A pattern-free wafer defect classification method based on multi-feature joint decision

Abstract

This invention discloses a method for classifying defects in patternless wafers based on multi-feature joint decision-making, belonging to the field of wafer defect detection technology. The method includes the following steps: irradiating the surface of a patternless wafer with a probe beam having a preset polarization state; acquiring spatial scattered light intensity information of the scattered light from the wafer surface under different combinations of polarization states based on a probe array distributed at multiple spatial angles; extracting local optical features of the candidate defect regions and constructing a feature tensor based on these local optical features; analyzing the candidate defect regions according to the polarization degradation evolution law of the feature tensor in the spectral dimension and outputting the defect classification result. This approach avoids the low classification accuracy caused by highly similar scattering features of small defects, significantly improving the accuracy of the detection results.

Description

Technical Field

[0001] This invention relates to the field of wafer defect detection technology, and in particular to a method for classifying patternless wafer defects based on multi-feature joint decision-making. Background Technology

[0002] In semiconductor manufacturing processes, wafer surface quality is a fundamental factor affecting device performance and manufacturing yield. Especially in the early stages of the manufacturing process, unpatterned wafers, such as bare silicon wafers or epitaxial wafers, have not yet formed circuit structures, and their surface micro-defects directly impact subsequent critical process steps such as oxidation, deposition, photolithography, and etching. Therefore, high-precision detection and classification of surface defects before the wafer enters the front-end processing has become a crucial part of the wafer fab's quality control system.

[0003] As semiconductor process nodes continue to advance to more advanced stages, the size of wafer surface defects is constantly decreasing, posing significant challenges to the sensitivity and resolution of detection systems. At the nanoscale, different types of defects often exhibit highly similar signal responses in their optical scattering characteristics, making it difficult for detection systems to reliably distinguish them based on a single optical signal.

[0004] To improve the detection capability of extremely small-scale defects, existing inspection systems typically enhance the detectability of optical signals by reducing laser spot size, increasing laser power, or decreasing scanning speed. However, reducing spot size and scanning speed directly prolongs the inspection time of a single wafer, while increasing laser power may affect system stability and equipment lifespan. Therefore, under advanced process conditions, how to quickly detect and distinguish defects while ensuring inspection reliability remains a pressing technical challenge in the field of patternless wafer defect inspection. Summary of the Invention

[0005] The purpose of this invention is to provide a patternless wafer defect classification method based on multi-feature joint decision-making, in order to solve the technical problem of low classification accuracy caused by the high similarity of scattering features of small defects in the prior art.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A method for classifying defects in patternless wafers based on multi-feature joint decision-making includes the following steps: A probe beam with a preset polarization state is irradiated onto the surface of a patternless wafer, the probe beam containing at least two preset wavelengths; Based on a detection array distributed at multiple spatial angles, spatial scattered light intensity information of the wafer surface under different polarization state combinations is collected. The spatial scattered light intensity information includes the same polarization response component and the cross polarization response component, and the defect candidate region is located accordingly. Local optical features of the candidate defect region are extracted and a feature tensor is constructed based on the local optical features. The dimensions of the feature tensor include at least: a spatial dimension composed of the spatial angles and a spectral dimension composed of the preset wavelengths. The defect candidate region is analyzed based on the polarization degradation evolution law of the feature tensor in the spectral dimension. By comparing the changing trend of the depolarization threshold under different preset wavelengths, the surface pit defects with multiple micro-reflection characteristics and surface particle defects with wavelength selective scattering characteristics are distinguished, and the defect classification results are output.

[0007] Further, the step of extracting the local optical features of the defect candidate region and constructing a feature tensor based on the local optical features includes: The same polarization response component and cross polarization response component acquired by the detection array at the first wavelength and the second wavelength are obtained; Calculate the ratio of the cross-polarization response component to the same-polarization response component, and construct a two-dimensional polarization matrix reflecting the defect polarization modulation capability by combining the corresponding spatial angles. The feature tensor is obtained by mapping and stacking the two-dimensional polarization matrices corresponding to different wavelengths along the spectral dimension.

[0008] Furthermore, based on the polarization degradation evolution law of the feature tensor in the spectral dimension, the candidate defect regions are analyzed, including: If the ratio of the feature tensor at the first wavelength and the second wavelength is greater than the preset depolarization threshold, then the defect type is determined to be a crystal native pit with multiple microscopic reflection characteristics. If the ratio of the feature tensor at the first wavelength is greater than the preset depolarization threshold, while the ratio at the second wavelength is not greater than the preset depolarization threshold, then the defect type is determined to be irregular surface particles with wavelength selective scattering characteristics.

