Bonded wafer defect detection method and system based on dual-channel scattered light

Abstract

The application discloses a bonded wafer defect detection method and system based on double-channel scattered light, which comprises the following steps: adopting incident light to irradiate a bonded wafer to be detected at an oblique angle; based on the spatial position difference of scattered light generated after the action of different defects on the bonded wafer relative to the light path of the incident light, using an optical system to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle in the scattered light to different receiving ends for collection respectively; obtaining the first intensity of the first scattered light and the second intensity of the second scattered light collected by the receiving end; based on the first intensity and the second intensity, calculating a characteristic parameter reflecting the relative strength relationship between the two; and determining the defect type of the bonded wafer according to the characteristic parameter. The above method realizes high-precision and artifact-free determination of different defect types such as bonded interface holes and surface pits, and significantly reduces the false positive rate and the missed detection rate of wafer detection.

Description

Technical Field

[0001] This invention relates to the field of wafer defect detection technology, and in particular to a method and system for detecting defects in bonded wafers based on dual-channel scattered light. Background Technology

[0002] In the field of semiconductor manufacturing technology, especially in the manufacturing process of third-generation semiconductors, wafer bonding technology (such as heterobonding of 3C-SiC and 4H-SiC) is widely used. The quality of the bonding interface directly affects the reliability of the final device. Therefore, high-precision non-destructive testing of the bonded wafers is required in the process to accurately identify potential surface defects (such as surface pits) and interface defects (such as bonding interface voids). At the same time, due to the specific grain boundary micromorphology of the wafer material itself, background crosstalk signals caused by the transmission of the underlying structure are often encountered during testing.

[0003] In existing bonded wafer inspection, common optical non-destructive testing methods include differential interferometry (DI) and photoluminescence (PRL) detection. DI typically characterizes the surface or near-surface morphology of the wafer by acquiring phase differences or brightness variations in the test area. PRL, on the other hand, typically excites the material to generate a light emission signal and determines material defects or interface anomalies based on the emission intensity or distribution. In practical applications, these methods generally image the test area as a single field of view or a single type of optical signal, and then identify defects using features such as grayscale, brightness, contrast, phase, or emission intensity.

[0004] For heterobonded 3C-SiC and 4H-SiC wafers, surface pits typically manifest as shallow surface morphology undulations, while bonding interface pores are located at the bonding interface and may be accompanied by steeper edge morphologies. Simultaneously, the microstructure of the 3C-SiC material itself, such as grains and grain boundaries, may generate transmitted or background signals under the influence of detection light. Since these signals may overlap in the imaging focal plane or the detection field of view, existing detection methods based on single intensity, phase, or luminescence information struggle to accurately determine the signal source, easily misinterpreting crosstalk signals caused by bottom-layer transmission as surface defects, or confusing surface pits with bonding interface pores. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting defects in bonded wafers based on dual-channel scattered light, which can increase the dimension of defect identification features and reduce the impact of underlying crosstalk.

[0006] To achieve the above objectives, the present invention provides a method for detecting defects in bonded wafers based on dual-channel scattered light, comprising: The bonding wafer to be tested is illuminated by incident light at an oblique angle; Based on the spatial position difference of the scattered light generated by different defects on the bonded wafer relative to the incident light path, an optical system is used to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle to different receiving ends for collection. The first intensity of the first scattered light and the second intensity of the second scattered light are obtained from the receiver. Based on the first intensity and the second intensity, calculate the characteristic parameters that reflect the relative strength of the two; The defect type of the bonding wafer is determined based on the characteristic parameters.

[0007] Preferably, when the feature parameter falls within a first preset value range, a hole defect located at the bonding interface of the bonding wafer is detected. When the feature parameter falls within the second preset value range, it is determined that a pit defect located on the surface of the bonded wafer has been detected. When the characteristic parameter falls into the third preset value range, it is determined to be a crosstalk signal caused by bottom layer transmission; wherein, the third preset value range is located between the first preset value range and the second preset value range.

[0008] Preferably, the characteristic parameter is obtained by calculating the ratio of the first intensity to the second intensity, or by calculating the difference between the first intensity and the second intensity.

