Wafer bonding method, system, and bonded wafer

Abstract

The present disclosure relates to the technical field of semiconductor manufacturing, and particularly relates to a wafer bonding method, system and bonded wafer. The method performs hydrophilic treatment on a silicon substrate to be bonded; in a state that a bonding surface of the silicon substrate after the hydrophilic treatment is spaced apart from a spectrum acquisition device, Fourier transform infrared reflection absorption spectrum of the bonding surface with a silicon oxide layer is acquired in an external reflection mode, and a target peak area of a target absorption peak is determined; a surface hydroxyl density of the bonding surface is determined according to the target peak area; when the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, direct hydrophilic bonding is performed on the silicon substrate; and when the surface hydroxyl density is less than the preset surface hydroxyl density threshold, the silicon substrate is subjected to hydrophilic treatment again, and the surface hydroxyl density is determined again. The present disclosure quantitatively distinguishes the surface state in the super-hydrophilic range, provides a basis for bonding release, and is conducive to improving the bonding quality.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor manufacturing technology, and in particular to a wafer bonding method, system and bonding wafer. Background Technology

[0002] Before direct bonding to silicon substrates, the bonding surfaces typically require a hydrophilic treatment to form surface silanol groups. The level of surface silanol groups is related to the hydrophilic state of the silicon substrate surface and can therefore serve as a basis for evaluating the surface state of the silicon substrate before bonding. Contact angle measurement is a common method for characterizing the hydrophilic state of the silicon substrate surface. This method acquires the lateral profile formed by a droplet on the silicon substrate surface and determines the contact angle based on the three-phase contact line formed by the droplet, the silicon substrate surface, and the surrounding medium.

[0003] After the silicon substrate surface undergoes superhydrophilic treatment, the contact angle can be less than 5°. At this point, the droplet spreads fully along the silicon substrate surface, reducing the accuracy of identifying the droplet side profile and the three-phase contact line. Therefore, relying solely on the contact angle is insufficient for further quantitative differentiation of the surfaces to be bonded within the superhydrophilic range, and it is also difficult to provide a stable basis for releasing the surface state in subsequent bonding steps. Consequently, silicon substrates with surface states that do not meet the requirements may enter the bonding process, affecting the bonding quality and bonding strength of the bonding interface. Summary of the Invention

[0004] This disclosure provides a wafer bonding method, system, and bonding wafer, which can improve the ability to determine the surface state to be bonded in the superhydrophilic range where the contact angle is difficult to further quantify and distinguish, reduce the number of silicon substrates with surface states that do not meet the requirements from entering the bonding process, thereby improving the bonding quality and bonding strength of the bonding interface.

[0005] The technical solution disclosed herein is implemented as follows: This disclosure provides a wafer bonding method. The method involves hydrophilicating the silicon substrate to be bonded, and then, with the bonding surface of the treated silicon substrate spaced apart from a spectral acquisition device, acquiring the Fourier transform infrared reflectance absorption spectrum of the bonding surface with the silicon oxide layer using external reflection. Based on the Fourier transform infrared reflectance absorption spectrum, the target peak area corresponding to the target absorption peak of the surface silanol groups is determined. The surface hydroxyl density of the bonding surface is determined based on the target peak area. When the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, direct hydrophilic bonding is performed on the silicon substrate; when the surface hydroxyl density is less than the preset surface hydroxyl density threshold, the silicon substrate undergoes another hydrophilic treatment, and the surface hydroxyl density is re-determined.

[0006] This disclosure also provides a wafer bonding system. The system includes a control and processing device, a spectral acquisition device, a cleaning device, and a bonding device. The cleaning device performs a hydrophilic treatment on the silicon substrate to be bonded; the spectral acquisition device, with the bonding surface of the hydrophilic-treated silicon substrate spaced apart from the spectral acquisition device, acquires the Fourier transform infrared reflectance absorption spectrum of the bonding surface by external reflection; the control and processing device determines the target peak area and the surface hydroxyl density of the bonding surface based on the spectrum; when the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, the control device controls the bonding device to perform direct hydrophilic bonding on the silicon substrate; when the surface hydroxyl density is less than the preset surface hydroxyl density threshold, it is determined that the silicon substrate needs to undergo hydrophilic treatment again, and after the second hydrophilic treatment, the control device re-acquires the Fourier transform infrared reflectance absorption spectrum to redetermine the surface hydroxyl density.

[0007] This disclosure also provides a bonding wafer. The bonding wafer comprises two silicon substrates bonded together by direct hydrophilic bonding and is fabricated using the above-described wafer bonding method; the surface hydroxyl density of the bonding surfaces of the two silicon substrates before bonding is greater than or equal to 4.6 OH / nm², and a bonding interface is formed between the two silicon substrates, the bonding energy of which is greater than or equal to 2 J / m².

[0008] This disclosure obtains Fourier transform infrared reflectance and absorption spectra while maintaining a distance between the spectral acquisition device and the bonding surface, thus preserving the integrity of the bonding surface during the measurement process. Furthermore, the target peak area corresponding to the target absorption peak of the surface silanol group is determined from the Fourier transform infrared reflectance and absorption spectrum, and this target peak area is defined as the silanol spectral response value. This transforms the characterization of superhydrophilicity from a geometric measurement of the droplet profile and three-phase contact line to a quantitative characterization of the spectral response of the surface silanol group. Therefore, it improves the ability to distinguish different bonding surface states within the superhydrophilic range where the contact angle is difficult to further differentiate, and provides a quantitative basis for determining the surface state in subsequent bonding steps. Furthermore, it reduces the number of silicon substrates with unsatisfactory surface states entering the bonding process, lowering the possibility of insufficient interfacial bonding due to inadequate bonding surface states, thereby improving the bonding quality and bonding strength of the bonding interface. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the wafer bonding system provided in this disclosure.

[0010] Figure 2 This is a flowchart of the wafer bonding method provided in this disclosure.

[0011] Figure 3 This is a flowchart of the external reflectance spectrum acquisition process provided in this disclosure.

[0012] Figure 4A flowchart for determining the target peak area provided in this disclosure.

[0013] Figure 5 This is a flowchart of water interference correction provided in this disclosure.

[0014] Figure 6 This is a schematic diagram of the infrared absorption response related to hydroxyl groups and adsorbed water provided in this disclosure.

[0015] Figure 7 This is a flowchart for determining the hydroxyl density on the surface of a silicon substrate, as provided in this disclosure.

[0016] Figure 8 This is a flowchart of the silicon substrate bonding process provided in this disclosure.

[0017] Figure 9 This is a schematic diagram of the bonding wafer structure provided in this disclosure. Detailed Implementation

[0018] The technical solutions in this disclosure will now be clearly and completely described with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic diagram of the structure of the wafer bonding system 100 provided in this disclosure. Figure 1 As shown, the wafer bonding system 100 includes a control processing device 110, a spectral acquisition device 120, a measurement chamber 130, a sample platform 140, a bonding device 170, and a cleaning device 180. The control processing device 110 may include a processor 112 for performing spectral data processing and bonding control. The cleaning device 180 may be located outside the measurement chamber 130 and is used for hydrophilic treatment of the silicon substrate. The spectral acquisition device 120 includes an incident optical assembly 122 and a receiving optical assembly 124. The sample platform 140 is located within the measurement chamber 130. The processor 112 is communicatively connected to both the spectral acquisition device 120 and the bonding device 170, for sending acquisition control information, receiving spectral data, and outputting bonding control information to the bonding device 170 when the surface hydroxyl density of the bonding surface meets a preset condition.

[0020] The silicon substrate 150 to be bonded is placed on the sample platform 140, with the back surface 154 of the silicon substrate 150 facing the sample platform 140 and the bonding surface 152 facing the incident optical component 122 and the receiving optical component 124. The vacuum adsorption structure 142 is disposed on one side of the bearing surface of the sample platform 140, and fixes the silicon substrate 150 through the back surface 154, so that the bonding surface 152 is kept at a distance from the spectral acquisition device 120.

[0021] The measurement chamber 130 forms a closed space accommodating the sample platform 140 and the silicon substrate 150, and is provided with a nitrogen inlet and a gas outlet. A motion mechanism is provided between the sample platform 140 and the measurement chamber 130, the motion mechanism including at least one of a translation mechanism and a rotation mechanism. The sample platform 140 changes the position of the silicon substrate 150 relative to the incident optical component 122 and the receiving optical component 124 through the motion mechanism, and the positioning structure restricts the initial orientation of the silicon substrate 150 within the support surface.