[0009] Further, acquiring the same-direction polarization response components and cross-polarization response components collected by the detection array at the first wavelength and the second wavelength includes: From the detection array, signals collected by the detection channels with pitch angles within the first pitch angle range are selected as a low-angle scattered signal set to represent the surface smooth defect characteristics. From the detection array, signals collected by detection channels with pitch angles within the second pitch angle range are selected as a high-angle scattered signal set to represent the characteristics of deep micro-pit defects. Wherein, the upper limit of the first pitch angle range is less than the lower limit of the second pitch angle range; Furthermore, for any signal acquired by any detection channel at any wavelength, its four polarization components are analyzed and obtained: the incident light is S-polarized, and the received light is the first co-polarized component. The incident light is P-polarized and the received light is the second component of the same polarization. The incident light is S-polarized and the received light is P-polarized, forming the first cross component. And the second cross component where the incident light is P-polarized and the received light is S-polarized. , Both the low-angle scattered signal set and the high-angle scattered signal set contain the corresponding four polarization components.

[0010] Furthermore, the acquisition of spatial scattered light intensity information of the wafer surface scattered light under different polarization state combinations based on the detection array distributed at multiple spatial angles includes: The detection channel is set with at least two discrete preset pitch angles, and at least one polarization analysis detection unit is configured at each preset pitch angle; The polarization analysis detection unit is used to synchronously or time-divisionally acquire the same-direction polarization response component and cross-polarization response component of the scattered light.

[0011] Furthermore, after extracting the local optical features of the defect candidate region, a background scattering suppression step is also included: Obtain the baseline surface roughness data around the candidate defect region; Calculate the dynamic depolarization noise threshold based on the surface roughness background data; The polarization response ratio extracted from the polarization state combination is denoised and corrected based on the dynamic depolarization noise threshold to obtain the true polarization components characterizing the defect body features.

[0012] Furthermore, the step of analyzing the candidate defect region based on the polarization degradation evolution law of the feature tensor in the spectral dimension further includes: If the ratio of the feature tensor at the first wavelength and the second wavelength is not greater than the preset depolarization threshold, and its spatial distribution is concentrated at low pitch angles, then the defect type is determined to be a surface scratch. If the ratio of the characteristic tensor at the first wavelength and the second wavelength exhibits periodic modulation related to the azimuth angle, then the defect type is determined to be lattice damage with an oriented structure.

[0013] Further, the step of analyzing the defect candidate region based on the polarization degradation evolution law of the feature tensor in the spectral dimension includes: calculating the defect candidate region at each preset wavelength. The overall polarization degradation index is as follows : ; in, and Let represent the elevation angle and azimuth angle of the i-th detection channel, respectively; and At wavelength ,angle( Cross-polarization and co-polarization response components acquired under ( ) conditions; The weighting function is angle-dependent, and the value of the weighting function varies with the pitch angle. Increase as it grows.

[0014] Further, the step of calculating the dynamic depolarization noise threshold based on the surface roughness background data includes: Based on the background scattering signals of the detection array at the first wavelength and the second wavelength, the surface roughness background at the first wavelength and the surface roughness background at the second wavelength are obtained respectively. Based on the scattering dependence of wafer surface micro-roughness and probe beam wavelength, the first dynamic depolarization noise threshold corresponding to the first wavelength surface roughness background and the second dynamic depolarization noise threshold corresponding to the second wavelength surface roughness background are calculated independently. Wherein, the first wavelength is a short wavelength, the second wavelength is a long wavelength, and the first dynamic depolarization noise threshold is used to characterize the strong random depolarization background of short wavelength excitation.

[0015] Further, the step of denoising and correcting the polarization response ratio extracted from the polarization state combination based on the dynamic depolarization noise threshold to obtain the true polarization components characterizing the defect body features includes: At the first wavelength, the first dynamic depolarization noise threshold is used for denoising to separate the abnormal cross-polarization component caused by Mie resonance at the sharp edge angle of the extremely small irregular surface particles from the strong random depolarization background. At the second wavelength, the second dynamic depolarization noise threshold is used for denoising to extract the dipole polarization-preserving components recovered by the irregular surface particles under long-wavelength irradiation.

[0016] The present invention provides a method for classifying defects in patternless wafers based on multi-feature joint decision-making, which has the following core advantages compared with the prior art: First, a probe beam containing at least two preset wavelengths and preset polarization states is irradiated onto the surface of a patternless wafer. Simultaneously, a probe array distributed at multiple spatial angles collects both co-polarized and cross-polarized scattering components. This expands the originally single-dimensional light intensity information into a multi-dimensional scattering response encompassing polarization state, spatial angle, and wavelength, thereby characterizing the differentiated physical behavior of defects under various electromagnetic excitation conditions. Based on this, a feature tensor containing both spatial and spectral dimensions is constructed for the defect candidate region. Multi-angle and multi-wavelength information is fused, transforming the scattering characteristics from discrete measurements into an intrinsically correlated expression, significantly enhancing the characterization ability of weak differential signals. Furthermore, by analyzing the spectral characteristics of the feature tensor... This invention investigates the evolution of polarization degradation across dimensions and compares the relationship between the changing trends of polarization response ratios at different wavelengths and the depolarization threshold. This transforms defect discrimination from relying on absolute light intensity to a mechanism-based judgment based on changing trends. Crystal surface pits undergo rapid polarization degradation due to multiple microscopic reflections, while surface particles exhibit relatively stable or selective polarization responses due to size-related scattering effects, thus forming distinguishable features at the trend level. Therefore, this invention achieves stable differentiation of minute defects even under complex process fluctuations without increasing optical power or reducing scanning speed. It fundamentally solves the technical problem of low classification accuracy caused by highly similar scattering characteristics of minute defects in existing technologies, significantly improving the accuracy of detection results. Attached Figure Description

[0017] Figure 1 This is a flowchart of a patternless wafer defect classification method based on multi-feature joint decision-making according to the present invention; Figure 2 This is a graph showing the evolution trend of the comprehensive polarization degradation index of different types of defects as a function of wavelength in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0019] Combined with appendix Figure 1 - Appendix Figure 2The specific embodiments of the present invention will be further described in detail below.

