Methods for diagnosing, inspecting, and measuring

Abstract

Embodiments of improved apparatus and systems, and related methods, for optical diagnostic semiconductor manufacturing processes are disclosed. In a plasma processing system, a hyperspectral imaging system is used to acquire radiometrically resolved images from the plasma. The acquired hyperspectral images can be used to determine the chemical composition of the plasma and the endpoint of the plasma process. Alternatively, the hyperspectral imaging system is used to acquire spectrally resolved images of a substrate before, during, or after processing to determine the characteristics of the substrate or layers and features formed on the substrate, including whether the process endpoint has been reached; or to inspect the substrate condition before or after processing.

Description

[Technical Field]

[0001] This invention relates to apparatus, systems, and related methods for optically diagnosing the condition of semiconductor manufacturing processes. More specifically, it relates to the use of hyperspectral imaging for diagnosing the condition of semiconductor manufacturing processes and processed substrates. [Cross-reference to related applications]

[0002] This application relates to and claims priority to co-pending U.S. Provisional Patent Application No. 62 / 851,756, entitled “OPTICAL DIAGNOSTICS OF A SEMICONDUCTOR PROCESS USING HYPERSPECTRAL IMAGING” (reference 181412US01), filed May 23, 2019, the entire contents of which are hereby incorporated. This application relates to co-pending U.S. Patent Application No. 16 / 820,032, entitled “ENHANCED RESOLUTION IN SEMICONDUCTOR FABRICATION DATA ACQUISITION INSTRUMENTS USING MACHINE LEARNING” (reference 181077US02), filed March 16, 2020, the entire contents of which are hereby incorporated. [Previous Technology]

[0003] Optical diagnostics has entered the mainstream semiconductor manufacturing process and is currently used to diagnose many different types of semiconductor manufacturing processes, with or without plasma. Processes in which optical diagnostics can be used to measure process parameters and process results include patterning processes such as photolithography, etching, deposition, and cleaning.

[0004] Many optical diagnostic methods utilize the measurement of optical spectra, which are typically acquired using on-tool spectrometers. In plasma processes, such as etching and atomic layer etching (ALE), the optical emission spectra from the etched plasma can be captured and analyzed for process endpoint detection (EPD) and to determine other parameters of the plasma etching process, such as the chemical composition of the plasma processing environment. Example techniques generally categorized under the term Optical Emission Spectroscopy (OES) are described in more detail in U.S. Patent Nos. 9,330,990 and 10,002,804, entitled “METHOD OF ENDPOINT DETECTION OF PLASMA ETCHING PROCESS 5 USING MULTIVARIATE ANALYSIS” (references TTI-240_US_ORD_1 and TTI-240_US_CON_2), and in U.S. Patent Application Nos. 15 / 469,303 and 15 / 469,317, entitled “CONTROLLING DRY ETCH PROCESS CHARACTERISTICS USING WAFERLESS DRY CLEAN OPTICAL EMISSION SPECTROSCOPY” (reference No. TEA-136_US_ORD_2). And TEA-136_US_ORD_3), and U.S. Patent No. 10,436,717, entitled “COMPOSITIONAL OPTICAL EMISSION SPECTROSCOPY FOR DETECTION OF PARTICLE INDUCED ARCS IN AN ETCH PROCESS” (reference number TEA-138_US_ORD_2), the contents of which are incorporated herein by reference in their entirety.

[0005] Optical emission spectroscopy (OES) endpoint signals indicate that a specific layer has been etched through and the underlying layer has been exposed during the etching process, as shown by changes in the optical emission spectrum of the etching plasma. As the feature size of semiconductor devices continues to shrink, low-aperture-area plasma etching processes (where the plasma spectral characteristics of the etch-through conditions become weaker) place increasingly higher demands on endpoint detection (EPD). This necessitates the use of spectrometers with progressively higher signal-to-noise ratios (SNR) to accurately capture the etching endpoint. Simultaneously, the etching process itself requires maintaining high throughput, meaning the spectrometer must acquire optical spectra at a sufficiently high rate to achieve accurate endpoint detection (EPD). These requirements and their resulting tradeoffs are further amplified in cyclic plasma etching processes, such as atomic layer etching (ALE), where plasma conditions vary extensively between steps in a cycle, and the transitions between each cycle are dominated by rapid transients in the plasma.

[0006] In another series of optical diagnostic methods, a spectrometer is used during the manufacturing process to measure the spectral content of a broadband beam reflected from the surface of a workpiece (e.g., a wafer, substrate, etc.). These optical diagnostic methods (classified as generally known scattering and reflection methods) typically do not rely on plasma optical emission (though they may), and are therefore applicable to a wider range of semiconductor manufacturing processes, both plasma and non-plasma. The broadband optical spectrum captured by reflecting and diffracting the beam of illumination from structures and features on a substrate can be used in several different applications. Examples include detecting the end of a process, measuring the thickness of layers formed on a substrate, measuring the dimensions of structures or features formed on a substrate, and measuring the properties of the materials used to form layers and features on a substrate. What all these methods have in common is that optical diagnostics are specifically targeted at the layers and features on the substrate to be processed in order to diagnose and control the process. Some example techniques are described in more detail in U.S. Patent No. 9,059,038, entitled “SYSTEM FOR IN-SITU FILM STACK MEASUREMENT DURING ETCHING AND ETCH CONTROL METHOD” (reference number TTI-139_US_ORD_1), U.S. Patent Application No. 15 / 472,494, entitled “ADVANCED OPTICAL SENSOR, SYSTEM, AND METHODOLOGIES FOR ETCH PROCESS MONITORING” (reference number TTI-15 249_US_PRI_1), and U.S. Patent Application No. 16 / 051,082, entitled “NORMALINCIDENT IN-SITU PROCESS MONITOR SENSOR” (reference number 180903US01), the contents of which are incorporated herein by reference in their entirety.

[0007] In conventional spectrometers with gratings or other dispersive optics, the signal-to-noise ratio (SNR) is generally highly inversely proportional to its output. To achieve a higher SNR, a stronger light signal needs to be captured at each pixel of the detector array, where the elements or pixels of the detector array correspond to the wavelength of the captured optical spectrum. Conventional methods for increasing the signal are (a) increasing the slit width at the spectrometer entrance and (b) increasing the acquisition time. The former method (usually used in conjunction with a focusing lens or mirror with a large numerical aperture (NA) and effectively increasing the amount of light reaching the detector array, thus improving the SNR) has the disadvantage of reducing spectral resolution. This results in the spectrometer being unable to resolve closely spaced spectral peaks. This lack of spectral information may render the captured spectrum unsuitable for diagnostic and control processes. The latter method is undesirable in terms of reducing the spectrometer output because increasing the acquisition time to allow more signal to be collected at the detector array pixels reduces the rate of optical spectrum acquisition. This reduction in output can be detrimental to methods such as End Point Detection (EPD), in which precise timing of the process endpoint is crucial.

[0008] Based on the current requirements for process diagnostics (especially etching diagnostics), the aforementioned replacements become more stringent for each new generation of devices and processes. Therefore, new spectrometer designs and optical diagnostic systems are needed for in-situ semiconductor process diagnostics and control that do not sacrifice output and spectral resolution to improve the signal-to-noise ratio (SNR) or minimize replacements.

[0009] To this end, the present invention proposes a system and method for optical diagnosis and control of semiconductor manufacturing processes that uses hyperspectral imaging technology to replace conventional spectrometers.

[0010] Because hyperspectral imaging is an imaging technique, it is applicable to a wide range of optical diagnostic methods in semiconductor processing. For example, optical diagnostics involving imaging a substrate or a region of a substrate to determine the substrate and the structure and layer state formed thereon (where the full-spectrum information provided by hyperspectral imaging will be helpful for diagnosis) are good options for using this technique. Examples of optical diagnostics that can employ hyperspectral imaging include many optical imaging inspection steps, such as post-development inspection (ADI) in photolithography patterning processes, many imaging defect detection steps, and detection of contamination on substrates. [Summary of the Invention]

[0011] This invention relates to several embodiments of methods, apparatus and systems for hyperspectral imaging to diagnose the state of a substrate during semiconductor processing, and / or to optically diagnose and control the state of the semiconductor manufacturing process itself.

