KR20240018583A - Time-domain optical metrology and inspection of semiconductor devices - Google Patents

Abstract

Method for measuring semiconductor devices. The method includes generating a time-domain representation of wavelength-domain measurement data of light reflected by a three-dimensional (3D) pattern structure of a semiconductor device, one or more relevant peaks of the time-domain representation and It may include selecting at least one unrelated portion. One or more relevant peaks occur during one or more relevant time periods and are associated with corresponding relevant reference peaks associated with different versions of the reference 3D pattern structure. The similarity between different versions of the reference 3D pattern structure exceeds the difference between different versions of the reference 3D pattern structure. At least two related reference peaks corresponding to the same related peak among the one or more related peaks are different from each other. The method may further include determining at least one parameter of the 3D pattern structure based on the one or more relevant peaks and the corresponding relevant reference peaks.

Description

Time-domain optical metrology and inspection of semiconductor devices

cross-reference

This application claims priority on U.S. Provisional Patent Application No. 63/142,971, filed on January 28, 2021, and U.S. Provisional Patent Application No. 63/199,884, filed on January 29, 2021. Priority is claimed on PCT patent application PCT/IB2022/050774 filed on , all of which are incorporated herein by reference in their entirety.

This application claims priority from U.S. Provisional Patent Application No. 63/202,279, filed June 3, 2021, which is incorporated herein by reference in its entirety.

Semiconductor devices, such as logic and memory devices, are commonly manufactured by depositing a series of layers on a semiconductor wafer, where some or all of the layers include patterned structures. Optical scatterometry is the process of characterizing the properties of a semiconductor device by measuring the light reflected by the various layers of the device and then interpreting the measured light spectrum with respect to a predefined model or other reference data. is often used to Optical scatterometers are particularly suitable for use in semiconductor devices that have only periodic pattern structures, as is typically the case in memory devices. However, because some types of semiconductor devices have upper layers with periodic patterned structures, such as memory circuits, as well as lower layers with non-quadrature structures, such as logic circuits, the properties of these devices can be characterized using conventional optical scatterometry techniques. It may be difficult or impossible to do.

Systems, methods, and non-transitory computer readable media for time-domain optical metrology and inspection of semiconductor devices are provided.

Embodiments will be more fully understood and appreciated from the following detailed description taken in conjunction with the accompanying drawings:
1A-1D are simplified conceptual diagrams of a system for time-domain optical metrology and inspection of semiconductor devices constructed and operating in accordance with an embodiment of the present invention.
Figures 2A-2B are simplified graphical illustrations useful for understanding embodiments of the invention.
Figures 3A-3D are simplified flow diagrams of example methods of operation of the systems of Figures 1A-1D.
Figure 4A illustrates an example of the method.
Figure 4b illustrates an example of the method.
Figure 4C shows an example of the steps of the method of Figure 4A.
Figure 5 illustrates an example of the method.
Figure 6 shows an example of a pattern structure that irradiates and reflects radiation.
7 illustrates an example of a time-domain signal, which is a time-domain representation of wavelength measurement data.
Figure 8 is an example of parameter and time-domain representation.
Figure 9 illustrates an example of a time-domain representation.
Figure 10 illustrates an example of a method.

In one aspect, a method is provided for semiconductor device metrology, the method comprising: generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device, the latter comprising: Selecting an earlier-in-time portion of the time-domain representation, excluding the later-in-time portion, and performing model-based processing using the earlier-in-time portion of the time-domain representation. and determining one or more measurements of one or more parameters of interest of the pattern structure by doing so.

In another aspect, a predefined model is configured to determine a time-domain representation of theoretical wavelength-domain measurement data of light expected to be reflected by a patterned structure relative to corresponding theoretical measurements of the patterned structure.

In another aspect, the predefined model models one or more top layers of the pattern structure corresponding to the electrical portion in a time-domain representation.

In another aspect, the predefined model models one or more top layers of the pattern structure to the exclusion of all other layers of the pattern structure.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, where generating includes generating a time-domain representation using both the spectral amplitude and the spectral phase.

In one aspect, a method is provided for semiconductor device metrology, the method comprising: generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device, the latter comprising: Selecting an earlier-in-time portion of the time-domain representation, excluding a later-in-time portion, the selected electrical portion of the time-domain representation is converted into a time-filtered wavelength-domain portion. Converting to measurement data, and performing model-based processing using the time-filtered wavelength-domain measurement data to determine one or more measurements of one or more parameters of interest of the patterned structure.

In another aspect, a predefined model is configured to determine theoretical wavelength-area measurement data of light expected to be reflected by a patterned structure relative to corresponding theoretical measurements of the patterned structure.

In another aspect, the predefined model models one or more upper layers of the patterned structure corresponding to the time-filtered wavelength-domain measurement data.

In another aspect, the predefined model models one or more top layers of the pattern structure to the exclusion of all other layers of the pattern structure.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, where generating includes generating a time-domain representation using both the spectral amplitude and the spectral phase.

In another aspect, a method is provided for semiconductor device metrology, the method comprising generating a first time-domain representation of first wavelength-domain measurement data of light reflected by a first target location on a patterned structure of the semiconductor device. , generating a second time-domain representation of the second wavelength-domain measurement data of light reflected by the second target location on the pattern structure of the semiconductor device, corresponding to the height of the first target location in the first time-domain representation. identifying a first point, identifying a second point in the second time-domain representation corresponding to a height of a second target location, and a height between the height of the first target location and the height of the second target location. Includes the step of deciding on a car.

In another aspect, the first wavelength-domain measurement data includes a spectral amplitude and a spectral phase associated with a first target location, and the second wavelength-domain measurement data includes a spectral amplitude and a spectral phase associated with a second target location; , generating the first time-domain representation includes using both the spectral amplitude and the spectral phase of the first wavelength-domain measurement data to generate the first time-domain representation, and generating the second time-domain representation includes: and generating a second time-domain representation using both the spectral amplitude and the spectral phase of the second wavelength-domain measurement data.

In another aspect, a method is provided for semiconductor device inspection, the method comprising generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device, the time-domain representation being converted to a reference pattern. Comparing to a reference time-domain representation of the light reflected by the structure, and if a difference between the time-domain representations exists, identifying a structural abnormality in the semiconductor device.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, where generating includes generating a time-domain representation using both the spectral amplitude and the spectral phase.

In another aspect, a system for semiconductor device metrology is provided, the system generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device, wherein the later portion of the time-domain representation is a spectral processing unit configured to select an excluded electrical portion of the time-domain representation, and to perform model-based processing using the electrical portion of the time-domain representation to determine one or more measurements of one or more parameters of interest of a pattern structure; and a metrology unit configured, wherein the spectral processing unit and the metrology unit are implemented in either a) computer hardware, and b) computer software implemented in a non-transitory computer-readable medium.

In another aspect, a predefined model is configured to determine a time-domain representation of theoretical wavelength-domain measurement data of light expected to be reflected by a patterned structure relative to corresponding theoretical measurements of the patterned structure.

In another aspect, the predefined model models one or more top layers of the pattern structure corresponding to the electrical portion in a time-domain representation.

In another aspect, the predefined model models one or more top layers of the pattern structure to the exclusion of all other layers of the pattern structure.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, wherein the spectral processing unit is configured to generate a time-domain representation using both the spectral amplitude and the spectral phase.

In another aspect, a system for semiconductor device metrology is provided, the system generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device, wherein the later portion of the time-domain representation is a spectral processing unit configured to select excluded electrical portions of the time-domain representation and convert the selected electrical portions of the time-domain representation into time-filtered wavelength-domain measurement data, and the time-filtered wavelength-domain measurement data. a metrology unit configured to determine one or more measurements of one or more parameters of interest of the pattern structure by performing model-based processing using It is implemented in any of the computer software implemented in a medium.

In another aspect, a predefined model is configured to determine theoretical wavelength-area measurement data of light expected to be reflected by a patterned structure relative to corresponding theoretical measurements of the patterned structure.