[0009] Preferably, the characteristic parameter R is calculated using the following formula: R = (I1 + α) / (I2 + β); Wherein, I1 is the first intensity, I2 is the second intensity, α is the first channel background compensation coefficient of the receiver receiving the first scattered light, and β is the second channel background compensation coefficient of the receiver receiving the second scattered light.

[0010] Preferably, the method of irradiating the bonded wafer to be inspected with incident light includes: The bonding wafer is illuminated using a dark field illumination method, so that the positively reflected light generated on the surface of the bonding wafer avoids the receiving range of the optical system, and only the scattered light enters the optical system.

[0011] Preferably, the methods for collecting the first scattered light and the second scattered light include: The two receiving ends are a first image sensor and a second image sensor, respectively; At the same time, the first image sensor is used to acquire a first dark field image composed of the first scattered light, and the second image sensor is used to acquire a second dark field image composed of the second scattered light.

[0012] Preferably, the bonding wafer includes a 3C-SiC layer and a 4H-SiC layer heterobonded on the 3C-SiC layer.

[0013] The present invention also provides a bonding wafer defect detection system, which includes a laser emitter, an optical system, two receiving ends, and a data processing module; The laser emitter is used to emit incident light to illuminate the bonded wafer to be inspected at an angle; The optical system is used to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle to the two receiving ends for collection, based on the spatial position difference of the scattered light generated by different defects on the bonded wafer relative to the incident light path. The data processing module is communicatively connected to the two receiving ends and is configured as follows: The first intensity of the first scattered light and the second intensity of the second scattered light collected by the two receiving ends are obtained; Based on the first intensity and the second intensity, a characteristic parameter reflecting the relative strength of the two is calculated, and the defect type of the bonded wafer is determined according to the characteristic parameter.

[0014] Preferably, when the data processing module determines the defect type of the bonded wafer based on the feature parameters, it is specifically configured to execute the following logic: When the feature parameter falls within the first preset value range, it is determined that a hole defect located at the bonding interface of the bonding wafer has been detected. When the feature parameter falls within the second preset value range, it is determined that a pit defect located on the surface of the bonded wafer has been detected. When the characteristic parameter falls into the third preset value range, it is determined to be a crosstalk signal caused by bottom layer transmission; wherein, the third preset value range is located between the first preset value range and the second preset value range.

[0015] Preferably, the two receiving ends respectively include a first image sensor and a second image sensor; the optical system includes a dark-field light-collecting objective and a beam-splitting component; The dark-field light-collecting objective lens is used to collect the scattered light; The beam splitting component is disposed in the rear optical path of the dark field light-collecting objective lens and is configured to physically separate the scattered light in space into a first detection channel corresponding to the first scattered light and a second detection channel corresponding to the second scattered light, and guide them to the first image sensor and the second image sensor respectively for synchronous imaging.

[0016] Compared with existing technologies, the bonding wafer defect detection method provided by the above technical solution utilizes the difference in the influence of different morphological defects on the scattered light angle by irradiating with oblique incident light. An optical system physically separates the first scattered light with a smaller scattering angle from the second scattered light with a larger scattering angle and guides them to different receiving ends for acquisition. This solution effectively decouples the originally severely aliased optical signals from the optical hardware front end. Furthermore, by calculating the relative strength characteristic parameters of the first and second intensities acquired through dual channels, it overcomes the limitation of existing technologies that rely solely on single-dimensional information. This allows for precise filtering of background crosstalk noise caused by transmission from the underlying structure, achieving high-precision, artifact-free identification of different defect types such as bonding interface holes and surface pits, significantly reducing the false alarm rate and false negative rate of wafer detection. Attached Figure Description

[0017] Figure 1 This is a flowchart of the defect detection method in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the defect detection system in an embodiment of the present invention.

[0019] Figure 3 This is a defect state diagram on a wafer in an embodiment of the present invention.

[0020] Figure 4 This is a scattering state diagram of scattered light caused by different defects in the embodiments of the present invention. Detailed Implementation

[0021] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0022] This embodiment discloses a method for detecting defects in bonded wafers based on dual-channel scattered light, applicable to online detection of the bonding interface quality of heterogeneous bonded wafers such as 3C-SiC and 4H-SiC in third-generation semiconductor manufacturing. This method achieves precise differentiation between subsurface voids and surface pits at the bonding interface through physical separation and differentiated acquisition of the spatial distribution of scattered light (e.g., ...). Figure 3 Please refer to the following: Figure 1 and Figure 2 The method mainly includes the following steps: S1: The bonding wafer to be tested is illuminated by incident light at an oblique angle.

