signal from the upper half and the signal from the lower half of the quadrant photodiode 108.

[0055] Furthermore, the angle data 124 includes a lateral / torsion signal that indicates the twist angle of the cantilever, i.e., the angle that changes as the cantilever twists. Thus, the lateral / torsion signal indicates the twist angle of the cantilever. The lateral / torsion signal can be determined by the difference between the signal from the left half and the signal from the right half of the quadrant photodiode 108.

[0056] Figure 3 shows the steps for measuring the side wall of sample 7. Sample 7 has an upper surface 202, a lower surface 204, and a side wall 206 between the upper surface 202 and the lower surface 204. The upper surface 202 intersects with the apex of the side wall 206 at a convex corner, and the lower surface 204 intersects with the bottom of the side wall 206 at a concave corner. The side wall 206 may form part of a structure in the sample, such as a well or projection.

[0057] During scanning of the upper and lower surfaces, the probe is moved closer to and further away from the sample 7 in a series of measurement cycles shown in Figure 4, with each approach and retraction phase constituting one cycle in which a single measurement point is acquired when the probe contacts the sample surface.

[0058] The probe is scanned laterally across the sample by an XY driver that drives the probe in a raster scanning pattern. Figure 4 shows the X-scanning direction of the raster scanning pattern. In this example, the movement of the cantilever mount is indicated by the arrows in Figure 4. In each measurement cycle, the cantilever mount moves vertically downward and then vertically upward. The horizontal movement is driven by the XY driver. The XY driver may move the probe continuously in the X-scanning direction, or it may move the probe in an intermittent motion (repeated stopping and starting) without movement in the X-scanning direction as the probe approaches and retracts.

[0059] A dither signal may be applied to the probe during the first (approximate) drive stage as a means of identifying contact with the lower surface 204. The dither signal is applied using a signal from the photothermal drive 33 to illuminate the rear of the cantilever 2 with a working beam 32. This working beam 32 can be used to vibrate the probe by dither vibration. In each measurement cycle, the dither vibration, measured by an interferometer or quadrant photodiode, is monitored to detect contact between the probe and the sample. For example, the phase or amplitude of the dither vibration may change, and contact can also be detected by detecting this change.

[0060] In an alternative embodiment, the deflection / bending signal can be monitored to detect contact between the probe and the sample. For example, the deflection / bending signal may change rapidly when the probe comes into contact with the sample, and this change can be detected to detect contact. In this case, the dither signal is not required, so the photothermal operating systems 33, 30 can be excluded, and the cantilever 2 does not need to have a thermal bimorph structure.

[0061] Figure 4 shows a series of four top surface measurements of the top surface 202. Each top surface measurement is acquired during a top surface measurement cycle, which includes an approach drive phase in which the cantilever mount is driven downward so that the probe is driven downward to the top surface 202, and a subsequent retraction drive phase in which the cantilever mount is driven upward so that the probe is driven upward away from the top surface 202. Surface measurements are acquired for each top surface measurement cycle by acquiring a height reading from the interferometer detector when contact with the top surface 202 is detected by monitoring dither vibrations. The approach drive phase can be terminated in response to the detection of contact between the probe and the sample. Optionally, the dither signal may not be applied to the probe during the retraction drive phase.

[0062] Figure 4 also shows a series of two bottom surface measurements of the bottom surface 204. Each bottom surface measurement is acquired during a bottom surface measurement cycle, which includes an approach drive phase in which the cantilever mount is driven downward so that the probe is driven downward to the bottom surface 204, and a subsequent retraction drive phase in which the cantilever mount is driven upward so that the probe is driven upward away from the bottom surface 204. Surface measurements are acquired for each bottom surface measurement cycle by acquiring a height reading from the interferometer detector when contact with the bottom surface 204 is detected by monitoring dither vibrations. The approach drive phase can be terminated in response to the detection of contact between the probe and the sample. Optionally, the dither signal may not be applied to the probe during the retraction drive phase.

[0063] As the probe approaches the sidewall 206, the probe experiences a pair of sidewall measurement cycles within an interaction region of the order of 5 nm, as shown in Figure 4. Each sidewall measurement cycle includes a pair of sidewall measurement drive stages. The pair of sidewall measurement drive stages includes a first drive stage in which the cantilever mount is driven downward, driving the probe downward (i.e., towards the base of the sidewall), and a second drive stage in which the cantilever mount is driven upward, driving the probe upward (i.e., away from the base of the sidewall).

