Method for detecting abnormality of alignment mark and method for improving overlay accuracy of photolithography process

Abstract

The application relates to the technical field of semiconductor manufacturing, in particular to an abnormality detection method of an alignment mark and a method for improving overlay accuracy of a photolithography process, which comprises the following steps: acquiring a measured mark signal of the alignment mark; splitting the measured mark signal into a sum of signals of multiple different orders, extracting signals of multiple specific orders from the sum, and respectively calculating values of the symmetry centers of the signals of the multiple specific orders, which are denoted as h(x); and calculating a difference value of h(x) and a predetermined standard value, and determining that the alignment mark is abnormal if the difference value is greater than a deviation threshold. In the application, the mark signal is split into the sum of the signals of the multiple different orders, the symmetry centers of the different order components are compared by using a high-order model analysis method, high-precision detection of the alignment mark can be realized, and the problem that a traditional Q-merit value detection method is invalid under an advanced process is solved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing technology, and in particular to a method for detecting anomalies in alignment marks and a method for improving the overlay accuracy of photolithography processes. Background Technology

[0002] In semiconductor manufacturing or photolithography, mark damage can prevent measurement systems from accurately reading the alignment mark's position, affecting alignment accuracy feedback and compensation. This ultimately leads to interlayer alignment deviations and impacts chip yield. This is a critical process parameter that requires strict control in semiconductor manufacturing. As technology advances, alignment marks are becoming increasingly smaller, and traditional measurement methods can no longer accurately identify abnormal alignment marks, failing to meet the demands of advanced processes. Therefore, there is an urgent need to develop faster and more accurate automatic identification technologies to solve the problem of align mark damage detection. Summary of the Invention

[0003] One of the objectives of this invention is to provide a method for detecting abnormalities in alignment marks and a method for improving the overlay accuracy of photolithography processes, so as to improve the detection accuracy of alignment mark damage.

[0004] To achieve the above objectives, the present invention provides an anomaly detection method for alignment marks, comprising the following steps: Obtain the measured mark signal of the alignment mark; The measured mark signal is decomposed into the sum of multiple signals of different orders. Multiple signals of specific orders are extracted from them, and the value of the center of symmetry of each signal of specific order is calculated and denoted as h(x); where x represents the order. Calculate the difference between h(x) and a predetermined standard value. If the difference is greater than the deviation threshold, the alignment mark is determined to be abnormal.

[0005] Furthermore, in the step of splitting the measured mark signal into a sum of signals of multiple different orders, a polynomial fitting method is used to split the measured mark signal into a sum of signals of multiple different orders.

[0006] Furthermore, the plurality of specific order signals include first-order signals, third-order signals, and fifth-order signals; or The multiple specific order signals include second-order signals, fourth-order signals, and sixth-order signals.

[0007] Furthermore, the standard value and deviation threshold are obtained by processing and calculating the raw signal after scanning multiple normal alignment marks with a measurement device in advance.

[0008] Furthermore, the method for pre-determining standard values ​​includes the following steps: Use a measurement device to scan multiple normal alignment marks, and use the original signal obtained from each scan as the standard mark signal; The standard mark signal is preprocessed to obtain the reference signal; The reference signal is decomposed into the sum of multiple signals of different orders. The signal of a specific order corresponding to h(x) is extracted from it, and the value of the center of symmetry of each specific order signal is calculated and denoted as f(x). The average value g(x) of f(x) corresponding to each reference signal is calculated as the standard value.

[0009] Furthermore, the step of preprocessing the standard mark signal to obtain the reference signal includes the following sub-steps: Each of the aforementioned standard mark signals is flattened to obtain a flattened signal; Each of the flattened signals is normalized to obtain a normalized signal; the normalized signal is used as a reference signal.

[0010] Furthermore, the deviation threshold is obtained through the following steps: Based on the fluctuation range of f(x) relative to the standard value corresponding to each reference signal, the fluctuation amplitude of f(x) is obtained, and the fluctuation amplitude A(x) of f(x) is used as the deviation threshold.

[0011] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is: to provide a method for improving the overlay accuracy of photolithography processes, comprising the following steps: During the photolithography process, measurement equipment is used to scan each alignment mark on the wafer to obtain the measured mark signal of each alignment mark; An abnormal alignment mark is detected according to any of the above-described methods for detecting alignment marks; After removing the abnormal alignment marks, calculate the OVL of the wafer and determine whether rework is required; If so, the wafer is returned to the previous process; otherwise, the process parameters are compensated according to the wafer's OVL.

[0012] Furthermore, the step of calculating the OVL of the wafer after removing the abnormal alignment marks and determining whether rework is required includes the following sub-steps: Output the coordinates of the abnormal alignment mark to the photolithography process analysis and control system; The photolithography process analysis and control system removes the corresponding alignment marks based on the coordinates to obtain the remaining alignment marks; Calculate the OVL of the wafer based on the remaining alignment marks; The photolithography process analysis and control system determines whether the OVL error is within the limits of the process specifications. If it is, the overlay accuracy is considered to be qualified; otherwise, the overlay accuracy is considered to be unqualified.

[0013] Furthermore, the method for compensating for process parameters based on the OVL error of the wafer includes the following sub-steps: The optimal compensation value is calculated based on the OVL error of the wafer; Send the calculated optimal complement value to the APC system; The APC system updates its internal prediction model based on the optimal compensation value and automatically fine-tunes the production parameters for subsequent wafer batches.

