Wafer level multi-material thin film deposition instability detection method

Abstract

The present disclosure provides a wafer-level multi-material thin film deposition instability detection method, comprising: arranging an ellipsometric characterization unit and an oxygen partial pressure probe in an inter-cavity transfer channel, discretizing a wafer surface into a plurality of measurement points, and constructing a transfer gap interface exposure matrix; obtaining the equivalent optical thickness and in-plane coverage of each measurement point, and constructing an interface degradation feature matrix; extracting the curvature change rate, fusing the curvature change rate and the interface degradation feature matrix, and constructing an interface bonding force distribution field; fusing the new and old interlayer mismatch stress and the interface bonding force distribution field, obtaining the peeling instability tendency index of each measurement point, and constructing a wafer-wide peeling instability tendency distribution; screening out the super-threshold measurement points with the peeling instability tendency index greater than the dynamic judgment threshold, spatially clustering the super-threshold measurement points, delineating the implicit instability region, and outputting the implicit instability detection result containing the position coordinates of the implicit instability region, the cause interface layer sequence number and the confidence.

Description

Technical Field

[0001] This invention relates to the field of thin film technology, and in particular to a wafer-level multi-material thin film deposition instability detection method. Background Technology

[0002] Wafer-level multi-material thin-film deposition is a fundamental process in the manufacturing of advanced logic devices, memory devices, and RF devices. In the stacked structures of these devices, metal layers, dielectric layers, barrier layers, and seed layers are stacked alternately according to the process formulation. A single wafer often requires the deposition of dozens or even hundreds of different materials. Because different materials require significantly different deposition methods, operating atmospheres, and operating temperatures, metal layers are typically deposited in physical vapor deposition (PVD) or electrochemical deposition (ECD) chambers, while dielectric layers are typically deposited in chemical vapor deposition (CVD) or atomic layer deposition (ALD) chambers. A single chamber cannot accommodate the deposition conditions of all materials. Therefore, mass production lines generally employ a multi-chamber cluster architecture, where a robotic arm in a central transfer module transfers the wafer back and forth between the deposition chambers, thus completing the continuous fabrication of multi-material stacks within a single wafer loading cycle. The bonding state of the interfaces within the stack directly determines whether the thin film will peel, flake, or develop interface cracks during subsequent chemical mechanical polishing (CMP), photolithography, and packaging processes. This is a key factor affecting the overall wafer yield and device reliability.

[0003] For instability detection during multi-material thin film deposition, existing technologies mainly rely on in-situ monitoring within the deposition chamber. A typical approach involves deploying an in-situ optical reflectance spectroscopy unit, a chamber temperature field sensing array, and a deposition rate monitoring component within the deposition chamber. This simultaneously acquires the reflection intensity, temperature, and instantaneous deposition rate at various measurement points on the wafer surface, constructing a multi-source raw signal matrix. Based on this, the signals are denoised and normalized. Material layers are segmented according to the time sequence of the deposition formulation. The film thickness at each measurement point is obtained by integrating the deposition rate, and the intra-layer film thickness variance is calculated. Then, combining the elastic modulus and thermal expansion coefficient of each layer, the interfacial mismatch stress between adjacent material layers is calculated using the Stony equation, yielding the stress distribution field and stress gradient across the entire wafer. Finally, the stress distribution field and film thickness variance are fused into an instability discrimination feature vector. The instability index at each measurement point is weighted and calculated. After comparison with a preset threshold, the instability region is clustered and located, outputting the detection result containing the instability type, location coordinates, and confidence level. All inputs for this type of scheme are taken from the in-situ acquisition channel within the deposition chamber. Its criteria are based on two macroscopically measurable types: film thickness anomaly and stress anomaly. It has a certain ability to detect instability morphologies that have become apparent, such as uneven film thickness, local peeling, and interface cracks.