[0064] During each drive phase of the sidewall measurement cycle, the probe is adjacent to the sidewall 206. Here, “adjacent” means close to the sidewall 206 and, in some cases, interacting with it, though not necessarily. The probe is sufficiently close to the sidewall and within the interaction region at least in one of the drive phases. The interaction region depends on the properties of the sample and the probe, e.g., whether the sample and / or the probe are charged. As a non-limiting example, during each sidewall measurement cycle, the probe tip may be located within 100 nm of the sidewall, within 50 nm of the sidewall, within 10 nm of the sidewall, or within 5 nm of the sidewall.

[0065] During the sidewall measurement cycle, the sidewall 206 applies a twisting force to the probe tip 3, and as a result, the probe tip 3 interacts with the sidewall 206 during one or both of the sidewall measurement drive stages.

[0066] In the example in Figure 5, the force is an attractive force arising from the van der Waals interaction, and the probe tip 3 interacts with the sidewall 206 during a second drive phase in which the probe is driven adjacent to the sidewall 206. A series of sidewall measurements are performed by measuring the angle of the cantilever as the probe tip interacts with the sidewall during the second sidewall measurement drive phase.

[0067] Figure 5 schematically illustrates the movement of the probe tip during a pair of sidewall measurement cycles. In the first drive phase, the probe tip moves vertically downward adjacent to the sidewall and then "snaps" to make contact with the sidewall. In the second drive phase, the probe tip is pulled up along the sidewall. As the van der Waals interaction decreases as the probe retracts, the cantilever twist is relieved, and the probe tip moves away from the sidewall as it moves upward.

[0068] If a dither signal is applied to the probe during the first drive phase of each sidewall measurement cycle, causing dither vibration of the probe, the dither vibration can be monitored to detect contact between the probe and the sample. The first drive phase can be terminated in response to the detection of contact between the probe and the sample. Optionally, the dither signal is applied to the probe during the first drive phase but not during the second drive phase. The absence of a dither signal in the second drive phase makes it easier to accurately measure the angle of the cantilever as it interacts with the sidewall.

[0069] In the example shown in Figure 5, at the end of each first driving phase, the probe contacts the lower surface 204, and detection of this contact triggers the reversal of the driver 4 and the retraction of the probe in the second driving phase. In other embodiments, for example, the inclined side wall 306b shown in Figure 6 may cause the probe to contact the side wall 306b without contacting the lower surface.

[0070] Figure 3 shows four regions constituting the second drive stage, starting from the base of the side wall 206. For each region, the lateral / torsional motion of the probe is shown on the left side of Figure 3, with the cantilever extending both inside and outside the figure. The deflection / bending motion of the probe is shown for each region on the right side of Figure 3, with the side wall behind or in front of the probe, both inside and outside the figure.

[0071] Starting from region 1, the probe is in contact with the lower surface 204 and is slightly pressed into this surface. Therefore, there is no or very little twisting of the probe, but a slight positive deflection occurs because the probe tip 3 is pressed into the lower surface 204.

[0072] Region 2 shows the lateral / torsional and deflection / bending motions when the probe is slightly retracted. The probe, no longer pushed further into the lower surface 204, does not bend and is pulled toward the side wall 206 by van der Waals forces. Therefore, some torsion occurs. However, since the probe is not bent, no deflection / bending occurs.

[0073] Region 3 shows the lateral / torsional and deflection / bending motions as the probe retracts further above the sidewall 206. The probe continues to be attracted to the sidewall 206 by van der Waals forces. As the piezoelectric driver moves the probe upward, the probe is pulled upward along the sidewall 206. The probe moves by sliding or stick / slip motion, making temporary contact with the sidewall due to the attractive force from the sidewall features, and then releasing as the force from the driver 4 moving the probe upward overcomes the attractive force. The torsional state of the probe continues, and the probe slides upward along the sidewall, but the probe repeatedly makes and releases contact with the sidewall. The cantilever 2 is negatively deflected when the probe tip 3 contacts the sidewall 206.

[0074] Region 4 shows the probe when it has retracted to a position where no or negligible attractive force is generated between the probe and the side wall 206. Here, no twisting or deflection of the probe occurs.

[0075] As the probe retracts, a series of measurements of lateral / torsion and deflection / bending signals are taken. By obtaining this information, it is possible to determine the properties of the side wall 206. These properties could be geometric properties such as the profile or shape of the side wall 206, or material properties.

[0076] Each set of sidewall measurements includes a dataset or "sidewall signature" that provides information about the characteristics of the sidewall.