[0014] In summary, the semiconductor device and its fabrication method of this invention have the following unexpected technical effects: By innovatively decomposing the mark signal into the sum of multiple signals of different orders, employing a high-order model analysis method to calculate the center of symmetry of different order components, and comparing it with standard values ​​to detect damage to the alignment mark, this method surpasses traditional intensity / contrast analysis, directly detecting the geometric integrity of the mark, thereby discovering hidden asymmetric damage that affects alignment accuracy. This achieves high-precision detection of the alignment mark, solving the problem of failure of traditional Q-merit value detection methods in advanced processes; and realizing a leap from macroscopic intensity monitoring to microscopic shape detection. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of a normal alignment mark with a Q-merit value of 0.9 obtained by measuring using existing technology.

[0016] Figure 2 It is scanning using measurement equipment Figure 1 A schematic diagram of the waveform of the mark signal obtained by aligning the marker.

[0017] Figure 3 This is a schematic diagram of an abnormal alignment mark with a Q-merit value of 0.9 obtained by measuring using existing technology.

[0018] Figure 4 It is scanning using measurement equipment Figure 3 A schematic diagram of the waveform of the mark signal obtained by aligning the marker.

[0019] Figure 5 This is a process flow diagram of an embodiment of the alignment mark anomaly detection method of the present invention.

[0020] Figure 6 This is a waveform diagram of an original signal near its center point.

[0021] Figure 7 This is a schematic diagram of the waveform of the first-order component near the center point.

[0022] Figure 8 This is a schematic diagram of the waveform of the third-order component near the center point.

[0023] Figure 9 This is a schematic diagram of the waveform of the fifth-order component near the center point.

[0024] Figure 10 A schematic diagram of the waveform of an even-order signal near the center point, which is a decomposition of a standard mark signal.

[0025] Figure 11 A schematic diagram of the waveform of an even-order signal near the center point, which is a fraction of a measured mark signal.

[0026] Figure 12 This is a process flow diagram of an embodiment of the alignment mark anomaly detection method of the present invention.

[0027] The diagrams in the instruction manual are labeled as follows: First rectangular area 100; First rectangular mark 101; Second rectangular area 200; Second rectangular mark 201. Detailed Implementation

[0028] The following disclosure provides various embodiments or examples for implementing different features of the invention. Specific examples of components and arrangements will be described below to simplify the invention. Of course, these are merely examples and are not intended to limit the invention. For example, in the following description, forming a first component above or on a second component may include embodiments where the first and second components are in direct contact, or embodiments where other components may be formed between the first and second components such that the first and second components are not in direct contact. Furthermore, reference numerals and / or characters may be repeated in various instances of the invention. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or configurations.

[0029] Furthermore, spatial relation terms such as "below," "under," "below," "above," and "above" may be used herein to readily describe the relationship between one element or component and another element (or component) or component (or component) as shown in the figure. In addition to the orientations shown in the figure, spatial relation terms will encompass various different orientations of the device in use or operation. The device may be positioned in other ways (rotated 90 degrees or in other orientations), and will be interpreted accordingly through the spatial relation descriptors used herein.

[0030] Furthermore, the technical parts described in this invention and the appended claims are primarily the improved technical parts of this invention, and do not limit the object protected by this invention to only having these technical parts. Other known essential components (structures and / or methods) and / or non-essential components of the object protected, besides the technical parts described in this invention and the appended claims, are not included in this invention and the appended claims because they do not fall within the scope of improvements of this invention; however, this does not mean that the object protected by this invention does not possess these known components.

[0031] The following embodiments illustrate preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

[0032] The formation of alignment marks is a micro-nano fabrication process that is synchronized with the chip's circuit pattern. It involves "carving" specific three-dimensional physical structures onto the wafer through photolithography, etching, and deposition processes during multiple key chip manufacturing steps. In semiconductor manufacturing, alignment mark formation is not an independent step but is deeply embedded and integrated into the critical layers of the front-end process. Its core principle is to generate them synchronously with the chip's functional pattern to ensure absolute correlation of spatial coordinates.

[0033] Specifically, this process begins in the initial stage of chip manufacturing, namely the first photolithography and etching of the active region or shallow trench isolation layer. In this step, the first set of circuit patterns on the wafer and the "zero layer" alignment mark, which serves as an absolute reference, are simultaneously defined on the silicon substrate. This alignment mark is permanent and stable because it is directly etched on the wafer substrate, providing a fixed spatial reference origin for all subsequent process layers.

[0034] As the manufacturing process progresses layer by layer, the formation of each new dielectric layer, polysilicon layer, or metal interconnect layer involves the alignment system identifying and aligning with the existing alignment marks of the previous layer during the specific photolithography and patterning steps of that layer. Then, using the photomask of the current layer, the functional circuit pattern of the current layer and the alignment mark pattern specific to that layer are simultaneously fabricated in a single exposure and subsequent etching or deposition process. This mechanism of interlayer alignment and intralayer synchronization ensures that the alignment marks within each layer maintain a precise and fixed relative position with the transistors, interconnects, and other functional components in the same layer.

[0035] Therefore, the formation of alignment marks is essentially an inherent component of each patterning step in the manufacturing process. Starting from the establishment of the zero-layer reference, it transmits and replicates spatial coordinate information losslessly to each subsequent process layer like a relay race, ultimately building a complete, precise physical coordinate network on the wafer that can be read by lithography machines and metrology equipment. This is the foundation for realizing nanoscale multilayer overlay.

[0036] In photolithography, alignment marks primarily serve the function of interlayer alignment, ensuring precise alignment between the current lithographic layer pattern and the previous layer. Even the slightest misalignment can lead to transistor connection errors or even chip failure; therefore, modern advanced processes require overlay precision controlled within a few nanometers. Secondly, in multiple exposure or self-aligned multiple imaging techniques, alignment marks coordinate the relative positions between multiple exposures within the same lithographic layer, a necessary means to achieve feature sizes below the resolution of the photolithography machine. Furthermore, in hybrid lithography scenarios, alignment marks also coordinate pattern layers drawn using different techniques, achieving system error matching and compensation through dedicated marks and models.