[0022] The incident optical assembly 122 includes an infrared light source, an interferometer, a polarizer, an optical modulator, a reflector, and an incident mirror arranged sequentially along the incident optical path; the receiving optical assembly 124 includes an exit mirror and a detector arranged along the receiving optical path. The incident mirror and the exit mirror are located above opposite sides of the sample platform 140, facing the central region of the support surface. The incident optical assembly 122 and the receiving optical assembly 124 are spaced apart from the bonding surface 152.

[0023] The processor 112 is connected to the memory 114, the spectral acquisition interface 116, and the bonding control interface 118. The spectral acquisition interface 116 is connected to the spectral acquisition device 120 and is used to transmit control information, device status information, and spectral data between the processor 112 and the spectral acquisition device 120. The bonding control interface 118 is connected to the bonding device 170 and is used to transmit bonding control information to the bonding device 170. The memory 114 includes a program storage area and a data storage area. The program storage area stores the program instructions executed by the processor 112, and the data storage area stores device parameters, task records, spectral data, and surface hydroxyl density data.

[0024] The control processing device 110, the spectral acquisition device 120, the measurement chamber 130, and the sample platform 140 can be configured as separate units or integrated units. The bonding device 170 can be integrated with the above components or can be a separate device that communicates with the control processing device 110. When a separate unit structure is used, the control processing device 110 is located outside the measurement chamber 130 and is connected to the sample platform 140, the spectral acquisition device 120, and the bonding device 170 via cables.

[0025] This disclosure establishes a wafer bonding system 100 that links pre-bonding surface measurement with subsequent bonding control through the aforementioned structure. A sample platform 140 supports a silicon substrate 150 from the back side 154, with the bonding surface 152 facing the optical component and spaced apart from it. A processor 112 determines the surface hydroxyl density of the bonding surface 152 based on spectral data output from a spectral acquisition device 120 to characterize the hydrophilic state of the bonding surface 152; and controls the bonding device 170 to perform direct hydrophilic bonding when the surface hydroxyl density reaches a preset condition.

[0026] Figure 2This is a flowchart of the wafer bonding method provided in this disclosure. The method is implemented based on a wafer bonding system 100, such as... Figure 2 As shown, the method includes the following steps: In step S210, the cleaning equipment 180 performs a hydrophilic treatment on the silicon substrate to be bonded. Specifically, the cleaning equipment 180 applies a hydrophilic treatment solution to the bonding surface of the silicon substrate, so that the hydrophilic treatment solution comes into contact with the silicon oxide layer on the bonding surface. By removing contaminants that affect the surface hydrophilicity and forming silanol groups on the surface of the silicon oxide layer, the hydrophilicity of the bonding surface is improved.

[0027] In step S220, the spectral acquisition device 120 acquires the spectrum of the bonding surface 152. With the bonding surface 152 of the hydrophilically treated silicon substrate 150 and the spectral acquisition device 120 spaced apart, the Fourier transform infrared reflection absorption spectrum of the bonding surface 152 is acquired by the spectral acquisition device 120 using external reflection. The bonding surface 152 has a silicon oxide layer.

[0028] The processor 112 sends acquisition control information to the spectral acquisition device 120. The spectral acquisition device 120 projects infrared light onto the bonding surface 152, and the incident infrared light enters the receiving optical path after being externally reflected by the bonding surface 152. The spectral acquisition device 120 maintains a distance from the bonding surface 152 throughout the acquisition process, so that the silicon substrate 150 remains in a surface state suitable for bonding after the measurement is completed.

[0029] The spectral acquisition device 120 performs photoelectric conversion and Fourier transform on the externally reflected infrared light to obtain a Fourier transform infrared spectrum (FTIR). This disclosure employs an external reflection acquisition method using infrared reflection absorption spectroscopy (IRRAS), ensuring that the obtained spectrum includes the absorption responses of chemical bonds and chemical groups on the bonded surface 152. The spectral acquisition device 120 sends the Fourier transform infrared reflection absorption spectrum to the processor 112 as input for determining the target peak area.

[0030] In step S230, processor 112 determines the target peak area. Processor 112 determines the target peak area corresponding to the target absorption peak of the surface silanol group based on the Fourier transform infrared reflectance absorption spectrum.

[0031] The processor 112 identifies the target absorption peak in the Fourier transform infrared reflectance absorption spectrum based on the correspondence between the surface silanol groups and the infrared absorption response. The wavenumber position of the target absorption peak reflects the vibrational characteristics of the surface silanol groups, and the absorption response of the target absorption peak varies with the level of surface silanol groups on the bonded surface 152.

[0032] The processor 112 integrates the absorption response of the spectral curve relative to the baseline within the wavenumber range corresponding to the target absorption peak to obtain the target peak area. The target peak area aggregates the response of the target absorption peak over the entire integrated wavenumber range, thus simultaneously reflecting the overall absorption amount formed by the peak height, peak width, and peak shape.

[0033] In step S240, processor 112 determines the surface hydroxyl density of bonded surface 152. Processor 112 determines the surface hydroxyl density of bonded surface 152 based on the target peak area obtained in step S230.

[0034] The target peak area characterizes the infrared absorption response corresponding to the surface silanol groups on the bonded surface 152, and the surface hydroxyl density refers to the number of surface silanol groups per unit area of ​​the bonded surface 152, used to characterize the hydrophilic state of the bonded surface 152. The processor 112 determines the surface hydroxyl density corresponding to the surface silanol response reflected by the target peak area, and uses this surface hydroxyl density as the quantification result of the superhydrophilicity of the bonded surface 152. Thus, the spectral response of the surface silanol groups can be converted into a quantitative parameter characterizing the hydrophilic state of the bonded surface 152.

[0035] In step S250, processor 112 determines whether the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold; if the surface hydroxyl density is greater than or equal to the preset surface hydroxyl density threshold, step S260 is executed; if the surface hydroxyl density is less than the preset surface hydroxyl density threshold, steps S210 to S250 are executed again to perform hydrophilic treatment on the substandard silicon substrate and redetermine its surface hydroxyl density.

[0036] In step S260, the bonding device 170 performs direct hydrophilic bonding to the silicon substrate 150. When the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, the processor 112 outputs bonding control information to the bonding device 170, controlling the bonding device 170 to perform direct hydrophilic bonding to the silicon substrate 150. Specifically, during bonding, the bonding surface 152 of the silicon substrate 150 can be positioned opposite to the corresponding bonding surface of another silicon substrate to be bonded, and direct hydrophilic bonding can be performed.

[0037] In one embodiment, steps S210 to S250 are executed no more than twice. When the re-determined surface hydroxyl density reaches a preset surface hydroxyl density threshold, step S260 is executed; if the re-determined surface hydroxyl density after two more hydrophilic treatments is still less than the preset surface hydroxyl density threshold, the corresponding silicon substrate is scrapped, and subsequent bonding steps on the silicon substrate are stopped.

[0038] Therefore, this disclosure obtains the spectral response corresponding to the surface silanol groups of the bonding surface 152 through non-contact external reflection spectroscopy, and determines the surface hydroxyl density accordingly. This enables quantitative characterization of the surface silanol group level of the bonding surface 152 within a superhydrophilic range where the contact angle is difficult to further distinguish, improving the ability to distinguish different surface states to be bonded and providing a stable quantitative basis for the silicon substrate to enter subsequent bonding steps. This reduces the number of silicon substrates with unsatisfactory surface states entering the bonding process, improving the bonding quality and bonding strength of the bonding interface after bonding. Simultaneously, the spectral acquisition device 120 maintains a distance from the bonding surface 152 during spectral acquisition, avoiding contact interference with the bonding surface 152 during the measurement process and helping to maintain the original state of the bonding surface after measurement.

[0039] Figure 3 This is a flowchart of the partitioned external reflectance spectral acquisition process provided in this disclosure. The processor 112 determines the measurement partition and measurement location and controls the positioning of the sample platform 140. The spectral acquisition device 120 acquires the partitioned spectrum at the corresponding measurement location. Figure 3 As shown, the specific steps include: In step S310, processor 112 determines the measurement partitions and measurement locations. Processor 112 divides the bonding surface 152 into at least two measurement partitions and determines the measurement locations in each of the at least two measurement partitions.

[0040] The measurement partitions correspond to different surface regions on the bonding surface 152. The processor 112 determines the number of measurement partitions based on the size of the silicon substrate 150 and the surface range to be evaluated, and sets at least one measurement position in each measurement partition so that the spectral acquisition positions are distributed within the evaluation range of the bonding surface 152.