[0077] In Figures 4 and 5, a series of sidewall measurements for each measurement cycle are obtained by measuring the angle of the cantilever as the probe interacts with the sidewall during the second drive phase, i.e., by acquiring a series of samples of the lateral / torsion and deflection / bending signals as the probe tip moves upward along the sidewall. Probe height measurements are also acquired simultaneously with each sidewall measurement. As described below, probe height measurements can be obtained from the piezoelectric drive signal 5 or from the interferometer.

[0078] Furthermore, a series of sidewall measurements for each measurement cycle may be obtained by measuring the angle of the cantilever as the probe interacts with the sidewall during the first drive phase, i.e., as the probe tip moves downward along the sidewall. In the first drive phase, the probe tip moves downward along the sidewall (i.e., towards the base of the sidewall) and a series of sidewall measurements are obtained. In the second drive phase, the probe tip moves upward adjacent to the sidewall, i.e., away from the base of the sidewall.

[0079] Figure 6 shows a typical well 300 on the sample surface, which has a pair of side walls 306a, 306b. In this example, the side walls 306a, 306b extend substantially linearly.

[0080] Trace 350 shows the deflection / bending signal plotted with the cantilever height. Trace 360 ​​shows the lateral / torsion signal plotted with the cantilever height.

[0081] Note that the scale in Figure 6 labeled "Vertical Deflection Extension (nN)" is associated with the deflection / bending signal and not with the lateral / torsional signal. This scale is based on multiplying the deflection / bending signal by the cantilever's bending spring constant, which imparts a force in nN units.

[0082] The other scale in Figure 6 (indicated by the label "Height (measured and smoothed) μm") is based on the piezoelectric drive signal 5 (which controls the piezoelectric driver 4 that drives the base of the cantilever 2). Therefore, this scale effectively indicates the height of the proximal or base of the cantilever, rather than the height of the distal end supporting the probe tip.

[0083] As the probe tip slides along the sidewall, the shape of the cantilever changes, and a series of sidewall measurements change in accordance with the change in the shape of the cantilever, as shown by traces 350 and 360.

[0084] The series of sidewall measurements represented by traces 350 and 360 typically include 100-second or 1000-second sidewall measurements, with each sidewall measurement containing a single sample from the quadrant photodiode.

[0085] When the probe is in region 1, the probe height is at its minimum (approximately 2.55 μm), and The cantilever bends as shown in Figure 3. Consequently, the deflection / bending signal indicated by trace 350 is also at its maximum (approximately 2 nN). When the piezoelectric driver 4 retracts, the cantilever does not bend until the probe tip is lifted off the underside (as indicated by section 351 of trace 350).

[0086] Note that in this example, a series of probe height measurements are performed to show the height of the base of the cantilever in the cantilever mount. That is, the height components of each trace 350, 360, based on the piezoelectric drive signal 5, effectively indicate the height of the proximal end or base of the cantilever in the cantilever mount, rather than the height of the distal end supporting the probe tip. This can be confirmed by the fact that in region 1, the height component of trace 350 changes even though the probe tip is in contact with the lower surface.

[0087] Optionally, a series of probe height measurements indicating the height of the free end of the cantilever can also be obtained. In the first example, such probe height measurements can be obtained by subtracting the deflection / bending signal from the piezoelectric drive signal 5. In the second example, such probe height measurements are instead based on the height signal 20 from the interferometer.

[0088] As the probe tip lifts at the end of region 1, van der Waals forces or other attractive forces cause the probe tip to snap into contact with the sidewall, and the lateral / torsion signal becomes sharply negative (as shown by section 361 of trace 360).

[0089] In region 3, the cantilever is gradually untwisted (as shown by section 362 of trace 360), and the cantilever becomes slightly curved (as shown by section 352 of trace 350).

[0090] Since both of these signals contain information about the sidewalls, a series of sidewall measurements in region 3 can be analyzed to determine the characteristics of the sidewalls.

[0091] The machine learning model-based method is shown in Figures 7 and 8.

[0092] Figure 7 shows how to train the machine learning model 402, which is a neural network, but any other type of model capable of machine learning can be used.

[0093] The method shown in Figure 7 provides multiple training samples 400a to c, each having different known characteristics.

[0094] For each training sample 400a-c, probe data 401a-c are obtained by measuring the interaction between probe 1 (or another similar probe) and the training sample. The training module 405 stores the probe data 401a-c and is configured to input some or all of the probe data 401a-c as training data into the machine learning model 402, thereby training the machine learning model 402 through machine learning.

[0095] Probe data 401a-c are acquired from training samples 400a-c in a series of measurement cycles. Traces 350 and 351 in Figure 6 show an example of probe data from a single measurement cycle that is acquired and input into the machine learning model 402.