[0037] In the field of process monitoring and measurement, alignment marks serve as both rulers and sensors. Specialized overlay measurement marks are used to measure and monitor the actual overlay error between process layers. Their design is more precise than real-time alignment marks, and the generated OVL (Overlay, or superposition error) map can be fed back to the APC (Advanced Process Control) system for process control. Simultaneously, alignment marks also serve as the positioning and focusing reference for measurement equipment such as CD-SEM (Critical Dimension Scanning Electron Microscope) and OCD (Oblique Dispersive X-ray Diode). These devices must first identify the marks to accurately locate the micro-circuit pattern under test. Furthermore, by monitoring the signal quality of the alignment marks, process stability can be assessed. Damage or signal degradation of the alignment marks is often an early indication of process problems such as CMP (Chemical Mechanical Planarization) over-polishing, etching load, and abnormal thin film stress.

[0038] In equipment and process integration, alignment marks enable grid matching between lithography machines and metrology stations. This allows different devices to share or have a mark system with defined positional relationships, ensuring that the pattern exposed on the lithography machine can be accurately located and measured on the metrology station. Simultaneously, alignment marks within the dicing lane are integrated with other functional marks such as wafer ID and test structures into a process information area, becoming a physical data carrier connecting design, manufacturing, metrology, and packaging.

[0039] With the rapid development of advanced packaging and heterogeneous integration, the application frontiers of alignment marks are constantly expanding. In wafer-to-wafer bonding, it achieves submicron-level face-to-face alignment through special infrared or pre-bonding markings, completing three-dimensional vertical interconnects. In chip-to-wafer bonding, it ensures that individual qualified chips are mounted onto the wafer with high precision. In through-silicon vias and redistribution layers, it supports interlayer alignment for high-density interconnects within the package, with precision requirements approaching those of front-end processes. Furthermore, in three-dimensional integrated circuits, alignment marks achieve back-side alignment through infrared or visible light-transmitting designs. In compound semiconductors and photonics integration, they are adapted to the special materials and structures of non-silicon substrates such as InP and GaAs, or silicon photonic chips, enabling the alignment of components such as optical waveguides and modulators.

[0040] In summary, alignment marks serve as a bridge for precision transfer, a sensor for process control, a carrier for information integration, and an enabler for technological development. Through alignment marks, the ultra-high coordinate precision of the laser interferometer inside the lithography machine can be transferred and locked to the physical position of each wafer and each layer of pattern. Their physical morphology and signal characteristics reflect the combined effects of multiple processes such as deposition, lithography, etching, and CMP. The marking system within the dicing channel connects the entire chain of design, manufacturing, measurement, and packaging. From 2D scaling and 3D stacking to heterogeneous integration, any advancement in semiconductor technology that relies on precise spatial relationships is inseparable from the synchronous evolution of alignment mark technology. Therefore, alignment marks are an indispensable cornerstone technology.

[0041] However, in semiconductor manufacturing, damage to alignment marks is one of the key issues leading to overlay errors and yield losses. This damage is complex and varied, occurring throughout the entire manufacturing process. Chemical mechanical polishing (CMP) can lead to asymmetric overpolishing, dish-shaped depressions, or edge erosion due to differences in polishing rates, uneven consumables, or narrow process windows. Dry or wet etching can cause loading effects, micro-loading effects, or uneven sidewall angles due to uneven distribution of reactive ions, polymer deposition, or etching selectivity issues. Thin film deposition can result in poor conformality, stress-induced deformation, or incomplete gap filling due to insufficient step coverage or intrinsic film stress. Photolithography and development can cause imaging distortion or incomplete / overdevelopment due to optical system defects or improper process parameters. Cleaning and wet processes can cause corrosion or particulate contamination due to improper chemical solution ratios, temperature, or time control. Equipment hardware and operation can also cause mechanical scratching or electrostatic discharge damage due to equipment malfunctions or improper operation. These multi-factor, multi-form systemic damage problems must be detected early, accurately, and automatically using diagnostic technologies to effectively ensure the yield of advanced processes.

[0042] In existing technologies, the Q-merit (Quality Merit) value is commonly used as a quantitative indicator to evaluate the quality of lithographic alignment marks. This indicator is mainly calculated by combining signal intensity and signal-to-noise ratio (SNR) collected by measurement equipment. Its value ranges from 0 to 1 and generally reflects the signal quality of the mark. Specific judgment criteria are as follows: when the Q-merit value > 0.8, the mark signal quality is excellent and can support high-precision alignment; when 0.8 > Q-merit value > 0.5, the mark has slight signal attenuation or noise interference, requiring the activation of process monitoring mechanisms to ensure production stability; the mark may suffer severe signal degradation due to etching residue, material deformation, or contamination, which will significantly affect overlay accuracy and requires immediate treatment.

[0043] The Q-merit calculation method is a key algorithm in semiconductor metrology for evaluating the quality of alignment marks. Its core is the signal-to-noise ratio (SNR) and contrast of the quantized mark signal. While there is no globally unified standard formula, its calculation principles and core elements are common to the industry. Q-merit is essentially a comprehensive score of signal quality, primarily based on a function of two core parameters: signal contrast and SNR. Its general approach can be expressed as: Q-merit = f(contrast, SNR).

[0044] In practice, the following two computational logics are employed: Method 1: Calculation based on peak-to-peak contrast. This is the most intuitive and commonly used method, directly utilizing the peak and trough values ​​of the signal. The specific process is as follows: First, locate a region corresponding to the complete alignment mark period on the acquired raw signal waveform. Then, find the global maximum value I_max and global minimum value I_min within this region. Finally, substitute them into the following formula to calculate: Q-merit=(I_max-I_min) / (I_max+I_min) Formula (1) In equation (1), I_max represents the maximum intensity value of the mark signal within one period or feature window. I_min represents the minimum intensity value of the mark signal within the corresponding period.