[0020] Reference Figure 1 In one exemplary embodiment, this application provides a method for classifying defects in patternless wafers based on multi-feature joint decision-making. It is mainly applicable to the detection and classification of surface micro-defects in patternless wafers such as bare silicon wafers or epitaxial wafers during semiconductor manufacturing. It is particularly suitable for high-precision differentiation of various types of defects such as particle defects, crystal native pits (COP), scratches, and lattice damage.

[0021] The above-mentioned method for classifying defects in patternless wafers specifically includes the following steps: Step 101: Irradiate a probe beam with a preset polarization state onto the surface of the patternless wafer, wherein the probe beam contains at least two preset wavelengths.

[0022] Specifically, in this embodiment, the preset polarization state refers to the electric field vibration direction of the incident light being artificially controlled to a certain state, such as S-polarization perpendicular to the incident surface or P-polarization parallel to the incident surface. This setting is used to excite the scattering differences of the defect under different polarization conditions. The at least two preset wavelengths refer to the probe beam containing light with different center wavelengths, such as a combination of short wavelengths and long wavelengths, which correspond to different scattering sensitivity scales, thereby making the defect exhibit differentiated responses in the spectral dimension.

[0023] Furthermore, the at least two preset wavelengths include a first wavelength and a second wavelength, wherein the first wavelength is a relatively short wavelength used to enhance the scattering sensitivity to small-sized particle defects, and the second wavelength is a relatively long wavelength used to enhance the response capability to deep structures or multiple reflection defects. This step enables multi-channel response excitation of defects under different electromagnetic excitation conditions.

[0024] Specifically, the illumination section can employ a tunable laser source or a combination of multiple laser sources to output short-wavelength and long-wavelength probe beams respectively. A polarization modulation module, such as a polarizer or electro-optic modulator, sets the incident light to different polarization states. By setting a polarization state control module, the incident light can be rapidly switched between S-polarization and P-polarization, or a dual-channel simultaneous emission method can be used to achieve polarization multiplexing, thereby acquiring complete polarization response information within the same scan cycle. Short-wavelength light, due to its smaller wavelength scale, has a stronger scattering response to tiny particles with sizes close to or smaller than its wavelength, while long-wavelength light penetrates surface microstructures more easily and is sensitive to multiple reflection paths. Therefore, the combination of the two can form complementary excitation detection for different types of defects. Furthermore, by employing time-division multiplexing or synchronous dual-polarization emission, the wafer surface is electromagnetically excited under different wavelengths and polarization state combinations, thereby stimulating differences in the scattering response of defects under different conditions, providing fundamental data for subsequent multi-dimensional feature extraction.

[0025] Step 102: Based on the detection array distributed at multiple spatial angles, collect the spatial scattered light intensity information of the scattered light on the wafer surface under different polarization state combinations, wherein the spatial scattered light intensity information includes the same polarization response component and the cross polarization response component, and locate the candidate defect region accordingly.

[0026] Specifically, in this embodiment, multiple spatial angles refer to the distribution of the detection array in different pitch and azimuth directions to obtain the angular distribution characteristics of the scattered light; the same polarization response component refers to the scattered component with the same incident and received polarization states, reflecting the defect's ability to maintain polarization; the cross polarization response component refers to the scattered component with different incident and received polarization states, reflecting the defect's ability to convert polarization; by simultaneously acquiring the above components, basic data reflecting the polarization modulation behavior of the defect can be formed; locating the defect candidate region refers to the preliminary identification of the potential defect location based on the abnormal abrupt change in the spatial distribution of scattering intensity.

[0027] Step 103: Extract the local optical features of the defect candidate region and construct a feature tensor based on the local optical features. The dimensions of the feature tensor include at least: a spatial dimension composed of the spatial angle and a spectral dimension composed of the preset wavelength.

[0028] Specifically, local optical features refer to the polarization scattering response at different spatial angles and wavelengths within the defect candidate region; the feature tensor is a high-dimensional data structure that provides a structured representation of the aforementioned multi-source data, where the spatial dimension corresponds to different angular positions and the spectral dimension corresponds to different wavelengths; this tensor can uniformly describe the spatial and spectral correspondence of scattering behavior.