[0004] However, during wafer transfer between multiple cavities, the deposited surface is exposed to a transitional environment, resulting in extremely thin interfacial oxidation and adsorption contamination. The transfer gaps do not fall within the in-situ sampling timeline of any deposition cavity. This degraded layer, with a thickness on the sub-nanometer scale, does not alter the macroscopic film thickness or cause significant changes in the stress distribution field, yet it weakens interlayer adhesion and becomes the starting point for subsequent peeling instability. Existing methods are completely insensitive to this type of interfacial degradation that occurs outside the cavity but determines subsequent instability within the cavity. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to resolve the aforementioned deficiencies and propose a wafer-level multi-material thin film deposition instability detection method and system.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of this invention discloses a wafer-level multi-material thin film deposition instability detection method, comprising: Elliptic characterization units and oxygen partial pressure probes are arranged in the intercavity transfer channel. The wafer surface is discretized into several measurement points. The exposure time, time-averaged effective oxygen partial pressure and elliptic reflection phase difference of each measurement point are collected to construct the transfer gap interface exposure matrix. The exposure time is the length of the exposure period of the measurement point from the opening of the gate valve of the previous deposition cavity to the closing of the gate valve of the next deposition cavity. Based on the exposure time and average effective oxygen partial pressure in the transfer gap interface exposure matrix, the estimated thickness of the degradation layer at each measurement point is obtained. Based on the elliptic reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of the degradation layer, the equivalent optical thickness and in-plane coverage of each measurement point are obtained, and an interface degradation feature matrix is ​​constructed. The curvature time sequence of each of the measurement points after the wafer enters the next deposition chamber is collected during the early stage of the growth of the new material layer. The curvature change rate is extracted and fused with the interface degradation feature matrix to obtain the interface bonding force attenuation factor of each measurement point, and the interface bonding force distribution field is constructed. Based on the curvature time sequence, the new and old interlayer mismatch stresses at each measurement point are obtained. The new and old interlayer mismatch stresses are fused with the interface bonding force distribution field to obtain the peeling instability tendency index at each measurement point, and the whole wafer peeling instability tendency distribution is constructed. A dynamic judgment threshold is generated based on the distribution of the whole wafer peeling instability tendency. Exceeding threshold measurement points with peeling instability tendency index greater than the dynamic judgment threshold are selected. Spatial clustering is performed on the exceeding threshold measurement points to delineate the hidden instability region. The hidden instability detection result containing the location coordinates of the hidden instability region, the index of the inducing interface layer and the confidence level is output.

[0008] Furthermore, the wafer surface is discretized into several measurement points, specifically including: The elliptic characterization unit and the oxygen partial pressure probe are arranged above the robotic arm's travel path along the intercavity transfer channel. The elliptic characterization unit uses a single-wavelength laser source, and the incident angle is set in the Brewster angle range of the metal and dielectric stack. As the wafer passes beneath the elliptic characterization unit at a constant speed with the robotic arm, the wafer surface is discretized into several measurement points according to the equal arc length and equal radius ring zone division method, and each measurement point is assigned a measurement point polar coordinate composed of the radius and the azimuth angle relative to the travel direction. The oxygen partial pressure probe is installed on the inner wall of the intercavity transfer channel, and the polar coordinates of the measuring points are used for calculating the exposure time of each measuring point and for collecting the ellipsoidal reflection phase difference.

[0009] Furthermore, each measuring point is assigned polar coordinates consisting of the radius and the azimuth angle relative to the direction of travel, specifically including: The exposure time of each measuring point is synthesized from three factors: the fixed reference time for the opening and closing of the gate valve and the start and stop of the robot arm, the ratio of the geometric projection position of each measuring point in the direction of travel to the linear velocity of the robot arm, and the edge stagnation correction factor linearly amplified according to the normalized radius of the measuring point. According to the exposure start and end time window of each measurement point, the channel oxygen partial pressure time sequence signal continuously collected by the oxygen partial pressure probe is extracted, and the arithmetic mean within the window is taken as the time-averaged effective oxygen partial pressure of that measurement point. The p-polarized reflected light intensity and s-polarized reflected light intensity are collected at each measurement point by the elliptic characterization unit. The elliptic reflection phase difference at the measurement point is solved by the analyzer rotation method. The elliptic reflection phase difference contains an integer number of phase order determined by the phase continuity expansion of adjacent measurement points.

