A method and system for extracting surface defects of semiconductor wafers using single-pixel imaging

Abstract

This invention discloses a method and system for extracting surface defects on semiconductor wafers using single-pixel imaging, relating to the field of wafer defect extraction technology. The method includes: performing dark current subtraction and brightness flat-field correction on a bright-field image of a defect-free wafer; generating a reference mask; dividing the image into tiles and establishing coordinate mapping; generating defect-sensitive pattern pairs; projecting them tile by tile and integrating them with a single-pixel detector; obtaining phase shift difference groups through complementary differential methods; demodulating the phase-locked loop (PLL) to obtain in-phase and quadrature components using a four-step phase-shifting method; estimating the global coordinates of the defect based on the phase and frequency slopes; and constructing a frequency domain suppression weight using the Fourier spectrum of the defect-free image to obtain saliency; and determining the defect type through adaptive adjustment of the distance between the output end and the tile and secondary differential saliency. This invention directly solves for the in-phase / quadrature components and amplitude-phase distribution using a four-step phase-shifting method, and obtains the defect coordinates using the phase-frequency slope, avoiding image reconstruction, reducing computational overhead, and improving positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of wafer defect extraction technology, and in particular to a method and system for extracting surface defects of semiconductor wafers using single-pixel imaging. Background Technology

[0002] In recent years, semiconductor wafer surface inspection has evolved from bright / dark field microscopy and threshold segmentation to a computational optical path that combines structured light modulation, single-pixel detection, and phase shift demodulation. Coherent lasers generate spatially frequency-controllable illumination through digital micromirror devices, and in-phase / quadrature components can be directly obtained by combining the four-step phase shift method. The Fourier shift theorem is used to characterize spatial displacement with phase-frequency slope to achieve sub-pixel-level positioning.

[0003] However, conventional processes often involve reconstructing a 2D image before performing anomaly extraction or convolutional segmentation, which is computationally intensive and sensitive to non-uniform illumination, dark current drift, and ambient light. Defect type identification often relies on empirical thresholds and texture heuristics, making it susceptible to background periodic structures and noise, thus compromising both localization accuracy and throughput. There is an urgent need for a robust localization and type determination method that can be performed without image reconstruction under low signal-to-noise ratio and non-uniform illumination conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a single-pixel imaging method for extracting defects on the surface of semiconductor wafers to solve the problem of inaccurate defect localization under non-uniform illumination and background drift conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for extracting defects on the surface of a semiconductor wafer using single-pixel imaging, comprising,

[0008] The surface grayscale image of the defect-free semiconductor wafer is preprocessed to obtain a reference mask. The semiconductor wafer under test is divided into several test tiles, and a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test is constructed.

[0009] The coherent laser is modulated to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, a defect-sensitive pattern pair is generated.

[0010] The defect-sensitive pattern pairs are projected onto the tile under test, and the reflected signals are collected using a single-pixel detector. The phase shift difference groups are obtained through complementary differential operations.

[0011] Phase-locked demodulation is performed on the phase shift difference groups to obtain the in-phase and quadrature components, and the frequency domain amplitude and phase distribution are calculated. Based on the linear relationship between phase and frequency, the coordinates of the defect location are determined.

[0012] By constructing a frequency domain suppression weighting function, the amplitude of the reflected light from the tile under test is weighted and calculated to obtain the defect saliency. By adaptively adjusting the distance between the output light port and the tile under test, the defect saliency sequence at different distances is obtained. The second difference of the defect saliency is calculated to identify the defect type.

[0013] In a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the modulation of the coherent laser comprises the following specific steps.

[0014] Based on a two-dimensional sine function, a two-dimensional sine grayscale image is generated by computer and then loaded by a digital micromirror device;

[0015] Coherent laser light is reflected by a digital micromirror device and its spatial amplitude is modulated according to a two-dimensional sinusoidal grayscale image to obtain modulated coherent laser light.

[0016] As a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging according to the present invention, the specific steps for generating defect-sensitive pattern pairs are as follows:

[0017] The reference mask is multiplied pixel by pixel with the light field distribution function to obtain the defect-sensitive pattern;

[0018] A complementary pattern is generated for each frame of the defect-sensitive pattern to obtain a defect-sensitive pattern pair.