In another aspect, the predefined model models one or more upper layers of the patterned structure corresponding to the time-filtered wavelength-domain measurement data.

In another aspect, the predefined model models one or more top layers of the pattern structure to the exclusion of all other layers of the pattern structure.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, wherein the spectral processing unit is configured to generate a time-domain representation using both the spectral amplitude and the spectral phase.

In another aspect, a system for semiconductor device metrology is provided, the system generating a first time-domain representation of first wavelength-domain measurement data of light reflected by a first target location on a patterned structure of the semiconductor device, the system comprising: a spectral processing unit configured to generate a second time-domain representation of second wavelength-domain measurement data of light reflected by a second target location on the pattern structure of the semiconductor device, and of the first target location in the first time-domain representation; Identify a first point corresponding to a height, identify a second point in the second time-domain representation corresponding to a height of a second target location, and identify a height between the height of the first target location and the height of the second target location. and a metrology unit configured to determine the difference, wherein the spectral processing unit and the metrology unit are implemented in either a) computer hardware, and b) computer software implemented in a non-transitory computer-readable medium.

In another aspect, the first wavelength-domain measurement data includes a spectral amplitude and a spectral phase associated with a first target location, and the second wavelength-domain measurement data includes a spectral amplitude and a spectral phase associated with a second target location; , the spectral processing unit is configured to generate a first time-domain representation using both the spectral amplitude and the spectral phase of the wavelength-domain measurement data associated with the first target location, wherein the spectral processing unit is configured to generate a first time-domain representation of the wavelength-domain measurement data associated with the second target location. and configured to generate a second time-domain representation using both the spectral amplitude and the spectral phase of the area measurement data.

In another aspect, a system for semiconductor device inspection is provided, the system comprising: a spectral processing unit configured to generate a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of the semiconductor device; and a structural anomaly detector configured to compare the representation to a reference time-domain representation of light reflected by the reference pattern structure and identify structural anomalies in the semiconductor device if a difference between the time-domain representations exists, the spectral processing unit and the structural anomaly detector is implemented in any of a) computer hardware, and b) computer software implemented in a non-transitory computer-readable medium.

In another aspect, the wavelength-domain measurement data includes spectral amplitude and spectral phase, wherein the spectral processing unit is configured to generate a time-domain representation using both the spectral amplitude and the spectral phase.

Reference is made to FIGS. 1A-1D which are simplified conceptual diagrams of a system for time-domain optical metrology and inspection of semiconductor devices constructed and operating in accordance with an embodiment of the present invention. In the system of FIG. 1A, an optical measurement tool 100, such as PRIZM®, commercially available from Nova Measurement Instruments, Ltd., Rehovot, Israel, or otherwise described in U.S. Pat. No. 10,161,885, is used. According to conventional techniques, the light reflected by the patterned structure 102 of the semiconductor device 104, such as on the semiconductor wafer 106, is measured, including both the spectral amplitude and the spectral phase of the reflected light. It is desirable to generate corresponding wavelength-domain measurement data 108.

Optical metrology tool 100 measures light reflected by pattern structure 102 at any selected point during or after fabrication of pattern structure 102.

An example of wavelength-domain measurement data 108 is shown in FIG. 2A , where a spectral reflectance graph 200 such as that of patterned structure 102 is shown. Also shown is a spectral reflectance graph 202 of the comparative pattern structures that serves as a standard against which the pattern structures 102 are compared. The comparison pattern structure may also be a “test” pattern structure 110 positioned on the semiconductor device 104, where spectral reflectance graph 202 is generated in the same manner as spectral reflectance graph 200. The graphs are virtually identical up to about 430 nm, but are quite different after that.

1A also shows a spectral processing unit 112, which is preferably integrated into the optical metrology tool 100. The spectral processing unit 112 preferably produces a time-domain representation ( 114).

FIG. 2B shows a time-domain representation 200' of the spectral reflectance graph 200 showing the time at which reflected light is received by the optical metrology tool 100 after illuminating the patterned structure 102. A time-domain representation 202' of the spectral reflectance graph 202 is also shown for comparison. Here the graph is substantially identical along the It indicates that the upper layer of and the upper layer of the test pattern structure 110 are also substantially identical.

The spectral processing unit 112 of FIG. 1A is configured to select the early portion 116 of the time-domain representation 114, preferably excluding the late portion of the time-domain representation 114. The selection may be displayed to the spectrum processing device 112 by a human operator, or the electrical portion 116 of the time-domain representation 114 - n may be any predefined value - containing only the first n femtoseconds of reflected light. ) can be automatically performed by the spectrum processing device 112 according to predefined criteria, such as selecting. Thus, for example, spectral processing unit 112 may select the early portion 204 of the time-domain representation 200' of FIG. 2B, excluding the late portion 206 of the time-domain representation 200'. You can.

Figure 1a also shows a metrology unit 118, which is preferably integrated into the optical metrology tool 100. In one embodiment, the metrology unit 118 performs model-based processing using selected electrical portions 116 of the time-domain representation 114 of the wavelength-domain measurement data 108 to determine the pattern structure 102 . and configured to determine one or more measurements of a parameter of interest (e.g., OCD, SWA, height, etc.). In this embodiment, the predefined model 120 determines a time-domain representation of the theoretical wavelength-domain measurement data of the light expected to be reflected by the patterned structure 102 with respect to the corresponding theoretical measurements of the patterned structure 102. It is configured to do so. Predefined model 120 preferably models one or more upper layers of pattern structure 102 corresponding to selected electrical portions 116 of time-domain representation 114, and predefined model 120 Preferably excluding all other layers of pattern structure 102. Model-based processing preferably uses model fitting techniques, such as those commonly used in the semiconductor metrology field, to determine a set of theoretical measurements of the patterned structure 102 using a predefined model 120, which Given a set of theoretical measurements, we will actually determine the measurements of the patterned structure 102 by deriving a model-based time-domain representation of the theoretical wavelength-domain measurement data of the light expected to be reflected by the patterned structure 102, and the model The underlying time-domain representation is identical within a predefined tolerance to the selected electrical portion 116 of the time-domain representation 114.

In another embodiment shown in FIG. 1B , spectral processing unit 112 converts selected electrical portions 116 of time-domain representation 114 into time-filtered wavelength-domain measurement data 122. Metrology unit 118 then performs model-based processing using time-filtered wavelength-domain measurement data 122 to determine one or more measurements of patterned structure 102.

In this embodiment, the predefined model 120 is configured to determine theoretical wavelength-area measurement data of light expected to be reflected by the patterned structure 102 relative to the corresponding theoretical measurements of the patterned structure 102 . The predefined model 120 preferably models one or more upper layers of the patterned structure 102 corresponding to the time-filtered wavelength-domain measurement data 122, and the predefined model 120 preferably models All other layers of pattern structure 102 are excluded.

In another embodiment shown in Figure 1C, as described above, the optical metrology tool 100 measures the light reflected by the first target location 124 on the pattern structure 102 and produces corresponding wavelength-domain measurement data ( 126). The optical metrology tool 100 is then used to measure the light reflected by the second target location 128 on the pattern structure 102 and generate corresponding wavelength-domain measurement data 130, as described above. .

The spectral processing unit 112 generates a first time-domain representation 132 of the first wavelength-domain measurement data 126 of the light reflected by the first target location 124 and a second target location 128. Generating a second time-domain representation 134 of the second wavelength-domain measurement data 130 of the reflected light. When the first target position 124 and the second target position 128 have different heights, the light reflected by them assumes that the positions of the reference mirrors are the same when measuring the two target positions 124 and 128. When doing so, it will appear at different points in the time-domain representation. The metrology unit 118 generates a first point corresponding to the height of a first target location 124 in the first time-domain representation 132 and a second target location 128 in the second time-domain representation 134. and is configured to identify a second point corresponding to the height of. The measurement unit 118 then determines the height difference between the height of the first target location and the height of the second target location, and this information can be used to control the CMP of the ONO step 208.