[0045] The results can be expressed as decimals or percentages, with 100% representing ideal contrast. The above method is simple to calculate, has clear physical meaning, and directly reflects the degree of "black and white" distinction in the markings.

[0046] Method 2: Signal-to-noise ratio (SNR) calculation. This method focuses more on the strength of the signal relative to the background noise. The specific process is as follows: First, define the signal region (ROI, i.e., the location of the alignment mark) and the background region (the adjacent unmarked region). Then, calculate the signal statistics (mean and standard deviation) for both regions. Finally, substitute them into the following formula to calculate the Q value in the form of SNR: Q=(Signal_Level-Noise_Mean) / Noise_Std Formula (2) In Equation (II), Signal_Level represents the average intensity of the mark signal in the marked feature region; Noise_Mean represents the average intensity of the mark signal in the background region; and Noise_Std represents the standard deviation of the mark signal in the background region (representing the magnitude of noise fluctuation). This calculation method can better distinguish between weak signals and high noise conditions.

[0047] However, the Q-merit method is insensitive to asymmetric damage. Even with a severely asymmetric mark, as long as I_max and I_min don't change significantly, the Q-merit value can still be high, but its true center has shifted, leading to alignment errors. The Q-merit method is also susceptible to global process variations. For example, when the film thickness changes across the entire wafer, it can simultaneously raise or lower I_max and I_min, which may not affect the Q-merit value but alters the absolute level of the signal.

[0048] However, as alignment mark sizes continue to shrink, the limitations of Q-merit values ​​become increasingly apparent. Due to insufficient sensitivity in detecting spatial defects such as edge damage and asymmetric deformation, it is prone to "false negatives"—that is, the measurement system misjudges the alignment mark as normal when it actually exists and affects alignment accuracy. This misjudgment leads to a decrease in lithographic alignment accuracy, thereby affecting chip yield. Therefore, Q-merit is a useful but limited rapid screening tool. In advanced processes, it cannot provide a comprehensive and accurate assessment of the health status of alignment marks and cannot meet nanometer-level overlay accuracy requirements. More advanced automatic identification technologies need to be developed to replace traditional Q-merit measurement methods.

[0049] Please see Figure 1 , Figure 1 This is a schematic diagram of a normal alignment mark with a detected Q-merit value of 0.9. Figure 1 In this configuration, the white first rectangular marker 101 in the first rectangular area 100 at the four corners constitutes the alignment marker for the upper layer, and the black second rectangular marker 201 in the four second rectangular areas 200 near the center point constitutes the alignment marker for the current layer. When both layers of alignment markers are functioning correctly, the current layer is considered aligned with the upper layer if the center point of the alignment marker in the upper layer coincides with the center point of the alignment marker in the current layer.

[0050] Please see Figure 2 , Figure 2 Scanning using measurement equipment Figure 1 The diagram shows the waveform of the mark signal obtained after alignment. Figure 2 In the test, the waveform of the mark signal was basically normal. The detected Q-merit value was 0.9, which is within the normal range. Therefore, the correct conclusion can be drawn about the mark based on its Q-merit value.

[0051] Please see 3. Figure 3 An anomalous alignment marker with a detected Q-merit value of 0.9 can be seen. Figure 3 The alignment mark in the image shows obvious damage: all the first rectangular marks 101 in the first rectangular area 100 in the upper left corner are damaged, resulting in blurred marks. The first rectangular marks 101 in other first rectangular areas 100 are also damaged. It can be clearly identified as an abnormal mark by manual inspection. However, its detected Q-merit value is 0.9. If judged solely by the Q-merit value, the alignment mark would be judged as a normal mark. This example clearly reflects the limitations of the Q-merit index in advanced processes.

[0052] Please see Figure 4 , Figure 4 Scanning using measurement equipment Figure 3 The waveform of the mark signal obtained after aligning the marker is shown. It can be seen that the amplitude difference between the left and right parts of the waveform is significant. Therefore, an anomaly in the alignment marker can be identified through the waveform. It is evident that judging the anomaly of the focus marker based on the waveform shape is more accurate than judging it based on the Q-merit value.

[0053] Please see Figure 5 , Figure 5 A process flow diagram of an anomaly detection method for alignment marks is shown. In the illustrated embodiment, the anomaly detection method for alignment marks includes the following steps: S100. Acquire the measured mark signal of the alignment mark. In semiconductor manufacturing, the mark signal is the physical response signal generated after scanning the alignment mark on the wafer by a detection system such as optics, electron beam, or X-ray. Essentially, it is a quantifiable electrical signal that is received by the sensor and converted into a quantifiable electrical signal after the physical morphological features of the alignment mark (such as height, edge, and periodic structure) interact with the detection beam.

[0054] During the scanning of alignment marks using measurement equipment, various optical effects occur when a probe beam (usually a laser or broadband light) illuminates the alignment mark, such as diffraction, scattering, reflection, and interference. The geometry of the alignment mark modulates the amplitude, phase, polarization, or propagation direction of the incident light. This modulated light is received by a sensor, forming the original analog signal. Measurement equipment (such as an optical or diffraction signal sensor) converts the received optical signal into an electrical signal (usually a waveform of voltage or current changing with time / position). This waveform is the original Mark signal.

[0055] When the probe beam forms a spot in the blank area of ​​the alignment mark, the reflected light intensity is high and stable. When the spot moves to the edge of the line (raised or recessed), the reflected light intensity changes drastically due to enhanced diffraction / scattering, forming a steep edge signal. Scanning the entire alignment mark yields a periodically changing waveform. Its peaks, valleys, rising / falling edge positions, symmetry, and other characteristics encode information about the alignment mark's position, shape, quality, and other features.