[0019] As a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps for determining the defect location coordinates based on the linear relationship between phase and frequency are as follows:

[0020] Along the frequency monotonic path and The phase set at different frequencies along the coordinate direction is continuously expanded point by point to obtain the expanded phase;

[0021] Calculate the sample mean of the expanded phase and frequency, and perform least squares linear fitting to obtain the slope of the linear relationship between phase and frequency;

[0022] The local coordinates of the defect in the tile under test are calculated based on the slope of the linear relationship between phase and frequency.

[0023] Based on the coordinate mapping function between the tile under test and the semiconductor wafer under test, the local coordinates on the tile under test are converted into global coordinates.

[0024] In a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps for constructing the frequency domain suppression weight function are as follows:

[0025] A two-dimensional Fourier transform is performed on the defect-free image after dark current subtraction, brightness flat field correction and global normalization to obtain the defect-free spectrum;

[0026] A frequency domain suppression weighting function is constructed based on the amplitude of the defect-free spectrum.

[0027] As a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps of adaptively adjusting the distance between the output light port and the tile under test are as follows:

[0028] Starting from the initial distance between the output light port and the tile under test, a monotonically decreasing distance sequence is set, and the defect saliency corresponding to each distance in the distance sequence is calculated and summarized to obtain the defect saliency sequence.

[0029] Calculate the normalized increment of defect significance at two adjacent distance points in the distance sequence;

[0030] Using defect-free tiles as the calibration object, an increment threshold is set based on the mean and standard deviation distribution of the normalized increment of defect significance of defect-free tiles.

[0031] If the normalized increment of the test tile at adjacent distance points is not greater than the increment threshold, then the complete defect saliency sequence and distance sequence are output. If the normalized increment of two adjacent distance points in the test tile is greater than the increment threshold, then distance point compensation is performed.

[0032] In a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps for setting the incremental threshold are as follows:

[0033] Calculate the defect significance and normalized increment of defect-free tiles at the same distance sequence as the tile under test;

[0034] Calculate the mean and standard deviation of the normalized increment of defect-free tiles, and set the single-distance increment threshold based on the three-times-standard-deviation principle;

[0035] Take the maximum value among all single-distance increment thresholds as the global increment threshold.

[0036] In a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps for identifying the defect type are as follows:

[0037] Calculate the quadratic difference of each defect significance in the complete defect significance sequence;

[0038] If the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is greater than zero, it is determined to be a point defect; if the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is equal to zero, it is determined to be a linear defect; if the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is less than zero, it is determined to be a planar defect.

[0039] In a preferred embodiment of the semiconductor wafer surface defect extraction method for single-pixel imaging described in this invention, the specific steps for obtaining the reference mask are as follows:

[0040] Acquire surface grayscale images of defect-free semiconductor wafers as defect-free image outputs;

[0041] Dark current subtraction and brightness flat field correction are performed on the defect-free image, and global normalization is performed and the reciprocal is taken to obtain the reference mask.

[0042] In a second aspect, the present invention provides a semiconductor wafer surface defect extraction system for single-pixel imaging, comprising,

[0043] The reference mask module is used to preprocess the surface grayscale image of the defect-free semiconductor wafer to obtain the reference mask, divide the semiconductor wafer under test into several test tiles, and construct a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test.

[0044] The laser modulation module is used to modulate the coherent laser to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, a defect-sensitive pattern pair is generated.

[0045] The laser projection module is used to project defect-sensitive pattern pairs onto the tile under test, collect the reflected signals using a single-pixel detector, and obtain phase shift difference groups through complementary differential operations;

[0046] The defect location module is used to perform phase-locked demodulation on the phase shift difference grouping to obtain the in-phase and quadrature components, calculate the frequency domain amplitude and phase distribution, and determine the defect location coordinates based on the slope relationship between the phase and frequency.

[0047] The defect type determination module is used to construct a frequency domain suppression weight function to calculate the amplitude of the reflected light from the tile under test, obtain the defect significance, and obtain the defect significance sequence at different distances by adaptively adjusting the distance between the output light port and the tile under test. The module then calculates the second difference of the defect significance to determine the defect type.

[0048] The beneficial effects of this invention are as follows: by forming a reference mask, the defect-free response is suppressed while the defect disturbance is optically amplified; the four-step phase-shifting method is used to directly solve the in-phase / quadrature components and amplitude-phase distribution; the defect coordinates are obtained by phase-frequency slope, avoiding image reconstruction, reducing computational overhead and improving positioning accuracy; the defect saliency sequence is obtained by distance adaptive scanning; and the gain criterion is clearly defined by quadratic difference to determine the point / line / surface type, which is more physically interpretable and robust than the empirical threshold method. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of a method for extracting surface defects on semiconductor wafers using single-pixel imaging.