[0029] The method of collecting spatial scattered light intensity information of the wafer surface scattered light under different polarization state combinations based on a detection array distributed at multiple spatial angles includes: setting detection channels with at least two discrete preset pitch angles, and configuring at least one polarization analysis detection unit at each preset pitch angle; the polarization analysis detection unit is used to synchronously or time-divisionally acquire the same polarization response component and cross polarization response component of the scattered light.

[0030] In some specific embodiments of the present invention, a first pitch angle scattering component in the range of 10°-30° is obtained by a low pitch angle detection array to capture surface smooth defect features; a second pitch angle scattering component in the range of 60°-80° is obtained by a high pitch angle detection array to capture micro-pit defect features; and an azimuth angle scattering component in the range of 0°-360° is obtained by an omnidirectional detection array.

[0031] Further, the step of extracting the local optical features of the defect candidate region and constructing a feature tensor based on the local optical features includes: obtaining the co-polarization response component and the cross-polarization response component acquired by the detection array at the first wavelength and the second wavelength; calculating the ratio of the cross-polarization response component to the co-polarization response component, and constructing a two-dimensional polarization matrix reflecting the polarization modulation capability of the defect by combining the corresponding spatial angle; mapping and stacking the two-dimensional polarization matrices corresponding to different wavelengths along the spectral dimension to obtain the feature tensor.

[0032] Based on the polarization degradation evolution law of the feature tensor in the spectral dimension, the defect candidate region is analyzed, including: if the ratio of the feature tensor at the first wavelength and the second wavelength is greater than a preset depolarization threshold, the defect type is determined to be a crystal native pit with multiple microscopic reflection characteristics; if the ratio of the feature tensor at the first wavelength is greater than the preset depolarization threshold, but the ratio at the second wavelength is not greater than the preset depolarization threshold, the defect type is determined to be an irregular surface particle with wavelength selective scattering characteristics.

[0033] Specifically, the ratio of the cross-polarization response component to the same-polarization response component can be defined as: R = / The ratio R directly characterizes, in physics, the degree to which the incident polarization state is disrupted during scattering, where... Corresponding polarization-preserving components, Corresponding to the polarization conversion component, a larger R value indicates a stronger depolarization effect during scattering. Based on this definition, the distribution of R values ​​at different spatial angles within the same defect region reflects the modulation method of the polarization state by the defect geometry: for defects with continuous surfaces or weakly undulating structures, scattering is dominated by specular or quasi-spectral components. The dominant polarization factor results in a generally low and gently distributed R value across all angles; however, for defects with sharp edges or multiple reflection paths, significant local polarization coupling occurs, causing... It is significantly enhanced at a specific angle, thus forming a local high-value region in the two-dimensional polarization matrix.

[0034] Furthermore, R is expanded in space to form a two-dimensional polarization matrix M, constructing the angle-polarization response distribution of the defect under single-wavelength excitation. The numerical distribution in matrix M not only contains intensity information, but more importantly, its spatial structural characteristics: for example, the concentration of high-value regions, their distribution location, and their extension range in the pitch direction correspond to the local curvature of the defect, scattering directionality, and the presence of deep structures, respectively.

[0035] Based on this, the two-dimensional polarization matrices at different wavelengths are stacked along the spectral dimension to obtain the characteristic tensor T, which is essentially a unified expression of the scattering behavior of the same defect under electromagnetic excitation at different scales. Since different defect types have different response mechanisms to wavelength, the tensor exhibits significant differences in its spectral dimension: for particulate defects, their scattering intensity and polarization conversion capability are relatively sensitive to wavelength changes, typically showing a generally high R-value at short wavelengths and a significant decrease at long wavelengths, resulting in a decaying distribution of the tensor in the spectral dimension; while for crystal native pits, due to the scale insensitivity of polarization disruption caused by multiple internal reflection paths, their R-value remains high at different wavelengths, resulting in a stable high-value distribution of the tensor in the spectral dimension.

[0036] Further, acquiring the same-polarization response components and cross-polarization response components collected by the detection array at the first wavelength and the second wavelength includes: selecting signals collected by detection channels with pitch angles within the first pitch angle range from the detection array as a low-angle scattering signal set to represent surface smooth defect features; and selecting signals collected by detection channels with pitch angles within the second pitch angle range from the detection array as a high-angle scattering signal set to represent deep micro-pit defect features. Wherein, the upper limit of the first pitch angle range is less than the lower limit of the second pitch angle range; Furthermore, for any signal acquired by any detection channel at any wavelength, its four polarization components are analyzed and obtained: the incident light is S-polarized, and the received light is the first co-polarized component. The incident light is P-polarized and the received light is the second component of the same polarization. The incident light is S-polarized and the received light is P-polarized, forming the first cross component. And the second cross component where the incident light is P-polarized and the received light is S-polarized. .

[0037] Both the low-angle scattered signal set and the high-angle scattered signal set contain the corresponding four polarization components.