[0012] In one embodiment, an optical detector is provided for detecting optical signals from a semiconductor processing system. The optical detector is configured to detect optical signals transmitted through a window mounted in the wall of the semiconductor processing system. The optical detector includes collecting optics configured to collect and transmit the optical signals transmitted through the window; a wavelength-tunable filter for tunably selecting the wavelength of the transmitted optical signal; an array detector for detecting wavelength-filtered optical signals; and a controller for controlling at least the wavelength-tunable filter and the array detector, and for storing and processing images captured by the array detector. The optical detector can be used to detect optical signals including plasma emission, such as in optical emission spectroscopy (OES) applications, including, for example, endpoint detection (EPD). The optical detector can also be used to detect diffracted optical signals, wherein the diffracted optical signals are generated by reflecting and diffracting an illumination beam from a substrate surface in the semiconductor processing system. The illumination beam and diffracted light signal can have normal incidence on the substrate, or the optical detector can be configured for oblique (non-normal) incidence on the substrate. Collecting lenses, deflectors, apertures, optical fibers, polarizers, etc., can constitute the collecting optics. The wavelength-tunable filter can include, but is not limited to, tunable Fabry-Perot resonators, acousto-optic tunable filters, and liquid crystal tunable filters. Depending on the application requirements, the array detector can be one-dimensional or two-dimensional, and can include, but is not limited to, CCD detector arrays, CMOS detector arrays, photodiode arrays, etc. The controller can be configured to selectively adjust the passband wavelength of the wavelength-tunable filter within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. The controller can be further configured to capture image cubes, each image cube consisting of a set of images captured by an array detector, each image in the set being captured at a wavelength defined by an instantaneous passband wavelength adjusted by a wavelength-tunable filter. In one embodiment, an optical detector can be used in a semiconductor substrate inspection system to inspect (through imaging) the substrate during or after a processing step. The inspection system can be part of a semiconductor processing system, or it can be a stand-alone kiosk-type inspection system.

[0013] Further embodiments include a semiconductor processing system and a semiconductor substrate inspection system equipped with the aforementioned optical detector. The semiconductor processing system may be a plasma or non-plasma processing system, or a hybrid chamber processing system. The semiconductor substrate inspection system may be part of the semiconductor processing system, or it may be a stand-alone kiosk-type inspection system.

[0014] A further embodiment includes a method for diagnosing plasma process steps in a plasma processing system, comprising: igniting plasma in a plasma processing chamber of the plasma processing system; collecting plasma light emission signals through a window installed in the wall of the plasma processing chamber and through collecting optics; guiding the plasma light emission signals from the collecting optics to a wavelength-tunable filter to tunably select the wavelength of the collected plasma light emission signals; detecting the wavelength-filtered plasma light emission signals using an array detector; and storing and processing the images captured by the array detector in a controller. The method may further include selectively adjusting the passband wavelength of the wavelength-tunable filter within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. The method may also include forming an image cube, which consists of a set of images captured by an array of detectors, each image being captured at a wavelength defined by an instantaneous passband wavelength adjusted by a wavelength-tunable filter. This method can be used to determine the endpoint of a plasma process step from a series of plasma photoemission spectra formed from a series of image cubes captured during the plasma process. It can also be used to determine the relative concentration of the plasma's chemical composition from the plasma photoemission spectra formed from the image cubes.

[0015] A further embodiment includes a method for measuring the characteristics of a layer or structure formed on a substrate disposed in a semiconductor processing system, comprising: initiating a process step in the semiconductor processing system; irradiating the substrate with an irradiation beam, the irradiation beam forming a diffracted light signal by a portion reflected and diffracted from the substrate; collecting the diffracted light signal through a window mounted in the wall of the semiconductor processing system and through a collecting optics; guiding the diffracted light signal from the collecting optics to a wavelength-tunable filter to tunably select the wavelength of the collected diffracted light signal; detecting the wavelength-filtered diffracted light signal using an array detector; and storing and processing the image captured by the array detector in a controller. The characteristics of the layer or structure on the substrate may be a top profile critical dimension (CD), a bottom profile critical dimension (CD), a middle profile critical dimension (CD), a profile sidewall angle, a layer thickness, layer optical properties, or the remaining thickness of an etched layer, etc. Furthermore, the endpoint of the process step can be determined from a series of diffraction spectra obtained from a series of image cubes captured during the process step.

[0016] The method may further include determining the spatial correspondence between a position on the substrate and a pixel position in an image of an image cube, and obtaining a diffraction spectrum from the image cube. The diffraction spectrum may be obtained from a single selected pixel position within all images or a subset of images in one or more image cubes. Alternatively, the diffraction spectrum may be obtained from at least one selected pixel position region within all images or a subset of images in one or more image cubes. Finally, the diffraction spectrum may be obtained from all pixel positions within all images or a selected subset of images in one or more image cubes.

[0017] In plasma diagnostic applications, the same three steps described above for extracting diffraction spectra from one or more image cubes can also be used to extract plasma emission spectra from one or more image cubes. In that case, the pixel positions in the image cube do not correspond to positions on the substrate, but rather to regions of the probe volume within the plasma processing chamber from which the plasma emission signal is captured.

[0018] These same steps can be further used to extract optical spectra from one or more image cubes in a substrate inspection system.

[0019] In a substrate inspection system, an embodiment of a method for inspecting a substrate includes: placing the substrate on a platform in the substrate inspection system; irradiating the substrate with an illumination beam; collecting the light signal caused by irradiating the substrate using a collecting optics; guiding the collected light signal from the collecting optics to a wavelength-tunable filter to tunably select the wavelength of the collected light signal; detecting the wavelength-filtered signal using an array detector; and storing and processing the image captured by the array detector in a controller. The method may further include determining at least one characteristic of the substrate or a layer or structure formed on the substrate from at least one optical spectrum obtained from an image cube according to the previously described method for extracting spectra from an image.

Implementation Method

[0032] In the following description, specific details, such as the particular geometry of optical detectors, plasma processing systems, non-plasma semiconductor processing systems, and optical imaging detection systems, as well as descriptions of numerous components and processes, are set forth for the purpose of promoting a thorough understanding of the invention and for purposes of explanation rather than limitation. However, it should be understood that the invention may be practiced in other embodiments departing from these specific details.

[0033] Throughout this specification, the reference to "an embodiment" or "an embodiment" means that a particular feature, structure, material, or characteristic described with respect to that embodiment is included in at least one embodiment of the invention, but does not mean that it exists in every embodiment. Therefore, the appearance of the terms "in an embodiment" or "in an embodiment" in various places throughout this specification does not necessarily refer to the same embodiment of the invention. Furthermore, particular features, structures, materials, or characteristics may be incorporated in any suitable manner in one or more embodiments.

[0034] The various operations will be described sequentially as multiple separate operations in a manner most conducive to understanding the invention. However, the order of description should not be construed as implying that these operations must be performed in sequence. In particular, these operations need not be performed in the order presented. The operations may be performed in a different order than in the described embodiments. In additional embodiments, numerous additional operations may be performed and / or the operations may be omitted.

[0035] Figures 1A and 1B illustrate a semiconductor processing system according to an embodiment of the present invention. Figure 1A illustrates an exemplary semiconductor processing system that utilizes plasma to process a substrate, such as an etching system, an atomic layer etching (ALE) system, a deposition system, etc., hereinafter referred to as plasma processing system 100. Plasma processing system 100 includes a plasma processing chamber 110, which may further include an inductively coupled plasma source (ICP), a capacitively coupled plasma source (CCP), a transformer-coupled plasma source (TCP), a microwave plasma source, a surface wave plasma source (SWP), etc. (not shown). Together with a gas supply system (not shown), the plasma source is used to strike and hold the plasma within the plasma processing chamber 110, above a substrate 120 disposed on a support member 115. The support member 115 may be an electrostatic chuck (ESC), a base, etc.

[0036] The optical detector 140 is used to monitor the state of the plasma within the plasma processing chamber 110. A window 130 is provided in the side wall of the plasma processing chamber 110 to allow an optical signal (in this case, the plasma light emission signal 170) to leave the plasma processing chamber 110 and enter the optical detector 140. In an alternative embodiment, the window 130 may be mounted on the top wall of the plasma processing chamber 110, and the optical detector 140 may in this case be mounted above the plasma processing chamber 110, or the plasma light emission signal 170 may be transmitted along a path to the remote optical detector 140 using, for example, an optical fiber. The window 130 may be made of a suitable material that is transparent to the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. For example, the window 130 may include various types of glass, such as borosilicate glass, quartz, sapphire, etc. A device that can protect the window 130 from excessive deposition of material from plasma, such as a perforated grid, a flushing gas injection system, etc. (not shown).

[0037] The optical detector 140 includes collecting optics 145, a wavelength-tunable filter 150, and an array detector 160, which together constitute a hyperspectral imaging (HSI) system. The optical detector and its components are controlled by a controller 180 that communicates with at least the wavelength-tunable filter 150 and the array detector 160. The wavelength-tunable filter may include a Fabry-Perot cavity, an acousto-optic tunable filter, a liquid crystal tunable filter, or any other tunable narrowband optical filter device capable of changing the passband wavelength, as selected and controlled by the controller 180. Depending on the application, the array detector 160 may be a one-dimensional array detector (i.e., a line detector) or a two-dimensional array detector. If required by the application, the array detector 160 may include a CCD detector array, a CMOS detector array, a photodiode array, or a combination of various types of detectors for different parts of the array detector 160. For example, due to their inherent differences in wavelength sensitivity, different array detector types can be used side-by-side or stacked for different wavelength ranges.