In another embodiment shown in Figure 1D, as described above, optical metrology tool 100 measures light reflected by pattern structure 102 of semiconductor device 104 and spectral processing unit 112 measures the time- It is used to generate corresponding wavelength-domain measurement data 108, which generates an area representation 114. A structural anomaly detector 136, preferably integrated into the optical metrology tool 100, compares the time-domain representation 114 to a reference time-domain representation 138, e.g., of light reflected by a reference pattern structure, It is configured to identify a structural anomaly, such as a void or other structural defect, in the semiconductor device 104 if a difference exists between the time-domain representations 114 and 138.

Reference is now made to Figure 3A, which is a simplified flow diagram of an exemplary method of operation of the system of Figure 1A operating in accordance with an embodiment of the present invention. In the method of Figure 3A, an optical metrology tool is used to measure the light reflected by the patterned structure of the semiconductor device and generate corresponding wavelength-domain measurement data comprising both the spectral amplitude and the spectral phase of the reflected light (step 300). A time-domain representation of the wavelength-domain measurement data is generated using both the spectral amplitude and the spectral phase of the wavelength-domain measurement data (step 302). The early portion of the time-domain representation is selected, excluding the late portion of the time-domain representation (step 304). Measurements of the pattern structure are determined by performing model-based processing using selected electrical portions of the time-domain representation (step 306).

Reference is now made to Figure 3B, which is a simplified flow diagram of an exemplary method of operation of the system of Figure 1B operating in accordance with an embodiment of the present invention. In the method of FIG. 3B, an optical metrology tool is used to measure the light reflected by the patterned structure of the semiconductor device and generate corresponding wavelength-domain measurement data comprising both the spectral amplitude and the spectral phase of the reflected light (step 310) . A time-domain representation of the wavelength-domain measurement data is created using both the spectral amplitude and the spectral phase of the wavelength-domain measurement data (step 312). The earlier portion of the time-domain representation is selected, excluding the later portion of the time-domain representation (step 314). The selected electrical portion of the time-domain representation is converted to time-filtered wavelength-domain measurement data (step 316).

Measurements of the pattern structure are determined by performing model-based processing using the time-filtered wavelength-domain measurement data (step 318).

Reference is now made to Figure 3C, which is a simplified flow diagram of an exemplary method of operation of the system of Figure 1C operating in accordance with an embodiment of the present invention. 3C, measuring light reflected by first and second target locations on the patterned structure of the semiconductor device and measuring corresponding first and second wavelength-domains including both spectral amplitude and spectral phase of the reflected light. An optical metrology tool is used to generate data (step 320). First and second time-domain representations for the first and second wavelength-domain measurement data are generated using both the spectral amplitude and the spectral phase of the wavelength-domain measurement data (step 322). A first point in the first time-domain representation and a second point in the second time-domain representation are identified corresponding to the heights of the first and second target locations (step 324). Thereafter, the height difference between the height of the first target location and the height of the second target location is determined (step 326).

Reference is now made to Figure 3D, which is a simplified flow diagram of an exemplary method of operation of the system of Figure 1D operating in accordance with an embodiment of the present invention. In the method of Figure 3D, an optical metrology tool is used to measure the light reflected by the patterned structure of the semiconductor device and generate corresponding wavelength-domain measurement data comprising both the spectral amplitude and the spectral phase of the reflected light (step 330). A time-domain representation of the wavelength-domain measurement data is created using both the spectral amplitude and the spectral phase of the wavelength-domain measurement data (step 332). The time-domain representation is compared to a reference time-domain representation (step 334). If differences between the time-domain representations exist (step 336), structural abnormalities in the semiconductor device are identified (step 338).

Optical critical dimension (OCD) metrology is a mainstream approach for dimensional characterization of semiconductor devices during the manufacturing process.

OCD is based on optical scatterometry - high-quality measurements of the optical reflection properties of measured patterns under various conditions (wavelength, polarization, angle of incidence, etc.) and interpretation of the reflectance information using advanced algorithms, modeling and machine learning methods.

The main challenge of using the OCD method is also one of its main advantages: the light penetrates and interacts deep with the structure being measured, providing sensitivity to dimensions and materials throughout the stack. Because modern SC structures are very thin, light typically penetrates deep into several layers of the patterned structure and is affected by the dimensions and material properties of all these regions. It is often very difficult to separate the sensitivity of the different parts of the structure being measured.

This sensitivity has several implications: (a) model setup complexity and solution implementation time, (b) sensitivity to various parameters and addressing parameter correlations, and (c) solution robustness, especially to sublayer variations. , (d) OCD for R&D environments providing resilience to design changes, (e) ability to use machine learning (ML) solutions, reducing reference count, (f) metrology for complex structures, especially in-die. Issues related to instrumentation, (g) a single recipe for different sites sharing a single common upper area, and (h) transfer from short loops to full loops.

Model setup complexity and time-to-solution (TTS).

a. The first step in building an OCD solution involves establishing the OCD ‘recipe’. During this step, a simulation representation of the structure being measured (or some simplified version close enough to correctly represent its scattering properties) is established. A fit/regression approach can then be used to interpret the measurements, with the simulated structure updated and improved based on measured information and reference properties until the simulation correctly represents the key properties of the stack being measured. Key attributes may be attributes of interest, for example, attributes that significantly affect the signal. The properties may be physical properties and/or geometric properties.

b. For complex, multi-layered structures, building and refining these simulation representations can take days, and in some cases weeks. Often the solutions obtained are inaccurate due to the extremely high complexity associated with these structures and require the use of simplifying assumptions. Often, highly skilled engineers are needed to converge this process into a useful solution.

Resolve sensitivity and parameter correlation for various parameters

a. Complex structures are typically described by numerous parameters, some of which can be difficult to resolve due to the sensitivity of the measured signal to parameter changes (weak parameters).

b. When evaluating multiple parameters, the spectral characteristics for changes in two or more parameters may be similar. In these cases, it is difficult to determine exact values for individual parameters (parameter correlations).

Solution robustness, especially solution robustness to sublayer variations

a. OCD solutions are invalid if used on structures that are significantly different from those used for development. Obviously, the addition of structural elements (layers, significant shape changes), material changes, and even significant dimensional changes render the developed simulation description or machine learning-based description of the stack invalid.

b. The wide sensitivity of OCD means that such changes in any part of the measured stack can result in detrimental degradation of measurement quality. In this respect, OCD solutions are very delicate and various defense mechanisms are commonly used to identify these departures from the 'verification' zone.

c. In some cases, the base layer (or any other layer) is aperiodic or does not respect other constraints necessary to have an OCD solution. In this case, even if the base layer (or other layer) does not contain the parameter of interest, the parameter of interest in the desired layer cannot be obtained.

OCD for R&D environments, providing resilience to design changes.

a. The R&D environment presents even more extreme situations that further exacerbate the previous challenges. Design changes frequently occur during manufacturing process development. Each of these changes typically renders any associated OCD solution inapplicable and requires the solution to be redeveloped (or at least adapted) to adapt to the new design.

b. As explained, even when design changes occur in buried sublayers, the metrology solutions for the layers above often need to be changed due to the noted wide range of sensitivities of these optical methods.

Reduce the number of references and ability to use machine learning (ML) solutions.

a. In recent years, ML solutions have played an increasingly important role in OCD metrology interpretation. In these methods, a set of OCD measurements is performed on samples with accurate reference data (obtained through other means, e.g., different metrology methods, such as TEM). The ML solution then uses this 'training' information to learn how to interpret measurements from similar structures (either using just the measurements and references or using additional information from a simulation tool).

b. Although incredibly powerful ML tools have been developed to date, they inherently require a significant amount of 'training' measurements and referencing. This extensive set is needed for ML solutions to learn how to decode and separate variations of the parameter of interest (POI) from variations of other parameters in the stack. If an ML solution is used on a stack that is significantly different from the one on which it was learned - for example, involving variations in parameters that did not change during the training session - it will generally fail.

c. Again, the wide sensitivity of OCD introduces significant complexity in that it requires a training dataset containing a very large number of samples covering a wide range of dimensional parameter variations. The ability to limit sensitivity to specific regions of the stack significantly reduces the number of reference data points. Moreover, this reduced sensitivity will greatly strengthen and stabilize ML tools because there will be no need to find mathematical filtering solutions to remove this sensitivity.