[0041] The processor 112 determines the measurement position by combining the diameter of the focused spot, so that the focused spot is located within the corresponding measurement zone. Each measurement position corresponds to a Fourier transform infrared reflectance and absorption spectrum, and the processor 112 stores the correspondence between the measurement zone, the measurement position, and the spectrum.

[0042] In step S320, the processor 112 controls the sample platform 140 to fix the silicon substrate 150. After the silicon substrate 150 is placed on the sample platform 140, the processor 112 controls the vacuum adsorption structure 142 to establish an adsorption state, and vacuum adsorbs the back side 154 of the silicon substrate 150 onto the sample platform 140.

[0043] The vacuum adsorption structure 142 applies an adsorption force to the back surface 154 of the silicon substrate 150 through the adsorption position on the bearing surface of the sample platform 140, thereby fixing the silicon substrate 150 relative to the sample platform 140. The adsorption position is located on one side of the back surface 154 of the silicon substrate 150, so the fixing process does not contact the bonding surface 152. After the silicon substrate 150 is fixed, the bonding surface 152 faces the external reflected light path where the incident optical component 122 and the receiving optical component 124 are located, and the bonding surface 152 is spaced apart from the spectral acquisition device 120.

[0044] After the silicon substrate 150 is fixed, it moves with the sample platform 140. The processor 112 controls the sample platform 140 to sequentially send the measurement positions determined in step S310 into the external reflection optical path. When the sample platform 140 reaches the corresponding position and remains stable, the spectral acquisition device 120 begins to acquire data.

[0045] In step S330, the processor 112 controls the sample platform 140 to be positioned sequentially, and the spectral acquisition device 120 acquires the partitioned spectra. The spectral acquisition device 120 focuses infrared light into spots at the measurement positions and acquires the infrared light reflected once by the bonding surface 152, thereby obtaining the Fourier transform infrared reflection and absorption spectra corresponding to the measurement positions.

[0046] The processor 112 controls the positioning of the sample platform 140 according to a predetermined sequence of measurement positions. After scanning is completed at one measurement position, the sample platform 140 moves to the next measurement position until the spectral acquisition of all measurement zones is completed.

[0047] Infrared light emitted from the infrared source in the incident optical assembly 122 passes sequentially through the interferometer, polarizer, optical modulator, reflector, and incident mirror within the assembly before reaching the bonding surface 152, forming a focused spot with a diameter of 2–5 mm at the corresponding measurement position. The infrared light, after a single reflection from the bonding surface 152, passes through the exit mirror in the receiving optical assembly 124 and enters the detector, where it undergoes phase-locked demodulation to generate corresponding spectral data. The angle between the incident direction of the infrared light and the normal to the bonding surface 152 is 65° to 75°, meaning the infrared light illuminates the bonding surface 152 with near-grazing incidence. The spectral acquisition range is 400–4000 cm⁻¹. -1 Spectral resolution of 4 cm -1 The number of scans at the corresponding measurement positions is 256 to 512.

[0048] The focused light spot defines the bonding surface region corresponding to a single spectral acquisition. The same incident angle, spectral range, spectral resolution, and number of scans are used at each measurement position to ensure consistent acquisition conditions for the partitioned spectra. Before spectral acquisition, the wafer bonding system 100 purges the measurement chamber 130 with nitrogen for 10 min. During spectral acquisition, the temperature of the measurement chamber 130 is maintained at 22±1 °C, and the relative humidity is below 30%. The spectral acquisition device 120 outputs the corresponding spectrum after completing a predetermined number of scans at each measurement position. The processor 112 saves the spectral data according to the measurement positions, forming a set of partitioned spectra for the bonding surface 152.

[0049] Segmented acquisition allows for the generation of spectral data with spatial location meaning at different measurement locations, reducing the impact of a single local area on the overall evaluation. The focused spot defines the surface range of a single acquisition, and a single external reflection adapts to the flat bonding surface of the silicon substrate 150, obtaining the spectral response of each segment while maintaining the integrity of the bonding surface 152.

[0050] Figure 4 A flowchart for determining the target peak area provided in this disclosure. (See attached flowchart.) Figure 4 As shown, the process for determining the target peak area includes the following steps: In step S410, the spectral acquisition device 120 acquires the background spectrum of the measurement chamber 130. The background spectrum records the spectral response formed by the atmosphere and optical system within the measurement chamber 130. The background spectrum and the corresponding measurement spectrum use the same resolution, number of scans, and optical path conditions; in one example, both the background spectrum and the measurement spectrum are scanned 256 times.

[0051] The spectral acquisition device 120 sends the background spectrum to the processor 112. The processor 112 selects a background spectrum with the same acquisition conditions as the measurement spectrum, ensuring that the wavenumber ranges and wavenumber sampling points of the two spectra correspond to each other. In addition, the processor 112 verifies the wavenumber ranges, wavenumber sampling intervals, and arrangement directions of the background spectrum and the measurement spectrum, and pairs data points at the same wavenumber positions to establish a data correspondence for point-by-point background subtraction.

[0052] In step S420, processor 112 performs background subtraction and full-spectrum baseline correction. Processor 112 performs background subtraction and baseline correction on the Fourier transform infrared reflectance and absorption spectrum based on the background spectrum, and performs multi-point rubber band baseline correction on the Fourier transform infrared reflectance and absorption spectrum after background subtraction.

[0053] Processor 112 first reads the corresponding measurement spectrum and background spectrum, and arranges the two sets of spectra in the same wavenumber direction. When both the measurement spectrum and background spectrum are arranged in an increasing wavenumber direction, processor 112 directly matches data points at the same wavenumber position; when the arrangement directions of the data points are different, processor 112 adjusts the arrangement direction of one set of spectra and then matches data points at the same wavenumber position. Processor 112 performs background subtraction based on the measurement spectral response and background spectral response at the corresponding wavenumber position to obtain the Fourier transform infrared reflectance and absorption spectrum with background subtraction.

[0054] Background subtraction is used to remove the background response formed by the measurement chamber 130 and the optical system, but the spectrum after background subtraction may still have baseline fluctuations that vary slowly with wavenumber. Processor 112 operates in the range of 4000–900 cm⁻¹. - ¹ Baseline support positions are determined sequentially within the baseline correction wavenumber range, and adjacent baseline support positions are connected to form a multi-point rubber band baseline. The baseline support positions are distributed outside the target absorption region and the associated silanol absorption region, and are used to characterize the 4000–900 cm⁻¹ region. - ¹ Slow baseline fluctuations within the range. Processor 112 obtains the spectrum after multi-point rubber band baseline correction based on the spectral response at the corresponding wavenumber position of the multi-point rubber band baseline correction.

[0055] In step S430, processor 112 determines the integration wavenumber range. Based on the second derivative spectrum of the Fourier transform infrared reflectance absorption spectrum after background subtraction, processor 112 determines the starting and ending inflection points of the target absorption peak, and defines the wavenumber range between the starting and ending inflection points as the integration wavenumber range.

[0056] The target absorption peak is approximately 3745 cm⁻¹. - The isolated surface silanol absorption peak at ¹. Processor 112 at approximately 3745 cm⁻¹. - ¹A target peak search interval is set nearby, and the second derivative is calculated on the spectral data within the target peak search interval to obtain the second derivative spectrum corresponding to the target peak search interval. The second derivative spectrum reflects the change in spectral curvature on both sides of the target absorption peak, and the processor 112 uses this to find candidate inflection points located on the low wavenumber side and the high wavenumber side of the target absorption peak, respectively.

[0057] This disclosure determines the target peak search range and candidate inflection point screening range using scanning data from 10 identical silicon substrates. Processor 112 statistically analyzes the distribution of candidate inflection points on both sides of the target absorption peak for each spectrum, forming a unified search range and screening conditions. When processing the spectrum of the current silicon substrate 150, processor 112 calculates the second derivative within this search range and determines the starting and ending inflection points based on the curvature change positions of the current spectrum.

[0058] The processor 112 determines the candidate inflection point on the low wavenumber side of the target absorption peak as the starting inflection point and the candidate inflection point on the high wavenumber side as the ending inflection point. The wavenumber range between the starting and ending inflection points constitutes the integral wavenumber range of the current target absorption peak. When the peak position changes, the two inflection points are determined according to the peak shape of the current spectrum.

[0059] For silicon substrates 150 of the same type and under the same acquisition conditions, processor 112 uses the same target peak search range and inflection point screening conditions; for spectra at different measurement locations, processor 112 determines the actual integration boundary based on the second derivative of the corresponding spectrum. Thus, the determination of the integration wavenumber range adopts a unified judgment condition while preserving the actual differences in the position and shape of the target absorption peak in different spectra.