[0038] In operation, the plasma emission signal 170 is collected by collecting optics 145 and introduced into a wavelength-tunable filter 150 as a transmitted optical signal 172. Collecting optics 145 may include any combination of collecting lenses, deflectors, apertures, polarizers, optical fibers, etc., required to form the transmitted optical signal 172. Figure 1A shows an example of a Fabry-Perot resonator where two semi-reflective mirrors 152 are separated by a controllable gap Gi, thereby forming an optical cavity 155. As the gap Gi is varied by a controller 180, the Fabry-Perot resonator selectively transmits different passband wavelengths to form a wavelength-filtered optical signal 175, which is incident on an array detector 160. A suitable exemplary Fabry-Perot resonator is the HinaLea 4200 hyperspectral imager, available from TruTag Technologies. This Fabry-Perot resonator (integrated with a 2.3 MegaPixel array detector) has a spectral range of 200 to 1000 nm and up to 600 spectral bands.

[0039] In operation, controller 180 controls the controllable gap Gi to perform a scan of the passband wavelength, while array detector 160 simultaneously captures images subsequently received by controller 180, wherein the images may be stored and / or processed in controller 180. Each image captured by array detector 160 is captured at the instantaneous passband wavelength selected by wavelength-tunable filter 150 as the passband wavelength of wavelength-tunable filter 150 changes. In this way, the hereinafter “image stack” or “image cube” refers to each scan of the passband wavelength of wavelength-tunable filter 150, which is then received by controller 180 (see Figures 5A-C). Passband wavelength scans can be repeated to capture consecutive image cubes. Controller 180 can be programmed, for example, to perform continuous scans of the passband wavelength of wavelength-tunable filter 150 (from minimum to maximum passband wavelength). However, in reality, only certain portions of the optical spectrum may be of interest. Therefore, the controller 180 can be programmed to selectively adjust the passband wavelength within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. In this way, the output is improved because only the wavelength range of interest is captured.

[0040] For a Fabry-Perot resonator, the typical time for switching and setting the passband wavelength is around 0.1 ms, thus allowing, for example, the selection of 100 passband wavelengths within 10 ms and the formation of a corresponding image by the array detector 160. Although this spectral acquisition speed may be slower than some high-performance spectrometers, the advantage of a hyperspectral imaging (HSI) system is that the amount of light received through the plasma emission signal 170 is not limited by the slit width of the spectrometer or the numerical aperture (NA) of the collecting optics. Furthermore, a large effective aperture can be used in the collecting optics 145, resulting in the array detector 160 detecting a strong signal, thus significantly improving the signal-to-noise ratio (SNR) compared to conventional spectrometers. Due to the typical 2 nm spectral resolution of a Fabry-Perot resonator, a single optical detector 140 employing a hyperspectral imaging (HSI) system combines high SNR and good spectral resolution. If the output of a single optical detector 140 is insufficient for the application, multiple optical detectors 140 operating in parallel can be used, each capturing a portion of the plasma emission spectrum.

[0041] Plasma optical diagnostics relies on the acquisition of plasma photoemission spectra, as previously described. Figures 5A-C illustrate the process of extracting plasma photoemission spectra from exemplary image cubes 900, 902, and 905. In Figure 5A, image cube 900 is captured at n passband wavelengths, varying from λ1 through λi to λn. In this simplest embodiment, the plasma photoemission spectrum is extracted by extracting light intensity from the same single selected pixel 940 across all images of image cube 900. When the extracted light intensities are arranged in ascending wavelength order, the plasma photoemission spectrum 950 of Figure 5A is obtained, which can be further utilized in the same manner as spectra obtained by conventional spectrometers.

[0042] Alternatively, in FIG. 5B, the plasma emission spectrum is extracted by extracting and averaging the light intensity within at least one selected pixel region 940 of all images in the image cube 902. Multiple discontinuous pixel regions 940 can be selected and averaged, depending on the application, which specifies which parts of the image are retained and which are discarded. For each image captured at each wavelength λi, various pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within the selected pixel region 940. Once averaged, the averaged extracted light intensities are arranged in ascending wavelength order to form the plasma emission spectrum 950 of FIG. 5B.

[0043] Alternatively, in FIG. 5C, the plasma emission spectrum is extracted by extracting and averaging the light intensity within all pixels 940 of the captured image of the image cube 905. For each image captured at each wavelength λi, a variety of pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within the pixels 940. Once averaged, the average extracted light intensities are arranged in ascending wavelength order to form the plasma emission spectrum 950 of FIG. 5C.

[0044] When forming the plasma photoemission spectrum 950 as shown in Figures 5A-C, all images in image cubes 900, 902, or 905 can be used to form the plasma photoemission spectrum 950. Alternatively, only a subset of the images in image cubes 900, 902, and 905 can be used to form the plasma photoemission spectrum 950, thus removing spectral portions not needed for a given application. For example, in endpoint detection (EPD) applications, only certain narrow portions of the plasma photoemission spectrum may be of interest. This approach of first reducing the number of images captured in the image cube (i.e., the number of passband wavelengths) and then further selecting only a subset of the images in the image cube for plasma photoemission spectrum extraction can significantly improve the computational efficiency of the endpoint detection (EPD) algorithm. The same efficiency gain can be seen in other possible applications, such as actinometry applications, where plasma photoemission spectroscopy is used to determine the relative concentration of at least one chemical component in plasma.

[0045] In plasma optical diagnostics, due to the configuration of the collecting optics 145 in FIG1A, the pixel positions in the two-dimensional images of the image cubes 900, 902, and 905 in FIG5A-C generally correspond to different regions in the plasma, which are imaged onto the individual pixels of the array detector 160. One of the methods in FIG5A and 5B can be advantageously selected to accurately select the region in the plasma to be detected, thereby achieving, for example, more reliable endpoint detection (EPD). At the same time, using the methods in FIG5A and 5B allows the removal of a portion of the plasma emission that is strong, for example, due to interference (from the structure or substrate 120 in the plasma processing chamber 110). Furthermore, if a one-dimensional array detector 160 is used, the same principle described above for plasma emission spectrum extraction is applied to the case of one-dimensional images. In plasma optical diagnostics where binning and averaging across many pixels in one direction are not required, a one-dimensional array detector 160 (i.e., a line detector) can be advantageously used to reduce the cost of the optical detector 140 and increase processing speed and output.

[0046] If a higher spectral resolution (higher than that obtainable through a wavelength-tunable filter of a selected type 150) is required, a spectral deconvolution method can be used to reconstruct a high-resolution spectrum from the captured plasma emission spectrum 950. An example spectral deconvolution method is disclosed in M. Morháč's "Deconvolution methods and their applications in the analysis of gamma-ray spectra" (ACAT2005, May 22-27, Zeuthen, Germany). Alternatively, machine learning techniques disclosed in co-pending U.S. Patent Application No. 16 / 820,032 (titled "ENHANCED RESOLUTION IN SEMICONDUCTOR FABRICATION DATA ACQUISITION INSTRUMENTS USING MACHINE LEARNING" (reference number 181077US02, filed March 16, 2020) can be used to reconstruct a higher resolution spectrum from the captured plasma photoemission spectrum 950.

[0047] In Figure 1B (where the same reference numerals denote the same elements as in Figure 1A), a plasma processing system 200 is shown that differs from the plasma processing system 100 of Figure 1A, except that an optical fiber 135 is used to feed the plasma light emission signal 170 into the optical detector 140. The use of the optical fiber 135 provides additional flexibility in positioning the window 130 (which can be mounted on the side or top wall of the plasma processing chamber 110). The optical fiber 135 allows for further flexibility in remotely positioning the optical detector 140. The optical fiber 135 couples the plasma light emission signal 170 and transmits it to the collecting optics 145, and may include coupling optics, multiple fiber bundles, etc., to ensure proper coupling of the plasma light emission signal 170 and optimal transmission along the fiber. In all other respects, the plasma processing system 200 functions and operates similarly to or equivalent to the plasma processing system 100 of Figure 1A as described above.

[0048] Figures 2A and 2B illustrate a semiconductor processing system according to a further embodiment of the present invention. Figure 2A depicts an exemplary semiconductor processing system 300. The semiconductor processing system 300 may or may not utilize plasma for processing and may include or be part of a photolithographic track system, etching system, atomic layer etching (ALE) system, deposition system, wet or dry cleaning system, diffusion furnace system, or any other semiconductor processing system, wherein the progress of the process steps needs to be monitored to ensure that the substrate meets specifications and for process step control purposes. The semiconductor processing system 300 includes a process module 210 having a support member 215 for supporting a substrate 220 during process steps. The support member 215 may be part of an electrostatic chuck (ESC), a base, a movable stage, a platform / substrate arm, etc. The process steps are performed on the substrate 220 within the process module 210, during which layers or features may be formed or modified on the substrate 220. For example, in an etching system, layers and features on the substrate may be etched. In photolithography systems, photoresist coatings can be applied, baked, or removed. In deposition systems, layers can be deposited onto substrates. In wet or dry cleaning systems, features, layers, processing solutions, contaminants, etc., can be removed from substrates.

