Method and System for Reducing Systematic Error in Electron Beam Measurement

Abstract

To provide methods and systems for correcting a system error in electron beam measurement.SOLUTION: Embodiments may include methods and systems for correcting a response function of an electron beam tool. The correcting may include: modulating an electron beam parameter having a frequency; emitting an electron beam based on the electron beam parameter towards a specimen, thereby scattering electrons, wherein the electron beam is described by a source wave function having a source phase and a landing angle; detecting a portion of the scattered electrons at an electron detector, thereby yielding electron data including an electron wave function having an electron phase and an electron landing angle; determining, using a processor, a phase delay between the source phase and the electron phase, thereby yielding a latency; and correcting, using the processor, the response function of the electron beam tool using the latency and a difference between the source wave function and the electron wave function.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present disclosure generally relates to improvements in semiconductor wafer metrology. More specifically, the present disclosure generally relates to methods and systems for reducing systematic errors in electron beam metrology.

Background Art

[0002] (Cross - reference to Related Applications) This application claims priority based on U.S. Provisional Patent Application No. 62 / 779,485, filed on December 14, 2018, U.S. Provisional Patent Application No. 62 / 781,412, filed on December 18, 2018, and U.S. Provisional Patent Application No. 62 / 785,164, filed on December 26, 2018, and accordingly incorporates their disclosures herein by reference.

[0003] As the semiconductor manufacturing industry develops, the demands for yield management, particularly for metrology and inspection systems, are increasing. Not only are the critical dimensions continuously shrinking, but the industry also demands a reduction in the time required to achieve high - yield and high - value - added production. Reducing the total time from detecting a yield problem to correcting it has become a determining factor in the return on investment for semiconductor manufacturers.

[0004] Measuring individual semiconductor inter - layer overlays is important for ensuring the functionality of integrated circuits (ICs). As the structural size scales down, electron beam (EB) systems based on the principle of scanning electron microscopes, such as EB - based inspection, review, and metrology systems, are becoming increasingly attractive due to their high resolution. However, in EB systems, unfortunately, the theoretically achievable accuracy is reduced by the influence of systematic errors, and the measurement results may be perturbed.

[0005] One of the conventional methods relies on the individual offline measurement of internal latencies and corrections related to them. Changes in those latencies are not covered by the measurement results. Based on the internal latency, shifts and asymmetries may occur in the detected image, which may cause errors in the distance measurement results. Changes in latency on a short time scale can act as an additional noise source. Changes in latency on a long time scale can appear as a drift-like effect.

[0006] Detecting secondary electrons (SE) is beneficial because precise measurement can be performed with high spatial resolution since the locations where they are generated within the specimen are in a small interaction volume near the surface. However, the SE signal is subject to perturbations by other electron sources, mainly backscattered electrons (BSE) with a broadband energy distribution. Even with current band-pass energy filters, it is possible to discriminate high-energy BSE. Low-energy or medium-energy BSE usually pass through the band-pass filter, so an additional background signal is added to the SE signal. Therefore, the detector signal obtained in the measurement represents a combination of the SE signal and the parasitic low- to medium-energy backscattered electron signal. In such prior art solutions, the image may be blurred due to the influence of the parasitic BSE, thereby causing significant systematic errors.

[0007] Among the conventional methods, as depicted in FIGS. 1 and 2A - 2C, there are those that statically pre-align the beam axis with respect to the optical axis by aligning the image symmetry of a special target structure or a special sensor. However, this beam alignment directly affects the measurement of distance or position in specimens of different heights. In addition, since the orientation of the specimen may change slightly for each load, realignment of the beam is required. Leaving it in the old orientation will lead to tilt-induced errors.

Prior Art Documents

Patent Documents

[0008]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0009] Therefore, it is necessary to improve the method and system for reducing systematic errors in EB measurement.

Means for Solving the Problems

[0010] The embodiments disclosed herein have the following functions: (1) reducing systematic errors in high-resolution EB measurement by discriminating, tracking, and correcting in-tool responses; (2) reducing systematic errors in high-resolution EB measurement by improving the secondary electron (SE) detection method... The embodiments are not limited to overlay measurement and can be used for both semiconductor metrology and inspection, and (3) reducing systematic errors in EB measurement by measuring and aligning the beam direction based on the surface normal of the sample.

[0011] According to the embodiments, the response function can be corrected by modulating beam parameters, detector, or filter parameters, or by correcting beam tilt using asymmetry detection.

[0012] According to the methods and systems of the embodiments, it is possible to detect the part of the system response caused by the perturbation of the system, determine the correction index for the response, and apply the correction index. The measurement device used for detecting the system response can be an electron beam tool for metrology, inspection, and review. The system response caused by perturbation can be defined as inherent to the system and detected by an index, induced by intentional perturbation of beam parameters, or induced by an external source. The perturbation of the beam parameters includes periodic ones with a given specific frequency. The perturbation includes inherent ones, such as those caused by beam tilt. The perturbation includes external ones, such as those caused by a magnetic field.

[0013] Some embodiments may include a method for correcting the response function of an electron beam tool. The method may modulate an electron beam parameter having a frequency, emit an electron beam described by a source wave function having a source phase and a landing angle toward a sample based on the electron beam parameter and scatter electrons therein, detect a portion of the scattered electrons with an electron detector, thereby generating electron data including an electron wave function having an electron phase and an electron landing angle, use a processor to determine a phase delay between the source phase and the electron phase, thereby obtaining a latency, and use the same processor and the latency and a difference between the source wave function and the electron wave function to correct the response function of the electron beam tool.

[0014] The scattered electrons may include backscattered electrons; the method may further detect a portion of the backscattered electrons with a backscattered electron detector using the electron beam tool, thereby generating backscattered electron data.

[0015] When correcting the response function, the backscattered electron data may be subtracted from the electron data.

[0016] When correcting the response function, the processor may extrapolate the spectral distribution of the backscattered electron data toward the low energy side to generate extrapolated backscattered electron data, and the processor may subtract the extrapolated backscattered electron data from the electron data.

[0017] When correcting the response function, the processor may model the energy distribution function of the backscattered electron data to generate a modeled energy distribution function, and may calibrate the modeled energy distribution function using one or more measurement results collected using the same processor and the electron beam tool.

[0018] When correcting the response function, it may be possible to determine the relative ratio that the measurement parameters and the sample characteristics, which are based on the backscattered electron data, account for with respect to the sum of the backscattered electron data and the electron data.

[0019] The relative ratio may be stored on an electron data storage unit.

[0020] When correcting the response function, a machine learning model may be used, and the machine learning model may be trained using the stored relative ratio stored on the electron data storage unit.

[0021] This method may further be such that a second portion of the backscattered electrons is detected by the contrast electron detector with the electron beam tool, thereby generating contrast electron data, generating tilt data by comparing the electron data with the contrast electron data using the processor, and changing the tilt of the electron beam using the tilt data.

[0022] Embodiments may include a system including an electron beam tool and a controller that communicates electronically with the electron beam tool.

[0023] The electron beam tool can be made to have an electron beam emitter, a stage, and an electron detector. The electron beam emitter can be configured to emit electrons in the form of an electron beam. The electron beam can be described by a source wave function having a source phase and a landing angle. The stage can be configured to hold a sample on the path of the electron beam. The electron detector can be configured to detect a portion of the electrons scattered when the electron beam collides with the sample, thereby generating electron data including an electron wave function having an electron phase and an electron landing angle.