[0051] Figure 2 This is a flowchart for generating coherent laser modulation and defect-sensitive pattern pairs.

[0052] Figure 3 This is a flowchart of the projection and phase-locked demodulation of a defect-sensitive pattern.

[0053] Figure 4 This is a flowchart for distance adaptive adjustment and defect type determination. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] Reference Figures 1-4As one embodiment of the present invention, this embodiment provides a method for extracting defects on the surface of a semiconductor wafer using single-pixel imaging, comprising the following steps:

[0058] S1. Preprocess the surface grayscale image of the defect-free semiconductor wafer to obtain a reference mask. At the same time, divide the semiconductor wafer under test into several test tiles and construct a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test.

[0059] A bright-field microscopy imaging device was used to acquire surface grayscale images of defect-free semiconductor wafers under uniform white light illumination, which were then output as defect-free images.

[0060] Dark current subtraction is performed on defect-free images using dark field image subtraction to eliminate background current drift. Specifically, under completely dark conditions, dark field images of defect-free semiconductor wafers are acquired using the same exposure time, and dark current is subtracted pixel by pixel from the defect-free images using the dark field images.

[0061] Brightness flat field correction is performed on the defect-free image after dark current subtraction using the white board reference normalization method. Specifically, under the same optical conditions, a reference image of a uniform standard white board is acquired. The defect-free image after dark current subtraction and the reference image are normalized to eliminate the non-uniformity caused by the light source or bright field microscopy imaging device, so that the brightness distribution of the corrected defect-free image is uniform and only the surface features of the semiconductor wafer itself are retained.

[0062] For a defect-free image that has undergone dark current subtraction and brightness flat field correction, global normalization is performed and the reciprocal is taken to obtain the reference mask;

[0063] It should be noted that by global normalization and taking the reciprocal, the optical response of the defect-free area is naturally weakened, while the local disturbance introduced by the defect cannot be completely canceled, thus forming an optical magnification effect on the defect during the detection stage. This is fundamentally different from the existing technology that relies solely on image post-processing.

[0064] The surface of the semiconductor wafer under test is divided into several square tile regions. Each tile overlaps with its adjacent boundaries at a preset ratio, such as 10% overlap, to avoid missing edge regions. A coordinate mapping function between each tile and the global coordinate system of the semiconductor wafer is established.

[0065] S2. Modulate the coherent laser to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, generate a pair of defect-sensitive patterns.

[0066] After collimation and beam expansion, the coherent laser beam is incident on a digital micromirror device (DMM). Spatial modulation of the coherent laser beam is then performed on a two-dimensional plane to obtain the optical field distribution function of the modulated coherent laser. Specifically, based on a two-dimensional sine function, a two-dimensional sine grayscale image is generated by a computer and loaded by the DMM. After the coherent laser irradiates the DMM, spatial amplitude modulation is performed according to the two-dimensional sine grayscale image to obtain the modulated coherent laser. The optical field distribution function of the modulated coherent laser is calculated as follows:

[0067]

[0068]

[0069] in, Indicates frequency and phase Optical field distribution of downmodulated coherent laser, and They are two-dimensional coordinates, For frequency, For phase, and These represent the modulated coherent laser in coordinates. and coordinates Frequency components in the direction;

[0070] Based on the reference mask and the light field distribution function, a defect-sensitive pattern and its complementary pattern are generated to obtain a defect-sensitive pattern pair. The calculation formula is as follows:

[0071]

[0072]

[0073] In the formula, As a reference mask, For defect-sensitive patterns, It is a complementary pattern to a defect-sensitive pattern.

[0074] S3. Project the defect-sensitive pattern pair onto the tile to be tested, collect the reflected signal using a single-pixel detector, and obtain the phase shift difference grouping through complementary differential operation.

[0075] The two-dimensional moving stage moves the semiconductor wafer under test into the projection range, and keeps the tile under test positioned within the conjugate field of view of the projection optical path and the single-pixel detector;

[0076] The defect-sensitive pattern and the complementary pattern are projected onto the tile under test one by one using a digital micromirror device. The intensity of reflected light after being irradiated onto the tile under test is integrated and collected by a single pixel detector to obtain the positive electrical signal value and the complementary electrical signal value.