[0060] In step S440, processor 112 performs local baseline correction. From the Fourier transform infrared reflectance absorption spectrum after multi-point rubber band baseline correction, processor 112 determines a local baseline endpoint on each side of the local baseline segment on both sides of the target absorption peak, and performs two-point linear baseline correction on the target absorption peak within the integral wavenumber range based on the two local baseline endpoints.

[0061] After completing the multi-point rubber band baseline correction, processor 112 searches for flat spectral segments adjacent to the low-wavenumber side and high-wavenumber side of the target absorption peak, respectively, outside the integration wavenumber range determined in step S430. The spectral response in the local baseline segments varies with wavenumber less than that of the main body of the target absorption peak, and therefore can be used to characterize the local baseline state on both sides of the target absorption peak.

[0062] The processor 112 determines the intermediate positions in two local baseline segments and defines the data points corresponding to the two intermediate positions as local baseline endpoints. The processor 112 connects the two local baseline endpoints to form a local linear baseline spanning the integral wavenumber range. The target absorption peak within the integral wavenumber range is then corrected using this local linear baseline as a reference. The local linear baseline only applies to the integral wavenumber range corresponding to the target absorption peak and does not alter the spectral data outside this range.

[0063] Approximately 3400 cm - The broad absorption band of the associated silanol at position ¹ may form a slowly varying response extending to the vicinity of the target absorption peak on the high wavenumber side, causing the local baselines on both sides of the target absorption peak to rise or tilt. Two-point linear baselines are established based on the actual spectral responses on both sides of the target absorption peak to compensate for the influence of this slowly varying response on the integral reference of the target peak, ensuring that subsequent processing mainly corresponds to approximately 3745 cm⁻¹. - The target absorption peak at position ¹.

[0064] When the target absorption peak and the broad absorption band of associated silanol significantly overlap, the processor 112 uses a Gaussian function or a Lorentz function to perform bimodal fitting, with two peak shapes representing the absorption response of associated silanol and the absorption response of isolated surface silanol respectively, and the fitting peak corresponding to the isolated surface silanol is included in the target peak integral.

[0065] In step S450, processor 112 calculates the target peak area. Within the integration wavenumber range, processor 112 integrates the target absorption peak after baseline correction to obtain the target peak area.

[0066] Specifically, processor 112 calculates the area between the spectral curve of the target absorption peak and the local linear baseline within the integrated wavenumber range. For two adjacent wavenumber points, processor 112 calculates the area of ​​the corresponding interval based on the adjacent wavenumber interval and the absorption response of the two wavenumber points relative to the local linear baseline, and then accumulates the areas of each adjacent wavenumber interval within the integrated wavenumber range. When the spectral data is stored in ascending or descending order of wavenumbers, processor 112 calculates the area based on the absolute value of the wavenumber interval, ensuring that the wavenumber arrangement direction does not change the positive aperture of the target peak area. The target peak area is denoted as A. 3745 The data was recorded as the spectral response value of the silanol group. Processor 112 simultaneously recorded approximately 3745 cm⁻¹. - The peak height H of the target absorption peak at ¹ 3745 Peak height is saved as an auxiliary spectral parameter.

[0067] After the target peak area is calculated, the processor 112 saves the starting inflection point, the ending inflection point, the local baseline endpoints, and the corrected target peak curve so as to verify the basis for the calculation of the target peak area.

[0068] This disclosure first removes the background response formed by the measurement chamber 130 and the optical system using the background spectrum, and then corrects the slow baseline fluctuations over a wide wavenumber range using a multi-point rubber band baseline. Based on this, the processor 112 uses the second derivative of the current spectrum to determine the actual integration boundary of the target absorption peak, and establishes a two-point linear baseline based on the local baseline states on both sides of the target absorption peak, thereby reducing the influence of the tailing response of the broad absorption band of associated silanol on the integration reference of the target peak. The target peak area is determined by a unified search interval and inflection point selection conditions, and the integration boundary is determined according to the actual peak position and shape of the current target absorption peak. Therefore, the target peak area at different measurement locations and on different silicon substrates 150 has a consistent processing basis, while preserving the actual spectral differences of the corresponding target absorption peaks.

[0069] Figure 5 This is a flowchart of the water disturbance correction process provided in this disclosure. Figure 5 As shown, the water interference correction process includes the following steps: In step S510, processor 112 determines the area of ​​the interfering peak. Processor 112 determines the area of ​​the interfering absorption peak corresponding to the adsorbed water based on the Fourier transform infrared reflectance absorption spectrum.

[0070] The interference absorption peak corresponding to adsorbed water is located at approximately 1640 cm⁻¹. - At position ¹, this absorption peak corresponds to the HOH bending vibration of water molecules. Processor 112 calculates the area of ​​interfering peaks over the integral range of this absorption peak to characterize the adsorbed water response in the current spectrum. Adsorbed water is located at approximately 3400 cm⁻¹. - ¹ Near the OH stretching vibration absorption response, this response overlaps with the broad absorption band of associated silanol groups and can change by approximately 3745 cm⁻¹. - ¹Local baseline near the target absorption peak. Approximately 1640 cm⁻¹ - ¹The area of ​​the interference peak is used to determine the contribution of the water absorption.

[0071] Figure 6 This is a schematic diagram of the infrared absorption response related to hydroxyl groups and adsorbed water provided in this disclosure. The absorption response in the Fourier transform infrared reflectance absorption spectrum is characterized by absorbance. Figure 6 Plotting wavenumber on the x-axis and absorbance on the y-axis, the image schematically shows the wavelength range of 2800–3800 cm⁻¹. - ¹The total absorption response across the wavenumber range, and the contributions of free water, bound water, surface silanol groups (i.e., surface -OH), and internal hydroxyl groups (i.e., internal -OH) to the total absorption response. The component curves are used to represent the wavenumber regions in which the corresponding absorption responses are located and the overlap between different absorption responses.

[0072] Depend on Figure 6 It can be seen that within the wavenumber range corresponding to the O–H stretching vibration, the absorption response of adsorbed water overlaps with the absorption response of surface silanol groups and other hydroxyl groups. Therefore, approximately 3400 cm⁻¹ - The broad absorption response near ¹ is influenced by both water and hydroxyl components, making it difficult to distinguish the contribution of adsorbed water based solely on this broad absorption response. Therefore, processor 112 employs an approximately 1640 cm⁻¹ [structure / mechanism / temperature]. - The interference absorption peak at position ¹ corresponding to the HOH bending vibration of water molecules is used as an independent criterion for adsorbed water, and the degree of water interference in the current spectrum is judged based on the area of ​​the interference absorption peak.

[0073] Adsorbed water at approximately 3400 cm - The OH stretching vibration absorption response formed in the vicinity may also change the local baseline near the target absorption peak. Therefore, the area of ​​the interfering peak can be used to determine whether to perform subsequent water interference correction on the current spectrum.

[0074] Processor 112 is still at 2850-2960 cm -¹The absorption response corresponding to the stretching vibration of organic CH is extracted within the range. When the CH peak area reaches the organic pollution screening condition, the corresponding measurement data is classified into the organic pollution sample category and statistically analyzed separately from the normal process samples.

[0075] In step S520, processor 112 performs water interference correction. Processor 112 performs water interference correction based on the area of ​​the interference peak, and determines the target peak area based on the target absorption peak obtained after completing the water interference correction.

[0076] The preset water disturbance threshold can be set at approximately 1640 cm. - ¹Baseline noise in the wavenumber range where the interference absorption peak is located is determined. For example, the preset water interference threshold is set to three times the standard deviation of the peak area fluctuation corresponding to the baseline noise within the same integration range. Under different acquisition conditions, the threshold can be recalibrated according to the polarization mode, whether atmospheric compensation is performed, the target peak band and integration range, and the number of scans. Preferably, the polarization difference subtraction method of polarization-modulated infrared reflection absorption spectroscopy (PM-IRRAS) can be used to suppress gas phase water vapor interference. The processor 112 compares the area of ​​the interference peak with the preset water interference threshold. When the area of ​​the interference peak is less than or equal to the preset water interference threshold, no water interference correction is required. The processor 112 uses the current Fourier transform infrared reflection absorption spectrum and follows the... Figure 4 The process shown determines the target peak area; when the interference peak area is greater than the preset water interference threshold, the processor 112 performs water interference correction, which includes a drying retest method or a proportional deduction method.