[0049] During the process steps, the optical detector 240 uses reflection to monitor the state of the substrate within the semiconductor processing system 300 and the process module 210. The illumination beam 234 is generated by a broadband illumination light source 232 and guided to the substrate 220 via a semi-reflective mirror 233 and a window 230 positioned relative to the substrate 220, with normal incidence (zero angle of incidence). Some process modules 210 may not have walls, meaning they may not be enclosed or isolated from their surroundings; in such cases, the window 230 may not be necessary. The illumination light source can be a continuous wave (CW) or pulsed light source, and it can cover the wavelength range of the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. The illumination light source 232 may include various incandescent and gas discharge light sources, flash lamps, LEDs, lasers, laser-driven plasma light sources (LDLS), etc. If a particularly wide range of light wavelengths is used for the reflective method, the illumination source 232 may include multiple sources, each covering a different range of light wavelengths, and suitable beam combiner optics may be used to combine the beams. The illumination source 232 may also include a shutter (not shown) to adjust the illumination beam 234 so that interference from plasma or other captured radiation can be subtracted from the measured reflectometer signal when the shutter is closed. The window 230 may be made of a suitable material that is transparent to the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. For example, the window 230 may include various types of glass, such as borosilicate glass, quartz, sapphire, etc. Facilities may be provided to protect the window 230 from excessive material deposition from plasma or other corrosive environments within the process module 210, such as perforated grilles, flushing gas injection systems, etc. (not shown).

[0050] When incident on the substrate 220, the irradiation beam 234 is reflected and diffracted due to the presence of layers, structures, or features formed on the substrate 220, thereby forming a diffracted light signal 236. The diffracted light signal 236 includes a zero-order diffracted light signal reflected from the substrate 220 at a normal (zero) reflection angle. As discussed previously, the spectrum of the diffracted light signal 236 contains information from which the characteristics of the layers, structures, or features on the substrate 220 can be determined. The diffracted light signal 236 passes through the window 230 (if present) and is reflected by the semi-reflective mirror 233 into the optical detector 240.

[0051] Optical detector 240 includes collecting optics 245, a wavelength-tunable filter 250, and an array detector 260, which together constitute a hyperspectral imaging (HSI) system. The optical detector and its components are controlled by a controller 280 that communicates with at least the wavelength-tunable filter 250 and the array detector 260. The wavelength-tunable filter may include a Fabry-Perot resonator, an acousto-optic tunable filter, a liquid crystal tunable filter, or any other tunable narrowband optical filter device capable of changing the passband wavelength, as selected and controlled by controller 180. Depending on the application, array detector 260 may be a one-dimensional array detector (i.e., a line detector) or a two-dimensional array detector. If required by the application, array detector 260 may include a CCD detector array, a CMOS detector array, a photodiode array, or a combination of various types of detectors for different parts of array detector 260. For example, due to their inherent differences in wavelength sensitivity, different array detector types can be used side-by-side or stacked for different wavelength ranges.

[0052] In operation, the diffracted optical signal 236 is collected by the collecting optics 245 and introduced into the wavelength-tunable filter 250 as the transmitted optical signal 172. The collecting optics 245 may include any combination of collecting lenses, deflectors, apertures, polarizers, optical fibers, etc., required to form the transmitted optical signal 272. Figure 2A shows an example of a Fabry-Perot resonator, in which two semi-reflective mirrors 252 are separated by a controllable gap Gi to form an optical cavity 255. When the gap Gi is varied by the controller 280, the Fabry-Perot resonator selectively transmits different passband wavelengths to form a wavelength-filtered diffracted optical signal 275, which is incident on the array detector 260.

[0053] In operation, the controller 280 controls the controllable gap Gi to perform a scan of the passband wavelength, while the array detector 260 simultaneously captures images subsequently received by the controller 280. These images can be stored and / or processed within the controller 280. As the passband wavelength of the wavelength-tunable filter 250 changes, each image captured by the array detector 260 is captured at the instantaneous passband wavelength selected by the wavelength-tunable filter 250. In this manner, the hereinafter "image stack" or "image cube" refers to each scan of the passband wavelength of the wavelength-tunable filter 250, which is then received by the controller 280 (see Figures 5A-C). Passband wavelength scans can be repeated to capture consecutive image cubes. The controller 280 can be programmed, for example, to perform continuous scans of the passband wavelengths of the wavelength-tunable filter 250 (from minimum to maximum passband wavelength). However, in reality, only certain portions of the optical spectrum may be of interest. Therefore, the controller 280 can be programmed to selectively adjust the passband wavelength within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. In this way, the output is improved because only the wavelength range of interest is captured.

[0054] For a Fabry-Perot resonator, the typical time for switching and setting the passband wavelength is around 0.1 ms, thus allowing, for example, the selection of 100 passband wavelengths within 10 ms and the formation of a corresponding image by the array detector 260. Although this spectral acquisition speed may be lower than that of some high-performance spectrometers, the advantage of a hyperspectral imaging (HSI) system is that the amount of light received through the diffracted light signal 236 is not limited by the slit width of the spectrometer or the numerical aperture (NA) of the collecting optics. Furthermore, a large effective aperture can be used in the collecting optics 245, resulting in the array detector 260 detecting a strong signal, thus significantly improving the signal-to-noise ratio (SNR) compared to conventional spectrometers. Due to the typical 2 nm spectral resolution of a Fabry-Perot resonator, a single optical detector 240 employing a hyperspectral imaging (HSI) system combines high SNR and good spectral resolution. If the output of a single optical detector 240 is insufficient for the application, multiple optical detectors 240 operating in parallel can be used, each capturing a portion of the diffracted light signal 236 (which can be pre-splitted using a suitable beam splitter or other wavelength-selective optics).

[0055] The reflection method relies on the acquisition of diffraction spectra, as previously described. Figures 5A-C illustrate the process of extracting diffraction spectra from exemplary image cubes 900, 902, and 905. In Figure 5A, image cube 900 is captured at n passband wavelengths, varying from λ1 through λi to λn. In this simplest embodiment, the diffraction spectrum is extracted by extracting light intensity from the same single selected pixel 940 in all images of image cube 900. When the extracted light intensities are arranged in ascending wavelength order, the diffraction spectrum 950 of Figure 5A is obtained, which can be further utilized in the same manner as spectra obtained by conventional spectrometers.

[0056] Alternatively, in FIG. 5B, the diffraction spectrum is extracted by extracting and averaging the light intensity within at least one selected pixel region 940 of all images in the image cube 902. Multiple discontinuous pixel regions 940 may be selected and averaged, depending on the application, which specifies which parts of the image are retained and which are discarded. For each image captured at each wavelength λi, a variety of pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within the selected pixel region 940. Once averaged, the averaged extracted light intensities are arranged in ascending wavelength order to form the diffraction spectrum 950 of FIG. 5B.

[0057] Alternatively, in FIG. 5C, the diffraction spectrum is extracted by extracting and averaging the light intensity within all pixels 940 of the captured image of the image cube 905. For each image captured at each wavelength λi, a variety of pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within the pixels 940. Once averaged, the average extracted light intensities are arranged in ascending wavelength order to form the diffraction spectrum 950 of FIG. 5C.

[0058] When forming the diffraction spectrum 950 as shown in Figures 5A-C, all images in image cubes 900, 902, or 905 can be used to form the diffraction spectrum 950. Alternatively, only a subset of the images in image cubes 900, 902, and 905 can be used to form the diffraction spectrum 950, thus removing spectral portions not needed for a given application. For example, in endpoint detection (EPD) applications, only certain narrow portions of the diffraction spectrum may be of interest. This approach of first reducing the number of images captured in the image cube (i.e., the number of passband wavelengths) and then further selecting only a subset of the images in the image cube for diffraction spectrum extraction can significantly improve the computational efficiency of the endpoint detection (EPD) algorithm. The same efficiency gain can be seen in other possible applications, such as CD measurement, feature profile measurement, and determination of the optical properties of upper layers of a substrate.

[0059] In the reflection method, due to the arrangement and position of the collecting optics 245, the semi-reflective mirror 233, etc., relative to the substrate 220, the pixel positions in the two-dimensional images of the image cubes 900, 902, and 905 in Figures 5A-C correspond to different positions on the substrate 220, which are imaged onto the individual pixels of the array detector 260. This spatial correspondence between the pixel positions and the positions on the substrate 220 can be determined through optical modeling or by using captured images to verify the image field of view on the substrate 720. Utilizing the known spatial correspondence, one of the methods in Figures 5A and 5B can be advantageously used to accurately select the area on the substrate 220 to be diagnosed, thereby achieving, for example, more reliable endpoint detection (EPD). This allows for the removal of a portion of the substrate surface if these portions do not have a meaningful impact on the reflection method signal (e.g., unetched areas, areas with different layers or features and their optical properties, scribing, interference process module structures, etc.).

[0060] In one embodiment, the sequentially captured images of the image cube (for the same passband wavelength λi) can be analyzed for pixels or pixel regions where the light intensity changes rapidly between consecutive images (representing, for example, etching at these locations). In one embodiment, the rate of change of light intensity of the pixel or pixel region can be compared with a pre-selected light intensity threshold rate of change to determine whether the pixel or pixel region is experiencing a rapid light intensity change, thereby enabling processing. These pixels or pixel regions with rapid light intensity changes can be used as regions 940 for extracting diffraction spectra 950. This method eliminates the need to know the spatial correspondence between pixel positions and positions on the substrate, and eliminates the need to pre-select pixel regions for diffraction spectrum extraction. Simultaneously, this method ensures that only the processed areas on the substrate are diagnosed, while ignoring all other areas, thus improving diagnostic accuracy.