[0023] In this step, the bonded wafer to be inspected is fixed on a support platform, and a collimated laser beam is emitted by a laser emitter, so that the laser beam is incident on the area of ​​the wafer to be inspected at a preset tilt angle θ. The selection of θ should be such that there is a certain angle between the incident light path and the normal of the wafer surface, so that the scattered light generated by the defects on the wafer surface forms a clear angular distribution in space.

[0024] In this embodiment, θ can be any angle within the range of 30° to 75°, such as 45°, 60°, etc. Those skilled in the art should understand that the tilt angle mentioned here is not limited to the above range. As long as the incident light avoids the positive reflection direction and enters the receiving end, and the scattered light produces a spatially resolvable angular distribution, it is acceptable. It can be adjusted according to the morphological characteristics of the actual object being detected and the numerical aperture of the optical system.

[0025] The incident light source can be a continuous or pulsed laser with wavelengths of 405nm, 488nm, 532nm, or 633nm, or it can be collimated light formed by a broadband light source after passing through a narrowband filter. The selected wavelength should balance the appropriate penetration capability of the upper 4H-SiC material with the resolvability of the transmission through the grain boundaries of the lower 3C-SiC.

[0026] S2: Based on the spatial position difference of the scattered light relative to the incident light path caused by different defects on the bonded wafer, an optical system is used to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle to different receiving ends for collection.

[0027] The Pit on the surface of the bonded wafer and the Void at the bonding interface have fundamentally different geometric morphologies, and therefore have different effects on incident light.

[0028] Specifically, such as Figure 4 Void is a deep cavity formed by the rupture of bubbles at the bonding interface. Its edges are steep and accompanied by a crater-like protrusion. The steep sidewalls affect the incident light, causing the scattered light to deflect at a smaller angle. The spatial distribution of the scattered light is closer to the original path of the incident light. This part of the scattered light is defined as the first scattered light in this paper.

[0029] Pit is a shallow pit located on the surface of the wafer. Its sidewalls are relatively flat. The effect of the gently sloping sidewalls on the incident light is that the scattered light is deflected at a larger angle and the spatial distribution of the scattered light is further away from the original optical path of the incident light. This part of the scattered light is defined as the second scattered light in this paper.

[0030] Based on the aforementioned differences in scattering angles, the optical system is equipped with a first receiving end and a second receiving end at different spatial positions of its light outlet.

[0031] The first receiving end is specifically used to receive the first scattered light; the second receiving end is specifically used to receive the second scattered light. Separating scattered light at different angles directly through their own propagation directions and guiding them to different receiving ends is a function inherent in the optical system itself. That is, by properly configuring the positional relationship of the objective lens, aperture, and beam splitter, spatial physical separation of the scattering angles can be achieved.

[0032] S3: Obtain the first intensity of the first scattered light and the second intensity of the second scattered light collected by the receiver.

[0033] The first and second receivers can employ photoelectric conversion elements such as back-illuminated CMOS image sensors, EMCCDs, photomultiplier tubes (PMTs), or avalanche photodiodes (APDs). During the scanning process, the two receivers output light intensity signals at corresponding locations, which are then converted from analog to digital to form spatial distribution data of the first and second intensities.

[0034] S4: Based on the first and second intensities, calculate the characteristic parameters that reflect the relative strength of the two.

[0035] This step involves calculating the first and second intensities at the same spatial coordinates in the data processing module to obtain characteristic parameters reflecting the relative strength of the two channels. Since the background crosstalk signal formed by the transmission of the underlying 3C-SiC grains and grain boundaries has approximately the same intensity contribution in both channels, the relative calculation of the dual-channel intensities can physically filter out the common-mode background, thereby highlighting the scattering characteristics truly dominated by the defect morphology.

[0036] S5: Determine the defect type of the bonded wafer based on the characteristic parameters.

[0037] The data processing module compares the feature parameters of each pixel location with a preset threshold according to the pre-established judgment rules, and then outputs the defect category label (Void, Pit or background noise) at that location, and finally generates a defect distribution map of the entire wafer.