[0077] The proportional subtraction method uses a pre-calibrated area ratio. During calibration, a gold-plated mirror substrate with a silica film can be used as the IRRAS quantitative substrate; for example, the silica film thickness is approximately 200 Å. A calibration relationship between peak area and water content is established using an ultrapure water standard with known water content. The raw IRRAS signal can be processed according to… Converted to absorbance and calibrated. In the formula, A represents the standard absorbance. Represents the difference signal after demodulation , and These represent the intensities of p-polarized and s-polarized reflected light, respectively. Representation and signal , norm represents the normalization of the ratio of the difference channel signal to the sum channel signal, and 0.0223 is the coefficient used in this conversion formula.

[0078] Under conditions of eliminating interference from vapor phase water, fixing the temperature and water content windows, and unifying the integration range and acquisition parameters, approximately 1640 cm - ¹The absorption peak area of ​​water bending vibration is approximately 3400 cm⁻¹ - ¹The ratio of the absorption peak area of ​​water stretching vibration to A 1640 / A 3400 A reference coefficient of approximately 0.22 can be used. The processor 112 is approximately 1640 cm². - The estimated area of ​​the interference peak¹ is approximately 3400 cm². - ¹Water absorption contribution in the broad absorption band is calculated and subtracted. After proportional subtraction, the approximate 3745 cm² area is considered. - ¹The target absorption peak is locally baseline corrected and integrated to obtain the target peak area after water interference correction.

[0079] When using the dry retesting method, in response to the interference peak area exceeding the preset water interference threshold, the silicon substrate 150 is dried, and the measurement chamber 130 is purged with nitrogen. Specifically, the wafer bonding system 100 performs a drying process on the silicon substrate 150, causing adsorbed water on the bonding surface 152 to detach from the surface; subsequently, the gas containing moisture in the measurement chamber 130 is replaced by a nitrogen purging channel. After drying and nitrogen purging are completed, the silicon substrate 150 returns to its original measurement position, and the bonding surface 152 continues to maintain a distance from the spectral acquisition device 120, entering the retesting Fourier transform infrared reflectance absorption spectroscopy acquisition stage.

[0080] The spectral acquisition device 120 acquires the retested Fourier transform infrared reflectance and absorption spectrum. The spectral acquisition device 120 also acquires the retested Fourier transform infrared reflectance and absorption spectrum of the silicon substrate 150, and obtains the corresponding retested background spectrum. Specifically, the spectral acquisition device 120 first acquires the retested background spectrum after nitrogen purging, and then acquires the retested Fourier transform infrared reflectance and absorption spectrum of the silicon substrate 150 at the original measurement position. The retested background spectrum and the retested Fourier transform infrared reflectance and absorption spectrum use the same spectral range, resolution, number of scans, and optical path conditions.

[0081] Subsequently, the processor 112 performs background subtraction and baseline correction on the re-measured Fourier transform infrared reflectance and absorption spectrum based on the re-measured background spectrum.

[0082] The processor 112 first checks the wavenumber range and wavenumber sampling interval of the remeasured background spectrum and the remeasured Fourier transform infrared reflectance and absorption spectrum, and ensures that the two sets of spectra are arranged in the same wavenumber direction. Then, it matches data points at the same wavenumber position, and uses the remeasured background spectrum to subtract the measurement chamber background response and optical system background response from the remeasured Fourier transform infrared reflectance and absorption spectrum, thus forming a remeasured Fourier transform infrared reflectance and absorption spectrum with background subtraction.

[0083] Processor 112 at 4000–900 cm -1 Baseline support positions were selected within the baseline correction wavenumber range, and multi-point rubber band baseline correction was performed on the re-measured Fourier transform infrared reflectance and absorption spectra after background subtraction. The processor 112 then performed baseline correction at approximately 3745 cm⁻¹. -1 Local baseline endpoints are determined on both sides of the target absorption peak. Two local baseline endpoints are connected to form a local linear baseline, and the target absorption peak is corrected using the local linear baseline as the integration reference.

[0084] Finally, the processor 112 determines the target peak area based on the target absorption peak in the remeasured Fourier transform infrared reflectance absorption spectrum after background subtraction and baseline correction.

[0085] Processor 112 calculates the remeasured Fourier transform infrared reflectance absorption spectrum at approximately 1640 cm⁻¹. - ¹The area of ​​the re-measured interference peak is compared with the preset water interference threshold. When the area of ​​the re-measured interference peak is less than or equal to the preset water interference threshold, it is determined according to… Figure 4 The process shown is for approximately 3745 cm - ¹The absorption peak of isolated surface silanol groups is integrated, and the resulting peak area is used as the target peak area for the current measurement location. If the area of ​​the re-measured interfering peak exceeds the preset water interference threshold, the corresponding measurement location is dried and purged with nitrogen, and the background spectrum and Fourier transform infrared reflectance absorption spectrum are re-acquired. Organic CH absorption peak screening and adsorbed water interference correction are performed separately. The former is used for sample classification, and the latter is used to select the correction path for the target peak area.

[0086] This disclosure is based on approximately 1640 cm. -1 The area of ​​the interference peak at a given location characterizes the adsorbed water response in the current spectrum, and a preset water interference threshold is used to determine whether water interference correction is needed. When water interference correction is required, either a drying retest or a proportional subtraction method can be used. The drying retest path reduces the moisture content in the bonding surface 152 and the measurement chamber 130, while the proportional subtraction path corrects approximately 3400 cm⁻¹. -1 Water absorption contribution in the broad absorption band. The background spectrum was re-acquired after drying and nitrogen purging to ensure the re-measured Fourier transform infrared reflectance absorption spectrum corresponded to the same measurement environment as the re-measured background spectrum. Therefore, processor 112 determined the target peak area based on the spectrum where water interference was within acceptable limits or where water absorption contribution correction was completed, reducing the adsorbed water pair at approximately 3745 cm⁻¹. -1 The influence of local baseline generation on the absorption peak of silanol groups on isolated surfaces.

[0087] Figure 7 This is a flowchart illustrating the determination of hydroxyl density on the surface of a silicon substrate, as provided in this disclosure. Figure 7As shown, the process includes a calibration stage and a measurement stage. In the calibration stage, conversion factors are established using at least two reference silicon substrate samples with gradient distributions of hydroxyl density on their surfaces. In the measurement stage, the normalized hydroxyl signal value is determined using the target peak area and the reference peak area in the same Fourier transform infrared reflectance absorption spectrum, and then the surface hydroxyl density of the silicon substrate (150) is determined using the conversion factors. Both stages use the same target absorption peak, reference absorption peak, and peak area for calculation. The process specifically includes the following steps: In step S710, processor 112 acquires calibration data. Processor 112 acquires the calibration target peak area, calibration reference peak area, and reference surface hydroxyl density for at least two calibration silicon substrate samples, respectively. The reference surface hydroxyl density exhibits a gradient distribution.

[0088] This disclosure uses 5 to 7 identical silicon substrate samples as calibration silicon substrate samples. The calibration silicon substrate samples have the same material type and silicon oxide layer type, and their surface hydroxyl density covers the expected process range and forms a gradient distribution. The reference surface hydroxyl density of each calibration silicon substrate sample can be obtained independently using deuterium exchange-binding mass spectrometry: isotopic exchange occurs between surface silanol hydrogens and deuterium oxide (D2O) at approximately 150 °C, converting gaseous water species into D2, H2, and HD. The gaseous isotopic composition at equilibrium is then determined using mass spectrometry, from which the number of surface hydroxyl groups can be inferred. This reference measurement is traceable to the quantitative conservation of substances upon the addition of D2O and the quantitative determination of the infrared absorbance of the isotopic mixture.

[0089] Different surface treatment conditions result in varying levels of isolated and associated silanol groups on the calibrated silicon substrate samples. The spectral acquisition device 120 uses the same spectral acquisition conditions as the silicon substrate 150 to obtain the calibration spectrum, and the processor 112 uses the same target peak area and reference peak area calculation parameters as in the measurement stage to process the calibration spectrum. Each calibrated silicon substrate sample generates a target peak area, a reference peak area, and a reference surface hydroxyl density; these three factors together constitute the calibration data for the corresponding calibrated silicon substrate sample.

[0090] In step S720, processor 112 calculates the calibration normalized hydroxyl signal value. Processor 112 determines the calibration normalized hydroxyl signal value as the ratio of the calibration target peak area to the calibration reference peak area of ​​the same calibration silicon substrate sample.

[0091] The calibrated target peak area corresponds to approximately 3745 cm². - The isolated surface silanol absorption peak at ¹ corresponds to a calibration reference peak area of ​​approximately 1070 cm⁻¹. - The absorption peak of silicon-oxygen bond stretching vibration at position ¹. The processor 112 reads the two peak areas of the same calibration silicon substrate sample according to the calibration sample identifier, calculates the ratio between the two, and writes the calculation result into the corresponding calibration sample record.