[0061] Another advantage of being able to select only certain pixel regions 940 in the image of the image cube is that it allows for a relatively large illumination point and field of view to be used on the optical detector 240 on the substrate 220. This simplifies the optical setup, relaxes dimensional tolerances, and reduces the cost of the optical detector 240 and the entire semiconductor processing system 300. Finally, it relaxes the positioning tolerances of the substrate 220 on the support member 215, because the required pixel region 940 can always be found in the image of the image cube, as long as it is within the field of view of the optical detector 240.

[0062] If a higher spectral resolution (higher than that obtainable through a wavelength-tunable filter 250 of a selected type) is required, a spectral analysis method can be used to reconstruct a high-resolution spectrum from the acquired diffraction spectrum 950. An example spectral analysis method is disclosed in M. Morháč's "Deconvolution methods and their applications in the analysis of gamma-ray spectra" (ACAT2005, May 22-27, Zeuthen, Germany). Alternatively, machine learning techniques disclosed in co-pending U.S. Patent Application No. 16 / 820,032 (titled "ENHANCED RESOLUTION IN SEMICONDUCTOR FABRICATION DATA ACQUISITION INSTRUMENTS USING MACHINE LEARNING" (reference number 181077US02, filed March 16, 2020) can be used to reconstruct a higher-resolution spectrum from the acquired diffraction spectrum 950.

[0063] Characteristics that can be determined by the reflection method include the remaining thickness of the film on the substrate structure, layer thickness, critical dimensions (CD) (e.g., top, middle, and bottom critical dimensions), contour dimensions (e.g., contour height, sidewall angles, etc.), and optical properties of the upper layer of the substrate. The remaining thickness of the film is a measured quantity used to determine the end point of the etching process. Furthermore, the inventors have demonstrated that by using the reflection method with a hyperspectral imaging (HSI) system, the control of the remaining thickness in the FINFET etching process has been improved by five times throughout the timed etch.

[0064] In Figure 2B (where the same reference numerals denote the same elements as in Figure 2A), a semiconductor processing system 400 is shown that differs from the semiconductor processing system 300 of Figure 2A, except that an optical fiber 235 is used to feed a diffracted optical signal 236 into an optical detector 240. The use of the optical fiber 235 provides additional flexibility in positioning the optical detector 240. The optical fiber 235 couples the diffracted optical signal 236 and transmits it to a collecting optics 245, and may include coupling optics, multiple fiber bundles, etc., to ensure proper coupling of the diffracted optical signal 236 and optimal transmission along the optical fiber. In all other respects, the semiconductor processing system 400 functions and operates similarly to or equivalent to the semiconductor processing system 300 of Figure 2A as described above.

[0065] Figures 3A and 3B illustrate a semiconductor processing system according to a further embodiment of the present invention. Figure 3A depicts an exemplary semiconductor processing system 500. The semiconductor processing system 500 may or may not utilize plasma for processing and may include or be part of a photolithographic track system, etching system, atomic layer etching (ALE) system, deposition system, wet or dry cleaning system, diffusion furnace system, or any other semiconductor processing system, wherein the progress of the process steps needs to be monitored to ensure that the substrate meets specifications and for the purpose of process step control. The semiconductor processing system 500 includes a process module 310 having a support member 315 for supporting a substrate 320 during the process steps. The support member 315 may be part of an electrostatic chuck (ESC), a base, a movable stage, a platform / substrate arm, etc. The process steps are performed on the substrate 320 within the process module 310, during which layers or features may be formed or modified on the substrate 320. For example, in an etching system, layers and features on the substrate may be etched. In photolithography systems, photoresist coatings can be applied, baked, or removed. In deposition systems, layers can be deposited on substrates. In wet or dry cleaning systems, features, layers, processing solutions, contaminants, etc., can be removed from substrates.

[0066] During the process steps, the optical detector 340 uses reflection to monitor the state of the substrate within the semiconductor processing system 500 and the process module 310. The illumination beam 334 is generated by a broadband illumination light source 332 and guided to the substrate 320 through the illumination window 330 in an oblique incidence manner. The incident angle θi of the illumination beam 334 can vary from 0.1° to 89°, and the incident angle is selected based on several factors, including the sensitivity of the reflectometer to the type and geometry of features and structures formed on the substrate 320, and the constraints imposed by the geometry of the process module 310. Some process modules 310 may not have walls, meaning they may not be enclosed or isolated from their surrounding environment; in such cases, the illumination window 330 may not be necessary. The illumination light source can be a continuous wave (CW) or pulsed light source, and it can cover the wavelength range of the deep ultraviolet (DUV), ultraviolet (UV), visible light (VIS), and infrared (IR) portions of the optical spectrum. The illumination source 332 may include various incandescent and gas discharge light sources, flash lamps, LEDs, lasers, laser-driven plasma light sources (LDLS), etc. If a particularly wide range of light wavelengths is used for the reflection method, the illumination source 332 may include multiple light sources, each covering a different range of light wavelengths, and suitable beam combiner optics are used to combine the beams. The illumination source 332 may also include a shutter (not shown) to adjust the illumination beam 334 so that interference from plasma or other captured radiation can be subtracted from the measured reflectometer signal when the shutter is closed. The illumination window 330 may be made of a suitable material that is transparent to the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. For example, the illumination window 330 may include various types of glass, such as borosilicate glass, quartz, sapphire, etc. Facilities may be provided to protect the irradiation window 330 from excessive deposition of materials from plasma or other corrosive environments within the process module 310, such as perforated grids, flushing gas injection systems, etc. (not shown).

[0067] When incident obliquely onto the substrate 320, the illumination beam 334 is reflected and diffracted due to the presence of layers, structures, or features formed on the substrate 320, to form a diffracted light signal 336. The diffracted light signal 336 is reflected from the substrate 320 at a reflection angle θr (which is equal to the incident angle θi of the illumination beam 334 and can vary from 0.1° to 89°). The diffracted light signal 336 may include light signals of one or more diffraction orders reflected from the substrate. As discussed previously, the spectrum of the diffracted light signal 336 contains information from which the characteristics of layers, structures, or features on the substrate 320 can be determined. The diffracted light signal 336 passes through a reflection window 338 (if present) and is introduced into the optical detector 340. The reflection window 338 is generally located on one side of the process module 310 relative to the illumination window 330, and may include the same or different material as the illumination window 330. Facilities may be provided to protect the reflective window 338 from excessive deposition of materials from plasma or other corrosive environments within the process module 310, such as perforated grilles, flushing gas injection systems, etc. (not shown).

[0068] The optical detector 340 includes collecting optics 345, a wavelength-tunable filter 350, and an array detector 360, which together constitute a hyperspectral imaging (HSI) system. The optical detector and its components are controlled by a controller 380 that communicates with at least the wavelength-tunable filter 350 and the array detector 360. The wavelength-tunable filter may include a Fabry-Perot resonator, an acousto-optic tunable filter, a liquid crystal tunable filter, or any other tunable narrowband optical filter device capable of changing the passband wavelength, as selected and controlled by the controller 380. Depending on the application, the array detector 360 may be a one-dimensional array detector (i.e., a line detector) or a two-dimensional array detector. If required by the application, the array detector 360 may include a CCD detector array, a CMOS detector array, a photodiode array, or a combination of various types of detectors for different parts of the array detector 360. For example, due to their inherent differences in wavelength sensitivity, different array detector types can be used side-by-side or stacked for different wavelength ranges.

[0069] In operation, the diffracted optical signal 336 is collected by the collecting optics 345 and introduced into the wavelength-tunable filter 350 as the transmitted optical signal 372. The collecting optics 345 may include any combination of collecting lenses, deflectors, apertures, polarizers, optical fibers, etc., required to form the transmitted optical signal 372. Figure 3A shows an example of a Fabry-Perot resonator, in which two semi-reflective mirrors 352 are separated by a controllable gap Gi to form an optical cavity 355. When the gap Gi is varied by the controller 380, the Fabry-Perot resonator selectively transmits different passband wavelengths to form a wavelength-filtered diffracted optical signal 375, which is incident on the array detector 360.

[0070] In operation, the controller 380 controls the controllable gap Gi to perform a scan of the passband wavelength, while the array detector 360 simultaneously captures images subsequently received by the controller 380, which can be stored and / or processed in the controller 380. Each image captured by the array detector 360 is captured at the instantaneous passband wavelength selected by the wavelength-tunable filter 350 as the passband wavelength of the wavelength-tunable filter 350 changes. In this way, the hereinafter “image stack” or “image cube” refers to each scan of the passband wavelength of the wavelength-tunable filter 350, which is then received by the controller 380 (see Figures 5A-C). Passband wavelength scans can be repeated to capture consecutive image cubes. The controller 380 can be programmed, for example, to perform continuous scans of the passband wavelengths of the wavelength-tunable filter 350 (from minimum to maximum passband wavelength). However, in reality, only certain portions of the optical spectrum may be of interest. Therefore, the controller 380 can be programmed to selectively adjust the passband wavelength within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. In this way, the output is improved because only the wavelength range of interest is captured.