[0096] Figure 9 shows another example of the probe tip trajectory between two measurement cycles 500 and 501 in a relatively flat portion of the training sample 400a.

[0097] The first measurement cycle 500 includes a first drive stage 520 in which the probe mount 4 is driven toward the learning sample 400a, followed by a second drive stage 502 in which the probe mount 4 is driven toward the learning sample. In the first part of the first drive stage 520, the probe tip moves along a linear trajectory 503 without interacting with the learning sample 400a. In the second part of the first drive stage 520, the probe tip interacts with the learning sample, causing the probe tip to follow a more complex curved trajectory 504 in response to the interaction.

[0098] At position 505, the surface detection unit detects the interaction between the probe tip and the training sample 400a and triggers a second drive phase based on the detection. This initiates a second drive phase 502 in which the probe mount 4 is driven away from the training sample 400a. In the first part of the second drive phase, the probe tip continues to interact with the training sample, causing the probe tip to follow a complex curved trajectory 506 in response to the interaction. In the second part of the second drive phase, the probe tip moves along a straight trajectory 507 without interacting with the training sample 400a.

[0099] In the first measurement cycle 500 in Figure 9, there is little to no adhesion between the probe tip and the learning sample 400a. In the second measurement cycle 501 in Figure 9, due to tip adhesion, the probe tip follows a more complex path during the second drive phase.

[0100] In the first part of the first driving stage, the probe tip moves along a linear trajectory 511 without interacting with the learning sample. In the second part of the first driving stage, the probe tip interacts with the learning sample 400a, and in response to the interaction, the probe tip follows a more complex curved trajectory 512.

[0101] At position 513, the surface detection unit detects the interaction between the probe tip and the learning sample and triggers a second drive phase based on the detection. This initiates a second drive phase 501 in which the probe mount 4 is driven away from the learning sample 400a. The probe tip remains in contact with the sample until the retraction of the cantilever mount 4 overcomes the adhesive force due to the adhesion of the probe tip and the probe tip rapidly separates from the learning sample 400a. This causes a complex oscillating twist / bending motion of cantilever 2, which in turn causes the probe tip to follow a complex trajectory 515, schematically shown in Figure 9, despite the fact that there is little to no interaction between the probe tip and the learning sample at this point.

[0102] Figure 10 shows an example of the probe tip trajectory during a measurement cycle as the probe tip scans across features of the learning sample 400a, which may be trenches, holes, wells, or other depressions.

[0103] The learning sample 400a has an upper surface 410 that abuts against the respective side walls 411 and 412 at each convex corner. The characteristic lower surface 413 abuts against the respective side walls 411 and 412 at each concave corner.

[0104] Top surface probe data is acquired in a series of top surface measurement cycles (three cycles are shown in Figure 10) in which the probe tip interacts with the top surface 410. Each top surface measurement cycle includes an approach drive phase in which the probe is driven toward the top surface 410, followed by a retraction drive phase in which the probe is driven toward the top surface 410. Top surface probe data is acquired by measuring one or more parameters of probe 1, for example, the height of the probe measured by interferometry and / or the angle of the cantilever measured by the segmented quadrant photodiode 108.

[0105] The next measurement cycle includes a first sidewall measurement cycle having a first drive phase in which the probe is driven downward at a position adjacent to the first sidewall 411, and a second drive phase in which the probe is driven upward at a position adjacent to the first sidewall 411. During one or both of these drive phases, the probe tip interacts with the first sidewall 411, for example, by sliding interaction or stick / slip interaction. Sidewall probe data is obtained by measuring one or more parameters of the probe during the first sidewall measurement cycle. It will be done.

[0106] Next, bottom surface probe data is acquired in a series of bottom surface measurement cycles (three cycles are shown in Figure 10) in which the probe tip interacts with the bottom surface 413. Each bottom surface measurement cycle includes an approach drive phase in which the probe is driven toward the bottom surface 413, followed by a retraction drive phase in which the probe is driven toward the bottom surface 413. The probe tip is driven to move in and out of the feature during each bottom surface measurement cycle.

[0107] The next measurement cycle is a second sidewall measurement cycle, comprising a first drive phase in which the probe is driven downward at a position adjacent to the characteristic second sidewall 412, and a second drive phase in which the probe is driven upward at a position adjacent to the second sidewall 412. During one or both of the drive phases, the probe tip interacts with the second sidewall 412, for example, by sliding interaction or stick / slip interaction. Sidewall probe data is obtained by measuring one or more parameters of the probe during the second sidewall measurement cycle.