[0024] The controller may be configured to include a processor. The controller may be configured to command the electron beam emitter to modulate an electron beam parameter, which is an electron beam parameter of the electron beam and has a frequency, to determine a source phase - electron phase inter - phase delay, thereby obtaining a latency, and to correct the response function of the electron beam tool using the latency and the difference between the source wave function and the electron wave function.

[0025] The scattered electrons may include back - scattered electrons; the system may further include a back - scattered electron detector configured to generate back - scattered electron data by detecting a part of the back - scattered electrons.

[0026] The controller may be configured to correct the response function by subtracting the back - scattered electron data from the electron data.

[0027] The controller may be configured to extrapolate the spectral distribution of the back - scattered electron data towards the low - energy side to generate extrapolated back - scattered electron data, and to correct the response function by subtracting the extrapolated back - scattered electron data from the electron data.

[0028] The controller may be configured to model the energy distribution function of the back - scattered electron data to generate a modeled energy distribution function, and to calibrate the modeled energy distribution function using one or more measurement results collected using the electron beam tool, thereby correcting the response function.

[0029] The controller may be configured to correct the response function by determining the relative ratio of the measurement parameters and sample characteristics based on the back - scattered electron data to the sum of the back - scattered electron data and the electron data.

[0030] The system may further include an electronic data storage unit configured to store the relative ratio.

[0031] The controller may be configured to correct the response function using a machine learning model. When training the machine learning model, the stored relative ratio stored on the electronic data storage unit may be used.

[0032] The system may further include a reference electron detector configured to generate reference electron data by detecting a second portion of the backscattered electrons. The controller may be configured to generate tilt data by comparing the electron data with the reference electron data, and use the tilt data to change the angle of the electron beam emitter, thereby changing the tilt of the electron beam.

[0033] Embodiments may include a non - transitory computer - readable storage medium comprising one or more programs, which cause one or more information processing devices to execute steps including correcting the response function of an electron beam tool. When correcting the response function, modulate an electron beam parameter having a frequency, and instruct the electron beam tool to emit an electron beam described by a source wave function having a source phase and a landing angle toward the sample based on the electron beam parameter and scatter electrons therein. Receive, from an electron detector that detects a portion of the scattered electrons, electron data including an electron wave function having an electron phase and an electron landing angle, determine a phase delay between the source phase and the electron phase, thereby obtaining a latency, and correct the response function of the electron beam tool using the latency and the difference between the source wave function and the electron wave function.

[0034] The steps may include training a machine learning algorithm using the electron beam parameter, the electron data, and the response function.

[0035] For a more complete understanding of the nature and purpose of this disclosure, reference should be made to the following accompanying drawings in conjunction with the detailed description set forth below.

Brief Description of the Drawings

[0036]

Figure 1

Figure 2A

Figure 2B

Figure 2C

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7A

Figure 7B

Figure 7C

Figure 8

Figure 9

Modes for Carrying Out the Invention

[0037] The subject matter of the claims is described by specific embodiments, but other embodiments are included within the technical scope of the present disclosure, including embodiments in which not all of the benefits and features described in the present application are provided. Various structural, logical, processing step, and electronic changes can be made without departing from the technical scope of the present disclosure. Accordingly, the technical scope of the present disclosure is determined solely by reference to the appended claims.

[0038] Embodiments of the present disclosure include methods, systems, and apparatuses for correcting response functions of electron beam inspection tools. In these embodiments, real-time detection and correction of potential latency can be performed, and systematic errors in measurement results can be reduced. That is, the measurement results can be made more accurate.

[0039] As an example, FIG. 3 shows a first method for correcting, for example, a distorted sine wave, which is a response function of an electron beam type inspection tool. In Method 1, by modulating the source characteristics having one or more specific frequencies at a specific amplitude, the inherent static or dynamic response, for example, latency, can be discriminated and corrected. In Method 1, first, frequency 2 can be generated and modulated. Frequency 2 can be generated using a frequency generator. Using the frequency generated at 2 as an input, beam parameter 3, for example, an electron beam current, can be obtained. The beam parameter 3 can be used as the parameter of the primary beam 4. The beam parameter 3 can be modulated. The primary beam 4 can be directed to a sample 5, for example, a wafer or a die on the wafer. By colliding a part of the primary beam 4 with the wafer 5 and scattering a part of the electrons near the surface of the wafer 5, electrons, including, for example, secondary electrons and backscattered electrons, can be obtained. A part of those electrons can be detected by the detector 6 to generate corrected raw data 7. The modulation of those electrons is due to the modulation of the beam parameter 3. This modulation can be analyzed with respect to the excitation modulation, for example, by using the phase delay representing its latency. The modulation can be removed from the detected image by filtering and does not perturb the measurement result. By using the corrected raw data 7, for example, that representing the signal received by the detector 6, in combination with the beam parameter(s) 3, the phase delay can be discriminated to obtain the latency 9. By using the filter 10 for filtering the corrected raw data 7 using the latency 9, the corrected response function 11 can be obtained.

[0040] The primary beam 4 may be switched on and off, or may be transferred to a dump-like structure to modulate its beam deflection. For example, the dump-like structure may be used as a beam dump, or may have a shielding panel or a metal sheet.

[0041] When analyzing the corrected raw data 7, the sine wave for controlling the beam parameter 3 may be compared with the measured distorted sine wave.

[0042] The system according to an alternative example can include an electron beam inspection tool. The electron beam inspection tool can be provided with an electron beam emitter configured to emit electrons in the form of an electron beam. The electron beam can be described by a source electron wave function having a certain source electron phase.

[0043] The electron beam inspection tool can further include a stage. The stage can be configured to hold a sample. The sample can be, for example, a wafer. The wafer can include one or more dies. The sample can be held by the stage on the path of the electron beam.

[0044] The electron beam inspection tool can further include a secondary electron detector, for example, an Everhart-Thornley detector or a solid-state detector, or can be incorporated with a low-pass filter (group). The secondary electron detector can be configured to detect a part of the secondary electrons generated as a result of the electron beam colliding with the sample. Then, electron data, that is, data representing the signal of the (primary, secondary, or backscattered) electrons received by the detector 6, can be obtained as the output after electron detection. The electron wave function can have an electron phase.

[0045] The inspection tool and the controller can perform electronic communication. The controller can be configured to command, for example, an electron beam emitter to modulate the frequency of the electron beam parameters of an electron beam. The controller can obtain a source electron phase and an inter-electron phase delay, thereby obtaining latency. By using this latency in combination with the difference between the source wave function and the electron wave function, the controller can correct the response function of the electron beam inspection tool. The method of correcting the response function can depend on which aspect of the response function needs correction. If beam tilt modulation is required to correct the response function, beam position correction may be performed. If beam current modulation is required to correct the response function, image intensity correction may be performed.

[0046] Another embodiment is a non-transitory computer-readable storage medium, which can be provided with one or more programs for performing those initial steps of Method 1.