[0056] After receiving the Mark signal, the measurement equipment extracts the center position of the alignment mark using algorithms (such as correlation matching, edge detection, centroid calculation, Fourier analysis, etc.). This calculated position is the coordinate used for lithography alignment. Therefore, the Mark signal is a response generated by the physical structure of the alignment mark under specific detection energy, which can be received by the sensor. Its waveform characteristics directly reflect the morphology and state of the alignment mark, and it is the original information source upon which the lithography machine relies to achieve nanometer-level alignment accuracy.

[0057] For example, before the exposure sequence of each lithography process begins, after the wafer is fed into the lithography machine by a robotic arm and undergoes rough pre-alignment, the system enters the fine alignment stage. The lithography machine's measurement system rapidly scans multiple pre-designed alignment marks distributed on the wafer to obtain the measured mark signals.

[0058] S200. The measured mark signal is decomposed into a sum of signals of different orders. Multiple signals of specific orders are extracted from these signals, and the value of the center of symmetry for each specific order signal is calculated, denoted as h(x); where x represents the order. Theoretically, any complex waveform can be infinitely approximated by a weighted sum of a series of basic waveforms (basis functions). In this embodiment, a polynomial fitting method is preferably used to decompose the measured mark signal into a sum of signals of different orders.

[0059] In this embodiment, the extracted signals of specific orders are preferably first-order, third-order, and fifth-order signals, i.e., x = 1, 3, 5. If necessary, signals of other orders can be added to the first-order, third-order, and fifth-order signals; for example, a seventh-order signal can be added. Of course, the signals of specific orders can also be second-order, fourth-order, and sixth-order signals, or signals of other orders can be added to the second-order, fourth-order, and sixth-order signals.

[0060] Please see Figure 6 , Figure 7 , Figure 8 and Figure 9 Polynomial fitting decomposes complex raw signals into a superposition of components of different orders (i.e., signals of different orders), each representing a specific shape pattern in the signal. Common orders include first-order (linear functions), third-order (cubic functions), and fifth-order (quintic functions). Higher orders capture more signal details (such as asymmetry and local curvature). For each decomposed component, its center of symmetry can be calculated, which is the theoretical center point of the waveform at that order. This center of symmetry is a mathematical calculation, not a direct measurement, allowing for more precise and robust alignment.

[0061] When alignment marks are damaged by processes such as CMP and etching, the shape of their mark signals becomes asymmetrical. The center of the original signal (simply taking the maximum / minimum point) may no longer be accurate. Furthermore, alignment mark damage (such as unilateral collapse or contamination) is often more pronounced in higher-order signal components (such as third-order or fifth-order signals). For example: Please continue reading. Figure 7 The first-order component of the mark signal represents the linear trend of the signal. The overall shape is linearly tilted. Its center of symmetry can be understood as the overall tilt trend of the signal. Its center of symmetry (usually determined by the coefficient of the first term) reflects the position of the center of gravity of the signal, that is, the approximate position of the aligned mark.

[0062] Please continue reading. Figure 8 The third-order component of the mark signal represents the signal's asymmetry or distortion, exhibiting an overall "S"-shaped distortion. If the left and right halves of the third-order component signal differ in shape (e.g., one side is steeper, the other gentler), a strong third-order component will be generated. Therefore, the third-order component can extremely sensitively detect left-right asymmetric damage to the alignment mark. For example, uneven CMP polishing causing collapse on the aligned side of the mark will be reflected in the third-order component of the mark signal.

[0063] Please continue reading. Figure 9The fifth-order components of a mark signal represent local detail undulations or complex shape modulations, with the overall shape exhibiting more complex asymmetric undulations. It can capture finer, non-monotonic morphological anomalies, such as local roughness at the edges of the mark and minute periodic defects.

[0064] Similarly, the shape of the alignment mark can also be analyzed using the second-order, fourth-order, and sixth-order components of the mark signal. For example: The second-order component of the mark signal represents the overall symmetrical expansion or contraction of the signal, reflecting a global, smooth shape change trend. The overall waveform of the second-order component is parabolic. Although the shape of the second-order component is symmetrical and does not directly cause an offset in the alignment center calculation as in odd-order asymmetry, it will distort the relationship between the measured alignment position and the actual circuit pattern position.

[0065] The fourth-order component of the mark signal is used to diagnose the complex symmetrical fluctuations in the edge region or plateau of the alignment mark, revealing more localized and higher-frequency symmetrical deformations than the second-order component. The overall waveform of the fourth-order component is an "M" / "W" shaped fluctuation, which is a symmetrical fluctuation superimposed on the main signal, and may change the shape of the mark's "shoulders" or "top".

[0066] The sixth-order component of the mark signal is used to capture nanoscale details and higher-order coupling effects. The overall waveform of the sixth-order component is a more complex multi-peak oscillation, thus exhibiting a more refined, multi-peak symmetrical oscillation, reflecting the shape modulation of the mark signal at the microscale.

[0067] S300. Calculate the difference between h(x) and a predetermined standard value. If the difference is greater than the deviation threshold, the alignment mark is determined to be abnormal. In this step, the standard value and the deviation threshold are obtained by processing and calculating the original signal after scanning multiple normal alignment marks with a measurement device.

[0068] Specifically, the method for calculating the standard value may include the following steps: S310. Use a measurement device to scan multiple normal alignment marks, and use the raw signal obtained from each scan as a standard mark signal. The purpose of collecting multiple standard mark signals is to establish a mark signal database representing normal alignment marks, which will serve as a reference baseline for subsequent analysis of the measured mark signals.