Metrology of complex structures, especially in-die metrology.

a. Typically, OCD solutions are used in dedicated areas, i.e. sacrificial areas of the wafer that are not used for final functions (typically the wafer 'scribe line'). This allows some simplification of the measured patterns, allowing for reliable OCD solutions and better TTS, and importantly, it allows control over which patterns are placed underneath the measured structures, as these patterns have no functional role and therefore the OCD It can be designed so that it does not conflict with or confuse measurements. Often a flat metal buffer layer, a simple unpatterned layer, or an underlayer containing a simplified pattern is used.

b. In recent years, it has become increasingly necessary to measure OCD at locations more representative of the actual process, especially inside the die, i.e. in the functional areas of the SC die, for actual patterns that are later used for functional operation. This trend has been driven by increasingly stringent process control requirements, requiring instrumentation closer to the device of interest. Unlike scribe line measurements, these areas cannot be simplified as they are inherently built into the underlying layers, with complexity dependent on the device design.

c. The sensitivity of OCD to these underlying structures is often too high to allow such in-die metrology operations. This is especially true when the underlying layer is aperiodic, in which case model-based OCD solutions are not possible.

A single recipe for different sites sharing a common top area.

a. As explained, during manufacturing the semiconductor stack becomes a multilayer structure with different functional elements placed on top of each other. This is particularly typical for multiple metal-interconnect levels, and is most noticeable in logic interconnects, where 14 or more levels are common.

b. Due to the described sensitivity to subfloor, multiple sites sharing the same top floor layout but with different underlying structures require separate dedicated OCD solutions. The ability to use one solution for these multiple sites is of great value as it provides both flexibility and generality to the OCD solution. The greatly reduced sensitivity to the lower layers (although this sensitivity is not completely eliminated) makes it easier to transfer recipes between different sites with minimal adjustments.

Transferring solution from short loop to full loop.

a. The term 'short-loop' relates to the fabrication of specific layers on a bare (or simple) substrate with the final product lying on top of the underlying layers. The use of short loop wafers allows for fast cycle times during R&D and provides an important way to optimize fabrication protocols without completely fabricating these sublayers.

b. However, OCD short loops and full loop stacks typically produce very different reflectance data due to light interaction with the underlying layer. As a result, OCD solutions are typically developed separately for short and full loops, requiring significant investment overhead.

It may be beneficial to provide one or more additional ways to select signals associated with one or more semiconductor device (SD) portions of interest. Selection can be used in a variety of cases, including for example when the methods mentioned above are inaccurate.

References are made to layers in various text segments, but these are only examples of the SD portion. The plurality of SD portions may be located at different z-axis positions, the plurality of SD portions may include a pattern structure, may include one or more layers and/or may include one of a non-layer portion; , two SD parts can be located at the same z-axis position, etc.

In various text segments, reference is made to one or more upper layers (lower layers) that are relevant SD parts and one or more lower layers (lower layers) that are non-related SD parts, but these are only non-limiting examples of relevant and non-related SD parts.

A solution is provided that can reduce the measurement sensitivity to sublayers and reduce the correlation between interpreted results and undesired layer (e.g. sublayer) properties. This solution can be used to increase sensitivity to weak parameters (weak - have a minor effect on the spectrum - dependent on application and illuminated structural elements) and to resolve parameter correlations in a general sense.

This solution can reduce metrology costs by increasing sensitivity to relevant semiconductor parts and reducing sensitivity to unrelated semiconductor parts, eliminating or at least reducing the use of actual semiconductor devices as a reference.

The solution may include selecting one or more relevant portions and selecting one or more non-relevant portions of a time-domain representation of the wavelength-domain measurement data of light reflected by the patterned structure of the semiconductor device.

Measurement data may include reflected spectral amplitude and spectral phase over a broad wavelength range. Alternatively, the measurement data can contain reflected spectral amplitudes and the spectral phase can be estimated in some way.

Contributions associated with layers of different depths (or different heights or different z-axis values) within the semiconductor device are typically located at different times in the time-domain representation.

When the measured structure height is a few microns, reflections from the top and bottom regions are well separated in the time-domain representation.

However, the problem addressed by the current solution is also applicable to advanced structures that require much better vertical (z-axis) resolution, allowing separation of contributions from layers spaced apart by tens of nanometers to less than a few tens of nanometers. At least one of the layers may be a thin layer with a depth on the nanometer scale.

The method may select one or more relevant portions ("TD portions") of the time-domain representation - e.g., of an unrelated semiconductor device ("SD") portion (e.g., an unrelated underlying layer) - to achieve one or more goals. By reducing sensitivity to attributes, one reduces sensitivity to correlations between one or more related SD parts and at least one unrelated SD part. Additionally or alternatively - selections may be made to minimize reduction in sensitivity to one or more properties of one or more relevant SD portions.

Any trade-off may be provided between the sensitivity of the solution to one or more properties of one or more related SD parts and the sensitivity of the solution to one or more properties of at least one unrelated SD part.

It should be noted that all techniques detailed below may be implemented separately for each measurement feature (and/or applied separately to different wavelength ranges of illumination and/or collection). Examples of measurement characteristics may include at least one of polarization, incident angle, collection angle, azimuth, etc.

Selection may be made by applying one or more selection criteria that may be obtained in one or more ways.

For example - simulations can be used - applying model-based decisions can be used to describe semiconductor materials using simulation tools (such as finite element rigorous coupled-wave analysis (RCWA)) to account for light-matter interactions. The expected reflected field from the device can be calculated. Several simulations are run to provide reflection spectra (and corresponding time-domain results) for various dimensions and material properties of the semiconductor device (different dimensions and material properties are represented by different reference semiconductor devices). This may include performing several calculations that can be run for different subfloor designs. The results of the simulation are evaluated to provide one or more selection criteria. The evaluation may include finding a separation method that can achieve one of the previously mentioned goals.

As another example, the decision may be based on actual measurements. One or more selection criteria may be identified using a set of semiconductor devices with reference data. For example, a set of semiconductor devices may share related top layers and the bottom layers may be different from each other.

Determination of one or more selection criteria may be based on simulations and actual measurements.

Solutions can be chosen from more than a single related TD part.

Each associated TD portion may allow individual interpretation of a different portion and/or aspect of the semiconductor device. For example, one relevant part provides excellent selectivity for dimensional properties placed at the top of the stack and enables high-quality measurements of those parameters. Once this first parameter is interpreted, the second TD portion is selected to provide lower sensitivity to the properties of the semiconductor device, but with slightly increased sensitivity to the underlying layers.

Additional analyzes are now possible, but this second analysis can be performed much more reliably and robustly by injecting the solution using the upper SD part already known from the previous step.

Another method for selecting the relevant TD portion can be based on the radiation pattern of light impinging on the semiconductor device.

Assume that the radiation pattern includes a main lobe and one or more side lobes, and that the top layer (or part of the top layer) of the semiconductor device is first illuminated by a particular side lobe and only then by the main lobe. At the detector - when reflections from a particular sidelobe appear, there may be no other reflections from other layers of the semiconductor device. At the detector - When reflections from the main lobe appear, other signals from other layers of the semiconductor may also appear, so the detector detects the sum of the signals from several layers.

Even though the reflection from a particular sidelobe is weaker than the reflection from the main lobe, using the sidelobes to illuminate the top layer, choosing a TD that includes the reflected light everywhere and excludes the reflection from the main lobe is effectively the uppermost layer. Only information from the floor can be provided.