[0010] Furthermore, based on the ellipsoidal reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of the degradation layer growth, the equivalent optical thickness and in-plane coverage of each measurement point are obtained, specifically including: According to the self-limiting growth law, the estimated thickness of the degradation layer is synthesized by the self-limiting saturation thickness, the reference oxidation rate constant, the exposure time and the power ratio of the time-averaged effective oxygen partial pressure to the calibrated reference oxygen partial pressure. Using the estimated thickness of the degradation layer as the initial value for iteration, the elliptic reflection phase difference is substituted into the thin-film dielectric phase delay model. The equivalent optical thickness is solved according to the refractive index of the degradation layer, the refractive index of the channel atmosphere, the refractive index of the substrate of the underlying deposited film, and the elliptic incident angle. The iteration convergence criterion is set to the thickness change between two adjacent iterations being less than a given convergence amount. The ratio of the equivalent optical thickness to the estimated thickness of the degradation layer is taken as the in-plane coverage, and this ratio is limited to the range of zero to one.

[0011] Furthermore, the curvature time sequence of each measurement point after the wafer enters the next deposition chamber is collected during the initial stage of new material layer growth, and the rate of curvature change is extracted, specifically including: The multi-beam laser curvature meter in the subsequent deposition cavity scans the array of measurement points that are registered one by one with the measurement points point by point, and calculates the local curvature from the relative change of the spacing between the reflected light spots to obtain the curvature time sequence of each measurement point. A first-order linear least squares fit is performed on the curvature time sequence of each measurement point within the acquisition window. The initial sampling point caused by chuck adsorption jitter is removed, and the slope of the fitted straight line is taken as the curvature change rate of the measurement point. For measurement points whose standard deviation of the fitted residual exceeds the first residual threshold, the curvature change rate is made up by the average curvature change rate of the adjacent measurement points.

[0012] Furthermore, the curvature change rate is fused with the interface degradation feature matrix to obtain the interface bonding force attenuation factor for each of the measurement points, specifically including: In the interface degradation feature matrix, measurement points with equivalent optical thickness less than a first thickness threshold and in-plane coverage less than a first coverage threshold are selected to form a reference measurement point subset. The arithmetic mean of the curvature change rate of each measurement point in the reference measurement point subset is taken as the non-degradation reference curvature change rate. The bonding loss ratio is formed by the product of the in-plane coverage and the exponential attenuation rate of the equivalent optical thickness relative to the bonding barrier feature thickness, and the barrier path is raised to the power of the residual bonding area ratio obtained by subtracting the bonding loss ratio. The mechanical path power of the ratio of the curvature change rate to the non-degradable reference curvature change rate is taken, and the power is multiplied by the barrier path power of the residual bonding area ratio to obtain the interface bonding force attenuation factor.

[0013] Furthermore, the interlayer mismatch stress and the interfacial bonding force distribution field are fused to obtain the peeling instability tendency index for each measuring point, specifically including: The interface bonding force attenuation factor of each measuring point in the interface bonding force distribution field is sorted from large to small, and the arithmetic mean of the high-value measuring points after sorting is taken as the benchmark bonding force attenuation factor. The quotient of the product of the mismatch stress between the old and new layers, the attenuation factor of the interfacial bonding force, and the intrinsic critical peel stress of the interface is taken as the overload ratio, and the overload ratio is assigned an overload path weight. The sharpening power of the ratio of the interface bonding force attenuation factor to the benchmark bonding force attenuation factor is taken as the normalized bonding force loss, and a relative loss path weight is assigned to the normalized bonding force loss. The weighted overload ratio is added to the weighted normalized binding force loss to obtain the stripping instability tendency index.