[0038] In this embodiment, due to the adoption of the above-mentioned dual-channel collaborative detection scheme based on the spatial angle difference of scattered light, the surface morphology signal, subsurface hole signal and bottom grain boundary transmission crosstalk signal, which were originally severely mixed in the single channel of DIC (differential interference) or PL (photoluminescence), are physically separated at the front end of the optical hardware. Decoupling can be completed without relying on complex back-end algorithm inversion, thereby fundamentally reducing the false judgment rate caused by signal aliasing and providing a reliable detection method for online control of bonding interfaces of high-quality SiC power devices.

[0039] In another embodiment, the data processing module pre-sets three numerical ranges, namely a first preset numerical range, a second preset numerical range, and a third preset numerical range, and the third preset numerical range is located between the first preset numerical range and the second preset numerical range.

[0040] Specifically, when the feature parameter R(x,y) falls into the first preset value range, it indicates that the first scattered light at that location is dominant relative to the second scattered light, that is, the component with a smaller deflection angle of the scattered light is significantly enhanced, which is consistent with the scattering characteristics of the steep sidewall and volcanic morphology corresponding to the Void. At this time, it is determined that there is a void defect located at the bonding interface at that location.

[0041] When the characteristic parameter R(x,y) falls into the second preset value range, it indicates that the second scattered light at that position is dominant relative to the first scattered light, that is, the component with a larger deflection angle of the scattered light is significantly enhanced, which is consistent with the scattering characteristics of the Pit gently tilted sidewall. At this time, it is determined that there is a pit defect located on the wafer surface at that position.

[0042] When the characteristic parameter R(x,y) falls into the third preset value range, it indicates that the intensity of the scattered light in the two channels is relatively close. That is, the scattered signal at this location does not exhibit the specific deflection characteristics of Void or Pit. It can be inferred that it originates from the common-mode background crosstalk signal formed by the transmission of the underlying 3C-SiC grains and grain boundaries. The signal at this location is removed and not included in the defect statistics.

[0043] Taking the feature parameter R as the ratio of the first intensity I1 to the second intensity I2 as an example, in a specific application scenario, the first preset value range can be set to R>1.8, the second preset value range can be set to R<0.6, and the third preset value range can be set to 0.6≤R≤1.8.

[0044] When I1=850 and I2=420 are detected at a certain pixel position during the scanning process, R≈2.02 is calculated, which falls into the first preset value range and is determined to be Void; When I1=310 and I2=780 are detected, R≈0.40, which falls into the second preset value range and is determined to be Pit; When I1=520 and I2=500 are detected, R≈1.04, which falls into the third preset value range, and is determined to be background grain boundary transmission crosstalk and filtered out.

[0045] By introducing a three-segment interval determination, the data processing module can not only distinguish between two types of real defects, Void and Pit, but also actively identify and eliminate false defect signals caused by the transmission of the underlying structure, thus avoiding the misjudgment of background noise as defects and further reducing the false alarm rate.

[0046] In one implementation, the feature parameter can be either the ratio of the two channel intensities or the difference between the two channel intensities.

[0047] When taking the ratio, the influence of common-mode systematic factors such as incident light intensity fluctuation and CMOS gain drift on the judgment result can be effectively suppressed, because such common-mode factors act on both the numerator and denominator and are automatically canceled in the ratio calculation.

[0048] When taking the difference, the computation is small and it is easy to process in real time in parallel with hardware such as FPGA, making it suitable for high-throughput detection scenarios with high requirements for processing latency.

[0049] In another implementation, the feature parameter can also be taken as a normalized difference: R=(I1-I2) / (I1+I2), the value of which is limited to [-1,1], which facilitates the uniform calibration of the threshold.

[0050] To further improve the robustness of the feature parameter in the low signal-to-noise ratio region, this embodiment provides a feature parameter calculation formula with background compensation, namely: the feature parameter R is calculated using the following formula: R=(I1+α) / (I2+β) Where α is the background compensation coefficient of the first channel for receiving the first scattered light, and β is the background compensation coefficient of the second channel for receiving the second scattered light.