[0071] For a Fabry-Perot resonator, the typical time for switching and setting the passband wavelength is around 0.1 ms, thus allowing, for example, the selection of 100 passband wavelengths within 10 ms and the formation of a corresponding image by the array detector 360. Although this spectral acquisition speed may be slower than some high-performance spectrometers, the advantage of a hyperspectral imaging (HSI) system is that the amount of light received through the diffracted light signal 336 is not limited by the slit width of the spectrometer or the numerical aperture (NA) of the collecting optics. Furthermore, a large effective aperture can be used in the collecting optics 345, resulting in the array detector 360 detecting a strong signal, thus significantly improving the signal-to-noise ratio (SNR) compared to conventional spectrometers. Due to the typical 2 nm spectral resolution of a Fabry-Perot resonator, a single optical detector 340 employing a hyperspectral imaging (HSI) system combines high SNR and good spectral resolution. If the output of a single optical detector 240 is insufficient for the application, multiple optical detectors 340 operating in parallel can be used, each capturing a portion of the diffracted light signal 336 (which can be pre-splitted using a suitable beam splitter or other wavelength-selective optics).

[0072] If a higher spectral resolution (higher than that obtainable through a wavelength-tunable filter 350 of a selected type) is required, a spectral analysis method can be used to reconstruct a high-resolution spectrum from the captured diffraction spectrum 950 (Figures 5A, 5B, and 5C). An example spectral analysis method is disclosed in M. Morháč's "Deconvolution methods and their applications in the analysis of gamma-ray spectra" (ACAT2005, May 22-27, Zeuthen, Germany). Alternatively, machine learning techniques disclosed in co-pending U.S. Patent Application No. 16 / 820,032 (titled "ENHANCED RESOLUTION IN SEMICONDUCTOR FABRICATION DATA ACQUISITION INSTRUMENTS USING MACHINE LEARNING" (reference number 181077US02, filed March 16, 2020) can be used to reconstruct a higher-resolution spectrum from the captured diffraction spectrum 950.

[0073] In Figure 3B (where the same reference numerals denote the same elements as in Figure 3A), a semiconductor processing system 600 is shown that differs from the semiconductor processing system 500 of Figure 3A in that it uses an optical fiber 335 to feed a diffracted optical signal 336 into an optical detector 340. The use of the optical fiber 335 provides additional flexibility in positioning the optical detector 340. The optical fiber 335 couples the diffracted optical signal 336 and transmits it to a collecting optics 345, and may include coupling optics, multiple fiber bundles, etc., to ensure proper coupling of the diffracted optical signal 336 and optimal transmission along the optical fiber. In all other respects, the semiconductor processing system 600 functions and operates similarly to or equivalent to the semiconductor processing system 500 of Figure 3A as described above.

[0074] Semiconductor processing systems 300, 400, 500, and 600 all utilize reflection methods to diagnose process results on a substrate, and their differences include details of optical setup and incident and reflection angles. Therefore, the foregoing discussion of the methods used by semiconductor processing systems 300 and 400 for extracting diffraction spectra 950 (Figures 5A, 5B, and 5C), establishing the spatial correspondence between pixel positions in the image of the image cube and their positions on the substrate, selecting pixel regions 940 (Figures 5A, 5B, and 5C) to extract diffraction spectra 950, the characteristics of determinable layers and features, and their associated advantages are also applied to semiconductor processing systems 500 and 600 using the oblique incident angle reflection method and the associated optical detector 340 and controller 380.

[0075] In addition to monitoring the plasma state in the plasma processing chamber and using reflection methods to monitor the processing of the upper layers and features of the substrate (during the process, i.e., in situ and after the process is completed), hyperspectral imaging (HSI) systems can be used to enhance other optical diagnostic methods with spectral and / or imaging properties. For example, hyperspectral imaging (HSI) can be used for multispectral inspection of substrates. Applications of multispectral inspection of substrates can include common inspection steps, such as post-development inspection (ADI) and CD measurement / inspection in photolithography and etching, as well as many defect detection techniques, such as the detection of contamination, pattern collapse, improper self-assembly of DSA block copolymer layers, etc. What all these techniques have in common is the image acquisition of the substrate or substrate area, which can be performed at many wavelengths using a hyperspectral imaging (HSI) system, thereby greatly increasing the amount of data, which can be used to detect defects in low signal-to-noise ratio conditions.

[0076] Figures 4A and 4B illustrate a substrate inspection system according to a further embodiment of the present invention. Figure 4A depicts an exemplary semiconductor substrate inspection system 700. The substrate inspection system 700 may be part of a semiconductor processing system, or it may be part of a stand-alone inspection system (e.g., a kiosk-type inspection system).

[0077] The substrate inspection system 700 includes a platform 715 for supporting the substrate 720 during inspection. The platform 715 may be an electrostatic chuck (ESC), a base, a movable platform (e.g., an XY, XY-θ, XYZ, or XYZ-θ platform), a platform / substrate arm, etc. The substrate inspection system 700 may include other subsystems, such as a housing (not shown), a flushing gas system (not shown) for maintaining a controlled and clean environment around the substrate 720, etc.

[0078] During the detection step or process, the optical detector 740 is used to image the substrate within the substrate detection system 700. An illumination source (not shown) can be used to illuminate the surface of the substrate 720. The illumination source can be configured for bright-field illumination (from above the substrate 720) or for dark-field illumination (the illumination beam is substantially parallel to the surface of the substrate 720). The illumination source can be a continuous wave (CW) or pulsed light source, and it can cover the wavelength range of the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. The illumination source can include various incandescent and gas discharge light sources, flash lamps, LEDs, lasers, laser-driven plasma light sources (LDLS), etc., and can include other optics required for uniformly illuminating the surface of the substrate 720. If a particularly wide range of light wavelengths is used for detection, the illumination source can include multiple light sources, each covering a different range of light wavelengths, and a suitable beam combiner optics is used to combine the beams. The illumination source may also include a shutter (not shown) to adjust the illumination beam so that interference from the captured radiation can be subtracted from the measured light signal when the shutter is closed.

[0079] When incident on the substrate 720, a portion of the illumination beam is reflected from the substrate 720 to form an optical signal 770. As previously discussed, the spectrum of the optical signal 770 contains information from which the characteristics of a layer, structure, or feature on the substrate 720 can be determined during detection. The optical detector 240 includes collecting optics 748, a wavelength-tunable filter 750, and an array detector 760, which together constitute a hyperspectral imaging (HSI) system. The optical detector and its components are controlled by a controller 780 that communicates with at least the wavelength-tunable filter 750 and the array detector 760. The wavelength-tunable filter may include a Fabry-Perot resonator, an acousto-optic tunable filter, a liquid crystal tunable filter, or any other tunable narrowband optical filter device capable of changing the passband wavelength, as selected and controlled by the controller 780. Depending on the application, the array detector 760 may be a one-dimensional array detector (i.e., a line detector) or a two-dimensional array detector. If required by the application, the array detector 760 may include a CCD detector array, a CMOS detector array, a photodiode array, or a combination of various types of detectors used for different parts of the array detector 760. For example, due to their inherent differences in wavelength sensitivity, different array detector types may be used side by side or stacked for different wavelength ranges.

[0080] In operation, the optical signal 770 is collected by the collecting optics 748 and introduced into the wavelength-tunable filter 750 as the transmitted optical signal 772. The collecting optics 748 may include any combination of collecting lenses, deflectors, apertures, polarizers, optical fibers, etc., required to form the transmitted optical signal 772. Figure 4A shows an example of a Fabry-Perot resonator, in which two semi-reflective mirrors 752 are separated by a controllable gap Gi to form an optical cavity 755. When the gap Gi is varied by the controller 780, the Fabry-Perot resonator selectively transmits different passband wavelengths to form a wavelength-filtered optical signal 775, which is incident on the array detector 760.

[0081] In operation, the controller 780 controls the controllable gap Gi to perform a scan of the passband wavelength, while the array detector 760 simultaneously captures images subsequently received by the controller 780, wherein the images can be stored and / or processed in the controller 780. Each image captured by the array detector 760 is captured at the instantaneous passband wavelength selected by the wavelength-tunable filter 750 as the passband wavelength of the wavelength-tunable filter 750 changes. In this way, the hereinafter “image stack” or “image cube” refers to each scan of the passband wavelength of the wavelength-tunable filter 750, which is then received by the controller 780 (see Figures 5A-C). Passband wavelength scans can be repeated to capture consecutive image cubes. The controller 780 can be programmed, for example, to perform a continuous scan of the passband wavelength (from minimum to maximum passband wavelength) of the wavelength-tunable filter 150. However, in reality, only certain portions of the optical spectrum may be of interest. Therefore, the controller 780 can be programmed to selectively adjust the passband wavelength within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum. In this way, the output is improved because only the wavelength range of interest is captured.