[0014] The second aspect of this invention discloses a wafer-level multi-material thin film deposition instability detection system, applied to the method described in the first aspect, comprising: The interface exposure acquisition module is used to arrange elliptic characterization units and oxygen partial pressure probes in the intercavity transfer channel, discretize the wafer surface into several measurement points, and acquire the exposure duration, time-averaged effective oxygen partial pressure and elliptic reflection phase difference of each measurement point to construct the interface exposure matrix of the transfer gap. The exposure duration is the length of the exposure period of the measurement point from the opening of the gate valve of the previous deposition cavity to the closing of the gate valve of the next deposition cavity. The interface degradation inversion module is used to obtain the estimated thickness of degradation layer growth at each measurement point based on the exposure time and time-averaged effective oxygen partial pressure in the transfer gap interface exposure matrix, and to obtain the equivalent optical thickness and in-plane coverage at each measurement point based on the elliptic reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of degradation layer growth, thereby constructing an interface degradation feature matrix. The bonding force attenuation estimation module is used to collect the curvature time sequence of each of the measurement points after the wafer enters the next deposition chamber during the early stage of new material layer growth, extract the curvature change rate, fuse the curvature change rate with the interface degradation feature matrix to obtain the interface bonding force attenuation factor of each of the measurement points, and construct the interface bonding force distribution field. The peeling tendency fusion module is used to obtain the interlayer mismatch stress of the new and old layers at each of the measurement points based on the curvature time sequence, fuse the interlayer mismatch stress of the new and old layers with the interface bonding force distribution field to obtain the peeling instability tendency index of each of the measurement points, and construct the whole wafer peeling instability tendency distribution. The instability determination output module is used to generate a dynamic determination threshold based on the full wafer peeling instability tendency distribution, filter out the over-threshold measurement points whose peeling instability tendency index is greater than the dynamic determination threshold, perform spatial clustering on the over-threshold measurement points, delineate the latent instability region, and output the latent instability detection result containing the location coordinates of the latent instability region, the index of the inducing interface layer, and the confidence level.

[0015] A third aspect of the present invention discloses a terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0016] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method described in the first aspect.

[0017] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages: (1) This invention arranges elliptic characterization units and oxygen partial pressure probes in the intercavity transfer channel, incorporating the transfer gap, which was originally outside the in-situ acquisition sequence of any deposition cavity, into point-by-point characterization. The estimated thickness of the degradation layer growth is given by the exposure time and the time-averaged effective oxygen partial pressure as the initial value and constraint for optical inversion. Then, the equivalent optical thickness is solved by the elliptic reflection phase difference and the in-plane coverage is separated. Oxidation and adsorption contamination of extremely thin interfaces do not change the macroscopic film thickness, which cannot be perceived by existing schemes that rely on intracavity film thickness and stress signals. This invention transforms this type of interface degradation into comparable features with thickness and morphology dimensions, filling the acquisition gap of external interface degradation.

[0018] (2) This invention correlates the interface degradation characteristics collected outside the cavity with the rate of curvature change during the initial growth of the new material layer after the wafer returns to the deposition cavity, obtaining an interface bonding attenuation factor. This factor is then fused with the mismatch stress between the old and new layers to form a peeling instability tendency index. After dynamic threshold screening and spatial clustering, the detection results are output, including location coordinates, the index of the inducing interface layer, and confidence level. Thus, a type of latent instability, where the film thickness is normal, the stress does not exceed the conventional criteria, but the interface is weakened by an extremely thin degradation layer and is on the verge of peeling, can be located at the wafer coordinates and traced to the specific interface layer before the subsequent stacking deposition is completed. This enables early quantitative detection and tracing of instability caused by external factors. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the framework of a wafer-level multi-material thin film deposition instability detection system. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention discloses a wafer-level multi-material thin film deposition instability detection method, addressing the situation where the surface of a wafer is exposed to a transitional environment during the transfer between different deposition cavities in the fabrication of multi-material stacked devices, resulting in extremely thin interface oxidation or adsorption contamination. This type of interface degradation does not change the macroscopic film thickness, occurs outside the in-situ sampling sequence of any deposition cavity, but weakens interlayer bonding and becomes the starting point for subsequent peeling instability.