[0077] It should be noted that the reason why the integrated acquisition of reflected light intensity yields an electrical signal value is because single-pixel detectors, such as photodiodes and photomultiplier tubes, cannot directly output "light intensity values," but rather convert the light power incident on the tile into current or voltage values ​​for output.

[0078] The differential electrical signal value is obtained by performing a difference operation on the positive and complementary electrical signal values. The calculation formula is as follows:

[0079]

[0080] In the formula, Indicates frequency Phase and wavelength The differential electrical signal value under coherent laser light, In frequency Phase and wavelength The positive electrical signal value under coherent laser light, In frequency Phase and wavelength The complementary electrical signal values ​​under coherent laser light;

[0081] Among them, by solving the differential electrical signal value, the background constant interference and ambient light influence that are unrelated to the defect-sensitive pattern can be effectively removed, and only the electrical signal value caused by the defect of the tile under test is retained.

[0082] For each tile under test, differential electrical signal values ​​of four equally spaced phases are acquired at all frequencies of coherent laser to obtain the phase shift difference group of the tile under test. The differential electrical signal values ​​of the four equally spaced phases are acquired sequentially in the same order to ensure real-time differential operation and avoid excessive time drift. For example, the positive and complementary electrical signal values ​​are integrated and acquired in the order of 0, π / 2, π and 3π / 2, and differential operation is performed at the same phase.

[0083] It should be noted that the reason why only the differential electrical signal values ​​of the four equally spaced phases are used in the phase shift difference grouping is that in the subsequent phase-locked demodulation, the four equally spaced phases exactly cover the four key points of a complete sine cycle (0, π / 2, π and 3π / 2), and the DC component is naturally canceled out, so the in-phase component and the quadrature component can be obtained directly. Although other equally spaced phases can also be solved, the solution time and computational load increase because the DC component cannot be canceled out.

[0084] S4. Perform phase-locked demodulation on the phase shift difference group to obtain the in-phase component and quadrature component, calculate the frequency domain amplitude and phase distribution, and determine the defect location coordinates based on the slope relationship between phase and frequency.

[0085] Based on the differential electrical signal values ​​of the four equally spaced phases in the phase-shift differential grouping, the in-phase component and the quadrature component are calculated using the four-step phase-shift method, as shown in the following formula:

[0086]

[0087]

[0088] In the formula, Indicates frequency and wavelength The in-phase components below, Indicates frequency and wavelength Orthogonal components below, Pi , , and They represent the frequencies respectively. and wavelength Down , , and The differential electrical signal value;

[0089] Based on the in-phase and quadrature components, the amplitude and phase of the reflected light at the defect location on the tile under test are calculated using the following formula:

[0090]

[0091]

[0092] In the formula, This indicates that the light reflected from the tile under test has a frequency of and wavelength The amplitude below, This indicates that the light reflected from the tile under test has a frequency of and wavelength The lower phase;

[0093] According to the Fourier shift theorem, the phase of the defect location on the tile under test is linearly related to the frequency. The specific steps for calculating the local coordinates of the defect on the tile under test are as follows:

[0094] According to the Fourier shift theorem, when a defect exists within the tile under test, there is a local anomaly point within the tile that alters the propagation state of the incident light, i.e., a scattering center. The phase and frequency of the scattering center satisfy the following approximate linear relationship:

[0095]

[0096] In the formula, and Represents the local coordinates of the scattering center. This is a constant phase term independent of the defect location;

[0097] A linear fit is performed on the linear relationship between phase and frequency to obtain the slope of the phase-frequency linear relationship, and the local coordinates of the defect on the tile under test are calculated. Specifically, based on the... and The phase sets at different frequencies along the coordinate direction are used to perform least-squares linear fitting on the linear relationship between phase and frequency, and the slope of the linear relationship between phase and frequency is solved. Taking coordinate direction as an example, The phase set in the coordinate direction is: The phase set is continuously expanded point-by-point along a frequency monotonic path to eliminate the 2π periodic jumps, resulting in the expanded phase. The sample mean of the expanded phase and frequency is calculated, and a least-squares linear fit is performed to obtain the final phase. The slope of the linear relationship between phase and frequency in the coordinate direction is calculated as follows:

[0098]