[0047] According to other embodiments, the electron detectors in various inspection tools may be improved. By radiating an electron beam towards a sample, additionally, a portion of the electrons in the sample can be backscattered to obtain backscattered electrons. In this method, further, a portion of those backscattered electrons can be detected by a backscattered electron detector, and backscattered electron data, for example, that representing a signal received from the backscattered electrons, can be obtained.

[0048] According to the embodiment shown in FIG. 4, Method 100 can be used to improve the electron detectors in various inspection tools. Image blurring may occur due to the influence of parasitic backscattered electrons, which may cause significant systematic errors. According to Method 100, electron data 101 can be purified against the influence of parasitic backscattered electrons. By using high-frequency modulation 103 to sweep or periodically change the bias voltage of the existing high-pass filter 102 from a value near 0 (high-pass off) to several tens or hundreds of volts (high-pass on), discrimination and correction can be performed on the parasitic backscattered electron data. Whether the high-pass filter 102 is in the high-pass off mode or the high-pass on mode depends on the intensity of the filter voltage. The detected data corresponds to the data 106 of backscattered electrons and secondary electrons in the high-pass off mode, while in the high-pass on mode, it represents the remaining backscattered electron component 104. Optional spectral extrapolation 105 may be performed. By correcting 107 the data of backscattered electrons and secondary electrons with the backscattered electron data, pure secondary electron data 108 with corrected backscattered electrons can be obtained.

[0049] By such a method, the secondary electron detector can be improved. By using the secondary electron detector, data can be mainly received from its upper layer in an overlay application. Another detector, such as a backscattered electron detector, may be used to receive data from the buried layer.

[0050] By operating such an embodiment, the data can be corrected in real time or offline.

[0051] When correcting the response function, the backscattered electron data may be subtracted from the secondary electron data.

[0052] When correcting the response function, the spectral distribution of the backscattered electron data (the one similar to the optical spectrum) may be extrapolated to the low energy side. To perform this extrapolation, for example, a function may be fitted to the spectrum and the function may be used to extrapolate to other measurement data missing spectral positions. By doing this, extrapolated backscattered electron data can be obtained. By subtracting the extrapolated backscattered electron data from the secondary electron data, a corrected response function can be obtained.

[0053] When correcting the response function, a modeled energy distribution function of the backscattered electron data may be obtained by modeling the energy distribution function. The modeled energy distribution function can be calibrated using one or more sets of measurement results collected using the electron beam inspection tool.

[0054] A relative ratio may be obtained when correcting the response function. Based on the relative ratio, sample characteristics and measurement parameters based on the backscattered electron data can be associated with the sum of the backscattered electron data and the secondary electron data.

[0055] The relative ratio may be stored on the electronic data storage unit. One or more obtained relative ratios can be stored in a lookup table for later use.

[0056] A machine learning model may be used when correcting the response function. The training of the machine learning module may be performed using the stored relative ratios stored on the electronic data storage unit. The machine learning model may be, for example, an artificial neural network such as a convolutional neural network, and the artificial neural network may be applied to new measurement results.

[0057] In such embodiments, detection is performed by measuring backscattered electron data and data of backscattered electrons and secondary electrons on a pixel-by-pixel basis. In another type of embodiment, detection is performed by measuring backscattered electron data and backscattered electron data and secondary electron data for each group of pixels (e.g., for each line or for each frame). For backscattered electrons and for data of backscattered electrons and secondary electrons, the measurement may be performed at the same position. Calibration between individual detectors can be dispensed with.

[0058] Some embodiments may include hardware that controls the sweeping or periodic variation of the filter bias voltage and synchronizes it with other system time scales (e.g., pixel clock or frame rate).

[0059] In this method, further, a second portion of the secondary electrons may be detected by a contrast secondary electron detector. By this detection, contrast secondary electron data can be obtained. By comparing the secondary electron data with this contrast secondary electron data, tilt data, i.e., that indicating the alignment of the beam axis with respect to the optical axis, can be generated. The tilt of the electron beam can be changed using the tilt data. To change the tilt of the electron beam, the electron beam emitter may be changed.

[0060] The secondary electron data may be compared with the contrast secondary electron data using a dark field image. The dark field image can be used for contrast enhancement and / or topographic measurement. By using such an image, the beam tilt can be measured and controlled.

[0061] According to an embodiment, the beam tilt of an electron beam inspection system can be discriminated and corrected. By using a multi-channel detector and comparing reference channels with each other, a tilt-related value may be measured. The comparison result of data, such as differential data, may be used for real-time control of the beam tilt by a feedback loop. Such a feedback loop may be used for real-time discrimination and correction of the beam tilt by obtaining a tilt-related value using a group of reference channels provided in the multi-channel detector and a beam tilt correction target.

[0062] According to an embodiment, a special target using a high-Z substrate and an aperture array can be realized. The high-Z substrate may be selected so that backscattered electrons are generated thereby. For example, it may be a Pelco (trademark) silicon nitride support film.

[0063] Such embodiments of the present disclosure can be implemented by the system or the non-transitory computer-readable storage medium according to the present disclosure.

[0064] A system 12 of an embodiment is depicted in FIG. 5. In the system 12, the primary beam components 41 to 43 can be appropriately deflected by the deflection unit 13 and directed toward the sample 5. Second electrons and various portions 44 to 46 of backscattered electrons generated by the collision of the primary beam components 41 to 43 with the sample 5 can be detected by the detector 6. The detector 6 may have one or a plurality of portions, for example, portions 61 to 65, and be configured to detect second electrons or backscattered electrons at various angles based on the flatness or slope of the surface. It should be noted that the number of the portions provided in the embodiment is not limited to the portions 61 to 65 and can be various numbers. The detector 6 can be configured to supply one or a plurality of dark field images 14 to the analysis unit 15. Examples of the dark field images 14 include those shown in FIGS. 7A to 7C such as images 201 and 202.

[0065] FIG. 6 shows another embodiment of the system 16, in which a perforated membrane is disposed between the sample 5 and the detectors 66-68. It should be noted that the number of detectors included in the embodiment is not limited to the detectors 66-68 and can be various numbers. The perforated membrane may be, for example, 200 nm thick and / or made of silicon. However, depending on the process requirements, a different thickness or material may be used. The perforated membrane 17 can have one or more holes, and the holes can be circular or other shapes. For example, the holes provided in the perforated membrane 17 may be circular holes having a diameter in the range of, for example, 2500 nm to 5000 nm. The sample 5 may be a high-Z target that generates strong backscattering data. The perforated membrane 17 may be disposed above the sample 5 at a distance h. For example, the distance h can be in the range of 2 to 10 μm, but a different distance h may be used depending on the process requirements. The portions 47 and 48 of the backscattered electrons can be brought about by the primary beam 4 colliding with the sample 5.

[0066] Conversely, in FIG. 5, in the analysis unit 15, for example, by using image processing technology for conversion (e.g., by comparing several side channels), a dark field image can be analyzed, and the inclination of the sample can be determined. For example, from the dark field image 201 shown in FIG. 7A, it can be read that the backscattered electron emitting surface is flat. Further, for example, from the group of dark field images 202 shown in FIG. 7B, it can be read that the backscattered electron emitting surface is inclined. In this way, the inclination of the surface, that is, the inclination corresponding to the inclination of the sample, can be measured and fed back into the system for correction.