[0069] Due to inherent variations in front-end processes such as CMP, etching, and deposition, alignment marks themselves exhibit process variations. Their center of symmetry is not a fixed value but rather varies randomly within a certain statistical distribution range. The mark signal itself is affected by these variations, causing f(x) to fluctuate within a certain range. Under normal circumstances, even with variations, the fluctuation range of f(x) should be within the process specifications, and the fluctuations of the centers of symmetry for each order of signal should be consistent. However, when alignment marks are damaged, the fluctuation range of f(x) will exceed the normal range. Therefore, by monitoring the fluctuation range of f(x), anomalies in alignment marks can be identified.

[0070] S320. Preprocess the standard mark signal to obtain the reference signal. This step may include the following sub-steps: S321. Flatten each of the standard mark signals to obtain a flattened signal. The original acquired signal (i.e., the standard mark signal) contains unwanted background fluctuations in addition to the features of the alignment mark itself (such as peaks or depressions corresponding to the edges). These fluctuations may be caused by uneven illumination (e.g., bright in the center and dark around the edges), thin-film interference effects on the wafer surface, inherent vignetting of the optical system, etc. Flattening can remove these global background variation trends, making the signal baseline flatter, thereby making the features of the alignment mark itself more prominent, facilitating subsequent accurate comparison and analysis.

[0071] S322. Normalize each of the flattened signals to obtain a normalized signal; use the normalized signal as a reference signal. Normalization typically scales the signal amplitude (intensity) to a uniform standard range (e.g., between 0 and 1, -1 and 1, etc.). This is done to eliminate absolute signal intensity differences caused by factors such as slight differences in illumination intensity between different alignment marks or different measurement batches. This results in a set of normal mark signals with flat amplitude baselines and uniform scale, which can be used as reference signals to establish a clean and standard reference signal template for subsequent high-sensitivity detection of whether the alignment marks are damaged.

[0072] S330. The reference signal is split into a sum of multiple signals of different orders. From these signals, signals of specific orders (first-order, third-order, and fifth-order signals in this embodiment) corresponding to h(x) in step S200 are extracted, and the value of the center of symmetry of each specific order signal is calculated and denoted as f(x). The method for splitting the reference signal is the same as the method for splitting the standard signal in step S200.

[0073] S340. Calculate the average value of f(x) corresponding to each reference signal, denoted as g(x), and use g(x) as the standard value. That is, each specific order signal corresponds to a standard value. In this embodiment, g(x) includes g(1), g(3), and g(5). Among them, g(1) is the standard value of the first-order signal, g(3) is the standard value of the third-order signal, and g(5) is the standard value of the fifth-order signal.

[0074] Of course, since the values ​​of g(1), g(3) and g(5) are very close, we can also take the average value of g(1), g(3) and g(5) or specify a value among g(1), g(3) and g(5) as the standard value for signals of each specific order.

[0075] According to this embodiment, the deviation threshold can be obtained through the following steps: S350. The fluctuation range of f(x) relative to the standard value is obtained according to the fluctuation range of f(x) corresponding to each reference signal, and the fluctuation range A(x) of f(x) is used as the deviation threshold. For example, the deviation threshold may include: the fluctuation range A(1) of f(1) relative to g(1), the fluctuation range A(3) of f(3) relative to g(3), and the fluctuation range A(5) of f(5) relative to g(5). If the measured mark signal simultaneously satisfies |h(1)-f(1)|<A(1), |h(3)-f(3)|<A(3), and |h(5)-f(5)|<A(5), then the alignment mark corresponding to the measured mark signal is determined to be normal; otherwise, the alignment mark is determined to be abnormal. Of course, a value can also be uniformly determined based on the values ​​of A(1), A(3), and A(5) as the unified deviation threshold for signals of each specific order.

[0076] When the extracted signal of a specific order is an even-order signal (i.e., an even function), comparisons can also be made based on the value of the center of symmetry. See also... Figure 10 , Figure 10 A waveform diagram of an even-order signal f(x) decomposed from the standard mark signal near the center of symmetry; please refer to [link to diagram]. Figure 11 , Figure 11 This is a waveform diagram of an even-order signal h(x) decomposed from the measured mark signal near the center of symmetry. By calculating the difference between the average values ​​of h(x) and f(x) and comparing this difference with a deviation threshold, it is possible to determine whether the alignment mark is abnormal. The method for determining this is consistent with that for signals of a specific order that are odd-order signals.

[0077] For ideal signals of each order component, a position representing its center of symmetry is calculated using mathematical methods (such as finding the peak value of its autocorrelation function, calculating the energy center, or using polynomial zeros). For markers with asymmetric damage, the centers of their higher-order components (such as third-order and fifth-order components) will deviate significantly. In traditional methods, the Q-merit value mainly reflects the overall contrast and signal-to-noise ratio of the mark signal. Even if the aligned markers are not symmetrical, as long as the contrast remains, the Q-merit value may not decrease significantly, and it may still be judged as acceptable. However, the detection method of "center drift" in this embodiment is a strong indicator of the presence of damage, which is much more sensitive than simply looking at the signal strength (Q-merit value), thus providing diagnostic accuracy far exceeding that of traditional methods.

[0078] Normal alignment marks have smooth, symmetrical mark signals, and the coefficients of higher-order (e.g., third- and fifth-order) models are very small. Damaged alignment marks (especially those with asymmetric damage) exhibit distorted mark signal shapes. Even if the overall intensity of the mark signal doesn't change significantly, the coefficients of its higher-order models will inevitably increase significantly, and the centers of symmetry calculated by these higher-order models will deviate markedly. Therefore, higher-order center-of-symmetry analysis can capture asymmetric changes in the mark signal shape. These changes may not significantly affect the overall contrast of the mark signal, but they will severely impact the alignment algorithm results based on the symmetry assumption.