[0092] Let the first i The target peak area for each calibration silicon substrate sample is A. 3745,i The calibrated reference peak area is A 1070,i The normalized hydroxyl signal value was calibrated. N i The normalized hydroxyl signal value is then determined according to the following formula:

[0093] In the formula, i This indicates the serial number of the calibrated silicon substrate sample. Each calibrated silicon substrate sample generates a normalized hydroxyl signal value, which, together with the corresponding reference surface hydroxyl density, forms the calibration data. The target peak area and the reference peak area are derived from the same calibration spectrum. The ratio of the two peak areas reduces the influence of optical path transmission status and overall signal intensity variations on the calibration results.

[0094] In step S730, processor 112 performs forced zero-crossing linear regression. Processor 112 performs forced zero-crossing linear regression with the calibrated normalized hydroxyl signal value as the independent variable and the hydroxyl density of the reference surface as the dependent variable.

[0095] Processor 112 reads the normalized hydroxyl signal value and the reference surface hydroxyl density from each valid calibration sample record, and combines them into regression data points according to the same calibration sample identifier. Let the first... i The calibrated normalized hydroxyl signal value for each regression data point is... N i The hydroxyl density of the reference surface is σ i The number of calibrated silicon substrate samples included in the regression was m , m It is an integer greater than or equal to 2.

[0096] Forced zero-crossing linear regression sets the intercept of the regression line to zero, making the regression relationship express as follows: σ = kN In the formula, N This indicates the calibrated normalized hydroxyl signal value. σ This represents the surface hydroxyl density obtained from the regression relationship. k This represents the slope of the regression line. In this regression relationship, a zero calibrated normalized hydroxyl signal value corresponds to a zero surface hydroxyl density obtained from the regression. The processor 112 uses m sets of regression data points to obtain the regression line, so that the conversion coefficient reflects the overall correspondence between the calibrated normalized hydroxyl signal value and the reference surface hydroxyl density.

[0097] Processor 112 calculates the slope of the forced zero-crossing linear regression according to the following formula:

[0098] Processor 112 substitutes the slope into σ=kN to calculate the regression surface hydroxyl density corresponding to each calibration data point, and evaluates the calibration relationship by the deviation between the regression surface hydroxyl density and the reference surface hydroxyl density. During actual calibration, at least five different activation or oxidation condition points are selected within the expected process window, and A is recorded simultaneously at each condition point. 3745 A 1070 A 3745 / A 1070 The reference surface hydroxyl density, obtained by independent measurement, was recorded, along with the determination coefficient R², residual distribution, and confidence interval for forced zero-crossing regression.

[0099] In step S740, processor 112 determines the conversion factor. Processor 112 determines the slope k of the forced zero-crossing linear regression as the conversion factor. The conversion factor is used to convert the dimensionless normalized hydroxyl signal value to OH / nm. 2 This represents the surface hydroxyl density. Therefore, the conversion factor is expressed in OH / nm. 2 After the processor 112 completes the regression, it associates the conversion coefficients with the corresponding regression records and forms calibration records.

[0100] The conversion factor is stored together with the calibration of the silicon substrate sample type, silicon oxide layer type, spectral acquisition parameters, and the calculation aperture of the two peak areas. For samples before bonding with different ratios of isolated surface silanol groups and associated silanol groups, group regression can be performed separately, and the slope consistency of the conversion factor k in each group can be compared. When the preset consistency requirements are met, the corresponding conversion factor is used to convert the surface hydroxyl density during the measurement stage.

[0101] In step S750, processor 112 determines the reference peak area. Processor 112 determines the reference peak area of ​​the reference absorption peak based on the Fourier transform infrared reflectance absorption spectrum. The reference absorption peak corresponds to the stretching vibration of the silicon-oxygen bonds in the silicon oxide layer.

[0102] The reference absorption peak is located at approximately 1070 cm⁻¹ -1 The processor 112 extracts approximately 1070 cm⁻¹ from the Fourier transform infrared reflectance absorption spectrum of the current silicon substrate 150. -1 The absorption response of the silicon-oxygen bond stretching vibration at the point is calculated, and the integral area of ​​this absorption response is calculated within the preset reference peak integration range to obtain the reference peak area A. 1070 The processor 112 uses the same reference absorption peak position, reference peak integration range, and baseline processing aperture as in the calibration phase to determine the reference peak area.

[0103] In step S760, processor 112 determines the normalized hydroxyl signal value. Processor 112 determines the ratio of the target peak area to the reference peak area as the normalized hydroxyl signal value.

[0104] Processor 112 reads the target peak area A from the same Fourier transform infrared reflectance and absorption spectrum according to the silicon substrate identifier, measurement location identifier, and spectral identifier. 3745 and reference peak area A 1070 Processor 112 calculates the ratio of the two peak areas to obtain the normalized hydroxyl signal value N. OH :

[0105] In the formula, A 3745 It indicates approximately 3745 cm -1 The target peak area of ​​the isolated surface silanol target absorption peak, A 1070 It indicates approximately 1070 cm -1 The reference peak area is the same as the reference absorption peak of the silicon-oxygen bond stretching vibration. Since both peak areas originate from the same spectrum, they share the same measurement location, optical path configuration, and acquisition environment.

[0106] For the spectrum formed by the partitioned measurement, the processor 112 calculates the target peak area, the reference peak area, and the normalized hydroxyl signal value at each effective measurement position, and keeps the three corresponding to the corresponding measurement position.

[0107] In step S770, processor 112 determines the surface hydroxyl density. Processor 112 determines the surface hydroxyl density of the bonded surface based on the product of the normalized hydroxyl signal value and a pre-calibrated conversion factor.

[0108] Processor 112 selects a conversion factor k that matches the current silicon substrate material type, silicon oxide layer type, spectral acquisition parameters, and peak area calculation aperture. The surface hydroxyl density is denoted as σ. OH And determine according to the following formula:

[0109] In the formula, σ OH The unit is OH / nm 2 The processor 112 multiplies the normalized hydroxyl signal value at each valid measurement location by the conversion factor k to obtain the surface hydroxyl density at the corresponding measurement location.

[0110] This disclosure normalizes the target peak area using the reference peak area of ​​the silicon-oxygen bond stretching vibration in the same Fourier transform infrared reflectance absorption spectrum. Changes in the optical path state and the overall signal intensity jointly affect the target peak area and the reference peak area; the ratio of these two areas weakens the impact of this combined change on the evaluation results of silanol groups. After establishing the correspondence between the normalized hydroxyl signal value and the hydroxyl density of the reference surface through forced zero-crossing linear regression, a conversion factor transforms the dimensionless normalized hydroxyl signal value into a value with OH / nm. 2The surface hydroxyl density of the unit allows the bonding surfaces of different silicon substrates to be compared according to a uniform spectral processing aperture and density unit, and the surface hydroxyl density is used as a quantitative input for subsequent bonding steps.

[0111] Figure 8 This is a flowchart of the silicon substrate bonding process provided in this disclosure. Figure 8 As shown, the process includes the following steps: Step S810: The bonding surface 152 is activated using a plasma treatment device. The thickness of the silicon oxide layer is greater than 10 Å. The plasma treatment device uses any one of argon, oxygen, and nitrogen to generate plasma, with a plasma power of 100–1000 W and a treatment time of 10–120 s. The plasma acts on the bonding surface 152, changing its surface state and forming reaction sites, providing conditions for the subsequent hydrophilic treatment to form surface silanol groups.

[0112] In step S820, the cleaning equipment 180 performs a hydrophilic treatment on the bonding surface 152. The hydrophilic treatment solution is Standard Clean 1 (SC-1) or a 0.5% ammonia solution. The cleaning equipment 180 uses a tank-type cleaning device or a single-plate cleaning device to bring the plasma-activated bonding surface 152 into contact with the hydrophilic treatment solution containing hydroxide ions, forming surface silanol groups on the bonding surface 152.

[0113] In step S830, the wafer bonding system 100 determines the surface hydroxyl density of the bonding surface. The processor 112 uses the end of the hydrophilic treatment as the starting point for the measurement window, and the spectral acquisition device 120 completes Fourier transform infrared reflectance and absorption spectral acquisition within 10 minutes after the hydrophilic treatment ends. The spectral acquisition range is 400–4000 cm⁻¹. -1 Spectral resolution of 4 cm -1 During spectral acquisition, the temperature inside the measurement chamber 130 was 22±1 ℃ and the relative humidity was below 30%. Before spectral acquisition, nitrogen gas was introduced into the measurement chamber 130 through the nitrogen purging channel and purged continuously for 10 min.