[0067] System 800 according to an embodiment is shown in FIG. 8. This system 800 has an optical-based subsystem 801. Generally, the optical-based subsystem 801 is configured to generate an optical-based output regarding the sample 802 by directing light at (or scanning the light over) the sample 802 and detecting the light from the sample 802. The sample 802 according to one embodiment is a wafer. The wafer can include any wafer known in the art of this technology. The sample according to another embodiment is a reticle. The reticle can include any reticle known in the art of this technology.

[0068] In the system 800 of the embodiment shown in FIG. 8, the optical-based subsystem 801 has an illumination subsystem configured to direct light at the sample 802. At least one light source is provided in this illumination subsystem. For example, the illumination subsystem shown in FIG. 8 has a light source 803. The illumination subsystem according to one embodiment is configured to direct light at the sample 802 with one or more incident angles, for example, one or more oblique angles and / or one or more orthogonal angles. As shown in FIG. 8, for example, the light from the light source 803 is directed at the sample 802 at an oblique incident angle via the optical element 804 and then the lens 805. This oblique incident angle can be any oblique incident angle if suitable, and can be changed according to the characteristics of the sample 802, for example.

[0069] The light source 803, that is, the beam source, can be one having a broadband plasma light source, a lamp, or a laser. According to certain embodiments, the light or photons emitted by the beam source can also be in the form of infrared light, visible light, ultraviolet light, or X-rays.

[0070] The optical base subsystem 801 may be configured to direct light at the sample 802 at different angles of incidence at different points in time. For example, the optical base subsystem 801 may be configured to vary one or more characteristics of one or more elements provided in the illumination subsystem, and thus direct light at the sample 802 at an angle of incidence different from that shown in FIG. 8. In some such examples, the optical base subsystem 801 may be configured to move the light source 803, the optical element 804, and the lens 805, and thus direct light at the sample 802 at another oblique angle of incidence or a perpendicular (or near-perpendicular) angle of incidence.

[0071] According to some examples, the optical base subsystem 801 can be configured to direct light at the sample 802 at multiple angles of incidence simultaneously. For example, a plurality of illumination channels may be provided in the illumination subsystem, with one of the illumination channels provided with a light source 803, an optical element 804, and a lens 805 as shown in FIG. 8, and another illumination channel (not shown) provided with a similar or different group of elements configured in a similar or different manner, i.e., at least one light source and possibly one or more other members as detailed in the present application. When the light from one and the light from the other are directed at the sample simultaneously, by differentiating one or more characteristics (e.g., wavelength, polarization, etc.) of the light provided in the light directed at the sample 802 at different angles of incidence, it becomes possible to distinguish from each other the light resulting from the illumination of the sample 802 at those different angles of incidence using that or those detectors.

[0072] According to another example, the light source provided in the illumination subsystem can be only one (e.g., the light source 803 shown in FIG. 8), and the light from the light source can be distributed to different optical paths by one or more optical elements (not shown) provided in the illumination subsystem (e.g., based on wavelength, polarization, etc.). Then, the light on these different optical paths can be directed to the sample 802 respectively. The plurality of illumination channels may be configured to direct light to the sample 802 simultaneously, or may be configured to direct light at different times (e.g., when sequentially illuminating the sample using different illumination channels). According to another example, the same illumination channel can be configured to direct light to the sample 802 with different characteristics at different times. For example, according to a certain example, the optical element 804 can be configured as a spectroscopic filter, and by changing the characteristics of the spectroscopic filter in various ways (e.g., by replacing the spectroscopic filter), light of different wavelengths can be directed to the sample 802 at different times. The illumination subsystem can have any other configuration known in the technical field suitable for directing light with different or the same characteristics to the sample 802 sequentially or simultaneously at different or the same incident angles.

[0073] According to an embodiment, the light source 803 can be one having a broadband plasma (BBP) light source. By doing so, the light generated by the light source 803 and directed to the sample 802 can include broadband light. However, the light source can include any other suitable light sources such as lasers and lamps. The lasers can include any lasers known and suitable in the technical field, and the lasers can be configured to generate light at any wavelength or wavelength group known and suitable in the technical field. In addition, the lasers can be configured to generate monochromatic or near-monochromatic light. By doing so, the lasers can be narrowband lasers. The light source 803 can also be one having a multi-color light source that generates light at multiple discrete wavelengths or wavelength bands.

[0074] The light from the optical element 804 may be focused onto the sample 802 by the lens 805. In FIG. 8, the lens 805 is shown as a single refractive optical element, but as can be understood, in reality, the lens 805 may be composed of a plurality of refractive and / or reflective optical elements, and the light may be focused from the optical element to the sample through their cooperation. In addition to this, the illumination subsystem shown in FIG. 8 and described in the present application may be provided with any suitable optical elements (not shown). Examples of such optical elements include, but are not limited to, deflection member(s), spectroscopic filter(s), spatial filter(s), reflective optical element(s), apodizer(s), beam splitter(s) (e.g., beam splitter 813), aperture(s), etc., and all kinds of optical elements known and suitable in the technical field of the present case may be included therein. In addition, the optical base subsystem 801 may be configured such that one or more of the elements included in the illumination subsystem can be modified based on the type of illumination used to generate the optical base output.

[0075] A scanning subsystem configured to scan the sample 802 with light may be provided in the optical base subsystem 801. For example, a stage 806 on which the sample 802 is placed during the generation of the optical base output may be provided in the optical base subsystem 801. The scanning subsystem may be provided with any suitable mechanical and / or robotic assembly (one having the stage 806), and the assembly may be configured to move the sample 802, and thus scan the sample 802 with light. In addition to or instead of this, the optical base subsystem 801 may be configured such that some light scanning can be performed on the sample 802 by one or more optical elements included in the optical base subsystem 801. The scanning of the sample 802 with light may be performed in any suitable manner, such as along a serpentine path or a helical path.

[0076] The optical base subsystem 801 further has one or more detection channels. At least one of the one or more detection channels has a detector, and the detector is configured to detect light provided from the sample 802 due to illumination of the sample 802 by the subsystem and to generate an output according to the detected light. For example, the optical base subsystem 801 shown in FIG. 8 has two detection channels, one of which is formed by a condenser 807, an element 808, and a detector 809, and the other is formed by a condenser 810, an element 811, and a detector 812. As shown in FIG. 8, the two detection channels are configured to collect and detect light at different collection angles. In certain examples, both detection channels are configured to detect scattered light, and the detection channels are configured to detect light scattered from the sample 802 at different angles. However, one or more of the detection channels may be configured to detect a different type of light (e.g., reflected light) from the sample 802.

[0077] Also as shown in FIG. 8, both detection channels are shown in a position within the plane of the paper, and the illumination subsystem is also shown in a position within the plane of the paper. That is, in the present embodiment, both detection channels are arranged (e.g., core-aligned) within the incident plane. However, one or more of the detection channels may be arranged outside the incident plane. For example, the detection channel formed by the condenser 810, the element 811, and the detector 812 may be configured to collect and detect light scattered outside the incident plane. Such detection channels may thus be generically referred to as "side" channels, and such side channels may be core-aligned in a plane substantially perpendicular to the incident plane.