[0079] Traditional methods (such as Q-merit) primarily consider the overall amplitude, contrast, and signal-to-noise ratio of the signal. These characteristics are mainly dominated by first-order and second-order (energy) components. A marker that is partially damaged but still has sufficient contrast may still pass the Q-merit test. However, in this embodiment, by decomposing the mark signal into the sum of multiple signals of different orders, a higher-order model analysis method is used to compare the centers of symmetry of different order components. This approach not only considers the strength of the mark signal but also the regularity of its shape, significantly improving detection accuracy.

[0080] In summary, the anomaly detection method for alignment marks in this application has the following unexpected technical effects: By analyzing the intensity and symmetry center of these model components, this application can surpass traditional intensity / contrast analysis and directly reveal the geometric integrity of the mark, thereby discovering hidden asymmetric damages that affect alignment accuracy. Therefore, by decomposing the mark signal using a high-order polynomial model and independently analyzing the consistency of the symmetry centers of each order component, this application can achieve high-precision detection of alignment marks, solving the problem of failure of traditional Q-merit value detection methods in advanced processes; it achieves a leap from "macroscopic intensity monitoring" to "microscopic shape detection." This is not merely a simple tool upgrade, but a paradigm shift in addressing the challenges of nanoscale topography control in advanced semiconductor manufacturing.

[0081] Please see Figure 12 , Figure 12 A process flow diagram of a method for improving OVL feedback accuracy is shown. In the illustrated embodiment, the method for improving OVL feedback accuracy includes the following steps: S910. During the photolithography process, a measurement device is used to scan each alignment mark on the wafer (multiple alignment marks will be formed on a wafer) to obtain the measured mark signal of each alignment mark.

[0082] For example, before the formal start of the exposure sequence for each lithography process, after the wafer is transferred into the lithography machine by a robotic arm and undergoes preliminary pre-alignment, the system immediately transitions to a high-precision fine alignment stage. During this stage, the dedicated measurement system integrated into the lithography machine rapidly scans multiple pre-designed alignment marks distributed on the wafer surface, acquiring and obtaining the optical or diffraction responses of these marks to obtain measured signal waveforms reflecting their actual positions.

[0083] S920. An abnormal alignment mark is detected according to the anomaly detection method of any of the above embodiments. For example: First, the measured mark signal is split into the sum of multiple signals of different orders, and the first-order signal, third-order signal and fifth-order signal are extracted from it. The values ​​of the center of symmetry of the first-order signal, third-order signal and fifth-order signal are calculated respectively to obtain h(1), h(3) and h(5). Then, the differences between h(1), h(3) and h(5) and the predetermined standard values ​​g(1), g(3) and g(5) are calculated respectively. If |h(1)-f(1)| < A(1), |h(3)-f(3)| < A(3), and |h(5)-f(5)| < A(5) are satisfied at the same time, the alignment mark is determined to be normal. Otherwise, the alignment mark is determined to be abnormal.

[0084] S930. After removing the abnormal alignment marks, calculate the OVL of the wafer and determine whether rework is required. If so, return the wafer to the previous process and re-perform photolithography or related processing; otherwise, compensate the process parameters based on the OVL of the wafer.

[0085] In this step, the method for calculating the OVL of the wafer after removing abnormal alignment marks and determining whether rework is required includes the following sub-steps: S931. Output the coordinates of the abnormal alignment markers to the lithography process analysis and control system. For example, the lithography process analysis and control system can use 5DA software (i.e., 5D Analyzer software). 5DA software is a professional tool in the field of semiconductor metrology and data analysis. It is a powerful software platform for advanced lithography process analysis and control, especially in lithography alignment and overlay error analysis. The core task of 5DA software is to process, analyze, and model the large amount of alignment and overlay data collected from the lithography machine and metrology equipment to achieve nanometer-level process control. 5DA software can also calculate the optimal correction parameters (i.e., the optimal OVL compensation value), generate a compensation file, and output it to the lithography machine for real-time compensation during the next exposure.

[0086] S932. The photolithography process analysis and control system removes the corresponding alignment marks according to the coordinates to obtain the remaining alignment marks (i.e., normal alignment marks). Since each layer has multiple alignment marks, after removing abnormal alignment marks, the number of remaining alignment marks is still sufficient to meet the alignment requirements.

[0087] S933. Calculate the OVL of the wafer based on the remaining alignment marks. That is, in subsequent data analysis, the 5DA software will automatically ignore or "filter out" the data of these damaged alignment marks to avoid abnormal and unreliable measurement data caused by damaged marks from polluting the overall analysis model, and to ensure that the calculated process compensation value is based on the signal of normal and reliable alignment marks.

[0088] S934. The photolithography process analysis and control system determines whether the OVL error is within the process specification limits. If so, the overlay accuracy is considered acceptable, rework is not required, and the wafer can continue to the next process. Otherwise, it indicates that the wafer's overlay error is still too large and cannot meet the yield requirements. The overlay accuracy is considered unacceptable, and the wafer needs to be reworked (i.e., returned to the previous process for photolithography or related processing) to meet the process requirements.

[0089] The method for compensating for process parameters based on the OVL error of the wafer includes the following sub-steps: S935, the photolithography process analysis and control system calculates the optimal compensation value (i.e., the parameter that can optimally compensate for the remaining OVL error observed this time) based on the OVL error of the wafer. That is, when an OVL deviation is detected, the 5DA software will calculate the optimal compensation parameter for subsequent process adjustments.

[0090] S936. The photolithography process analysis and control system sends the calculated optimal compensation value to the APC system. The APC system is an advanced process control system in semiconductor manufacturing. It can achieve closed-loop control of the process by collecting process data in real time, building models, and automatically adjusting process parameters.