Selection of one or more relevant TD portions may be followed by filtering the signals of the non-relevant TD portions, such as assigning less weight to the signals of the non-relevant TD portions.

For example, the time-domain cutoff need not be a step function, but rather a window filter in which contributions outside the region t ₂ <t<t ₁ are ignored. These window filters can increase sensitivity to desired attributes in relevant SD parts and reduce correlation to attributes in unrelated SD parts.

Multiple window filters with different widths and/or centers in the time-domain can be used to improve parameter sensitivity and resolve parameter correlations.

One or more features of the solution, such as the wavelength-domain to time-domain conversion itself, and the measurement parameters may be determined in any way to achieve one of the above-mentioned goals.

Selecting solution features can significantly improve vertical resolution, solution robustness, and/or separation between different SD parts.

Determination of the characteristics of the solution may be based on simulation and/or measurement basis. The effect of one or more feature values on the outcome of the solution may be evaluated and determined to comply with one or more goals of the solution.

Determining the characteristics may include, for example, preprocessing the spectrum of the emitted and/or detected light.

Different wavelength ranges in the spectrum can have different penetration depths into the stack and inherently provide some of the desired vertical selectivity. Therefore, methods to narrow the spectral range used for analysis can be used. Another possibility is to improve TD conversion performance by weighting the spectrum and applying filters, especially to the spectral edges (UV and IR parts).

Figure 4A is an example of a method 400 for semiconductor device metrology.

Method 400 may begin at step 410 of generating a time-domain representation of wavelength-domain measurement data of light reflected by a patterned structure of a semiconductor device.

This semiconductor device is also called a measured semiconductor device.

The wavelength-domain to time-domain conversion (“transformation”) used during step 410 may be determined in any manner. For example, conversion can be determined based on the penetration depth (within the measured semiconductor device) of various wavelength components of light.

Different penetration depths can be used to determine which wavelengths penetrate each part of the semiconductor device. For example, it may be advantageous to use a specific wavelength if that wavelength only transmits the relevant part. Penetration depths of different wavelengths are used to remove wavelengths (from light impinging on the semiconductor device being evaluated) and/or to weight (or otherwise increase or decrease importance of) different wavelengths in the wavelength-domain measurement data. can do.

Method 400 may further include step 430 of receiving and/or determining one or more selection criteria to be used during step 420.

Steps 410 and 430 may be followed by step 420.

Step 420 may include selecting one or more relevant TD portions and at least one unrelated TD portion.

Selecting one or more relevant TD portions may include applying one or more selection criteria. One or more selection criteria may be one or more rules, or may be applied using a machine learning process, neural network, etc.

The z-axis propagates along the depth of the semiconductor device. For example, different layers may be located on different z-axis coordinates.

Step 420 may include selecting any number of relevant TD portions.

Step 420 may include or be preceded by obtaining one or more selection criteria to be applied to select one or more relevant TD portions (step 430).

The selection made during step 420 may be based, at least in part, on one or more of the following:

a. Z-axis location of at least one associated SD portion among multiple SD portions in a semiconductor device. For example, the method may receive one or more z-axis positions of one or more relevant SD portions and perform selection accordingly. For example, in the previous example a cutoff based selection was made to select one or more upper layers of SD and ignore one or more lower layers of SD.

b. One or more properties of one or more SD partial devices.

c. Sensitivity of method 400 to one or more properties of one or more DS portions.

d. At least one parameter of one or more parameters of interest.

e. at least one measurement condition, which may be illumination conditions and/or collection conditions, such as polarization;

The selection made during step 420 may include applying one or more selection criteria. Selection criteria can be determined in any way, including simulations or actual measurements.

The properties may be received by method 400 or determined in any manner, such as in a simulation, based on actual measurements, or in any other manner.

Step 420 may be followed by step 490 of determining one or more measurements of one or more parameters of interest of the pattern structure by performing processing using one or more relevant TD portions.

Signals contained in one or more non-relevant TD parts may be ignored. Alternatively - these may be considered but are less important than the signals of one or more relevant TD parts.

4B is an example of a method 401 for semiconductor device metrology.

Method 401 differs from method 400 in that it includes a step 411 of receiving additional information. The additional information may be, for example, measurements of the semiconductor device that are not performed by applying step 410 or the like.

Steps 411 and 420 are followed by determining one or more measurements of one or more parameters of interest of the pattern structure by using one or more relevant portions of the time-domain representation and performing processing using additional information ( 492) follows.

Figure 4C shows an example of step 430 of method 400.

Step 430 may include determining 450 one or more selection criteria based on metrology simulations of one or more reference semiconductor devices.

The simulation may be for a reference semiconductor device that may differ from each other, for example by one or more properties of one or more SD parts (eg unrelated SD parts).

A different reference semiconductor device can be determined by introducing changes (in at least one non-relevant SD portion) to the model of the measured semiconductor device.

The changes may include, for example, changing the material of at least one SD part, changing at least one of the shape and size of one or more elements of at least one SD part, changing the position of at least one SD part, omitting one SD part, one It may include addition of the above elements.

By simulating various properties, the simulation can find the sensitivity of method 400 to one or more properties of one or more SD portions.

Step 450 includes (a) simulating different parameters of interest to be measured during step 490 (step 452), and (b) simulating different values of at least one measurement condition (step 454). It may include simulating the measurement by at least one of the following: Measurement conditions may be lighting conditions, collection conditions, or a combination thereof.

Step 452 and/or step 454 may be applied when simulating one or more reference semiconductor devices.

Any reference semiconductor device may differ from the measured semiconductor device in one or more aspects, but may be similar in at least one other aspect. For example, one or more related SD parts may remain the same, but one or more unrelated SD parts may have one or more differences introduced.

Step 430 may include determining 460 one or more selection criteria based on metrology measurements of different reference semiconductor devices that differ from each other in at least one attribute of the at least one unrelated SD portion. The different reference semiconductor device may include at least a portion of the pattern structure of the measured semiconductor device.

Step 430 may include determining 470 one or more selection criteria based on actual or estimated metrology measurements of different reference semiconductor devices that differ from each other in at least one attribute of the at least one unrelated SD portion. .

Different reference semiconductor devices may include short loop semiconductor devices and long loop semiconductor devices. Both short loop semiconductor devices and long loop semiconductor devices can include at least a portion of the pattern structure of the semiconductor device being measured. A short loop semiconductor device may consist essentially of (a) at least a portion of a pattern structure and a substrate of the semiconductor device. A long loop semiconductor device can comprise substantially the entire portion of the semiconductor device being measured.

Step 440 may include determining 480 one or more selection criteria based on actual or estimated reflections of different lobes of radiation from the semiconductor device.

The lobes may be simulated radiation lobes or actual radiation lobes. Light reflected by the patterned structure of a semiconductor device may include different lobes.

For example, step 480 may include determining selection criteria for selecting a relevant TD portion that includes measurement data of reflections of optical sidelobes from at least a portion of a pattern structure of the semiconductor device.

The sidelobes of light impact at least a portion of the pattern structure of the semiconductor device before the main lobe of light impacts at least a portion of the pattern structure of the semiconductor device.

The selection criteria may include ignoring the measurement data of the main lobe reflected by the pattern structure of the semiconductor device.

Figure 5 illustrates a method 500 of comparing between measured semiconductor devices.

Method 500 may include steps 510, 520, and 530.

Step 510 may include obtaining one or more measurements of one or more parameters of interest of the pattern structure of the first measuring semiconductor device. One or more measurements of one or more parameters of interest are generated using one of methods 400 and 401.

Step 520 may include obtaining one or more measurements of one or more parameters of interest of the pattern structure of the second measuring semiconductor device. One or more measurements of one or more parameters of interest are generated using one of methods 400 and 401.