[0114] The processor 112 determines the target peak area based on the acquired Fourier transform infrared reflectance absorption spectrum, normalizes the target peak area, and determines the surface hydroxyl density of the bonding surface 152 based on the normalization result. A preset surface hydroxyl density threshold can be 4.6 OH / nm². The processor 112 compares the determined surface hydroxyl density with the preset surface hydroxyl density threshold. For two silicon substrates forming the same bonding interface, the corresponding bonding surfaces can be measured separately in the above manner.

[0115] Step S840: Perform direct hydrophilic bonding or rework based on the surface hydroxyl density.

[0116] When the surface hydroxyl density of the bonding surfaces of the two silicon substrates forming the same bonding interface is greater than or equal to a preset surface hydroxyl density threshold, the bonding apparatus 170 performs direct hydrophilic bonding and annealing. In one embodiment, the surface hydroxyl density of the bonding surfaces of the two silicon substrates forming the same bonding interface before bonding is greater than or equal to 4.6 OH / nm². After the two bonding surfaces are positioned opposite each other, direct hydrophilic bonding is performed, and the resulting bonded wafer is annealed.

[0117] If the surface hydroxyl density measured in step S830 is less than the preset surface hydroxyl density threshold, the substandard silicon substrate undergoes the hydrophilic treatment in step S820 again, and then returns to step S830 to re-acquire the Fourier transform infrared reflectance and absorption spectrum to determine the surface hydroxyl density. The number of reworks is counted as the number of times step S820 is executed again, excluding the initial treatment; rework is limited to a maximum of two times. After the retest meets the threshold, direct hydrophilic bonding and annealing are performed once both bonding surfaces meet the threshold conditions; if the substrate still fails to meet the threshold after two reworks, the corresponding silicon substrate is scrapped, and subsequent bonding steps are stopped.

[0118] In step S850, the detection device measures the bonding energy at the bonding interface. A blade is inserted at a constant speed into the bonding interface between two bonded silicon substrates (e.g., silicon substrate A and silicon substrate B), and an infrared camera continuously acquires infrared images of the bonded wafer. After the crack length stabilizes, the detection device determines the distance from the blade tip to the crack tip and records this distance as the crack length L. Repeated measurements use the same blade tip reference and crack tip identification aperture.

[0119] Bond energy Determined according to the following formula:

[0120] In the formula, E1 and E2 represent the Young's modulus of silicon substrate A and silicon substrate B, respectively; t1 and t2 represent the thicknesses of silicon substrate A and silicon substrate B, respectively; h represents the blade thickness; and L represents the crack length. The parameters used in this disclosure include: E1 = E2 = 1.69 × 10⁻⁶. 11 N / m 2 t1=t2=775 μm, the measurement uncertainty of silicon substrate thickness is 20 μm, h=304 μm, the measurement uncertainty of blade thickness is 5 μm, and the measurement uncertainty of crack length is 1 mm.

[0121] The testing equipment substitutes the actual measured values ​​of the silicon substrate thickness t1, t2 and the blade thickness h into the bonding energy formula, and saves the values ​​of E1, E2, t1, t2, h, and L. Since the crack length L participates in the bonding energy calculation in the fourth power form, the measurement deviation of the crack length will have a significant impact on the bonding energy calculation result; the blade thickness h participates in the calculation in the square form, and the silicon substrate thicknesses t1 and t2 are calculated using the bending stiffness of the two silicon substrates.

[0122] When Young's modulus is considered a constant parameter, the measurement error of bond energy... Determined according to the following formula:

[0123] In the formula, This represents the measurement error of the bonding energy. and These represent the thickness measurement errors of silicon substrate A and silicon substrate B, respectively. This indicates the error in the blade thickness measurement. This indicates the error in crack length measurement. In this disclosure, , , .

[0124] To verify the relationship between surface hydroxyl density and bonding quality, verification samples with different surface hydroxyl densities were formed using different combinations of steps S810 and S820. Bonding tests were then performed on control samples, including those below the release threshold. This control test is distinct from the production process described above, which releases samples based on the threshold. The bonding verification results corresponding to different surface hydroxyl densities are shown in the table below: S1 S2 Surface hydroxyl density (OH / nm²) Bonding energy (J / m²) Bubbles after annealing N N 1.6 0.62 The two pieces did not bond after annealing. Y N 2.2 1.15 have Y N 2.9 1.43 have N Y 3.4 1.61 have N Y 3.9 1.72 have Y Y 4.2 1.83 none Y Y 4.6 2.15 none Y Y 4.9 2.43 none Y Y 5.2 2.53 none In the table, S1 indicates whether plasma activation was performed, S2 indicates whether hydrophilic treatment was performed, Y indicates that the corresponding step was performed, and N indicates that the corresponding step was not performed. Verification results show that when the surface hydroxyl density is 4.6 OH / nm², the bonding energy is 2.15 J / m², and no bubbles were observed after annealing. When the surface hydroxyl density continues to increase to 4.9 OH / nm² and 5.2 OH / nm², the bonding energies are 2.43 J / m² and 2.53 J / m², respectively.

[0125] 4.6 OH / nm² corresponds to the Kiselev-Zhuravlev constant, representing the average maximum silanol density of fully hydroxylated amorphous silica. This disclosure uses a bonding energy greater than or equal to 2 J / m² as the acceptance criterion for bonding quality. When this standard is met, the bonding strength at the interface satisfies the requirements of subsequent processes, which helps avoid delamination of the device layer during fabrication. Below this standard, there is a risk of bubble formation and interface separation in subsequent processes. Surface hydroxyl density affects bonding energy, which is used to evaluate bonding quality. Verification data from this disclosure shows that, under the indicated verification conditions, samples with a surface hydroxyl density reaching this level have a bonding energy exceeding 2 J / m², and no bubbles were observed after annealing. Bonding bubbles and bonding energy are also affected by factors such as film quality, surface roughness, silicon substrate warpage, and annealing process. Therefore, surface hydroxyl density is used to characterize the surface state before bonding, while bubble state and blade bonding energy are used to characterize the verification results after bonding.

[0126] This disclosure presents a process flow including plasma activation, hydrophilic treatment, measurement of surface hydroxyl density within a defined time period, and bonding to a silicon substrate. The bonding results are verified using blade bonding energy. For samples with a contact angle less than 5°, the contact angle meter measurement results may appear close to 0°, with no clear distinction between samples. In contrast, this disclosure provides further quantitative differentiation through FTIR target peak area, normalized hydroxyl signal value, and surface hydroxyl density, and establishes an experimental correspondence between the surface chemical state before bonding and the bonding results.

[0127] Figure 9 This is a schematic diagram of the structure of the bonding wafer 200 provided in this disclosure. The bonding wafer 200 is fabricated using the aforementioned wafer bonding method and includes two silicon substrates bonded together by direct hydrophilic bonding. Figure 9 The diagram shows a first silicon substrate 150A and a second silicon substrate 150B. A bonding interface 162 is formed between the opposing surfaces of the first silicon substrate 150A and the second silicon substrate 150B.

[0128] In one embodiment, both the bonding surfaces of the first silicon substrate 150A and the second silicon substrate 150B have a silicon oxide layer. Before bonding, the bonding surfaces of the first silicon substrate 150A and the second silicon substrate 150B are subjected to a hydrophilic treatment, and the surface hydroxyl density is determined according to the aforementioned external reflection FTIR method. After ensuring that the surface hydroxyl density of both bonding surfaces is greater than or equal to 4.6 OH / nm² before bonding, the two bonding surfaces are positioned opposite each other and directly hydrophilically bonded to form a bonded wafer 200.

[0129] The surface hydroxyl density is obtained through spectroscopic measurements before bonding. Specifically, the target peak area corresponding to the target absorption peak of the surface silanol group is determined based on the Fourier transform infrared reflectance absorption spectrum of the bonded surface, and then the corresponding surface hydroxyl density is determined by reference peak normalization and pre-calibrated conversion coefficients.

[0130] After forming the bonding wafer 200, a bonding interface 162 is formed between the two silicon substrates. The bonding quality of the bonding interface can be characterized by the bonding energy measured by the blade method. In this disclosure, the bonding energy of the bonding interface 162 between the two silicon substrates is greater than or equal to 2 J / m².

[0131] Bonding energy can be measured using the blade method and is determined based on the silicon substrate thickness, Young's modulus, blade thickness, and the crack length formed after blade insertion. Since the measured bonding energy is affected by measurement conditions and silicon substrate structural parameters, the same measurement method, test conditions, and parameter specifications should be used when comparing the bonding energies of different bonded wafers to improve the consistency of the comparison results.