[0078] The optical base subsystem 801 of the embodiment shown in FIG. 8 has two detection channels, but the optical base subsystem 801 may have a different number of detection channels (e.g., a single detection channel or three or more detection channels). According to this kind of example, while one side channel is formed by the detection channel formed by the condenser 810, the element 811, and the detector 812 as described above, a further detection channel (not shown) can be provided in the optical base subsystem 801 and formed as another side channel located on the opposite side of the incident surface. That is, a detection channel having the condenser 807, the element 808, and the detector 809 can be provided in the optical base subsystem 801, core-aligned within the incident surface, and configured to collect and detect light at the surface normal of the sample 802 or at (a plurality of) scattering angles close thereto. This detection channel can thus be generically referred to as the "top" channel, and two or more side channels configured as described above can also be provided in the optical base subsystem 801. That is, the optical base subsystem 801 can be made to have at least three channels (i.e., one top channel and two side channels), and each condenser provided in each of those at least three channels can be configured to collect light at a scattering angle different from any of the other condensers.

[0079] As described above, each detection channel within the optical-based subsystem 801 can be configured to detect scattered light. Thus, the optical-based subsystem 801 shown in FIG. 8 can also be configured to generate a dark field (DF) output related to the sample 802. On the other hand, the optical-based subsystem 801 may be provided, in addition to or instead of, with a detection channel(s) configured to generate a bright field (BF) output related to the sample 802. In other words, the optical-based subsystem 801 may be provided with at least one detection channel configured to detect the light specularly reflected from the sample 802. Thus, the optical-based subsystem 801 described in the present application can be configured for DF only, BF only, or both DF and BF imaging. In FIG. 8, each condenser is shown as a single refractive optical element, but as can be understood, each condenser may have one or more refractive optical dies and / or one or more reflective optical elements.

[0080] The detector provided in the one or more detection channels may be any detector known and suitable in the technical field of the present case. For example, those detectors may include a photomultiplier tube (PMT), a charge-coupled device (CCD), a time delay integration (TDI) camera, and any other detector known and suitable in the technical field of the present case. Those detectors may include non-imaging detectors and imaging detectors. When the detector is made a non-imaging detector in this manner, although each detector can be configured to detect certain scattered light characteristics such as intensity, it cannot be configured to detect the characteristics as a function of the position within the imaging plane. That is, the output generated by each detector incorporated within each detection channel of the optical-based subsystem cannot be image signals or image data, whether signals or data. In this kind of example, a processor, for example, processor 814, may be configured to generate an image of the sample 802 from the non-imaging output of those detectors. On the other hand, in another example, the detector may be an imaging detector, that is, a detector configured to generate an imaging signal or image data. Thus, the optical-based subsystem can be configured in various ways to generate an optical image and other optical-based outputs described in the present application.

[0081] Note that FIG. 8 is provided in the present application to roughly depict the configuration of the optical-based subsystem 801 that can be incorporated into the various system embodiments described in the present application or can generate an optical-based output used in the various system embodiments described in the present application. As is usually done when designing a commercial output acquisition system, the optical-based subsystem 801 having the configuration described in the present application may be modified so that the performance of the optical-based subsystem 801 is optimized. In addition, the various systems described in the present application may be implemented using an existing system (e.g., by adding the functions described in the present application to an existing system). In such systems of this kind, the various methods described in the present application may be provided as an optional function of the system (e.g., in addition to other functions of the system). Alternatively, the systems described in the present application may be designed as a completely new system.

[0082] A processor 814 according to an example communicates with the present system 800.

[0083] FIG. 9 is a block diagram of a system 900 according to an embodiment. The present system 900 has a tool, for example, a wafer inspection tool (one having an electron column 901), and is configured to generate an image of a sample 904, for example, a wafer or a reticle.

[0084] The tool has an output acquisition subsystem, and it also has at least an energy source and a detector. This output acquisition subsystem may be an electron beam-based output acquisition subsystem. For example, in a certain embodiment, the energy directed towards the sample 904 is electrons, and the energy detected from the sample 904 is in the form of electrons. In this configuration, the energy source may be an electron beam source. FIG. 9 shows one of such embodiments, where the output acquisition subsystem has an electron column 901, which is coupled to a computer subsystem 902. The sample 904 can be held by a stage 910.

[0085] As also shown in FIG. 9, the electron column 901 has an electron beam source 903 configured to generate electrons, and those electrons are focused by one or more elements 905 onto a sample 904. Examples of the electron beam source 903 may include a cathode type electron source and an emitter chip. Examples of the one or more elements 905 may include a gun lens, an anode, a beam limiting aperture, a gate valve, a beam current selection aperture, an objective lens, and a scanning subsystem, all of which can have any suitable elements known in the art of this technology.

[0086] Electrons (e.g., secondary electrons) returned from the sample 904 may be focused by one or more elements 906 onto a detector 907. An example of the one or more elements 906 may include a scanning subsystem, which may be the same as the scanning subsystem included in the element 905.

[0087] In addition, the electron column 901 can be provided with any suitable elements known in the art of this technology.

[0088] In FIG. 9, the electron column 901 is shown in a configuration where electrons travel towards the sample 904 at an oblique incident angle and are scattered from the sample 904 at another oblique angle. However, the angle at which the electron beam is directed towards the sample 904 and scattered therefrom can be any suitable angle. In addition, an electron beam type output acquisition subsystem may be configured to generate an image of the sample 904 using multiple modes (e.g., different illumination angles, collection angles, etc.). The multiple modes in the electron beam type output acquisition subsystem will have different image generation parameters in any of its output acquisition subsystems.

[0089] The computer subsystem 902 may be coupled to the detector 907 as described above. With the detector 907, an electron beam image of the sample 904 may be generated by detecting electrons returning from the surface of the sample 904. The electron beam image includes any suitable electron beam image. The computer subsystem 902 may be configured to perform any of the functions described in this application using the output of the detector 907 and / or the electron beam image. The computer subsystem 902 may be configured to perform any of the additional step(s) described in this application. The system 900 having the output acquisition subsystem shown in FIG. 9 may be further configured as described in this application.

[0090] Note that FIG. 9 is provided in this application to roughly depict the configuration of an electron beam-based output acquisition subsystem that can be used in the embodiments described in this application. As is usually done when designing a commercial output acquisition system, the performance of the electron beam-based output acquisition subsystem having the configuration described in this application may be optimized by modifying it. In addition, the systems described in this application may be implemented using an existing system (e.g., by adding the functions described in this application to the existing system). In such a system, the methods described in this application may be provided as an optional function of the system (e.g., in addition to other functions of the system). Alternatively, the systems described in this application may be designed as a completely new system.

[0091] The output acquisition subsystem has been described above as an electron beam-based output acquisition subsystem, but the output acquisition subsystem may be an ion beam-based output acquisition subsystem. Such an output acquisition subsystem may be configured as shown in FIG. 9, except that the electron beam source is replaced with any suitable ion beam source known in the art. In addition, the output acquisition subsystem may be incorporated into any other suitable ion beam-based output acquisition subsystem, such as a commercially available focused ion beam (FIB) system, a helium ion microscope (HIM) system, and a secondary ion mass spectrometry (SIMS) system.

[0092] Computer subsystem 902 has a processor 908 and an electronic data storage unit 909. The processor 908 may include a microprocessor, a microcontroller, and other devices.