Steps 510 and 520 are followed by (a) comparing one or more measurements of one or more parameters of interest of the pattern structure of the first semiconductor device to (b) one or more measurements of one or more parameters of interest of the pattern structure of the second semiconductor device. Step 530 follows. Comparison provides one or more comparison results.

Comparisons may be made between two or more measurements of one or more parameters of interest of patterned structures of three or more semiconductor devices.

The comparison results may be processed, for example, to determine differences between semiconductor devices, to indicate potential defects or failures, to indicate process variations, or for other purposes.

6 illustrates an example of a cross-section of a pattern structure 640 including vias 643 passing through top pattern 641 and bottom pattern 642, which are spaced apart from each other and multiple non-reflective layers of a semiconductor device. Passes (644).

A time-domain representation of the wavelength-domain measurement data of light reflected by the pattern structure consists of (i) a first set of peaks resulting from illumination of the upper pattern 641, and (ii) the upper pattern 641 and lower pattern 642. ) will include a second set of peaks spaced apart from the first set of peaks due to the distance between them.

In general, light (the light that illuminates the patterned structure) can interact with multiple features of the patterned structure (which can typically be periodic and/or contain a grating), which can have complex 3D shapes, and thus can be used in the time-domain. The returned measurement signal (spectrum/phase), which is further converted, will contain complex information, from which it will be possible to gather the parameter(s) of interest (eg TCD, etc.).

frequency extension

Significant improvements can be provided in methods using time-domain (TD) approaches for metrology. Specifically, the previously described method uses broadband field reflectance (i.e., complex reflectance) measurements to infer the time-domain impulse response (TDIR) of the structure.

Different portions of the TDIR can be associated with reflections from different regions of the measured stack, with reflections occurring at different times.

Using these methods, sensitivity to underlying layers is reduced and metrology of the region of interest is improved.

One property of this method is the vertical resolution achieved. In general, vertical resolution is determined (to a rough approximation) by the spectral bandwidth. It can be estimated as

here, is the effective refractive index associated with the structure, and are the minimum and maximum wavelengths. different methods can be employed to estimate .

vertical spacing Smaller regions of the stack cannot be easily separated using TD methods, greatly limiting their applicability. Using standard scatterometry using the UV-IR spectral range, it can be inferred that typical vertical separations are tens or even hundreds of nm, meaning that the thicknesses of many of today's high-end semiconductor structures are lower than the TD method's approach. .

The proposed solution provides an algorithmic approach to address the vertical limitation and improve the resulting resolution beyond the theoretical limits given above.

As explained, the resolution limit of the TD method is determined by the measurement spectral bandwidth (Equation 1). Current solutions can extend the measured spectrum through artificial extrapolation, which does not provide additional insight into the measured structure but provides improved vertical (TD) resolution.

It is necessary to provide accurate extrapolation to prevent incorrect spectral broadening, which can lead to failure of the entire TD method, which ultimately causes more harm than good.

It should be noted that although spectral extrapolation to both the UV and IR ranges is advantageous, it is much more important to extend the spectrum into the UV, as the expected resolution improvement will be noticeably more pronounced (Equation 1 reference).

There are several ways to obtain this spectral extrapolation, including model-based extrapolation, reference-based extrapolation, and physical argument-based extrapolation.

Model-based extrapolation may involve one or more methods that can take advantage of the fact that in most cases the structure being measured is a periodic array of (approximately) known geometric elements, where model-based reflection spectra can be inferred using light-matter simulations. You can.

Using a nominal stack description, it is possible to extend the simulated spectrum to wavelengths beyond the measured range and add the results to the measured data set. Extending the model to wavelengths beyond the measured bandwidth is straightforward (often only information is needed about the optical properties of the stack materials at these wavelengths, which is available or can be obtained from physical considerations).

Simple implementation of this approach may lead to discontinuities between the simulated and measured spectra. This unrealistic broadening can be prevented by implementing a smoothing algorithm that modifies the extrapolated spectrum to continuously broaden the measured spectrum (or the measured spectrum and its local derivatives).

The measurements themselves can be used to improve the model used in this extension, by running preliminary analyzes that provide approximate characteristics of the stack and using the derived dimensions as the basis for model-based extrapolation. Obviously, this first step suffers from 'contamination' by the underlying layer, but since only an approximation is needed, the induced errors may be limited.

It is also possible to do it recursively like this: A preliminary guess for some stack description is first used in the spectral extension calculation and added to the measurements. At each point in the analytical regression process (i.e. stack parameters are changed and the calculation results are compared with measurements) the spectrum is simulated over a very wide spectral range. Parts of the spectrum outside the measurement range are added to the measurement. Now both simulated and measured spectra (including measured spectra expanded by simulation) are compared via VTS. This approach can provide improved results.

Reference-based extrapolation - here a single extrapolation is used for all measurements ('Use nominal stack' as in the first case above). However, the selection of these nominals is performed at the recipe generation stage, where a large set of these 'nominal' stacks are computed and used to analyze the referenced set of samples. The selected nominal giving the best results can then be used for extrapolation.

References herein may refer to actual dimensional values. It may be simpler to have different samples or sites on the same wafer if only the underlying layer is intentionally changed (e.g., short/full loops). These too can act as a 'reference' (even without external dimensional measurements) and the 'optimal spectrum' for extrapolation is selected, giving the most similar results for all samples with the same sublayer (i.e. best sublayer independence).

Extrapolation based on physical arguments - it can be seen that the field reflectance must meet some specific properties. One of these requirements is that the real and imaginary parts of the reflected field must satisfy the Kramers-Kronig (KK) relationship. These integral relationships impose constraints on the extrapolated spectrum based on known data in the measured region. After obtaining an arbitrary extrapolation approximation (using the above method or other methods), the overall extrapolation error can be reduced by correcting the extrapolation using the KK relationship.

These KK relations can be used without additional input for spectral extrapolation: by projecting the measured dataset onto a set of basis functions that satisfy the KK relations, the extrapolation can be obtained simply (AB Kuzmenko, 'Kramers-Kronig Constrained Variational Analysis of Optical Spectra', Review of Scientific Instruments 76 , 083108 (2005)).

Some stacks, especially when extrapolating to the 'long wavelength' region, can be treated as approximate descriptions: in mid-IR, 'effective medium' or other simplified approaches can be quite accurate, but a detailed description of the stacks is not necessary. With this approach, we use the inherent low sensitivity of MIR as a means to obtain high-accuracy spectral extrapolation.

As described above, expanding the spectral range improves the vertical resolution of TD analysis, allowing for greater ability to isolate contributions from different parts of the stack. As extensively described in [see previous IP], these improvements can bring many benefits, including overall metrology performance, resilience to process changes, and even applicability to patterns with multiple sublayers using a single solution scheme.

One example of particular interest is the use of TD methods for 3D-NAND metrology. Here, the overall stack height is a few microns, so there are no resolution issues in separating reflections from the top of the stack and the bottom of the 3D-NAND structure. However, if one is interested in isolating reflections from the bottom of the 3D-NAND structure and the underlying CMOS structure, the required vertical resolution can be much better.

This functionality can be achieved using the methods described herein.

Optical scattering has proven to be a mainstream method for semiconductor process control. In this technique, the light reflected by the semiconductor structure is analyzed to obtain the measured dimensions and material properties of the sample. Interpretation of optical measurements is performed by comparing reflected light spectra with theoretical model calculations or through machine learning methods.

A specific type of semiconductor device is a memory cell, which is characterized by a high aspect ratio structure consisting of multiple layers and various patterns throughout the device, especially on the top and bottom.

Several challenges arise while utilizing optical metrology to characterize these memory devices.

a. Because the device depth is much greater than the optical wavelength that produces the effective etalon, the light spectrum is extremely oscillatory. Highly oscillatory signals are difficult to handle due to their extremely nonlinear response to slight sample-to-sample differences or changes in sample dimensions or material properties.

b. Model calculations for highly oscillatory signals are challenging due to unstable regression between model and measured spectra and nonlinear responses to structural parameters that are not robust.

c. If theoretical signal libraries are preferred, they are difficult to construct due to the same nonlinear nature of the signals.