[0051] The reason for introducing α and β is that, in actual operation, image sensors will produce a certain basic output due to factors such as dark current and readout noise, even in the absence of light. When the values ​​of I1 or I2 are small, if a ratio calculation is performed directly, the slight noise fluctuations will be significantly amplified, causing the R value to jump drastically and leading to unstable judgment results. By superimposing α and β on the numerator and denominator respectively, the denominator of the ratio calculation can be moved away from zero, effectively suppressing the noise amplification effect in small signal regions, and allowing the R value to maintain a relatively smooth response in weak signal regions.

[0052] The values ​​of α and β should be calibrated based on the dark current level and noise floor of each of the two channels. In a specific application scenario, if the average dark output of the first channel image sensor under the condition of turning off the light source is 10 LSB (Least Significant Bit) and that of the second channel is 12 LSB, then α can be set to 10-30 and β to 12-36. For example, if α=20 and β=25, when I1=850 and I2=420, R=(850+20) / (420+25)≈1.96; while when noise fluctuations of I1=15 and I2=18 occur in the defect-free area, R=(15+20) / (18+25)≈0.81, falling into the third preset value range and being judged as background, thus avoiding false alarms in the small signal area.

[0053] In another preferred embodiment, in order to further reduce the interference of positively reflected light on the scattered signal, the present invention also provides an incident method based on dark field illumination.

[0054] Specifically, by selecting a suitable tilted incident angle and setting a blocking aperture in the receiving optical path of the optical system, the positively reflected rays of the incident light generated on the wafer surface are emitted in a direction away from the optical axis of the objective lens, thereby avoiding the receiving range of the optical system. At this time, only the scattered light signal generated by the scattering of defects on the wafer surface and inside the wafer enters the optical system.

[0055] The reason for using dark-field illumination is that on a smooth wafer surface, most of the incident light energy is concentrated in the direction of specular reflection. If this portion of light is captured by the receiver, its energy will be much higher than the defect scattering signal, directly saturating the sensor and masking the weak Void and Pit scattering characteristics. By excluding the positively reflected light from the receiving range, each bright spot recorded by the receiver corresponds to a discontinuity on the wafer surface or interface, fundamentally improving the contrast of the defect scattering signal.

[0056] In some alternative implementations, dark field illumination can be achieved by means of a ring light source, side single-point incidence, multi-angle array light source, etc., as long as the positively reflected light can bypass the numerical aperture of the objective lens and only the scattered light enters the optical system.

[0057] In another embodiment, the two receiving ends are a first image sensor and a second image sensor, and the two image sensors use a synchronous triggering method to acquire images at the same time.

[0058] During the scanning process, the system simultaneously activates the exposure of the first and second image sensors via a unified synchronization trigger signal. At the same time, the first image sensor acquires a first dark-field image composed of first scattered light, and the second image sensor acquires a second dark-field image composed of second scattered light. The two dark-field images correspond one-to-one in space, meaning that the same pixel coordinates in the two images reflect two different scattering responses at the same physical location on the wafer.

[0059] By using hardware synchronization triggering, the time difference between the two channels can be controlled at the nanosecond level, thereby ensuring that I1 and I2 strictly correspond to the same physical point.

[0060] In some alternative implementations, the first image sensor and the second image sensor may be back-illuminated CMOS, sCMOS or TDI-CCD, and their resolution, pixel size and frame rate may be selected according to the wafer scanning speed and the minimum resolvable defect size.

[0061] In another preferred embodiment, the present invention also provides a bonding wafer defect detection system for implementing the detection methods of the above embodiments. Figure 2 The system in this embodiment includes a laser emitter, an optical system, two receivers, and a data processing module.

[0062] The laser emitter is mounted on an adjustable bracket above the support platform and is used to emit incident light and illuminate the bonded wafer to be inspected at an angle. The output wavelength, pulse width, and power of the laser emitter can be selected according to the material of the wafer to be inspected, and the pulse width of the pulsed laser can be adaptively modulated in the nanosecond range.

[0063] The optical system is positioned along the propagation path of scattered light between the laser emitter and the support platform. Based on the spatial positional differences between the scattered light (generated by different defects on the bonded wafer) and the incident light (generated by a larger scattering angle), it guides the first scattered light (with a smaller scattering angle) and the second scattered light (with a larger scattering angle) to two receiving ends, respectively. Internally, the optical system utilizes a rationally arranged array of objective lenses, apertures, and beam splitters to ensure that the first scattered light propagates along the first optical path to the first receiving end near the extension of the incident light, while the second scattered light propagates along the second optical path to the second receiving end away from the extension of the incident light.