[0023] This method deploys an ellipsometry characterization unit and an oxygen partial pressure probe within the intercavity transfer channel, discretizing the wafer surface into several measurement points. The exposure time, time-averaged effective oxygen partial pressure, and ellipsometry reflection phase difference are collected for each measurement point to construct an exposure matrix for the transfer gap interface. The intercavity transfer channel refers to a closed pathway connecting the preceding and following deposition chambers, allowing the wafer to travel back and forth via a robotic arm. The wafer does not undergo any deposition within this pathway. The ellipsometry characterization unit is an optical component installed within this pathway that emits polarized laser light onto the wafer surface and receives its reflected light. The oxygen partial pressure probe is a sensing component installed on the inner wall of this pathway that continuously measures the oxygen partial pressure within the pathway. The exposure time refers to the length of the exposure period for that measurement point from the opening of the gate valve in the preceding deposition chamber to the closing of the gate valve in the following deposition chamber. The time-averaged effective oxygen partial pressure refers to the time-averaged value of the oxygen partial pressure in the pathway during the exposure period at that measurement point. The ellipsometry reflection phase difference refers to the phase delay between the p-polarized and s-polarized reflection components at that measurement point. The transfer gap interface exposure matrix refers to the data structure formed by assembling the above three types of quantities row by row according to the measurement point number.

[0024] Specifically, an ellipsometry unit and an oxygen partial pressure probe are arranged above the robot's travel path along the intercavity transfer channel. The ellipsometry unit uses a single-wavelength laser source, specifically a 632.8nm helium-neon laser, with an incident angle set within the Brewster angle range of the metal-dielectric stack, specifically 70° to 75°. This angle range is chosen because it approximates the Brewster angle of most metal-dielectric stacks, where the p-polarized reflection component has the highest phase sensitivity to the surface-mount thin film layer. The oxygen partial pressure probe is mounted on the inner wall of the intercavity transfer channel, with a sampling period of 20ms to 30ms. As the wafer passes uniformly below the ellipsometry unit with the robot, the wafer surface is discretized into N measurement points using equal arc length and equal radius rings, where N ranges from 48 to 64. Each measurement point covers approximately 1 / N of the total wafer area. The wafer diameter D ranges from 200mm to 300mm. Each measuring point is assigned polar coordinates consisting of the radius r_i and the azimuth angle φ_i relative to the direction of travel, where i is the measuring point number, ranging from 1 to N. These polar coordinates are used for calculating the exposure time and acquiring the ellipsoidal reflection phase difference for each measuring point.

[0025] Understandably, this partitioning process transforms the transfer gaps, which originally had no acquisition timing coverage, into an array of measurement points with clear spatial resolution. The degradation of the external cavity interface thus yields a point-by-point locatable carrier, which is a prerequisite for all subsequent quantization.

[0026] Specifically, the exposure time of each measuring point is composed of three factors: the fixed reference time for the opening and closing of the gate valve and the start and stop of the robot arm; the quotient of the geometric projection position of each measuring point in the direction of travel and the linear velocity of the robot arm; and the edge stagnation correction factor linearly amplified by the normalized radius of the measuring point. The wafer edge is exposed before the center and then closed after the center. The actual exposure time of each measuring point varies depending on its projection position in the direction of travel and is calculated by the following formula: ; in, The exposure time at the i-th measurement point is expressed in seconds. The fixed reference time for the opening and closing of the gate valve and the start and stop of the robot arm is measured in seconds, ranging from 3.0s to 4.5s, and is obtained from the equipment action sequence calibration. The wafer diameter is measured in meters (m). The radius of the i-th measurement point from the center of the wafer is in meters. Let be the azimuth angle of the i-th measuring point relative to the direction of travel, in rad; The linear velocity of the robotic arm is expressed in m / s, ranging from 0.10 m / s to 0.15 m / s. The edge retention correction factor is dimensionless and ranges from 0.12 to 0.18. This range is chosen because the airflow within the channel forms a retention zone at the wafer edge, and the actual exposure time of the edge measurement point is longer than the purely geometrically calculated value; statistical measurements show that the excess is concentrated within this range. The core of this formula is that the first term on the right-hand side is the fixed operating time of the equipment, and the second term is the exposure time difference between the measurement point and the wafer center, obtained by dividing the geometric projection position of the measurement point in the direction of travel by the linear velocity. The correction factor within parentheses linearly amplifies the edge retention effect according to the normalized radius of the measurement point.

[0027] Specifically, the time-series oxygen partial pressure signals continuously acquired by the oxygen partial pressure probe are extracted according to the exposure start and end time window for each measuring point, and the arithmetic mean within this window is taken as the time-averaged effective oxygen partial pressure for that measuring point. The unit is Pa. The typical order of magnitude of the oxygen partial pressure in the channel is 1 × 10⁻³ Pa to 5 × 10⁻² Pa.