[0108] Bottom probe data, top probe data, and sidewall probe data are obtained by measuring one or more parameters of probe 1, for example, the probe height measured by interferometry and / or the cantilever angle measured by a segmented quadrant photodiode 108.

[0109] Figure 11 shows a portion of the learning sample 400b being scanned by the probe. Figure 11 shows the distal end of the probe tip, including its vertex 800. In this example, the probe tip has a ridge (small projection) 801 that extends laterally from the probe tip near the vertex 800 of the probe tip.

[0110] The learning sample 400b has sidewalls with notches 810-812. Each sidewall is measured by performing a series of measurements of the sidewall with a probe over one or more measurement cycles, as described above.

[0111] In this example, when the probe is driven by the piezoelectric driver 4 to a position adjacent to the side wall, the probe tip interacts with the side wall to perform a series of measurements.

[0112] At each point in time, the lateral position is acquired along with its associated vertical position. These positions can also be interpreted as representing the position of the apex 800 of the probe tip as it slides upward along the side wall.

[0113] The time-series results for such a position are shown by trace 820 on the left side of Figure 11 and can be considered a dataset or sidewall signature representing the trajectory of the probe tip vertex 800.

[0114] The vertical position of each point in the trace 820 may be a height measurement obtained by measuring the height of the cantilever when the probe tip interacts with the side wall. For example, the vertical position can be calculated from the extension of the piezoelectric driver 4 (which can be measured or estimated directly from the piezoelectric drive signal 5) and the deflection / bend signal indicating the bending angle of the cantilever (which can be measured by the vertical position of the first component 107 on the segmented quadrant photodiode 108, or by any other means). Alternatively, the vertical position of each point in the trace 820 may be measured directly by the height signal on the height detection line 20, which provides a direct interferometric measurement of the height of the free end of the cantilever.

[0115] The lateral position of each point on Trace 820 corresponds to the twist angle of the cantilever, i.e., the cantilever The sidewall probe measurement may be based on a lateral / torsion signal that indicates the angle that changes as the cantilever twists. Thus, the lateral / torsion signal indicates the twist shape of the cantilever. The lateral / torsion signal may be determined by the difference between the signal from the left half and the signal from the right half of the quadrant photodiode 108.

[0116] At the start of the drive phase, when the probe is driven in a position adjacent to the side wall, the probe tip snaps laterally, bringing the protrusion 801 into contact with the side wall as shown in 830. Then, as the piezoelectric driver 4 retracts, the protrusion 801 slides upward along the side wall and enters the first notch 810, bringing the first feature 831 to the trace 820. This process continues, with the probe tip driven in a position adjacent to the side wall, providing further features 832, 833 associated with the notches 811, 812.

[0117] Returning to Figure 7, the probe data 401a acquired during the scanning of training sample 400a, and the probe data 401b and 401c acquired during the scanning of the other training samples 400b and 400c, are input into the machine learning model 402, respectively. As a result, the machine learning model 402 is trained through machine learning, and it is converted into the trained machine learning model 402a shown in Figure 8.

[0118] The machine learning process in Figure 7 may be supervised or unsupervised. In the case of supervised machine learning, different known properties of training samples 400a-c can be used along with probe data 401a-c to train the machine learning model 402 by supervised machine learning. For example, training samples 400a-c may have trenches of known width and different widths, where the known width provides a known property used in the supervised machine learning process. In another example, training samples 400a-c may have the same profile but different known electrostatic charge states, where the known electrostatic charge states provide a known property used in the supervised machine learning process.

[0119] The training of the machine learning model 402 can be supplemented by inputting additional training data 403 from the computational model into the machine learning model 402.

[0120] Each measurement cycle generates a dataset of multiple measurements of one or more parameters of the probe, with each measurement in the dataset being taken during the same measurement cycle. The number of measurements per measurement cycle is determined by the ability to acquire and store large amounts of data, as well as the performance of the machine learning model 402 that processes such large amounts of data. For example, each measurement cycle may generate a dataset of 256, 512, 1024, 2048, or 4096 measurements.

[0121] The probe scans the entire training sample 400a, and as shown in Figures 9 to 11 as an example, Multiple measurement cycles are performed in which the lobe tip interacts with the sample. The number of measurement cycles is determined by the ability to acquire and store a large amount of data, as well as the performance of the machine learning model 402 that processes such a large amount of data. For example, measurement cycles for a 256×256 array may be performed, or measurement cycles for a 512×512 array may be performed.

[0122] In an example of a 512*512 array measurement cycle, where each measurement generates a dataset of 4096 values ​​and is stored as 16-bit data, the resulting dataset is approximately 2GB. A global dataset (i.e., a collection of datasets from all measurement cycles) can be generated.