[0093] The processor 814 or 908 or the computer subsystem 902 may be coupled to the respective components of the systems 800 or 900 in any suitable manner (e.g., via one or more transmission media, including wired and / or wireless transmission media), so that the respective processors 814 or 908 can receive an output. The processors 814 or 908 may be configured to perform a number of functions using the output. In systems 800 or 900, the respective processors 814 or 908 can receive instructions and other information. The processor 814 or 908 and / or the electronic data storage unit 815 or 909 may optionally be capable of communicating electronically with an additional wafer inspection tool, wafer metrology tool, or wafer review tool (not shown) to receive additional information or send instructions. For example, the processor 814 or 908 and / or the electronic data storage unit 815 or 909 can communicate electronically with a scanning electron microscope (SEM).

[0094] The processor 814 or 908 communicates electronically with a wafer inspection tool, such as detector 809 or 812 or detector 907, respectively. The processors 814 or 908 may be configured to process an image generated using measurement results provided by the detector 809 or 812 or detector 907, respectively. For example, the processor may execute embodiments of method 100, or portions of method 1 or systems 12 and 16.

[0095] The processor 814 or 908, or the computer subsystem 902, or other system(s) or other subsystem(s), as described in the present application, may be a part of various systems, such as a personal computer system, an image computer, a mainframe computer system, a workstation, a network device, an Internet device, and other devices. Any processor known and suitable in the technical field of the present application, for example, a parallel processor, may be provided in the or those subsystems or systems. In addition, the or those subsystems or systems may have a high-speed processing platform and software, whether stand-alone or a network connection tool.

[0096] The processor 814 or 908 and the electronic data storage unit 815 or 909 may each be disposed inside, etc., as a part of the system 800 or 900 or other devices, respectively. For example, the processor 814 or 908 and the electronic data storage unit 815 or 909 may each be provided as a part of a stand-alone control unit or inside a centralized quality control unit. A plurality of the processor 814 or 908 or the electronic data storage unit 815 or 909 may be used respectively.

[0097] The processor 814 or 908 may be implemented in any combination of hardware, software, and firmware. Also, the functions as described in the present application may be executed by a single unit or divided among a plurality of different members, and each of them may be implemented in any combination of hardware, software, and firmware. Program codes or instructions for causing the processor 814 or 908 to execute and implement various methods and functions may be stored in a readable storage medium, for example, in a memory in the electronic data storage unit 815 or 909 or other memories respectively.

[0098] When multiple processors 814 or 908 or computer subsystems 902 are provided in system 800 or 900, different subsystems may be coupled to enable transmission of images, data, information, instructions, etc. between those subsystems. For example, one subsystem may be coupled to another subsystem (or group of subsystems) by any suitable transmission medium, which may include any and all wired and / or wireless transmission media known and suitable in the relevant technical field. Two or more of such subsystems may be substantially coupled by a shared computer-readable storage medium (not shown).

[0099] Processor 814 or 908 may be configured to perform a number of functions using the output of system 800 or 900, respectively, and other outputs. For example, processor 814 or 908 may be configured to send its output to an electronic data storage unit 815 or 909, respectively, or other storage medium. Processor 814 or 908 may be further configured as described in the present application.

[0100] Processor 814 or 908 or computer subsystem 902 may be part of a defect review system, inspection system, metrology system, or any other type of system. That is, some of the configurations described in the embodiments disclosed herein can be tailored in various ways for systems with different capabilities, more or less suitable for different applications.

[0101] When multiple subsystems are provided in the present system, different subsystems may be coupled to enable transmission of images, data, information, instructions, etc. between those subsystems. For example, one subsystem may be coupled to another subsystem (or group of subsystems) by any suitable transmission medium, which may include any and all wired and / or wireless transmission media known and suitable in the relevant technical field. Two or more of such subsystems may be substantially coupled by a shared computer-readable storage medium (not shown).

[0102] The processor 814 or 908 may be configured in accordance with any of the embodiments described in the present application. Further, the processor 814 or 908 may be configured to perform other functions or additional steps using the outputs of the system 800 or 900 respectively or using images or data from other sources.

[0103] The processor 814 or 908 may be communicatively coupled in any manner known in the art to any of the various components or subsystems provided in the system 800 or 900 respectively. Further, the processor 814 or 908 may be configured to receive and / or acquire data or information from other systems (e.g., inspection results from an inspection system such as a review tool, design data in a remote database, etc.) via a transmission medium, for example, having wired and / or wireless segments. By doing so, the transmission medium may be made to function as a data link between the processor 814 or 908 and other subsystem groups of the system 800 or 900 respectively or system groups outside of the system 800 or 900 respectively.

[0104] The processor 814 or processor 908 according to an embodiment may be configured to perform the steps of an embodiment of method 100, or steps according to aspects of method 1 or systems 12 and 16.

[0105] The processor 814 or processor 908 according to an embodiment may be further configured to perform fine alignment with block translation; block translation means dividing a reference image into one or more reference image subsections, dividing a test image into one or more test image subsections, associating each test image subsection with one reference image subsection, and translating each test image subsection and aligning it with the corresponding reference image subsection.

[0106] The various steps, functions, and / or operations of the system 800 or system 900 and the various methods of the present disclosure are performed by one or more of an electronic circuit, a logic gate, a multiplexer, a programmable logic device, an ASIC, an analog or digital controller / switch, a microcontroller, or an information processing system. For example, a program instruction group for implementing the methods described in the present application may be transmitted on or stored in a carrier medium. The carrier medium may include storage media such as read-only memory, random access memory, magnetic or optical disks, non-volatile memory, solid-state memory, magnetic tape, etc. The carrier medium may include transmission media such as wires, cables, or wireless transmission links. For example, the various steps described throughout the present disclosure may be performed by a single processor 814 or a single processor 908 (or computer subsystem 902), or alternatively, by a plurality of processors 814 or a plurality of processors 908 (or a plurality of computer subsystems 902). Further, one or more information processing or logic systems may be incorporated into the various subsystems of the system 800 or system 900. Therefore, the above description should be construed as illustrative rather than limiting to the present disclosure.

[0107] One additional embodiment relates to a non-transitory computer-readable medium storing program instructions executable on a controller, particularly for executing a computer-implemented method for determining the height of an illumination region on the surface of sample 802 or 904 as disclosed in the present application. Specifically, as shown in FIG. 8 or FIG. 9, a non-transitory computer-readable medium containing program instructions executable on processor 814 or 908, respectively, may be incorporated into the electronic data storage unit 815 or 909 or other storage medium. The computer-implemented method may include any of the steps (groups) of any of the methods (groups) described in the present application, starting with the embodiments of method 100, or portions of method 1 or systems 12 and 16.

[0108] The methods, for example, program instructions for implementing those described in the present application, may be stored on a computer-readable medium, for example, in an electronic data storage unit 815, an electronic data storage unit 909, or other storage media. The computer-readable medium may be a storage medium, such as a magnetic or optical disk, magnetic tape, or any other non-transitory computer-readable medium known and suitable in the technical field of the present case.

[0109] Those program instructions may be implemented in any of various ways, especially including procedural-based technology, element-based technology, and / or object-oriented technology. For example, those program instructions may be implemented by optionally using ActiveX (registered trademark) controls, C++ objects, JavaBeans (registered trademark), Microsoft (registered trademark) FoundationClasses (MFC), Streaming SIMD Extensions (SSE), or other technologies or methodologies.