[0099] In the formula, for The slope of the linear relationship between phase and frequency in the coordinate direction. and These are the sample means of frequency and expanded phase, respectively. To unfold the phase;

[0100] Similarly, we can obtain The slope of the linear relationship between phase and frequency in the coordinate direction;

[0101] Based on the linear relationship between phase and frequency, the formula for calculating the local coordinates of the defect is as follows:

[0102]

[0103]

[0104] In the formula, and They represent wavelengths respectively The local transverse and longitudinal coordinate values ​​of the defect in the tile to be tested are shown below. The slope of the linear relationship between phase and frequency in the y-coordinate direction;

[0105] Based on the coordinate mapping function between the tile under test and the semiconductor wafer under test, the local coordinates on the tile under test are converted into global coordinates;

[0106] It should be noted that traditional single-pixel imaging methods often reconstruct a complete image through compressed sensing or matrix inversion, and then search for abnormal pixels in the image domain to locate defects. In contrast, this invention directly solves the differential electrical signal into frequency domain amplitude and phase using a four-step phase shift method. By utilizing the Fourier shift theorem, the spatial location of the defect is transformed into the slope relationship between the phase and frequency. The local coordinates of the defect are calculated by fitting, avoiding the image reconstruction process, improving detection speed, and reducing computational resource consumption. At the same time, it does not rely on spatial domain images, but solves the defect location through frequency domain phase gradient, resulting in higher accuracy in defect localization.

[0107] S5. By constructing a frequency domain suppression weighting function, the amplitude of the reflected light from the tile under test is weighted and calculated to obtain the defect significance. By adaptively adjusting the distance between the output light port and the tile under test, the defect significance sequence at different distances is obtained. The second difference of the defect significance is calculated to clarify the defect type.

[0108] Frequency domain suppression weights are generated using the defect-free spectrum. The amplitude of the reflected light from the tile under test is then weighted and calculated to obtain the defect significance. The specific steps are as follows:

[0109] A two-dimensional Fourier transform is performed on the defect-free image after dark current subtraction, brightness flat-field correction, and global normalization to obtain the defect-free spectrum. Based on the amplitude of the defect-free spectrum, a frequency domain suppression weight function is constructed, and the calculation formula is as follows:

[0110]

[0111] In the formula, For frequency Frequency domain suppression weights, For frequency The amplitude of the defect-free spectrum;

[0112] The magnitude of the defect is obtained by weighting and summing the amplitudes of the reflected light from the tile under test using frequency domain suppression weights.

[0113] Starting from the initial distance between the output light port and the tile under test, a monotonically decreasing distance sequence is set. For example, if the initial distance is 30 μm, the distance sequence is (30, 25, 20, 15, 10) μm. The defect saliency corresponding to each distance in the distance sequence is calculated and summarized to obtain the defect saliency sequence.

[0114] The normalized increment of defect significance at two adjacent distance points in the distance sequence is calculated using the following formula:

[0115]

[0116] In the formula, For the first The normalized increment of defect significance at each distance, and The first The and the first Defect saliency at a distance, The index of the distance point in the distance sequence;

[0117] Using defect-free tiles as the calibration object, the mean and standard deviation distribution of the normalized increment of defect significance of defect-free tiles are statistically analyzed, and an increment threshold is set. Specifically, using several defect-free tiles, the defect significance sequence under the same distance sequence as the tile to be tested is calculated, and the normalized increment of the defect-free tiles is calculated. The normalized increment of the defect-free tiles is a natural fluctuation caused by factors such as noise and micro-drift, and is not a true "defect". The mean and standard deviation of the normalized increment of the defect-free tiles are calculated respectively. Based on the three-standard-deviation principle, a single-distance increment threshold is set. Considering the noise unevenness of different distance segments, the maximum value among all single-distance increment thresholds is taken as the global increment threshold.

[0118] If the normalized increment at adjacent distance points is not greater than the increment threshold, the complete defect significance sequence and distance sequence are output. If the normalized increment at two adjacent distance points in the tile under test is greater than the increment threshold, distance point compensation is performed. Specifically, an intermediate distance point is added between adjacent distance points. For example, if the normalized increment at 30μm and 25μm is greater than the increment threshold, the midpoint between 30μm and 25μm, 27.5μm, is taken as the intermediate distance point, and the defect significance at the intermediate distance point of the tile under test is calculated until the normalized increment at all adjacent distance points is not greater than the increment threshold.