[0123] Preferably, all measurements of the training sample 400a (i.e., global dataset) Probe data from the cycle is input into machine learning model 402.

[0124] Probe data 401a can be input "on the fly" into the machine learning model 402 during the scanning of the training sample 400a, or in a post-processing batch procedure after the scanning of the training sample 400a is completed.

[0125] As shown in Figures 9 and 10, for most (or all) measurement cycles, some datasets are acquired when the probe tip is interacting with the sample, and some datasets are acquired when the probe tip is not interacting with the sample. Traditionally, probe data acquired when the probe tip is not interacting with the sample is either discarded or ignored, but such probe data can be input into the machine learning model 402 shown in Figure 7 to support the machine learning process. For example, the lengths of the linear trajectories 503, 507, and 511 in Figure 9 can provide useful information to the machine learning model. Also, the complex trajectory 515 may contain useful information about the adhesion interaction between the probe tip and the training sample, despite the fact that there is little or no interaction between the probe tip and the training sample during this complex trajectory 515.

[0126] Alternatively, if the machine learning model 402 cannot process a large amount of data, probe data acquired when the probe tip is not interacting with the sample may be discarded or ignored, and therefore only probe data acquired during the measurement cycle when the probe tip is interacting with the training sample is input to the machine learning model 402.

[0127] Furthermore, probe data acquired when the probe tip interacts with the sample can provide useful information about the sample's profile and / or other properties of the sample (e.g., material properties or electrostatic state). For example, the probe data set acquired when the probe tip follows the curved trajectories 504, 506, and 512 in Figure 9 can provide information about the sample's material properties, and the probe data set acquired in Figure 10 when the probe tip slides up and down the wall can provide information about the wall's profile or angle.

[0128] Similarly, a dataset in the form of sidewall signatures (as shown in traces 350, 360 in Figure 6 or trace 820 in Figure 11) can be input to the machine learning model 402. In this case, the machine learning process can train the machine learning model 402 to recognize sidewall characteristics, such as the angles of sidewalls 306a, b, the vertical spacing between notches 810 and 812, the depth between notches 810 and 812, or the width between notches 810 and 812.

[0129] The height measurements from Figures 6 and 11, along with the sidewall measurements, can be input into the machine learning model 402.

[0130] Training model 402, as shown in Figure 7, involves changing the weights between nodes in the neural network. This transforms the untrained model 402 into the trained model 402a, as shown in Figure 8.

[0131] The trained machine learning model 401a, after being generated by the machine learning process in Figure 7, can be used to analyze an unknown test sample 600 by the method in Figure 8.

[0132] The method for measuring the test sample 600 in Figure 8 is the same as the method for measuring the learning samples 400a to c illustrated in Figures 9 to 11. That is, probe 1 (or another similar probe) scans the entire test sample 600, and multiple times the probe tip interacts with the test sample 600. A measurement cycle is performed. In each measurement cycle, probe data is obtained by measuring one or more parameters of the probe, and each measurement cycle includes a first drive phase in which the probe is driven toward the test sample 600, followed by a second drive phase in which the probe is driven toward the test sample 600.

[0133] Figure 12 shows an example of the probe tip trajectory during a measurement cycle over the entire indentation feature of test sample 600, which is similar to the indentation feature of training sample 400a shown in Figure 10. The scanning of the feature in Figure 12 is the same as the scanning process described with reference to Figure 10, so it will not be repeated here.

[0134] Bottom surface probe data, top surface probe data, and side wall probe data are acquired from the test sample 600 as shown in Figure 12 and input into the trained machine learning model 402a as part of the dataset 601. The input of the probe data 601 into the trained machine learning model 402a is managed and executed by the input module 605.

[0135] Next, based on the probe data 601 input to the trained machine learning model 402a, the output 601 from the trained machine learning model 402a is received.

[0136] For example, output 601 includes information about the test specimen 600, such as an image, profile, dimensions (e.g., height or width of the recess feature in Figure 12), or performance indicators (e.g., a numerical value indicating the quality of the sidewall of the recess feature in Figure 12).

[0137] Alternatively, output 603 may include information about probe 1 used to scan the test sample 600. If the test sample 600 has known characteristics, probe data 601 can provide an indicator of the wear state of probe 1. Thus, the output 603 of the trained machine learning model 402a can include information indicating the wear state of probe 1. This allows probe 1 to be replaced when the wear state reaches a threshold.