[0082] For a Fabry-Perot resonator, the typical time for switching and setting the passband wavelength is around 0.1 ms, thus allowing, for example, the selection of 100 passband wavelengths within 10 ms and the formation of a corresponding image by the array detector 760. Due to the typical 2 nm spectral resolution of the Fabry-Perot resonator, a single optical detector 740 employing a hyperspectral imaging (HSI) system combines high signal-to-noise ratio (SNR) and good spectral resolution. If the output or passband wavelength range of a single optical detector 740 is insufficient for the application, multiple optical detectors 740 operating in parallel can be used, each capturing a portion of the optical signal 770 (which can be pre-splittered using a suitable beam splitter or other wavelength-selective optics).

[0083] Multispectral substrate detection relies on the acquisition of optical spectra, as previously described. Figures 5A-C illustrate the process of extracting optical spectra from exemplary image cubes 900, 902, and 905. In Figure 5A, image cube 900 is captured at n passband wavelengths, varying from λ1 through λi to λn. In this simplest embodiment, the diffraction spectrum is extracted by extracting light intensity from the same single selected pixel 940 in all images of image cube 900. When the extracted light intensities are arranged in ascending wavelength order, the optical spectrum 950 of Figure 5A is obtained.

[0084] Alternatively, in FIG. 5B, the optical spectrum is extracted by extracting and averaging the light intensity within at least one selected pixel region 940 of all images in the image cube 902. Multiple discontinuous pixel regions 940 may be selected and averaged, depending on the application, which specifies which parts of the image are retained and which are discarded. For each image captured at each wavelength λi, various pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within the selected pixel region 940. Once averaged, the averaged extracted light intensities are arranged in ascending wavelength order to form the optical spectrum 950 of FIG. 5B.

[0085] Alternatively, in FIG. 5C, the diffraction spectrum is extracted by extracting and averaging the light intensity within all pixels 940 of the captured image of the image cube 905. For each image captured at each wavelength λi, various pixel-weighted and unweighted averaging methods can be used to obtain the average light intensity within pixel 940. Once averaged, the average extracted light intensities are arranged in ascending wavelength order to form the optical spectrum 950 of FIG. 5C. In the case of multispectral substrate detection, the methods of FIG. 5A and 5B are most suitable because they retain the most image information: the method of FIG. 5A retains all pixels, while the method of FIG. 5B retains pixel regions.

[0086] When forming the optical spectrum 950 as described in Figures 5A-C, all images in image cubes 900, 902, or 905 can be used to form the optical spectrum 950. Alternatively, only a subset of the images in image cubes 900, 902, or 905 can be used to form the optical spectrum 950, thus removing spectral portions not needed for a given application. This approach of first reducing the number of images captured in the image cubes (i.e., the number of passband wavelengths) and then further selecting only a subset of the images in the image cubes for optical spectrum extraction can significantly improve the computational efficiency of the detection algorithm.

[0087] In multispectral substrate inspection, due to the configuration and position of the collecting optics 748 relative to the substrate 720, the pixel positions in the two-dimensional images of image cubes 900, 902, and 905 in Figures 5A-C correspond to different positions on the substrate 720, which are imaged onto the individual pixels of the array detector 760. This spatial correspondence between pixel positions and positions on the substrate 720 can be determined through optical modeling or by using captured images to verify the field of view of the image on the substrate 720. Utilizing the known spatial correspondence, one of the methods in Figures 5A and 5B can be advantageously used to accurately select the area on the substrate 720 to be diagnosed. This allows for the removal of a portion of the substrate surface if these portions do not have a meaningful impact on the detection results.

[0088] Another advantage of being able to select only certain pixel regions 940 in the image of the image cube is that it allows for a relatively large field of view to be used on the optical detector 740 on the substrate 720. This simplifies the optical setup, relaxes dimensional tolerances, and reduces the cost of the optical detector 740 and the entire substrate inspection system 700. Finally, it relaxes the positioning tolerances of the substrate 720 on the platform 715, and if the platform 715 is movable, it relaxes its positioning tolerances because the desired pixel region 940 can always be found in the image of the image cube, as long as it is within the field of view of the optical detector 740.

[0089] If a higher spectral resolution (higher than that obtainable through a wavelength-tunable filter 750 of a selected type) is required, a spectral analysis method can be used to reconstruct a high-resolution spectrum from the captured optical spectrum 950. An example spectral analysis method is disclosed in M. Morháč's "Deconvolution methods and their applications in the analysis of gamma-ray spectra" (ACAT2005, May 22-27, Zeuthen, Germany). Alternatively, machine learning techniques disclosed in co-pending U.S. Patent Application No. 16 / 820,032 (titled "ENHANCED RESOLUTION IN SEMICONDUCTOR FABRICATION DATA ACQUISITION INSTRUMENTS USING MACHINE LEARNING" (reference number 181077US02, filed March 16, 2020) can be used to reconstruct a higher-resolution spectrum from the captured optical spectrum 950.

[0090] In Figure 4B (where the same reference numerals denote the same elements as in Figure 4A), a substrate inspection system 800 is shown that differs from the substrate inspection system 700 of Figure 4A in that it uses an optical fiber bundle 735 to feed an optical signal 770 into an optical detector 740. The use of the optical fiber bundle 735 provides additional flexibility in positioning the optical detector 740. The optical fiber bundle 735 couples the diffracted optical signal 770 and transmits it to the collecting optics 745 of the optical detector 740, and may include an objective lens 748, coupling optics, multiple optical fiber bundles, etc., to ensure proper coupling of the diffracted optical signal 770 and optimal transmission along the optical fiber. In all other respects, the substrate inspection system 800 functions and operates similarly to or equivalent to the substrate inspection system 700 of Figure 4A as described above.

[0091] In FIG. 6A (where the same reference numerals denote the same elements as in FIG. 2B), a semiconductor processing system 1400 according to a further embodiment of the present invention is shown, wherein varying the wavelength of the illumination beam replaces wavelength selection applied before the array detector. The semiconductor processing system 1400 differs from the semiconductor processing system 400 in that a wavelength-tunable filter 250 is disposed in the illumination beam 234 between the illumination light source 232 and the semi-reflector 233. In this embodiment, a variation in the passband wavelength of the wavelength-tunable optical filter 250 results in a variation in the wavelength of the illumination beam 234. Therefore, as the passband wavelength of the wavelength-tunable filter 250 varies, each image captured by the array detector 260 will be captured at the instantaneous passband wavelength selected by the wavelength-tunable filter 250. As in the semiconductor processing system 400, the controller 280 of the semiconductor processing system 1400 controls the controllable gap Gi to perform passband wavelength scanning, and similarly, the captured images are received by the controller 280. In all other respects, the operation of semiconductor processing system 1400 is identical to that of semiconductor processing system 400 previously described, and the captured “image stack” or “image cube” is the same. The same modification can also be made to the position of the wavelength-tunable filter 250 in semiconductor processing system 300 of FIG. 2A.

[0092] In FIG. 6B (where the same reference numerals denote the same elements as in FIG. 3B), a semiconductor processing system 1600 according to a further embodiment of the present invention is shown, wherein varying the wavelength of the illumination beam is used instead of applying wavelength selection before the array detector. The semiconductor processing system 1600 differs from the semiconductor processing system 600 in that a wavelength-tunable filter 350 is disposed in the illumination beam 334 between the illumination light source 332 and the illumination window 330. In this embodiment, a change in the passband wavelength of the wavelength-tunable optical filter 350 causes a change in the wavelength of the illumination beam 334. Therefore, as the passband wavelength of the wavelength-tunable filter 350 changes, each image captured by the array detector 360 will be captured at the instantaneous passband wavelength selected by the wavelength-tunable filter 350. As in the semiconductor processing system 600, the controller 380 of the semiconductor processing system 1600 controls the controllable gap Gi to perform passband wavelength scanning, and similarly, the captured images are received by the controller 380. In all other respects, the operation of semiconductor processing system 1600 is identical to that of semiconductor processing system 600 previously described, and the captured “image stack” or “image cube” is the same. The same modification can also be made to the position of the wavelength-tunable filter 350 in semiconductor processing system 500 of Figure 3A.

[0093] Both semiconductor processing systems 1400 and 1600 utilize reflection methods to diagnose the process results on the substrate. Therefore, the foregoing discussion of the methods used in semiconductor processing systems 300, 400, 500, and 600 for extracting diffraction spectra 950 (Figs. 5A, 5B, and 5C), establishing the spatial correspondence between pixel positions in the image of the image cube and their positions on the substrate, selecting pixel regions 940 (Figs. 5A, 5B, and 5C) to extract diffraction spectra 950, the characteristics of determinable layers and features, and their associated advantages are also applied to semiconductor processing systems 1400 and 1600.