[0064] The two receiving ends are the first receiving end and the second receiving end, respectively. They use photoelectric detection elements to convert optical signals into electrical signals and output them.

[0065] The data processing module communicates with the two receiving ends via a data interface and can be implemented using an industrial control computer, an embedded processor combined with an FPGA array, or a GPU-based high-performance computing platform. The data processing module is configured to: acquire the first intensity of the first scattered light and the second intensity of the second scattered light collected by the two receiving ends; calculate characteristic parameters reflecting the relative strength of the two intensities based on the first and second intensities; determine the defect type of the bonded wafer based on the characteristic parameters and output the defect distribution results.

[0066] This embodiment adopts an architecture that combines physical decoupling at the hardware front end with relative intensity calculation at the back end. This allows the detection system to retain the advantages of non-contact and non-destructive optical detection, while significantly improving the accuracy of defect type differentiation through dual-channel collaboration. Furthermore, the system architecture is highly compatible with existing DIC or PL detection machines, facilitating industrial integration.

[0067] On the other hand, the optical system includes a dark-field light-collecting objective and a beam-splitting assembly.

[0068] The dark-field focusing objective is positioned above the support platform. Its numerical aperture and working distance are selected to cover the angular distribution range corresponding to the first and second scattered light, while avoiding the emission direction of the positively reflected light. The function of the dark-field focusing objective is to collect the scattered light generated by defects on the wafer surface and interface, and to converge it into a collimated or focused beam that is convenient for subsequent processing.

[0069] The beam-splitting component is located in the optical path behind the dark-field light-collecting objective lens and is configured to physically separate the scattered light in space into a first detection channel corresponding to the first scattered light and a second detection channel corresponding to the second scattered light, and guide them to the first image sensor and the second image sensor respectively for synchronous imaging.

[0070] In one specific implementation, the beam splitting component can adopt a partitioned aperture and lens group combination structure: a spatial aperture with inner and outer partitions is placed at the Fourier surface (back focal plane) of the dark field light collecting objective lens, wherein the inner circle region corresponds to the first scattered light with a smaller scattering angle, which is reflected or refracted to the first image sensor; the outer circle region corresponds to the second scattered light with a larger scattering angle, which is reflected or refracted to the second image sensor.

[0071] In another specific implementation, the beam splitting component can also adopt a prism beam splitting structure, a multi-faceted mirror group, or a digital micromirror array (DMD), as long as it can physically separate the scattered light in space according to the scattering angle and guide them to two independent image sensors respectively.

[0072] During the scanning process, the first and second image sensors receive a unified synchronous trigger signal, simultaneously performing exposure and readout, thus ensuring a strict correspondence between I1(x,y) and I2(x,y) obtained at the same physical location. The data processing module calculates the feature parameter R(x,y) pixel-by-pixel from the two dark-field images and, combined with a preset threshold range, outputs the Void and Pit distribution maps for the entire wafer.

[0073] Because the optical system employs an architecture that combines a dark-field light-collecting objective with a beam-splitting component, the system achieves spatial separation of scattered light at the hardware level, avoiding model errors that are inevitably introduced by back-end algorithm inversion. At the same time, the synchronous imaging method ensures the spatial consistency of the two-channel data, providing a physical basis for the accuracy of feature parameter calculation.

[0074] In summary, based on the fundamental differences in geometric morphology between surface pits (Pit) and bonding interface voids on bonded wafers, this invention proposes a detection approach that involves setting up dual-channel receivers at different spatial locations at the optical system's output port. By physically separating and separately acquiring the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle, this invention decouples the surface morphology signals, subsurface defect signals, and underlying grain boundary transmission crosstalk signals—which were originally severely mixed under single-channel DIC or PL—at the optical hardware front end. This ensures high-precision, artifact-free identification of voids and pits in heterogeneous bonded wafers such as 3C-SiC and 4H-SiC, significantly reducing the false alarm and false negative rates.