In some cases, only part of the measured structure is known, while some parts cannot be modeled (for example, the logical array beneath the memory cells). In these cases, it is extremely difficult to construct a model and compare the theoretical signal with the measured signal.

Solutions are presented that are based on the use of measured or calculated complex field reflection spectra, which are converted to other domains, in particular the “time domain”, using mathematical transformations. The new time-domain signal can be viewed as a temporal series reflecting from successive layers of the structure.

Time-domain signals can be manipulated in a variety of techniques to generate one or more “effective signals.” The valid signal can be used as is for measurement purposes or later converted back to the original spectral domain.

This approach provides the following methods and benefits:

The time-domain signal is more intuitive than the original spectral-domain signal because reflections at the top of the stack appear “earlier” in the time-domain than reflections at the bottom of the stack. This view allows optimal investigation of structure, depth, multiple layers and other aspects used to construct a computerized model of the stack, much easier than current techniques based on trial and error.

A variety of operations can be applied to time-domain signals. The simplest way is to perform regression analysis by comparing the measured signal with the theoretical signal and find the correct model parameters. This can be done using physical models or machine learning schemes.

More complex techniques allow the structural model to be simplified by removing some of the time-domain signal. For example, clearing the portion of the signal that occurs at the bottom of the stack not only erases the bottom structure from the model, but also makes construction and calculations simpler.

Advanced manipulation of a time-domain signal involves separating it into contributions from different parts of the stack (top, middle, bottom) and using each signal individually to separately regress and find the structural parameters of each part of the stack.

All or part of the time-domain signal can be parameterized. For example, a signal originating from a specific layer of the stack appears at a specific point in the time-domain series. This point can be used as a parameter that reflects the exact depth of the layer.

The shape of each reflection series can be parameterized such that regression on the model fit is performed using the extracted parameters rather than the signal itself. For example, as in the previous case, the viewpoint for the bottom reflection, the peak width or height of the time-domain signal.

Parameterized versions of time-domain signals, or "effective spectra," can be used to generate theoretical signal libraries that can be used for signal regression.

A parameterized version of the time-domain signal or “effective spectrum” can be used as a function of machine learning techniques.

The mathematical manipulations that produce the effective spectrum can in some cases be reversed. In these cases it is possible to transform the effective spectrum back to the original spectral region. In these cases, one can construct libraries or perform calculations in the effective spectral domain (which may be easier than in the inherently bad-behaving case) and then convert these signals or libraries back to the spectral domain for further use.

Examples of systems that can generate various representations of a signal are described in US patent application US2021/0247699 and US patent 10,161,885, which are incorporated herein by reference.

Figure 7 shows a typical time-domain signal 700 from a memory device with top and bottom structures separated by multiple intermediate layers.

Figure 7 shows a typical time-domain signal. The first set of peaks 701 in the left rectangle represents information of parameters in the top part of the stack. The second set of peaks 703, defined by the right rectangle, represents information from both the top and bottom structures of the stack. The distance (gap 702) between the two sets (arrows) represents the total height of the stack between the top and bottom portions.

8 shows an example of a parameter and time-domain representation 720.

Time-domain representation 720 includes a first peak 701, a second peak 722, a third peak 732, and a fourth peak 734. Time-domain representations are different versions of the same 3D pattern structure. The third and fourth peaks show significant differences between the time-domain representations. 8 also shows parameters that can be extracted from the time-domain representation, such as peak position (e.g., location of highest peak 731), peak intensity or height (e.g., H highest peak 732, H lowest peak 734). )), gap between peaks, etc. Parameters may include relationships between one or more peak attributes.

Parameters may be determined using a training cycle, searching for parameters that can distinguish between different versions of the same 3D pattern structure, etc.

9 shows an example of a time-domain representation 710 with five peaks 711-715, and an enlarged view of the fourth peak 714 (five curves 714 of five different versions of the reference 3D pattern structure). 1) - showing 714(5)) and an enlarged view of the fifth peak 715 (showing five curves 715(1) - 715(5) of five different versions of the reference 3D pattern structure. ) is shown.

[00185] Five different versions of the reference 3D pattern structure are more similar to each other, but different from each other. For example, a reference 3D pattern structure may include multiple layers and different versions may differ by only one or less than half of the layers in thickness, shape, depth, etc. within the reference 3D pattern structure (e.g., differing by a single layer). (or two of six or more layers can be different, etc.) Additionally or alternatively - the amount of difference (e.g. due to thickness, shape, depth, etc. within the reference 3D pattern structure) may be lower than a certain threshold - for example 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70% etc.

Differences can be differences between pairs of different versions, especially when pairs of different versions are the most similar versions. For example, if you select a specific version and order four different versions based on their similarity to that specific version, successive pairs of versions will be more similar to each other than different.

The differences and/or similarities between different versions of the reference 3D pattern structure may be differences and/or similarities between the operability and/or functionality of the different versions of the reference 3D pattern structure. Examples of operability include whether the difference caused a defect (which can be tolerated to a predefined degree, which can be determined in any way by any entity, such as a customer, manufacturer, etc.), whether the difference caused a defect This may include whether or not the fault was detected, the severity of the fault (e.g., critical or non-critical), and whether the differences resulted in excessive current consumption.

Similarity may be determined in any way using mathematical functions and/or comparison processes. The method of calculating similarity may be determined arbitrarily using a similarity evaluation process instructed by the client, manufacturer, etc.

Figure 10 shows an example of a method 900 for semiconductor device metrology.

Method 900 may begin at step 910 of obtaining a time-domain representation of wavelength-domain measurement data of light reflected by a three-dimensional (3D) patterned structure of a semiconductor device. 3D pattern structures are also called evaluated 3D pattern structures.

Step 910 may include generating a time domain representation, for example, by applying a wavelength domain to time domain transform, wherein the wavelength domain to time domain transform is based on the penetration depth of different wavelength components of light/light. It is set.

Step 910 may be followed by selecting 920 one or more relevant peaks of the time-domain representation and at least one unrelated portion of the time-domain representation. The selection is made based on the assumption or expectation that one or more relevant peaks may provide information regarding the 3D pattern structure being evaluated.

One or more relevant peaks occur during one or more relevant periods.

One or more related peaks are associated with corresponding reference peaks that are associated with different versions of the reference 3D pattern structure.

The similarity between different versions of the reference 3D pattern structure exceeds the difference between different versions of the reference 3D pattern structure.

At least two related reference peaks corresponding to the same related peak among the one or more related peaks are different from each other.

Step 920 may be followed by step 930 of determining at least one parameter of the 3D patterned structure based on one or more associated peaks and corresponding associated reference peaks.

One or more parameters can be determined to distinguish different versions of the reference 3D pattern structure.

Different versions of the reference 3D pattern structure may differ from each other in at least one of the layer location and the layer width.

The 3D pattern structure is ideally identical to one of the different versions of the reference 3D pattern structure. The 3D reference pattern structure may be a desired or expected 3D pattern structure to be obtained from a manufacturing process that produces the 3D pattern structure.

One or more differences between at least two related reference peaks corresponding to the same related peak may be indicative of one or more parameters of the 3D patterned structure.

Step 930 may be followed by step 940 in response to the decision. Responding may include performing another measurement, transmitting information about the decision, storing information about the decision, etc.

Method 900 may include obtaining reference signatures of different versions of the reference 3D pattern structure.