[0028] Specifically, the ellipsoidal characterization unit collects the p-polarized and s-polarized reflected light intensities at each measurement point, and then solves the ellipsoidal reflection phase difference at that measurement point using the analyzer rotation method. The analyzer rotation method instructs the analyzer to rotate continuously within a set angle range, and the polarization state is calculated from the change in light intensity with angle. This phase difference contains an integer number of phase orders determined by the phase continuity expansion of adjacent measurement points, and is calculated using the following formula: ; in, Let be the elliptic reflection phase difference at the i-th measuring point, in rad; and These are the p-polarized reflected light intensity and s-polarized reflected light intensity at the i-th measurement point, respectively, in W / m². The analyzer rotation angle is expressed in rad, with a scanning range of 0 to π / 2. The phase order integer of the i-th measurement point is determined by the phase continuity expansion of adjacent measurement points and can be either 0 or 1. The core of this formula is that the arctangent function contains the interference phase ratio of the two polarization components. Its numerator and denominator are both dimensionless light intensity, and the ratio is dimensionless. The arctangent result and the additional phase order term are both in radians.

[0029] Specifically, the exposure duration, time-averaged effective oxygen partial pressure, and ellipsometric reflection phase difference are aligned according to the measurement point number i, and assembled row by row into an N-row, 3-column interface exposure matrix. In this matrix, the first column of the i-th row represents the exposure duration, the second column represents the time-averaged effective oxygen partial pressure, and the third column represents the ellipsometric reflection phase difference. This matrix, along with the polar coordinates of the measurement points, is used for subsequent inversion of the equivalent optical thickness and in-plane coverage of the interface degradation layer.

[0030] This method uses the exposure duration and average effective oxygen partial pressure in the interface exposure matrix of the transfer gap to obtain the estimated thickness of the degradation layer at each measuring point. Based on the ellipsoidal reflection phase difference in this matrix and the estimated thickness of the degradation layer, the equivalent optical thickness and in-plane coverage of each measuring point are obtained, thus constructing an interface degradation feature matrix. The estimated thickness of the degradation layer refers to the prior value of the interface degradation layer thickness calculated based on the environmental exposure according to the self-limiting growth law. The equivalent optical thickness refers to the interface degradation layer thickness value obtained by inverting the ellipsoidal reflection phase difference; the difference between the two is that the former originates from optical measurements, while the latter is derived from environmental calculations. In-plane coverage refers to the area ratio occupied by the interface degradation layer within a single measuring point. The interface degradation feature matrix refers to the data structure formed by assembling the equivalent optical thickness and in-plane coverage row by row according to the measuring point number.

[0031] Specifically, the initial oxidation and adsorption of metal or dielectric surfaces at room temperature in a low oxygen partial pressure environment follows a self-limiting growth law, meaning that as the degradation layer thickens, the penetration resistance of oxygen atoms increases and the growth rate decreases. The estimated thickness of the degradation layer is synthesized based on the power ratio of the self-limiting saturation thickness, the baseline oxidation rate constant, the exposure time, and the average effective oxygen partial pressure relative to the calibrated baseline oxygen partial pressure, and is calculated using the following formula: ; in, Estimate the thickness of the degradation layer at the i-th measurement point, in nm. The self-limiting saturation thickness of the material surface at room temperature is expressed in nm, ranging from 1.8 nm to 2.4 nm. The reason for choosing this range is that the thickness of the primary oxide layer on the surface of common metals and dielectrics in a low-oxygen environment at room temperature generally converges to the order of 2 nm. The reference oxidation rate constant, in nm / s, ranges from 0.28 nm / s to 0.36 nm / s, and is obtained from offline ellipsometry calibration of wafers in the same batch under calibrated oxygen partial pressure. To calibrate the reference oxygen partial pressure, the unit is Pa, ranging from 8×10⁻³Pa to 1.0×10⁻²Pa, consistent with the channel atmosphere used in the calibration experiment; The oxygen partial pressure exponent is dimensionless and ranges from 0.42 to 0.52. This range is chosen because the oxidation rate in the low-pressure region exhibits a sublinear dependence on oxygen partial pressure, with the measured exponent concentrated around 0.5. The core of this formula lies in the fact that the area within square brackets represents the saturated response of self-limiting growth, the first factor within the exponent term represents the normalized growth process, and the power of the oxygen partial pressure ratio represents the environmental acceleration factor. When the exposure duration approaches zero, the estimated thickness is zero; when the exposure duration is sufficiently long, the estimated thickness converges to the self-limiting saturation thickness, consistent with the physical nature of self-limiting oxidation.