[0075] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for detecting defects in bonded wafers based on dual-channel scattered light, characterized in that, include: The bonding wafer to be tested is illuminated by incident light at an oblique angle; Based on the spatial position difference of the scattered light generated by different defects on the bonded wafer relative to the incident light path, an optical system is used to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle to different receiving ends for collection. The first intensity of the first scattered light and the second intensity of the second scattered light are obtained from the receiver. Based on the first intensity and the second intensity, calculate the characteristic parameters that reflect the relative strength of the two; The defect type of the bonding wafer is determined based on the characteristic parameters.

2. The method for detecting defects in bonded wafers according to claim 1, characterized in that, When the feature parameter falls within the first preset value range, it is determined that a hole defect located at the bonding interface of the bonding wafer has been detected. When the feature parameter falls within the second preset value range, it is determined that a pit defect located on the surface of the bonded wafer has been detected. When the characteristic parameter falls into the third preset value range, it is determined to be a crosstalk signal caused by bottom layer transmission; wherein, the third preset value range is located between the first preset value range and the second preset value range.

3. The method for detecting defects in bonded wafers according to claim 1, characterized in that, The characteristic parameter is obtained by calculating the ratio of the first intensity to the second intensity, or by calculating the difference between the first intensity and the second intensity.

4. The method for detecting defects in bonded wafers according to claim 1, characterized in that, The characteristic parameter R is calculated using the following formula: R = (I1 + α) / (I2 + β); Wherein, I1 is the first intensity, I2 is the second intensity, α is the first channel background compensation coefficient of the receiver receiving the first scattered light, and β is the second channel background compensation coefficient of the receiver receiving the second scattered light.

5. The method for detecting defects in bonded wafers according to claim 1, characterized in that, Methods that use incident light to irradiate the bonded wafer to be inspected include: The bonding wafer is illuminated using a dark field illumination method, so that the positively reflected light generated on the surface of the bonding wafer avoids the receiving range of the optical system, and only the scattered light enters the optical system.

6. The method for detecting defects in bonded wafers according to claim 1, characterized in that, The methods for collecting the first scattered light and the second scattered light include: The two receiving ends are a first image sensor and a second image sensor, respectively; At the same time, the first image sensor is used to acquire a first dark field image composed of the first scattered light, and the second image sensor is used to acquire a second dark field image composed of the second scattered light.

7. The method for detecting defects in bonded wafers according to claim 1, characterized in that, The bonded wafer includes a 3C-SiC layer and a 4H-SiC layer heterobonded onto the 3C-SiC layer.

8. A bonding wafer defect detection system, characterized in that, It includes a laser emitter, an optical system, two receivers, and a data processing module; The laser emitter is used to emit incident light to illuminate the bonded wafer to be inspected at an angle; The optical system is used to guide the first scattered light with a smaller scattering angle and the second scattered light with a larger scattering angle to the two receiving ends for collection, based on the spatial position difference of the scattered light generated by different defects on the bonded wafer relative to the incident light path. The data processing module is communicatively connected to the two receiving ends and is configured as follows: The first intensity of the first scattered light and the second intensity of the second scattered light collected by the two receiving ends are obtained; Based on the first intensity and the second intensity, a characteristic parameter reflecting the relative strength of the two is calculated, and the defect type of the bonded wafer is determined according to the characteristic parameter.

9. The bonding wafer defect detection system according to claim 8, characterized in that, When determining the defect type of the bonded wafer based on the feature parameters, the data processing module is specifically configured to execute the following logic: When the feature parameter falls within the first preset value range, it is determined that a hole defect located at the bonding interface of the bonding wafer has been detected. When the feature parameter falls within the second preset value range, it is determined that a pit defect located on the surface of the bonded wafer has been detected. When the characteristic parameter falls into the third preset value range, it is determined to be a crosstalk signal caused by bottom layer transmission; wherein, the third preset value range is located between the first preset value range and the second preset value range.

10. The bonding wafer defect detection system according to claim 8, characterized in that, The two receiving ends each include a first image sensor and a second image sensor; the optical system includes a dark-field light-collecting objective and a beam-splitting assembly; The dark-field light-collecting objective lens is used to collect the scattered light; The beam splitting component is disposed in the rear optical path of the dark field light-collecting objective lens and is configured to physically separate the scattered light in space into a first detection channel corresponding to the first scattered light and a second detection channel corresponding to the second scattered light, and guide them to the first image sensor and the second image sensor respectively for synchronous imaging.

Images