[0119] The quadratic difference of each defect significance in the complete defect significance sequence of the tile under test is calculated as the gain of the defect significance, as shown in the following formula:

[0120]

[0121] In the formula, For the first The second difference of defect significance at each distance;

[0122] The defect significance sequence of the tile under test, if all This indicates that the gain in defect significance increases rapidly as the distance decreases, classifying it as a point defect. If all... This indicates that the gain in defect significance remains constant as the distance decreases, and it is judged as a linear defect. If all This indicates that the gain in defect significance slows down as the distance decreases, and it is determined to be a planar defect.

[0123] This embodiment also provides a single-pixel imaging semiconductor wafer surface defect extraction system, including: a reference mask module, used to preprocess the surface grayscale image of a defect-free semiconductor wafer to obtain a reference mask, divide the semiconductor wafer under test into several test tiles, and construct a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test;

[0124] The laser modulation module is used to modulate the coherent laser to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, a defect-sensitive pattern pair is generated.

[0125] The laser projection module is used to project a pair of defect-sensitive patterns onto each tile under test. It uses a single-pixel detector to collect the reflected signals and obtains phase shift difference groups through complementary differential operations.

[0126] The defect location module is used to perform phase-locked demodulation on the phase shift difference grouping to obtain the in-phase and quadrature components, calculate the frequency domain amplitude and phase distribution, and determine the defect location coordinates based on the slope relationship between the phase and frequency. At the same time, it uses the defect-free spectrum to generate frequency domain suppression weights, performs weighted calculation on the frequency domain amplitude, and obtains the defect significance.

[0127] The defect type determination module is used to adaptively adjust the distance between the output light port and the tile under test, obtain the defect saliency sequence at different distances, calculate the second difference of the defect saliency, and determine the defect type.

[0128] This embodiment also provides a computer device applicable to the single-pixel imaging semiconductor wafer surface defect extraction method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the single-pixel imaging semiconductor wafer surface defect extraction method as proposed in the above embodiment.

[0129] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0130] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the semiconductor wafer surface defect extraction method for single-pixel imaging as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] In summary, this invention achieves the following: by forming a reference mask to suppress defect-free responses while optically amplifying defect disturbances; by employing a four-step phase-shifting method to directly solve for in-phase / quadrature components and amplitude-phase distributions; by obtaining defect coordinates using phase-frequency slope; by avoiding image reconstruction; by reducing computational overhead and improving positioning accuracy; by obtaining defect saliency sequences through distance adaptive scanning; and by using quadratic difference to clearly define gain criteria for point / line / surface types. Compared to empirical thresholding methods, this invention offers greater physical interpretability and robustness.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting surface defects of semiconductor wafers using single-pixel imaging, characterized in that: include, The surface grayscale image of the defect-free semiconductor wafer is preprocessed to obtain a reference mask. The semiconductor wafer under test is divided into several test tiles, and a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test is constructed. The coherent laser is modulated to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, a defect-sensitive pattern pair is generated. The defect-sensitive pattern pairs are projected onto the tile under test, and the reflected signals are collected using a single-pixel detector. The phase shift difference groups are obtained through complementary differential operations. Phase-locked demodulation is performed on the phase shift difference groups to obtain the in-phase and quadrature components. The frequency domain amplitude and phase distribution are calculated, and the defect location coordinates are determined based on the linear relationship between phase and frequency. The specific steps are as follows: Along the frequency monotonic path and The phase set at different frequencies along the coordinate direction is continuously expanded point by point to obtain the expanded phase; Calculate the sample mean of the expanded phase and frequency, and perform least squares linear fitting to obtain the slope of the linear relationship between phase and frequency; The local coordinates of the defect in the tile under test are calculated based on the slope of the linear relationship between phase and frequency. Based on the coordinate mapping function between the tile under test and the semiconductor wafer under test, the local coordinates on the tile under test are converted into global coordinates; By constructing a frequency domain suppression weighting function, the amplitude of the reflected light from the tile under test is weighted and calculated to obtain the defect significance. The distance between the output light port and the tile under test is adaptively adjusted. The specific steps are as follows. Starting from the initial distance between the output light port and the tile under test, a monotonically decreasing distance sequence is set, and the defect saliency corresponding to each distance in the distance sequence is calculated and summarized to obtain the defect saliency sequence. Calculate the normalized increment of defect significance at two adjacent distance points in the distance sequence; Using defect-free tiles as the calibration object, an increment threshold is set based on the mean and standard deviation distribution of the normalized increment of defect-free tiles. If the normalized increment of the test tile at adjacent distance points is not greater than the increment threshold, then the complete defect saliency sequence and distance sequence are output. If the normalized increment of two adjacent distance points in the test tile is greater than the increment threshold, then distance point compensation is performed. Obtain the defect saliency sequence at different distances, calculate the second difference of the defect saliency, and identify the defect type. The specific steps are as follows. Calculate the quadratic difference of each defect significance in the complete defect significance sequence; If the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is greater than zero, it is determined to be a point defect; if the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is equal to zero, it is determined to be a linear defect; if the quadratic difference of the significance of all defects in the defect significance sequence of the tile to be tested is less than zero, it is determined to be a planar defect.