[0038] Specifically, by dividing the detection channel into low-angle and high-angle scattering signal sets according to the elevation angle, it is equivalent to separating the surface response, dominated by specular reflection and weak perturbation, from the deep scattering response, dominated by structural abrupt changes and multiple reflections, during the acquisition phase. Combined with the analysis of the four polarization components of each channel, the polarization preservation capability and polarization conversion capability can be described respectively, thus enabling the same defect to exhibit differentiated polarization responses in different angular subspaces: for surface smooth defects, the low-angle region is dominated by the same-direction component and the cross component is weak, while for deep micro-pit defects, the cross component is significantly enhanced in the high-angle region and forms an opposition with the same-direction component. Therefore, the above-mentioned angular partitioning and polarization component joint characterization enable different defects to form separable response states, providing a discriminative input basis for subsequent polarization ratio calculation and feature tensor construction.

[0039] Step 104: Analyze the candidate defect regions according to the polarization degradation evolution law of the feature tensor in the spectral dimension. By comparing the relationship between the changing trend of the ratio at different preset wavelengths and the depolarization threshold, distinguish between crystal surface pit defects with multiple microscopic reflection characteristics and surface particle defects with wavelength selective scattering characteristics, and output the defect classification results.

[0040] Specifically, the polarization degradation evolution law refers to the trend of the ratio of cross polarization and same polarization with wavelength, which is used to reflect the behavior of the degree of damage of the polarization state by the defect with the change of excitation conditions; the depolarization threshold is a preset or adaptively determined discrimination boundary, which is used to distinguish between strong depolarization and weak depolarization states; by comparing the relationship between this ratio and the threshold at different wavelengths, defect classification based on the changing trend can be realized.

[0041] Specific Implementation Example 1: In the factory full inspection process of 300mm patternless bare silicon wafers, when the optical inspection system captures a weak light spot with an equivalent size of about 15nm on the wafer surface, due to its extremely low scattering signal amplitude, traditional single-wavelength light intensity detection equipment often cannot accurately determine whether the defect is an irregular metal debris attached to the surface or a crystal native pit.

[0042] In this embodiment, the method described in this invention simultaneously activates DUV (deep ultraviolet, e.g., 266 nm) as the first wavelength and VIS (visible light, e.g., 532 nm) as the second wavelength. For the 15 nm defect, in the DUV band, due to the nanometer-scale acute angles at the edges of the metal debris, a strong Mie resonance is induced, causing a drastic reversal in the polarization state of the scattered light, resulting in an extremely high cross-polarization ratio extracted in step 103. Based solely on single-wavelength data, it is easily misclassified as a COP with strong depolarization characteristics. However, by introducing data from the second wavelength (VIS), since the size of the 15 nm defect is much smaller than the VIS wavelength, it exhibits a classic Rayleigh scattering dipole model under long-wavelength excitation, with the cross-polarization signal significantly disappearing, demonstrating good polarization-preserving characteristics. Ultimately, based on the evolution of the characteristic tensor from strong depolarization to extremely weak depolarization in the spectral dimension, the defect can be determined to be an irregular surface particle, thus effectively eliminating the classification ambiguity under single-wavelength conditions.

[0043] In some specific embodiments of the present invention, after extracting the local optical features of the defect candidate region, a background scattering suppression step is further included: obtaining the surface roughness background data around the defect candidate region; calculating the dynamic depolarization noise threshold based on the surface roughness background data; and denoising and correcting the polarization response ratio extracted by the polarization state combination based on the dynamic depolarization noise threshold to obtain the true polarization component characterizing the defect body features.

[0044] Specifically, when extracting cross-polarization response components using particles and micropits, the following fatal interference occurs when the defects are extremely small (e.g., <10 nm): the micro-roughness (Haze) of the wafer surface itself also causes random scattering and depolarization of light waves. At this point, the extremely weak cross-polarization response generated by the micropits (COP) becomes ineffective. The signal is easily overwhelmed by the polarization noise generated by the haze. Furthermore, not all particles are perfect spherical dipoles. When encountering highly asymmetrical or irregular particles, their unique geometry can also cause the scattered light to undergo polarization state rotation, exhibiting strong polarization noise. The components thus masquerade as micro-pits (COPs) in the polarization matrix of a single wavelength, leading to classification failure.

[0045] To address this, a background scattering suppression step is introduced. The core of this step is to separate the effective scattering information generated by the defect itself from the actual measurement signal, and the random scattering background caused by the inherent roughness of the wafer surface. Specifically, firstly, by acquiring the surface roughness background data around the candidate defect region, a local scattering baseline corresponding to the current detection position is established. This baseline reflects the intrinsic depolarization level caused by the surface micro-undulation structure in the absence of significant defects. Since this type of background scattering typically exhibits strong randomness, wavelength dependence, and distribution differences at different spatial angles, it is modeled at different preset wavelengths, and a dynamic depolarization noise threshold is calculated accordingly. This threshold adaptively reflects the true noise level of the current detection area, avoiding misjudgment. Based on this, the ratio of cross-polarization to same-direction polarization response extracted from the polarization state combination is compared with the dynamic depolarization noise threshold. Random depolarization components below the threshold are suppressed, thereby effectively filtering out noise components. Simultaneously, abnormal polarization responses above the threshold are retained and enhanced, enabling them to more accurately express the defect's modulation capability of the polarization state. Through the above denoising correction process, the signal-to-noise ratio of the polarization response ratio can be significantly improved without changing the original data acquisition method.