[0132] Therefore, the product structure of the bonded wafer 200 corresponds to the surface state before bonding during its fabrication process: the two bonding surfaces forming the bonding interface 162 have a surface hydroxyl density that meets the preset conditions before bonding, and form a mutually bonded structure through the wafer bonding method.

[0133] The 4.6 OH / nm² value in this disclosure serves as a reference threshold for the surface hydroxyl density before bonding, corresponding to the average maximum density of silanol on a fully hydroxylated amorphous silica monolayer surface. It is understood that this value is used to define the surface state before bonding; additional hydroxyl groups may exist on the actual silicon substrate surface after reaching silanol saturation in a monolayer. Therefore, the surface hydroxyl density obtained according to the spectral calibration relationship of this disclosure can reach or exceed the reference threshold.

[0134] The technical features described in different embodiments of this disclosure can be used in combination where there is no technical conflict. Such combinations are based on the object relationships, processing conditions, and execution order of each technical feature already disclosed herein. Equivalent changes made by those skilled in the art to specific implementations based on the technical content described in this disclosure do not alter the fundamental technical relationships of the technical solutions described herein.

Claims

1. A wafer bonding method, characterized in that, include: The silicon substrate to be bonded is subjected to a hydrophilic treatment; With the bonding surface of the hydrophilic treated silicon substrate spaced apart from the spectral acquisition device, the Fourier transform infrared reflectance absorption spectrum of the bonding surface is acquired by the spectral acquisition device using external reflection. The bonding surface has a silicon oxide layer. Based on the Fourier transform infrared reflectance absorption spectrum, determine the target peak area of ​​the target absorption peak corresponding to the surface silanol group; The surface hydroxyl density of the bonded surface is determined based on the target peak area, and the surface hydroxyl density characterizes the hydrophilic state of the bonded surface. When the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, the silicon substrate is directly hydrophilically bonded. When the surface hydroxyl density is less than the preset surface hydroxyl density threshold, the silicon substrate is subjected to hydrophilic treatment again, and the surface hydroxyl density is re-determined.

2. The wafer bonding method according to claim 1, characterized in that, Determining the surface hydroxyl density of the bonded surface based on the target peak area includes: The reference peak area of ​​the reference absorption peak is determined based on the Fourier transform infrared reflectance absorption spectrum, and the reference absorption peak corresponds to the stretching vibration of the silicon-oxygen bond in the silicon oxide layer. The ratio of the target peak area to the reference peak area is determined as the normalized hydroxyl signal value; The surface hydroxyl density of the bonded surface is determined by multiplying the normalized hydroxyl signal value by a pre-calibrated conversion factor.

3. The wafer bonding method according to claim 2, characterized in that, The calibration of the conversion factor includes: The calibration target peak area, calibration reference peak area, and reference surface hydroxyl density are obtained for at least two calibration silicon substrate samples, respectively, wherein the reference surface hydroxyl density exhibits a gradient distribution; The ratio of the area of ​​the calibration target peak to the area of ​​the calibration reference peak for the same calibrated silicon substrate sample is determined as the calibrated normalized hydroxyl signal value; Forced zero-crossing linear regression was performed with the calibrated normalized hydroxyl signal value as the independent variable and the hydroxyl density of the reference surface as the dependent variable. The slope of the forced zero-crossing linear regression is determined as the conversion factor.

4. The wafer bonding method according to claim 1, characterized in that, The preset surface hydroxyl density threshold is 4.6 OH / nm. 2 .

5. The wafer bonding method according to claim 1, characterized in that, The acquisition of the Fourier transform infrared reflectance and absorption spectrum includes: The bonding surface is divided into at least two measurement zones, and measurement positions are determined in each of the at least two measurement zones; Infrared light reflected from the bonding surface is collected at the measurement positions to obtain Fourier transform infrared reflection and absorption spectra corresponding to the measurement positions.

6. The wafer bonding method according to claim 5, characterized in that, The acquisition of the Fourier transform infrared reflectance and absorption spectrum also includes: The back side of the silicon substrate is vacuum-adsorbed onto the sample platform; The infrared light is focused into spot at the measurement position, and the infrared light reflected once by the bonding surface is collected.

7. The wafer bonding method according to claim 1, characterized in that, The target absorption peak is the absorption peak of isolated surface silanol groups; The determination of the target peak area includes: Collect the background spectrum of the measurement chamber; Background subtraction and baseline correction are performed on the Fourier transform infrared reflectance and absorption spectrum based on the background spectrum; The target peak area is determined based on the Fourier transform infrared reflectance absorption spectrum after background subtraction and baseline correction.

8. The wafer bonding method according to claim 7, characterized in that, The determination of the target peak area includes: Based on the second derivative spectrum of the Fourier transform infrared reflectance absorption spectrum after background subtraction, the starting and ending inflection points of the target absorption peak are determined. The wavenumber range between the starting inflection point and the ending inflection point is defined as the integral wavenumber range; Within the specified integration wavenumber range, the target absorption peak after baseline correction is integrated to obtain the target peak area.

9. The wafer bonding method according to claim 8, characterized in that, The baseline correction includes: Multi-point rubber band baseline correction is performed on the Fourier transform infrared reflectance and absorption spectrum after background subtraction; From the Fourier transform infrared reflectance absorption spectrum after multi-point rubber band baseline correction, a local baseline endpoint is determined on each of the local baseline segments on both sides of the target absorption peak. Based on the local baseline endpoints on both sides of the target absorption peak, a two-point linear baseline correction is performed on the target absorption peak within the integral wavenumber range.

10. The wafer bonding method according to claim 7, characterized in that, The determination of the target peak area also includes: The area of ​​the interfering peak corresponding to the adsorbed water is determined based on the Fourier transform infrared reflectance absorption spectrum. Water interference correction is performed based on the area of ​​the interference peak, and the target peak area is determined based on the target absorption peak obtained after the water interference correction is completed.

11. The wafer bonding method according to claim 10, characterized in that, The water disturbance correction includes: In response to the interference peak area being greater than a preset water interference threshold, the silicon substrate is dried and the measurement chamber is purged with nitrogen. The remeasured Fourier transform infrared reflectance and absorption spectrum of the silicon substrate was acquired, and the remeasured background spectrum corresponding to the remeasured Fourier transform infrared reflectance and absorption spectrum was obtained. Background subtraction and baseline correction are performed on the re-measured Fourier transform infrared reflectance and absorption spectrum based on the re-measured background spectrum; The area of ​​the target peak is determined based on the target absorption peak in the re-measured Fourier transform infrared reflectance absorption spectrum after background subtraction and baseline correction.

12. A wafer bonding system, characterized in that, This includes control and processing devices, spectral acquisition equipment, cleaning equipment, and bonding equipment; The cleaning equipment is configured to perform a hydrophilic treatment on the silicon substrate to be bonded; The spectral acquisition device is configured to: acquire the Fourier transform infrared reflectance absorption spectrum of the bonding surface of the hydrophilically treated silicon substrate by means of external reflection, with the bonding surface having a silicon oxide layer, in a state where the bonding surface of the bonding surface of the hydrophilically treated silicon substrate is spaced apart from the spectral acquisition device. The control processing device is communicatively connected to both the spectral acquisition device and the bonding device, and is configured to: The target peak area corresponding to the target absorption peak of the surface silanol group is determined based on the Fourier transform infrared reflectance absorption spectrum. The surface hydroxyl density of the bonded surface is determined based on the target peak area, and the surface hydroxyl density characterizes the hydrophilic state of the bonded surface. When the surface hydroxyl density is greater than or equal to a preset surface hydroxyl density threshold, the bonding device is controlled to perform direct hydrophilic bonding on the silicon substrate; When the surface hydroxyl density is less than the preset surface hydroxyl density threshold, it is determined that the silicon substrate needs to undergo hydrophilic treatment again. After the hydrophilic treatment is performed again, the spectral acquisition device is controlled to re-acquire the Fourier transform infrared reflectance absorption spectrum to redetermine the surface hydroxyl density.

13. A bonding wafer, characterized in that, The bonding wafer comprises two silicon substrates bonded together by direct hydrophilic bonding and is fabricated using the wafer bonding method according to any one of claims 1 to 11; The surface hydroxyl density of the bonding surfaces of the two silicon substrates before bonding is greater than or equal to 4.6 OH / nm². The bonding energy at the bonding interface between the two silicon substrates is greater than or equal to 2 J / m 2 .

Images