[0110] An element (group) executed by a processor may include a deep learning module (e.g., a convolutional neural network (CNN) module). The deep learning module can be one having a configuration detailed in the present application. Deep learning, which is based on neural network technology, is a probabilistic graph model involving many neuron layers and is widely known as a deep architecture. In deep learning technology, information such as images, text, audio, etc. is processed in a hierarchical manner. When using deep learning in the present disclosure, learning from data is utilized to automatically achieve feature extraction. For example, features referred to when obtaining rotational and translational offsets can be extracted using the deep learning module based on one or more extracted features.

[0111] Put simply, deep learning (also known as deep structured learning, hierarchical learning, or deep machine learning) is a branch of machine learning that relies on a set of algorithms that attempt to model high-level abstractions in data. In a simple case, there would be two sets of neurons, namely those that receive signal inputs and those that send output signals. When an input is received at the input layer, a modified version of that input is passed on to the next layer. In a deep network, since there are many layers between the input and output, multiple processing layers composed of multiple linear and non-linear transformations can be used according to the algorithm.

[0112] Deep learning is a member of the extended family of machine learning methods that rely on data representation learning. Observation results (e.g., extraction target features for reference) can be represented in various ways, such as a vector of per-pixel intensity values or, in a more abstract way, a set of edges, specific shape regions, etc. Some of these representations are superior to others in simplifying the learning task (e.g., face recognition or expression recognition). According to deep learning, efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction can be provided.

[0113] In research in this field, attempts have been made to create better representations and generate models that learn those representations from large-scale data. Some of these representations have been inspired by the progress of neuroscience and loosely rely on neural coding, which attempts to define the information processing and communication patterns in the nervous system, for example, the relationship between various stimuli and the associated neuronal responses in the brain.

[0114] Neural networks with deep architectures have numerous variants depending on their probabilistic specifications and network architectures, including, but not limited to, deep belief networks (DBNs), restricted Boltzmann machines (RBMs), and autoencoders. Another type of deep neural network, namely CNN, can also be used for feature analysis. The actual implementation may vary depending on the size of the input image, the number of features to be analyzed, and the nature of the problem. The deep learning module may include other layers in addition to the neural network disclosed in the present application.

[0115] A deep learning model according to an embodiment is a machine learning model. Machine learning can generally be defined as a type of artificial intelligence (AI) that gives a computer the ability to learn without being explicitly programmed. In machine learning, the focus is on the development of computer programs that can educate and grow and change themselves when exposed to new data. In machine learning, the research and construction of algorithms that can learn from data and make predictions about the data are explored. Such algorithms make data-driven predictions and decisions through model construction from sample inputs, thereby overcoming strict adherence to static program instructions.

[0116] In certain embodiments, the deep learning model is a generative model. A generative model can generally be defined as a model of a probabilistic nature. In other words, a generative model performs forward simulation or a rule-based method. The learning of the generative model (and thus the learning of the various parameters) can be performed based on the data of an appropriate training set. A deep learning model according to an embodiment is configured as a deep generative model. For example, the model can be configured to exhibit a deep learning architecture, and thus multiple layers can be provided in the model and a number of algorithms or transformations can be performed thereby.

[0117] In another embodiment, the deep learning model is configured as a neural network. According to a further embodiment, the deep learning model can be a set of weighted deep neural networks that model the world according to data supplied for training itself. A neural network can generally be defined as an information processing technique based on relatively loosely modeling the way a biological brain solves problems with a relatively large cluster of biological neurons connected by axons with a relatively large population of neural units. Each neural unit can be connected to many other neural units, and their influence on the activation state of the connected neural unit group can be enhanced or suppressed by links. These systems are self-learning and trained rather than being explicitly programmed, and are excellent in fields where it has been difficult to present solution or feature detection strategies with traditional computer programs.

[0118] A neural network is usually composed of multiple layers, and its signal path runs from the front to the back. The goal of a neural network is to solve problems in the same way as a human brain would, but there are also quite abstract ones among neural networks. In current neural network projects, usually, thousands to millions of neural units and millions of connections are handled. The neural network may be any architecture and / or configuration known and suitable in the present technical field.

[0119] For one embodiment, the deep learning model used for the semiconductor inspection application disclosed in the present application is configured as AlexNet. For example, by providing a number of convolutional layers (e.g., 5) and a number of subsequent fully connected layers (e.g., 3) in AlexNet and configuring and training them in combination, features can be analyzed to obtain rotational and translational offsets. For another kind of embodiment, the deep learning model used for the semiconductor inspection application disclosed in the present application is configured as GoogLeNet. For example, layers such as convolutional layers, pooling layers, and fully connected layers in GoogLeNet, such as those detailed in the present application configured and trained to analyze features to obtain rotational and translational offsets, can be incorporated. The GoogLeNet architecture can have a relatively large number of layers (especially compared to some of the other neural networks described in the present application), while some of those layers can operate in parallel, and a collection of layers that function in parallel with each other is generally called an inception module. The other layers can be operated sequentially. Therefore, GoogLeNet is different from the other neural networks described in the present application in that all the layers are not arranged in a sequential structure. Those parallel layers can be similar to the Google (registered trademark) Inception network and other structures.

[0120] In a further kind of embodiment, the deep learning model used for the semiconductor inspection application disclosed in the present application is configured as a Visual Geometry Group (VGG) network. The VGG network is, for example, created by increasing the number of convolutional layers while keeping other parameters of its architecture fixed. By using a sufficiently small convolutional filter in all those layers, it is possible to add convolutional layers to increase the depth. Similar to the other neural networks described in the present application, the VGG network is generated and trained to analyze features to obtain rotational and translational offsets. In the VGG network, a fully connected layer also follows the convolutional layer.

[0121] In some exemplary embodiments, the deep learning model used for the semiconductor inspection applications disclosed in the present application is configured as a deep residual network. For example, similar to several other networks described in the present application, convolutional layers and subsequent fully connected layers can be incorporated into the deep residual network, and they can be combinedly configured and trained for feature extraction. The layers of the deep residual network are configured to learn a residual function with reference to the layer input instead of learning a non-reference function. Specifically, for each stack consisting of several layers, instead of expecting it to directly fit the desired underlying mapping, these layers can be explicitly made to fit the residual mapping, and it is realized by a feed-forward neural network with shortcut connections. A shortcut connection is a connection that skips one or more layers. To generate a deep residual network, a plain neural network structure with convolutional layers can be adopted and shortcut connections can be inserted, whereby the plain neural network can be adopted and converted into the corresponding residual learning network.

[0122] In a further exemplary embodiment, the deep learning model used for the semiconductor inspection applications disclosed in the present application is one having one or more fully connected layers configured to analyze features to obtain rotation and translation offsets. A fully connected layer can generally be defined as a layer in which each node is connected to each node in the previous layer. Classification can be performed in the one or more fully connected layers based on features extracted by a convolutional layer or group of convolutional layers configured as detailed in the present application. The one or more fully connected layers are configured for feature selection and classification. In other words, features are selected from the feature map by the one or more fully connected layers, and further, the input image or images thereof are analyzed based on the selected features. Among the selected features, all features in the feature map may be included (when appropriate), or only a part of the features in the feature map may be included.