[0046] By employing the above suppression steps, and by independently denoising different wavelength channels, the abnormal cross-polarization component and the dipole polarization-preserving component are mapped into the feature tensor to eliminate the interference of the pseudo-depolarization characteristics of the irregular surface particles on the determination of the crystal's original pits.

[0047] Further, the step of analyzing the defect candidate region based on the polarization degradation evolution law of the feature tensor in the spectral dimension includes: calculating the defect candidate region at each preset wavelength. The overall polarization degradation index is as follows : ; in, and Let represent the elevation angle and azimuth angle of the i-th detection channel, respectively; and At wavelength ,angle( Cross-polarization and co-polarization response components acquired under ( ) conditions; The weighting function is angle-dependent, and the value of the weighting function varies with the pitch angle. Increase as it grows.

[0048] Specifically, pseudo-defects, such as anomalous signals caused by surface roughness or random scattering, have cross-polarization components that primarily originate from uncorrelated scattering from the randomly rough surface. Lacking a stable angle-dependent structure across different detection channels, they exhibit low amplitude and disordered superposition during weighted summation, resulting in an overall low IPDI and insignificant variation across different wavelengths. Conversely, genuine defects, such as particles or pits, possess definite geometric structures or electromagnetic resonance paths, causing their cross-polarization components to stably enhance at specific angles and amplify under the weighting function, thus reflected in the IPDI. Furthermore, the scattering of pseudo-defects typically exhibits approximately uniform random attenuation or smooth variation with wavelength, while genuine defects show differentiated evolution trends related to structural scale. Therefore, by introducing an angle-weighted comprehensive polarization degradation index and combining it with its consistency across multiple wavelengths, pseudo-defect responses caused by random roughness can be effectively suppressed, enabling stable identification and differentiation of genuine defects.

[0049] Specific Implementation Example 2: During the inspection of a batch of epitaxial wafers, due to process characteristics, the wafer surface has high micro-roughness. When detecting minute defects, the extremely short wavelength of the DUV laser will excite extremely strong random depolarization background noise, causing the scattering polarization signal of the real defect to be easily submerged by the background, resulting in missed detections or false alarms.

[0050] In this embodiment, a background scattering suppression step is first performed. By acquiring the surface roughness background data around the defect candidate region, the dynamic depolarization noise threshold corresponding to the current detection position is calculated. In step 103, when extracting local optical features, a weighting function is used to spatially weight the signals from different detection channels. Since the spurious defect signals caused by surface roughness lack stable angle dependence between different spatial detection channels, they exhibit disordered superposition characteristics during weighted summation; while genuine crystal damage or micro-pit defects have stable polarization enhancement responses at specific high pitch angles. By calculating the comprehensive polarization degradation index and combining it with its evolution consistency discrimination at multiple wavelengths, genuine defect signals above the dynamic threshold can be successfully extracted, and spurious defect responses caused by random roughness can be effectively suppressed. Experimental results show that this method can still achieve stable identification and high-precision differentiation of crystal native pits and surface particles even in environments with strong background noise.

[0051] Reference Figure 2 The X-axis represents wavelength, ranging from 266 nm (deep ultraviolet DUV) to 532 nm (visible light VIS); the Y-axis represents the Integrated Polarization Degradation Index (IPDI), a dimensionless value ranging from 0 to 1. A depolarization threshold of 0.5 is set. IPDI > 0.5 indicates significant depolarization; IPDI < 0.5 indicates strong polarization retention. Figure 2 As can be seen, the intrinsic pits of the crystal, i.e., COP, have an IPDI of 0.85 at a short wavelength of 266 nm, which is greater than the threshold, and an IPDI of 0.92 at a long wavelength of 532 nm, which is also greater than the threshold. Furthermore, the IPDI remains above the threshold line throughout, showing a high and stable or slightly increasing trend. This proves the strong depolarization characteristic of COP defects across the entire wavelength band due to multiple microscopic reflections. In contrast, irregular surface particles have an IPDI of 0.78 at a short wavelength of 266 nm, which is greater than the threshold, and an IPDI of 0.25 at a long wavelength of 532 nm, which is less than the threshold. The starting point is above the threshold line, and the ending point falls below the threshold line, showing a clear decreasing trend. This further proves the strong depolarization of tiny particle defects under Mie resonance at short wavelengths and the polarization-preserving characteristics of Rayleigh scattering at long wavelengths. For surface scratches, the IPDI is 0.22 at a short wavelength of 266nm, which is below the threshold, and 0.18 at a long wavelength of 532nm, also below the threshold. Furthermore, the IPDI remains below the threshold throughout, showing a stable low-level trend. This verifies that scratch defects are mainly characterized by specular reflection or weak scattering, lacking multiple reflection paths and the ability to disrupt the polarization state. In addition, although background noise exists, the signal characteristics of genuine defects are preserved and distinguished through angle weighting, i.e., IPDI calculation, while spurious signals exhibit irregular fluctuations or low values ​​(not shown in the figure) and are thus eliminated. Therefore, Figure 2 This visually demonstrates the wavelength sensitivity of particles and the wavelength insensitivity of COP. The decreasing slope corresponds to the wavelength-selective scattering characteristics, while the stable high value corresponds to multiple microscopic reflection characteristics.