[0138] At a minimum, the probe data 601 input to the trained machine learning model 402a may consist of only a single dataset of multiple measurements from only one measurement cycle, such as the dataset (or signature) shown in the graph in Figure 11, which is a dataset of deflection / bending measurements represented by trace 350 in Figure 6. Alternatively, the dataset of lateral / torsion measurements is represented by trace 360 ​​in Figure 6. More typically, the probe data 601 input to the trained machine learning model 402a may include probe data from multiple measurement cycles, e.g., probe data from the nine measurement cycles in Figure 10, or the above 2GB global dataset from a larger number of measurement cycles.

[0139] The probe data 601 can be input "on the fly" into the trained machine learning model 402a during the scanning of the test sample 600 or in a post-processing batch procedure after the scanning of the test sample 600 is complete.

[0140] As explained with reference to the learning method in Figure 7, for most (or all) measurement cycles, some datasets of the measurement cycle are acquired when the probe tip is interacting with the sample, and some datasets are acquired when the probe tip is not interacting with the sample. As shown in Figure 8, inputting both types of probe data into the trained machine learning model 402a can assist in the process of characterizing the test sample 600 and / or the probe 1.

[0141] Alternatively, if the trained machine learning model 402a cannot process large amounts of data, the probe data acquired when the probe tip is not interacting with the sample is used as input. Module 605 may discard or ignore the data, and therefore, only the probe data acquired during the measurement cycle while the probe tip is interacting with the training sample 400a is input into the trained machine learning model 402a.

[0142] In the example above, for each measurement cycle, probe 1 is driven toward the sample by moving the cantilever mount toward the sample (by the expansion of the piezoelectric driver 4), and probe 1 is driven away from the sample by moving the cantilever mount away from the sample (by the contraction of the piezoelectric driver 4). This method of operation is preferred because it allows the probe to be moved without changing the angle of the probe.

[0143] In other embodiments of the present invention, the probe may be driven toward the sample by changing the shape of the cantilever 2 (e.g., by bending the cantilever 2), and the probe may be driven away from the sample by changing the shape of the cantilever 2 in the opposite direction (e.g., by unbending the cantilever). This bending and unbending of the cantilever 2 may be driven by the laser 30, by an electrically heated element in the cantilever 2, or by any other means. This method of operation may be preferred (compared to moving the cantilever mount 4) because it allows for faster scanning of the sample. Although the change in the shape of the cantilever may result in changes in the deflection / bending signal and / or lateral / twist signal, the machine learning models 402 / 402a are still considered to be able to distinguish between different samples.

[0144] In the example shown in Figure 1, the probe microscope includes an interferometer that measures the height of the distal or free end of the cantilever 2 supporting the probe tip 3, and a quadrant photodiode that measures the angle of the distal or free end of the cantilever 2. Probe data from the interferometer and / or probe data from the quadrant photodiode can be input into a machine learning model 402 / 402a. In another embodiment, the interferometer can illuminate the cantilever at multiple positions and measure the height of the cantilever at different points (e.g., two points at the free end of the cantilever, or one point toward the base of the cantilever and another point at the free end). In this case as well, the height measurements at all points can be input into the machine learning model to obtain better information about the dynamic behavior of the cantilever. Similarly, the angle of another part of the cantilever can be detected using a second optical probe having a second quadrant photodiode.

[0145] Although the present invention has been described above with reference to one or more preferred embodiments, various changes or modifications can be made without departing from the technical scope of the invention as defined in the appended claims.

Claims

1. A method for measuring a test sample with a probe, wherein the probe comprises a cantilever mount, a cantilever extending from the cantilever mount to a free end, and a probe tip supported by the free end of the cantilever, and the method is A step of acquiring probe data during a measurement cycle while the probe tip interacts with the test sample, wherein the measurement cycle includes a first drive step in which the probe is driven toward the test sample, and a second drive step in which the probe is driven away from the test sample, and the probe data is acquired by measuring the parameters of the probe. A step of inputting probe data into a trained machine learning model, wherein the probe data input into the trained machine learning model includes a dataset of multiple measurements of the probe's parameters acquired in the same measurement cycle. The steps include receiving the output from the trained machine learning model based on the probe data input to the trained machine learning model, A method characterized by including the following.

2. The method according to claim 1, characterized in that the output of the trained machine learning model includes images, profiles, dimensions, performance metrics, or information relating to the probe.