[0123] In some embodiments, the information discriminated by the deep learning model includes the characteristic features extracted by the deep learning model. One such embodiment is where the deep learning model has one or more convolutional layers. The convolutional layer or layers may have any configuration known and suitable in the art. By doing so, the deep learning model (or at least a part of the deep learning model) can be configured as a CNN. For example, by configuring the deep learning model as a CNN, usually a stack of convolutional layers and pooling layers, local features can be extracted. According to the embodiments described in the present application, the advantages of deep learning concepts such as CNN can be utilized to solve the representation inversion problem, which is usually intractable. The CNN configuration or architecture of the deep learning model may be any known in the art. The one or more pooling layers may also have any configuration known and suitable in the art (e.g., max pooling layer), and generally, they are configured to reduce the dimensionality of the feature maps generated by one or more convolutional layers while preserving the most important features.

[0124] Generally, the deep learning model described in the present application is a trained deep learning model. For example, the deep learning model can be pre-trained by one or more other systems and / or methods. After the deep learning model is generated and trained in advance, the function of the model can be determined as described in the present application, and one or more additional functions related to the deep learning model can be executed using it.

[0125] As described above, in the present application, a CNN is used to depict the architecture of the deep learning system, but the present disclosure is not limited to CNN. In embodiments, other variants of deep learning architectures may be used. For example, autoencoders, DBNs, and RBMs can be used. Random forests can also be used.

[0126] The training data may be input into model training (e.g., CNN training), and it may be executed in any suitable manner. For example, in the model training, the training data is input into the deep learning model (e.g., CNN), and one or more parameters of the model may be modified until the output of the model becomes the same as (or substantially the same as) the external verification data. One or more trained models can be generated by the model training, and they can be sent to model selection executed using the verification data. Regarding the verification data input into the one or more trained models, by comparing the results brought about by each of the one or more trained models with the verification data, it can be determined whether the model is the best model. For example, the model that brings about the result that most closely matches the verification data may be selected as the best model. Then, using the test data, the model evaluation of the selected model (e.g., the best model) may be performed. The model evaluation may be executed in any suitable manner. Also, the best model may be sent to model deployment, and from there, the best model may be sent to a semiconductor inspection tool for use (post-training mode).

[0127] The various steps of the methods described in the various embodiments and examples of the present disclosure are sufficient to implement the method of the present invention. That is, the method according to an embodiment is essentially composed of a combination of the steps of the methods disclosed in the present application. The methods according to other embodiments are composed of only those steps.

[0128] Although the present disclosure has been described based on one or more specific embodiments, as can be understood, other embodiments of the present disclosure can also be made without departing from the technical scope of the present disclosure.

Claims

1. A beam of electrons is emitted toward the sample, modulating the electron beam to obtain a beam signal, the modulation being based on a control wave that controls the beam signal, the modulation including adjusting a beam current reaching the specimen; detecting secondary electrons and / or backscattered electrons emitted from the sample with an electron detector to obtain electronic data defining a detection signal; determining, with a processor, a comparison between the control wave of the beam signal and the wave of the detection signal; filtering the detection signal based on the comparison using the processor; method.

2. 10. The method of claim 1 further comprising: applying a high pass filter to the detected signal to obtain backscattered electron data.

3. 3. The method of claim 2, A method, wherein filtering the detection signal includes subtracting backscattered electron data from the electron data to obtain secondary electron data.

4. 3. The method of claim 2, A method, wherein filtering the detection signal includes determining a relative ratio that is a ratio of an intensity of the backscattered electron data to an intensity of the electron data.

5. 3. The method of claim 2 further comprising: detecting secondary electrons and / or backscattered electrons emitted from the sample using a reference electron detector to obtain reference electron data; comparing, with the processor, the electronic data to the reference electronic data to generate slope data; varying the slope of the electronic data using the slope data; method.

6. 2. The method of claim 1 , The method, wherein the modulation comprises varying the beam current of the electron beam.

7. 2. The method of claim 1 , The method, wherein the modulating comprises switching the electron beam between on and off.

8. 2. The method of claim 1 , The method, wherein the modulating comprises directing the electron beam between a beam dump and the sample.

9. 2. The method of claim 1 , The method, wherein the modulating further comprises deflecting the electron beam across the sample.

10. an electron beam tool, an electron beam emitter configured to emit electrons in an electron beam toward the specimen; a stage configured to hold the sample; an electron detector configured to detect secondary electrons and / or backscattered electrons emitted by the sample to obtain electronic data defining a detection signal; having a controller in electronic communication with the electron beam tool, the controller having a processor; The controller: instructing the electron beam emitter to modulate the electron beam to obtain a beam signal, the modulation being based on a control wave that controls the beam signal, the modulation including adjusting a beam current reaching the specimen; determining a comparison between the control wave of the beam signal and the wave of the detection signal; filtering the detection signal using the comparison; The system being configured.

11. 11. The system of claim 10, further comprising: A system configured to apply a high pass filter to the detected signal to obtain backscattered electron data.

12. 12. The system of claim 11, wherein the controller: A system configured to filter the detected signal by subtracting backscattered electron data from the electron data to obtain secondary electron data.

13. 12. The system of claim 11, wherein the controller: A system configured to filter the detection signal by determining a relative ratio that is a ratio of an intensity of the backscattered electron data to an intensity of the electron data.

14. The system of claim 10, further comprising: a reference electron detector configured to detect secondary electrons and / or backscattered electrons emitted from the sample to obtain reference electron data. wherein the controller: comparing the electronic data to the reference electronic data to generate slope data; Varying the tilt of the electron beam by varying the angle of the electron beam using the tilt data. The system is configured to:

15. 11. The system of claim 10, wherein the controller: modulating the electron beam by commanding the electron beam emitter to vary the beam current of the electron beam; The system is configured to:

16. 11. The system of claim 10, wherein the controller: configured to modulate the electron beam by commanding the electron beam emitter to switch the electron beam between on and off. system.

17. 11. The system of claim 10, wherein the controller: A system configured to modulate the electron beam by commanding the electron beam emitter to direct the electron beam between a beam dump and the specimen.

18. 11. The system of claim 10, wherein the controller further comprises: A system configured to modulate the electron beam by directing the electron beam emitter to deflect the electron beam across the specimen.

19. A non-transitory computer-readable storage medium comprising one or more programs, the steps of which are executed on one or more information processing devices according to the programs, the steps comprising: commanding the electron beam tool to emit an electron beam toward the specimen; modulating the electron beam to obtain a beam signal, the modulation being based on a control wave that controls the beam signal, the modulation including adjusting a beam current reaching the specimen; receiving electronic data defining a detection signal from an electron detector that detects secondary electrons and / or backscattered electrons emitted from the sample; determining a comparison between the control wave of the beam signal and the wave of the detection signal; filtering the detection signal based on the comparing to obtain a filtered detection signal. A non-transitory computer readable storage medium comprising:

20. 20. The non-transitory computer readable storage medium of claim 19, wherein the one or more programs further comprise: A non-transitory computer readable storage medium configured to perform training of a machine learning algorithm using the beam current, the electronic data, and the filtered detection signal.

Images