Step 930 may include at least some of the following:

a. Determining one or more time-domain representation parameters based on the one or more relevant peaks.

b. Determining one or more parameters of the 3D pattern structure based on one or more time-domain representation parameters.

c. Determining one or more parameters of the 3D pattern structure by applying a function for one or more time-domain representation parameters.

d. Steps to apply machine learning processes.

e. The step of executing a decision without applying a machine learning process.

f. Selecting the best matching reference 3D pattern structure among different versions of the reference 3D pattern structure. One or more properties of the (evaluated) pattern structure may be one or more properties of the best matching reference 3D pattern structure.

g. Finding top N matching reference 3D pattern structures—and determining one or more properties of the (evaluated) pattern structures based on one or more properties of the top N matching reference 3D pattern structures. N may be 2, 3, etc.

h. Generating a time-domain representation signature based on one or more parameters associated with one or more relevant peaks.

i. Determining one or more parameters of the 3D pattern structure based on the time-domain representation signature.

j. Determining at least one parameter of the 3D pattern structure based on the time-domain representation signature and the reference signature of the different versions of the reference 3D pattern structure.

k. Searching for the best matching reference signature among different versions of the reference signature of the reference 3D pattern structure.

Any of the aspects described herein may be implemented in computer hardware and/or computer software embodied in a non-transitory, computer-readable medium according to prior art, wherein the computer hardware may include one or more computer processors, computer memory, I/O Includes network interfaces that interoperate with devices and existing technologies.

It should be appreciated that the terms “processor” or “device” as used herein are intended to include any processing device, such as, for example, including a central processing unit (CPU) and/or other processing circuitry. It should also be understood that the terms “processor” or “device” may refer to one or more processing devices and that various elements associated with a processing device may be shared by other processing devices.

As used herein, the term “memory” is intended to include processors or CPUs, such as RAM, ROM, fixed memory devices (e.g., hard drives), removable memory devices (e.g., diskettes), flash memory, and the like. Such memory may be considered a computer-readable storage medium.

Additionally, as used herein, the phrase "input/output device" or "I/O device" refers to, for example, one or more input devices (e.g., keyboard, mouse, scanner, etc.) and/or processing devices for inputting data. It is intended to include one or more output devices (e.g., speakers, displays, printers, etc.) for displaying the results associated therewith.

Embodiments of the invention may include systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium (or medium) having computer-readable program instructions for causing a processor to perform aspects of the invention.

A computer-readable storage medium may be a tangible device capable of maintaining and storing instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. A non-exhaustive list of more specific examples of computer readable storage media include: portable computer diskettes, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM). or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanically encoded device, such as a punch card or command The raised structures of these recorded grooves, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide, or an electrical signal transmitted through another transmission medium (e.g., a fiber optic cable) or wire.

The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing/processing device or to an external computer or via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network. Can be downloaded to an external storage device. The network may include copper transmission cables, optical transmission fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and/or edge servers. A network adapter card or network interface of each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage on a corresponding computer-readable storage medium.

Computer-readable program instructions for performing the operations of the present invention include assembler instructions, instruction-set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or Java, Smalltalk, C++, etc. It may be source code or object code written in a combination of one or more programming languages, including an object-oriented programming language and a traditional procedural programming language, such as the "C" programming language or a similar programming language.

The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to your computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or to an external computer (for example, using an Internet service provider to connect to the Internet). through). In some embodiments, electronic circuits, for example, including programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), can be computer-readable to personalize the electronic circuits to perform aspects of the invention. A computer-readable program instruction can be executed by utilizing the status information of the program instruction.

Aspects are described herein with reference to flow diagram illustrations and/or block diagrams of methods, devices (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flow diagram example and/or block diagram, and combinations of blocks in the flow diagram example and/or block diagram, may be implemented by computer readable program instructions.

These computer-readable program instructions may be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing device to produce a machine, wherein such instructions are executed by the computer's processor or other programmable data processing device and the device may be flowchart and/or create means to implement the functions/behaviors specified in the block diagram blocks. These computer-readable program instructions may also be stored on a computer-readable storage medium that can direct a computer, programmable data processing device, and/or other device to function in a particular way, and thus the computer-readable storage medium may include instructions This is stored and consists of a manufactured product containing instructions that implement aspects of the functions/operations specified in the flowchart and/or block diagram blocks.

Computer-readable program instructions may be loaded into a computer, other programmable data processing device, or other device to cause a series of operational steps to be performed by the computer, other programmable device, or other device, to be executed by the computer, other programmable device, or other device. The instructions may generate a computer-implemented process that implements the functions/operations specified in the flowchart and/or block diagram blocks.

The flow diagram illustrations and block diagrams in the drawings illustrate the architecture, functionality and operation of possible implementations of systems, methods and computer program products in accordance with various embodiments of the present invention. In this regard, each block in a flow diagram example or block diagram may represent a module, segment, or portion of computer instructions that includes one or more executable computer instructions for implementing specified logical function(s). In some alternative implementations, the functions shown in the blocks may occur outside of the order shown in the drawings. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order depending on the functions involved. It will also be noted that each block in the flowchart examples and block diagrams, and combinations of such blocks, may be implemented by special-purpose hardware-based and/or software-based systems that perform the designated function or operation.

The description of various embodiments of the invention has been presented for illustrative purposes, but is not intended to be exhaustive or limit the disclosed embodiments. For example, the systems and methods described herein are applicable to all types of structures on semiconductor wafers. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims (17)

A method for measuring semiconductor devices, the method comprising:
generating a time-domain representation of wavelength-domain measurement data of light reflected by a three-dimensional (3D) pattern structure of a semiconductor device,
Selecting one or more relevant peaks of the time-domain representation and at least one unrelated portion of the time-domain representation, wherein the one or more relevant peaks occur during one or more relevant time periods, and the one or more relevant peaks are in a reference 3D associated with corresponding relevant reference peaks associated with different versions of a pattern structure, wherein the similarity between different versions of the reference 3D pattern structure exceeds the difference between the different versions of the reference 3D pattern structure, and wherein the same At least two related reference peaks corresponding to the related peak are different from each other - , and
A method comprising determining at least one parameter of the 3D patterned structure based on one or more relevant peaks and corresponding relevant reference peaks. The method of claim 1, comprising determining one or more time-domain representation parameters based on the one or more relevant peaks. 3. The method of claim 2, wherein determining at least one parameter of the 3D pattern structure is based on one or more time-domain representation parameters. 3. The method of claim 2, wherein at least one parameter of the 3D pattern structure is a function of one or more time-domain representation parameters. The method of claim 1 , wherein different versions of the reference 3D pattern structure differ from each other in at least one of a layer location and a layer width. The method of claim 1, wherein the 3D pattern structure is ideally identical to one of the different versions of the reference 3D pattern structure. The method of claim 1 , wherein one or more differences between at least two related reference peaks corresponding to the same related peak are indicative of one or more parameters of the 3D patterned structure. The method of claim 1 , wherein determining at least one parameter of the 3D pattern structure comprises applying a machine learning process. The method of claim 1 , wherein determining at least one parameter of the 3D pattern structure is performed without applying a machine learning process. The method of claim 1, wherein determining at least one parameter of the 3D pattern structure includes selecting the best matching reference 3D pattern structure among different versions of the reference 3D pattern structure. The method of claim 1, wherein generating the time-domain representation comprises applying a wavelength-domain to time-domain transform, wherein the wavelength-domain to time-domain transform is based on penetration depths of different wavelength components of light. How to set up. The method of claim 1 , comprising generating a time-domain representation signature based on one or more parameters associated with one or more relevant peaks. 13. The method of claim 12, wherein determining at least one parameter of the 3D pattern structure is based on a time-domain representation signature. 13. The method of claim 12, comprising obtaining reference signatures of different versions of the reference 3D pattern structure. 15. The method of claim 14, wherein determining at least one parameter of the 3D pattern structure is based on a time-domain representation signature and a reference signature of a different version of the reference 3D pattern structure. 15. The method of claim 14, wherein determining at least one parameter of the 3D pattern structure includes searching for the best matching reference signature among reference signatures of different versions of the reference 3D pattern structure. 13. The method of claim 12, wherein one or more parameters are determined to distinguish different versions of the reference 3D pattern structure.