2. The method for extracting semiconductor wafer surface defects using single-pixel imaging as described in claim 1, characterized in that: The specific steps for modulating the coherent laser are as follows: Based on a two-dimensional sine function, a two-dimensional sine grayscale image is generated by computer and then loaded by a digital micromirror device; Coherent laser light is reflected by a digital micromirror device and its spatial amplitude is modulated according to a two-dimensional sinusoidal grayscale image to obtain modulated coherent laser light.

3. The method for extracting semiconductor wafer surface defects using single-pixel imaging as described in claim 1, characterized in that: The specific steps for generating the defect-sensitive pattern pair are as follows: The reference mask is multiplied pixel by pixel with the light field distribution function to obtain the defect-sensitive pattern; A complementary pattern is generated for each frame of the defect-sensitive pattern to obtain a defect-sensitive pattern pair.

4. The method for extracting semiconductor wafer surface defects using single-pixel imaging as described in claim 1, characterized in that: The specific steps for constructing the frequency domain suppression weight function are as follows: A two-dimensional Fourier transform is performed on the defect-free image after dark current subtraction, brightness flat field correction and global normalization to obtain the defect-free spectrum; A frequency domain suppression weighting function is constructed based on the amplitude of the defect-free spectrum.

5. The method for extracting semiconductor wafer surface defects using single-pixel imaging as described in claim 1, characterized in that: The specific steps for setting the incremental threshold are as follows: Calculate the defect significance and normalized increment of defect-free tiles under the same distance sequence of the tiles to be tested; Calculate the mean and standard deviation of the normalized increment of defect-free tiles, and set the single-distance increment threshold based on the three-times-standard-deviation principle; Take the maximum value among all single-distance increment thresholds as the global increment threshold.

6. The method for extracting semiconductor wafer surface defects using single-pixel imaging as described in claim 1, characterized in that: The specific steps for obtaining the reference mask are as follows. Acquire surface grayscale images of defect-free semiconductor wafers as defect-free image outputs; Dark current subtraction and brightness flat field correction are performed on the defect-free image, and global normalization is performed and the reciprocal is taken to obtain the reference mask.

7. A semiconductor wafer surface defect extraction system based on single-pixel imaging, wherein the semiconductor wafer surface defect extraction method based on single-pixel imaging is described in any one of claims 1 to 6, characterized in that: include, The reference mask module is used to preprocess the surface grayscale image of the defect-free semiconductor wafer to obtain the reference mask, divide the semiconductor wafer under test into several test tiles, and construct a coordinate mapping function between the test tiles and the global coordinates of the semiconductor wafer under test. The laser modulation module is used to modulate the coherent laser to obtain the optical field distribution function of the modulated coherent laser. Based on the reference mask and the optical field distribution function, a defect-sensitive pattern pair is generated. The laser projection module is used to project defect-sensitive pattern pairs onto the tile under test, collect the reflected signals using a single-pixel detector, and obtain phase shift difference groups through complementary differential operations; The defect location module is used to perform phase-locked demodulation on the phase shift difference grouping to obtain the in-phase and quadrature components, calculate the frequency domain amplitude and phase distribution, and determine the defect location coordinates based on the slope relationship between the phase and frequency. The defect type determination module is used to construct a frequency domain suppression weight function to calculate the amplitude of the reflected light from the tile under test, obtain the defect significance, and obtain the defect significance sequence at different distances by adaptively adjusting the distance between the output light port and the tile under test. The module then calculates the second difference of the defect significance to determine the defect type.

Images