3. A method for training a machine learning model using machine learning, A step of measuring a learning sample with a probe, wherein the probe has a cantilever mount, a cantilever extending from the cantilever mount to a free end, and a probe tip supported by the free end of the cantilever, A step of acquiring probe data during a measurement cycle while the probe tip interacts with the learning sample, wherein the measurement cycle includes a first drive step in which the probe is driven toward the learning sample, and a second drive step in which the probe is driven toward the learning sample, and the probe data is acquired by measuring the parameters of the probe. A step of inputting the probe data into a machine learning model and training the machine learning model by machine learning, wherein the probe data input into the machine learning model includes a dataset of multiple measured values ​​of the probe's parameters acquired in the same measurement cycle. A method characterized by including the following.

4. The method according to claim 3, further comprising the step of supplementing the learning of the machine learning model by inputting additional training data from a computational model into the machine learning model.

5. A method for measuring a test sample with a probe, comprising the steps of training a machine learning model by the method described in claim 3 or 4, thereby generating a trained machine learning model, and measuring the test sample by the method described in claim 1 or 2 using the trained machine learning model.

6. The method according to any one of claims 1 to 5, characterized in that the dataset input to the machine learning model is acquired in the first driving stage and / or the second driving stage.

7. The dataset input to the machine learning model is acquired in the first driving stage. The method according to any one of claims 1 to 6, characterized in that it is obtained.

8. The method according to any one of claims 1 to 6, characterized in that the dataset input to the machine learning model is acquired in the second driving stage.

9. The method according to any one of claims 1 to 6, characterized in that the dataset input to the machine learning model is acquired in the first driving stage and the second driving stage.

10. The method according to any one of claims 1 to 9, characterized in that a portion of the dataset input to the machine learning model is acquired when the probe tip is interacting with the sample, and a portion of the dataset is acquired when the probe tip is not interacting with the sample.

11. The method according to any one of claims 1 to 10, characterized in that a portion of the probe data is acquired when the probe tip is interacting with the sample, a portion of the probe data is acquired when the probe tip is not interacting with the sample, and only the probe data acquired in the measurement cycle in which the probe tip is interacting with the sample is input to the machine learning model.

12. The method according to any one of claims 1 to 11, wherein the sample has features including trenches, holes, wells, or other recesses, and the probe is driven to move in and out of the features during the measurement cycle.

13. The method according to any one of claims 1 to 12, characterized in that the parameters of the probe include a height parameter indicating the height of the probe, or the parameters of the probe include an angle parameter indicating the angle of the probe.

14. The method according to any one of claims 1 to 13, characterized in that the probe is driven toward the sample by moving the cantilever mount toward the sample, and the probe is driven away from the sample by moving the cantilever mount away from the sample.

15. The method according to any one of claims 1 to 14, characterized in that the probe is driven toward the sample by deformation of the shape of the cantilever (for example, by bending the cantilever), and the probe is driven toward the sample by reversing the deformation of the shape of the cantilever (for example, by straightening the bend of the cantilever).

16. The method according to any one of claims 1 to 15, characterized in that the dataset input to the machine learning model includes 10 or more measurements obtained in the same measurement cycle, or includes 100 or more measurements obtained in the same measurement cycle, or includes 1000 or more measurements obtained in the same measurement cycle.

17. The method according to any one of claims 1 to 16, characterized in that the dataset input to the machine learning model includes a plurality of measurements of the probe's parameters acquired in the first driving stage and / or a plurality of measurements of the probe's parameters acquired in the second driving stage.

18. The steps include: scanning the sample with the probe, and the tip of the probe interacting with the sample. The method according to any one of claims 1 to 17, further comprising: performing a plurality of measurement cycles while in operation, each measurement cycle comprising the step of acquiring probe data by measuring the parameters of the probe, each measurement cycle comprising a first drive step in which the probe is driven toward the sample, and a second drive step in which the probe is driven away from the sample; and inputting the probe data from the measurement cycles into the machine learning model, the probe data input into the machine learning model comprising a global dataset comprising a plurality of datasets of plurality of measurements of the parameters of the probe, each measurement in each dataset being acquired in the same measurement cycle.

19. A probe having a cantilever mount, a cantilever extending from the cantilever mount to a free end, and a probe tip supported by the free end of the cantilever, A drive system for driving the probe toward and away from the sample, A measurement system that acquires probe data during a measurement cycle while the probe tip interacts with the sample, the measurement cycle comprising a first drive phase in which the probe is driven toward the sample, and a second drive phase in which the probe is driven away from the sample, wherein the probe data is acquired by measuring the parameters of the probe, Machine learning models and, A module for inputting probe data into the machine learning model, wherein the data includes a dataset of multiple measured values ​​of the probe's parameters acquired in the same measurement cycle. A probe microscope system characterized by having the following features.

20. The system according to claim 19, characterized in that the module has a learning module that inputs the probe data to the machine learning model and learns the machine learning model by machine learning.

Images