[0052] Furthermore, the analysis of the candidate defect region based on the polarization degradation evolution law of the feature tensor in the spectral dimension further includes: if the ratio of the feature tensor at the first wavelength and the second wavelength is not greater than the preset depolarization threshold, and its spatial distribution is concentrated at low pitch angles, then the defect type is determined to be a surface scratch; if the ratio of the feature tensor at the first wavelength and the second wavelength exhibits periodic modulation related to the azimuth angle, then the defect type is determined to be lattice damage with an oriented structure.

[0053] Specifically, further, the analysis of the candidate defect region based on the polarization degradation evolution law of the feature tensor in the spectral dimension also includes: if the ratio of the feature tensor at the first wavelength and the second wavelength is not greater than the preset depolarization threshold, and its spatial distribution is concentrated at low pitch angles, then the defect type is determined to be a surface scratch; if the ratio of the feature tensor at the first wavelength and the second wavelength exhibits periodic modulation related to the azimuth angle, then the defect type is determined to be lattice damage with an oriented structure.

[0054] Specifically, when the polarization ratio does not exceed the depolarization threshold at either wavelength, it indicates that the scattering process is mainly polarization-preserving, lacking multiple reflections or strong disturbances. Simultaneously, its response is concentrated in the low elevation angle region, indicating that the scattering is mainly distributed along the near-mirror surface direction, corresponding to a continuous and relatively linear surface with minimal undulations; therefore, it can be identified as a surface scratch. Conversely, when the ratio exhibits periodic modulation with azimuth in space, it indicates that the scattering response has a clear directional selectivity, reflecting the presence of regularly oriented structural units within the defect, such as lattice distortion or orientational damage. Such structures anisotropically modulate the incident electromagnetic field, causing the polarization conversion efficiency to periodically change with the azimuth angle, thus forming a stable angular periodic characteristic. Based on this, it can be identified as lattice damage with an oriented structure.

[0055] In some other embodiments of the present invention, a detection system for classifying defects in patternless wafers is also provided, for implementing the patternless wafer defect classification method described above, including: An illumination module is used to emit a probe beam containing at least two preset wavelengths onto the surface of a patternless wafer, and can modulate the polarization state of the probe beam. A multi-channel data acquisition module includes detection units set at multiple different spatial angles. Each detection unit includes a polarization beam splitter and a photodetector, used to acquire the same polarization response component and cross polarization response component of the scattered light at the corresponding angle. The processing module is communicatively connected to the multi-channel data acquisition module and is used to perform the steps described above, such as defect candidate region localization, feature tensor construction, and defect classification analysis.

[0056] Some embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor of a signal processing device, can implement the steps of the method for detecting and classifying wafer defects provided in some embodiments of this application.

[0057] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0058] Furthermore, the signal processing modules in the various embodiments of this application can be integrated into one processing unit, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0059] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the transmission and reception methods described in the various embodiments of this application. The aforementioned storage medium can be a volatile or non-volatile computer-readable storage medium, including: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0060] It is understood that the signal processing modules in the embodiments described in this application can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, modules, units, submodules, subunits, etc., can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0061] The above descriptions are some embodiments of this application. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0062] For details of this embodiment of the application, please refer to the description of the foregoing method embodiments. To avoid repetition, the description will not be repeated here.

[0063] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0064] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the present invention. Furthermore, the present invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A method for classifying defects in patternless wafers based on multi-feature joint decision-making, characterized in that, Includes the following steps: A probe beam with a preset polarization state is irradiated onto the surface of a patternless wafer, the probe beam containing at least two preset wavelengths; Based on a detection array distributed at multiple spatial angles, spatial scattered light intensity information of the wafer surface under different polarization state combinations is collected. The spatial scattered light intensity information includes the same polarization response component and the cross polarization response component, and the defect candidate region is located accordingly. Local optical features of the candidate defect region are extracted and a feature tensor is constructed based on the local optical features. The dimensions of the feature tensor include at least: a spatial dimension composed of the spatial angles and a spectral dimension composed of the preset wavelengths. The defect candidate region is analyzed based on the polarization degradation evolution law of the feature tensor in the spectral dimension. By comparing the changing trend of the depolarization threshold under different preset wavelengths, the surface pit defects with multiple micro-reflection characteristics and surface particle defects with wavelength selective scattering characteristics are distinguished, and the defect classification results are output.

2. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 1, characterized in that, The step of extracting the local optical features of the defect candidate region and constructing a feature tensor based on the local optical features includes: The same polarization response component and the cross polarization response component are acquired by the detection array at the first wavelength and the second wavelength. Calculate the ratio of the cross-polarization response component to the same-polarization response component, and construct a two-dimensional polarization matrix reflecting the defect polarization modulation capability by combining the corresponding spatial angles. The feature tensor is obtained by mapping and stacking the two-dimensional polarization matrices corresponding to different wavelengths along the spectral dimension.

3. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 2, characterized in that, Based on the polarization degradation evolution law of the feature tensor in the spectral dimension, the defect candidate region is analyzed, including: If the ratio of the feature tensor at the first wavelength and the second wavelength is greater than the preset depolarization threshold, then the defect type is determined to be a crystal native pit with multiple microscopic reflection characteristics. If the ratio of the feature tensor at the first wavelength is greater than the preset depolarization threshold, while the ratio at the second wavelength is not greater than the preset depolarization threshold, then the defect type is determined to be irregular surface particles with wavelength selective scattering characteristics.

4. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 2, characterized in that, Acquiring the same-direction polarization response components and cross-polarization response components collected by the detection array at the first wavelength and the second wavelength includes: From the detection array, signals collected by the detection channels with pitch angles within the first pitch angle range are selected as a low-angle scattered signal set to represent the surface smooth defect characteristics. From the detection array, signals collected by detection channels with pitch angles within the second pitch angle range are selected as a high-angle scattered signal set to represent the characteristics of deep micro-pit defects. Wherein, the upper limit of the first pitch angle range is less than the lower limit of the second pitch angle range; Furthermore, for any signal acquired by any detection channel at any wavelength, its four polarization components are analyzed and obtained: the incident light is S-polarized, and the received light is the first co-polarized component. The incident light is P-polarized and the received light is the second component of the same polarization. The incident light is S-polarized and the received light is P-polarized, forming the first cross component. And the second cross component where the incident light is P-polarized and the received light is S-polarized. , Both the low-angle scattered signal set and the high-angle scattered signal set contain the corresponding four polarization components.

5. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 2, characterized in that, The detection array, distributed at multiple spatial angles, collects spatial scattered light intensity information of the wafer surface under different polarization state combinations, including: The detection channel is set with at least two discrete preset pitch angles, and at least one polarization analysis detection unit is configured at each preset pitch angle; The polarization analysis detection unit is used to synchronously or time-divisionally acquire the same-direction polarization response component and cross-polarization response component of the scattered light.

6. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 1, characterized in that, After extracting the local optical features of the defect candidate region, a background scattering suppression step is also included: Obtain the baseline surface roughness data around the candidate defect region; Calculate the dynamic depolarization noise threshold based on the surface roughness background data; The polarization response ratio extracted from the polarization state combination is denoised and corrected based on the dynamic depolarization noise threshold to obtain the true polarization components characterizing the defect body features.

7. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 2, characterized in that, The step of analyzing the candidate defect region based on the polarization degradation evolution law of the feature tensor in the spectral dimension further includes: If the ratio of the feature tensor at the first wavelength and the second wavelength is not greater than the preset depolarization threshold, and its spatial distribution is concentrated at low pitch angles, then the defect type is determined to be a surface scratch. If the ratio of the characteristic tensor at the first wavelength and the second wavelength exhibits periodic modulation related to the azimuth angle, then the defect type is determined to be lattice damage with an oriented structure.

8. The method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 1, characterized in that, The step of analyzing the defect candidate region based on the polarization degradation evolution law of the feature tensor in the spectral dimension includes: calculating the defect candidate region at each preset wavelength. The overall polarization degradation index is as follows : ； in, and Let represent the elevation angle and azimuth angle of the i-th detection channel, respectively; and At wavelength ,angle( Cross-polarization and co-polarization response components acquired under ( ) conditions; The weighting function is angle-dependent, and the value of the weighting function varies with the pitch angle. Increase as it grows.

9. A method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 8, characterized in that, The step of calculating the dynamic depolarization noise threshold based on the surface roughness background data includes: Based on the background scattering signals of the detection array at the first wavelength and the second wavelength, the surface roughness background at the first wavelength and the surface roughness background at the second wavelength are obtained respectively. Based on the scattering dependence of wafer surface micro-roughness and probe beam wavelength, the first dynamic depolarization noise threshold corresponding to the first wavelength surface roughness background and the second dynamic depolarization noise threshold corresponding to the second wavelength surface roughness background are calculated independently. Wherein, the first wavelength is a short wavelength, the second wavelength is a long wavelength, and the first dynamic depolarization noise threshold is used to characterize the strong random depolarization background of short wavelength excitation.

10. A method for classifying defects in patternless wafers based on multi-feature joint decision-making according to claim 9, characterized in that, The step of denoising and correcting the polarization response ratio extracted from the polarization state combination based on the dynamic depolarization noise threshold to obtain the true polarization components characterizing the defect body features includes: At the first wavelength, the first dynamic depolarization noise threshold is used for denoising to separate the abnormal cross-polarization component caused by Mie resonance at the sharp edge angle of the extremely small irregular surface particles from the strong random depolarization background. At the second wavelength, the second dynamic depolarization noise threshold is used for denoising to extract the dipole polarization-preserving components recovered by the irregular surface particles under long-wavelength irradiation.

Images