[0091] The S937 and APC systems update their internal prediction models based on the optimal compensation value and automatically fine-tune the production parameters of subsequent wafer batches (such as the alignment parameters of the lithography machine), achieving feedforward control. When the optimal OVL compensation value of the wafer is input as feedback data into the APC system, the system records the current process status, updates the process model, automatically adjusts the process parameters of subsequent wafers, and achieves continuous optimization and stable control of the process. Through the closed-loop feedback of the APC system, it continuously improves lithography alignment accuracy, increases product yield, and reduces process fluctuations. In this way, the system becomes increasingly "intelligent," proactively preventing the recurrence of similar errors.

[0092] In summary, the alignment mark anomaly detection method of this application has the following unexpected technical effects: This application perfectly embodies the intelligence and adaptability of modern semiconductor manufacturing, capable of identifying and eliminating invalid data (Mark signals generated by damaged alignment marks), ensuring the reliability of decision-making. Based on real-time measurement results, it automatically makes a "release" or "rework" decision, ensuring quality. It can also feed back the compensation value of qualified batches to the APC system, enabling the production system to learn and evolve, thereby continuously reducing process fluctuations and improving overall yield and consistency. It achieves a complete intelligent control closed loop from "problem identification" to "data cleaning," then to "quality judgment" and "process self-optimization," contributing to the realization of high-yield, high-efficiency standard operating procedures.

Claims

1. A method for detecting anomalies in alignment marks, characterized in that, Includes the following steps: The measured mark signal of the alignment mark is obtained; the mark signal is the physical response signal generated by the detection system after scanning the alignment mark on the wafer. It is a quantifiable electrical signal that is received by the sensor and converted into a quantifiable electrical signal after the physical morphology of the alignment mark interacts with the detection beam. The measured mark signal is decomposed into the sum of multiple signals of different orders. Multiple signals of specific orders are extracted from them, and the value of the center of symmetry of each specific order signal is calculated and denoted as h(x); where x represents the order. Calculate the difference between h(x) and a predetermined standard value. If the difference is greater than the deviation threshold, the alignment mark is determined to be abnormal.

2. The method for detecting anomalies in alignment marks according to claim 1, characterized in that: In the step of splitting the measured mark signal into a sum of signals of different orders, a polynomial fitting method is used to split the measured mark signal into a sum of signals of different orders.

3. The method for detecting anomalies in alignment marks according to claim 1, characterized in that: The plurality of specific order signals include first-order signals, third-order signals, and fifth-order signals; or The multiple specific order signals include second-order signals, fourth-order signals, and sixth-order signals.

4. The method for detecting anomalies in alignment marks according to claim 1, characterized in that: The standard value and deviation threshold are obtained by processing and calculating the raw signal after scanning multiple normal alignment marks with a measurement device in advance.

5. The method for detecting anomalies in alignment marks according to any one of claims 1 to 4, characterized in that, The method for pre-determining standard values ​​includes the following steps: Use a measurement device to scan multiple normal alignment marks, and use the original signal obtained from each scan as the standard mark signal; The standard mark signal is preprocessed to obtain the reference signal; The reference signal is decomposed into the sum of multiple signals of different orders. The signal of a specific order corresponding to h(x) is extracted from it, and the value of the center of symmetry of each specific order signal is calculated and denoted as f(x). The average value g(x) of f(x) corresponding to each reference signal is calculated as the standard value.

6. The method for detecting anomalies in alignment marks according to claim 5, characterized in that, The step of preprocessing the standard mark signal to obtain the reference signal includes the following sub-steps: Each of the aforementioned standard mark signals is flattened to obtain a flattened signal; Each of the flattened signals is normalized to obtain a normalized signal; the normalized signal is used as a reference signal.

7. The method for detecting anomalies in alignment marks according to claim 5, characterized in that, The deviation threshold is obtained through the following steps: Based on the fluctuation range of f(x) relative to the standard value corresponding to each reference signal, the fluctuation amplitude of f(x) is obtained, and the fluctuation amplitude A(x) of f(x) is used as the deviation threshold.

8. A method for improving the registration accuracy of photolithography processes, characterized in that, Includes the following steps: During the photolithography process, measurement equipment is used to scan each alignment mark on the wafer to obtain the measured mark signal of each alignment mark; The method for detecting anomalies in alignment marks according to any one of claims 1 to 7 detects abnormal alignment marks; After removing the abnormal alignment marks, calculate the OVL of the wafer and determine whether rework is required; If so, the wafer is returned to the previous process; otherwise, the process parameters are compensated according to the wafer's OVL.

9. The method for improving the registration accuracy of photolithography according to claim 8, characterized in that: The step of calculating the OVL of the wafer after removing the abnormal alignment marks and determining whether rework is required includes the following sub-steps: Output the coordinates of the abnormal alignment mark to the photolithography process analysis and control system; The photolithography process analysis and control system removes the corresponding alignment marks based on the coordinates to obtain the remaining alignment marks; Calculate the OVL of the wafer based on the remaining alignment marks; The photolithography process analysis and control system determines whether the OVL error is within the limits of the process specifications. If it is, the overlay accuracy is considered to be qualified; otherwise, the overlay accuracy is considered to be unqualified.

10. The method for improving the overlay accuracy of photolithography according to claim 8, characterized in that, The step of compensating for process parameters based on the OVL error of the wafer includes the following sub-steps: The optimal compensation value is calculated based on the OVL error of the wafer; Send the calculated optimal complement value to the APC system; The APC system updates its internal prediction model based on the optimal compensation value and automatically fine-tunes the production parameters for subsequent wafer batches.

Images