[0094] The foregoing discussion focuses on the use of Fabry-Perot resonators for hyperspectral imaging (HSI) because they offer good spectral resolution in the 2 nm range and can improve the signal-to-noise ratio (SNR) without sacrificing output. With these characteristics, Fabry-Perot resonators can be used as an alternative to fast, compact spectrometers in many semiconductor diagnostic applications. Other filters (such as acousto-optic tunable filters and liquid crystal tunable filters) can also be used in some applications, but their performance varies, particularly in spectral resolution and passband wavelength switching speed. Thin-layer bandpass filter banks may also be used in hyperspectral imaging (HSI), as in some remote sensing applications (such as military applications and agricultural and land management applications). These filters have spectral resolutions in the 1 to 10 nm range, which may be well-suited for some less critical applications where sampling many wavelengths or closely spaced wavelengths is not required.

[0095] The lenses used in the collecting optics 145, 245, 345, 745 and objective lens 748 in the foregoing embodiments can be telecentric or double telecentric (i.e., telecentric in both object space and image space) to ensure the parallelism of light rays, a fixed image magnification regardless of depth, or both. In the embodiments of Figures 1A and 1B, the double telecentric lens provides a better balance of plasma light emission (sampling across the entire field of view and depth) in the plasma processing chamber. In the embodiments of Figures 2A, 2B, 3A, 3B, 4A, 4B, 6A, and 6B, the double telecentric lens ensures equal magnification and equal contribution to different portions of the captured image at different positions and depths.

[0096] Those skilled in the art will appreciate that many modifications and variations are possible in light of the foregoing teachings. Those skilled in the art will recognize numerous equivalent combinations and substitutions for the various components shown in the figures. Therefore, it is intended that the scope of the invention is not limited by this detailed description, but rather by the appended claims. [Simplified Explanation of the Diagram]

[0020] A more complete understanding of the invention and its many incidental advantages will become apparent when the following detailed description is taken, especially in conjunction with the accompanying drawings.

[0021] Figure 1A is a schematic diagram of a plasma processing system with an optical detector according to an embodiment of the present invention.

[0022] Figure 1B is a schematic diagram of a plasma processing system with an optical detector according to another embodiment of the present invention.

[0023] FIG2A is a schematic diagram of a semiconductor processing system having an optical detector (configured for normal incidence) according to an embodiment of the present invention.

[0024] Figure 2B is a schematic diagram of a semiconductor processing system having an optical detector (configured for normal incidence) according to another embodiment of the present invention.

[0025] Figure 3A is a schematic diagram of a semiconductor processing system having an optical detector (configured for oblique incidence) according to an embodiment of the present invention.

[0026] Figure 3B is a schematic diagram of a semiconductor processing system with an optical detector (configured for oblique incidence) according to another embodiment of the present invention.

[0027] Figure 4A is a schematic diagram of a substrate detection system with an optical detector according to an embodiment of the present invention.

[0028] Figure 4B is a schematic diagram of a substrate detection system with an optical detector according to another embodiment of the present invention.

[0029] Figures 5A-C are schematic diagrams of the steps for extracting a spectrum from an image cube according to an embodiment of the present invention.

[0030] Figure 6A is a schematic diagram of a semiconductor processing system with an optical detector (configured for normal incidence) according to a further embodiment of the present invention.

[0031] Figure 6B is a schematic diagram of a semiconductor processing system with an optical detector (configured for oblique incidence) according to a further embodiment of the present invention.

Claims

1. A method for diagnosing plasma process steps in a plasma processing system, comprising: The plasma is ignited in one of the plasma processing chambers of the plasma processing system; Plasma light emission signals are collected through a window installed in one wall of the plasma processing chamber and through collecting optics; the plasma light emission signals are guided from the collecting optics to a wavelength-tunable filter to tunably select the wavelength of the collected plasma light emission signals; an array detector is used to detect the wavelength-filtered plasma light emission signals; the images captured by the array detector are stored and processed in a controller; the passband wavelength of the wavelength-tunable filter is selectively adjusted within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum; and at least one image cube is formed, which is composed of a set of images detected by the array detector, each image in the set being captured at a wavelength defined by the instantaneous passband wavelength to which the wavelength-tunable filter is adjusted, wherein the controller is configured to control at least the wavelength-tunable filter and the array detector.

2. The method as described in claim 1, further comprising: The endpoint of the plasma process step is determined by a series of plasma emission spectra formed from a series of image cubes captured during the plasma process step.

3. The method as described in claim 1, further comprising: The relative concentration of at least one chemical component of the plasma is determined from at least one plasma emission spectrum formed by the at least one image cube.

4. A method for measuring at least one characteristic of at least one layer or structure formed on a substrate disposed in a semiconductor processing system, comprising: A process step begins in this semiconductor processing system; A substrate is irradiated with an illumination beam, and a portion of the illumination beam reflected and diffracted from the substrate forms a diffracted light signal; the diffracted light signal is collected through a window installed in one wall of the semiconductor processing system and through collecting optics; a wavelength-tunable filter disposed in the illumination beam or the diffracted light signal is used to adjustably select the wavelength of the illumination beam or the diffracted light signal to form a wavelength-filtered diffracted light signal; an array detector is used to detect the wavelength-filtered diffracted light signal; the image captured by the array detector is stored and processed in a controller; In one or more overlapping or non-overlapping wavelength ranges of the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum, the passband wavelength of the wavelength-tunable filter is selectively adjusted; and at least one image cube is formed, the at least one image cube being composed of a set of images detected by the array detector, each image in the set being captured at a wavelength defined by the instantaneous passband wavelength to which the wavelength-tunable filter is adjusted, wherein the controller is configured to at least control the wavelength-tunable filter and the array detector.

5. The method as described in claim 4, further comprising: At least one characteristic of a layer or structure formed on the substrate is determined from at least one diffraction spectrum obtained from at least one image cube.

6. The method as claimed in claim 5, wherein the at least one characteristic of a layer or structure on the substrate is a top profile critical dimension (CD), a bottom profile critical dimension (CD), a middle profile critical dimension (CD), a profile sidewall angle, a layer thickness, a layer optical property, or the remaining thickness of the etched layer.

7. The method as described in claim 4 further includes: The endpoint of the process step is determined by a series of diffraction spectra obtained from a series of image cubes captured during the process step.

8. The method as described in claim 4, further comprising: Determine the spatial correspondence between the position on the substrate and the pixel position in the image of at least one selected image cube; And obtain at least one diffraction spectrum from the at least one selected image cube.

9. The method as described in claim 8, wherein the at least one diffraction spectrum is obtained from a single selected pixel location within all images or a selected subset of images in the at least one selected image cube.

10. The method as described in claim 8, wherein the at least one diffraction spectrum is obtained from at least one selected pixel location region within all images or a selected subset of one of the images in the at least one selected image cube.

11. The method as described in claim 8, wherein the at least one diffraction spectrum is obtained from all pixel locations within all images or a selected subset of images in the at least one selected image cube.

12. The method of claim 4, wherein at least two diffraction spectra are obtained from a single pixel location within a selected subset of all images or one of the images in at least two image cubes, and wherein the single pixel location is selected such that the rate of change of at least one light intensity between the two obtained diffraction spectra exceeds a pre-selected rate of change of light intensity threshold.

13. The method of claim 4, wherein at least two diffraction spectra are obtained from pixel location regions within all images or a selected subset of one of the images in at least two image cubes, and wherein the pixel location regions are selected such that the rate of change of at least one light intensity between the two obtained diffraction spectra exceeds a pre-selected rate of change of light intensity threshold.

14. A method for inspecting a substrate in a substrate inspection system, comprising: The substrate is placed on one of the stands in the substrate inspection system; the substrate is then illuminated with an illumination light source. The light signal caused by irradiation of the substrate is collected by a collecting optics; the collected light signal is guided from the collecting optics to a wavelength-tunable filter to adjust the wavelength of the collected light signal; and the wavelength-filtered signal is detected using an array detector. The controller stores and processes images captured by the array detector; selectively adjusts the passband wavelength of the wavelength-tunable filter within one or more overlapping or non-overlapping wavelength ranges in the deep ultraviolet (DUV), ultraviolet (UV), visible (VIS), and infrared (IR) portions of the optical spectrum; and forms at least one image cube composed of a set of images detected by the array detector, each image in the set being captured at a wavelength defined by the instantaneous passband wavelength to which the wavelength-tunable filter is adjusted, wherein the controller is configured to control at least the wavelength-tunable filter and the array detector.

15. The method as described in claim 14, further comprising: At least one characteristic of the substrate or a layer or structure formed on the substrate is determined from at least one optical spectrum obtained from the at least one image cube.

16. The method as described in claim 14, further comprising: Determine the spatial correspondence between the position on the substrate and the pixel position in the image of at least one selected image cube; And determine at least one optical spectrum from the at least one selected image cube.

17. The method as described in claim 16, wherein the at least one optical spectral system is obtained from a single selected pixel location within all images or a selected subset of images in the at least one selected image cube.

18. The method as described in claim 16, wherein the at least one optical spectral system is obtained from at least a selected pixel location region within all images or a selected subset of images in the at least one selected image cube.

19. The method as described in claim 16, wherein the at least one optical spectral system is obtained from all pixel locations within all images or a selected subset of images in the at least one selected image cube.

Images