[0032] Understandably, this estimated thickness is based on the physical prior of two environmental factors: exposure duration and average effective oxygen partial pressure. Ellipsometry inversion in the extremely thin layer region involves multiple couplings between thickness and refractive index; this prior provides initial convergence values ​​and reasonable constraints for subsequent optical inversion.

[0033] Specifically, using the estimated thickness of the degradation layer as the initial value for iteration, the elliptic reflection phase difference is substituted into the thin-film dielectric phase retardation model. The equivalent optical thickness is then calculated using the refractive index of the degradation layer, the refractive index of the channel atmosphere, the refractive index of the substrate with the underlying deposited film, and the elliptic incident angle, and is calculated as follows: ; in, The equivalent optical thickness at the i-th measurement point is expressed in nm. The wavelength of the ellipsoidal light source is in nm, and is taken as 632.8 nm. The refractive index of the degradation layer is dimensionless and ranges from 1.62 to 1.78. The reason for choosing this range is that the refractive index of the mixed layer of interfacial oxides and adsorbed contaminants is between that of dense oxides and organic adsorbents. Let be the refractive index of the channel atmosphere, dimensionless, and taken as 1.000; The refractive index of the substrate under which the film has been deposited is dimensionless and can be found in the optical constants table of the material in the deposition formulation. The elliptic incident angle, in rad, is taken as the radian value corresponding to 70° to 75°. The core of this formula lies in the fact that the difference between the effective longitudinal wave vectors of the p-polarized and s-polarized components within the thin layer is expressed in parentheses in the denominator, obtained by expanding the Snell's refraction relation. The numerator is the product of the phase difference and the wavelength. This formula converts the phase delay measured by ellipticization into the geometric thickness of the thin layer. The solution uses the estimated thickness of the degraded layer growth as the initial value, substitutes it into the forward ellipticization model, calculates the theoretical phase difference, compares it with the measured elliptic reflection phase difference, and adjusts the equivalent optical thickness using the Gauss-Newton method. The iterative convergence criterion is set to the thickness change between two adjacent iterations being less than a given convergence amount, which is taken as 0.018 nm to 0.022 nm. The reason for choosing this range is that this magnitude is less than 1 / 10 of the thickness of a single oxide molecular layer and higher than the thickness resolution limit of the ellipticizer. The maximum number of iterations is set to 40.

[0034] Specifically, the interface degradation layer does not completely cover the area measured; its equivalent optical thickness is the product of the local true thickness of the covered area and the ratio of the covered area. Using the predicted thickness of the degradation layer growth as the local true thickness of the covered area, the ratio of the equivalent optical thickness to the predicted thickness of the degradation layer growth is taken as the in-plane coverage. The value is dimensionless and is limited to the range of 0 to 1, with values ​​exceeding 1 being taken as 1.

[0035] Understandably, in-plane coverage separates the morphological information of whether the interface degradation layer is a uniform thin overlay or a locally thick island from a single thickness measure. Locally thick island degradation weakens interlayer bonding more strongly than uniform thin overlay, and this measure provides a morphological dimension in addition to the thickness dimension for subsequent estimation of interface bonding attenuation.

[0036] Specifically, the equivalent optical thickness and in-plane coverage are aligned according to the measurement point number i and assembled into an N-row, 2-column interface degradation feature matrix. The first column of the i-th row of this matrix represents the equivalent optical thickness, and the second column represents the in-plane coverage. This matrix, along with the polar coordinates of the measurement points, is used for subsequent calculations of the interface bonding force attenuation factor.

[0037] This method acquires the curvature time series of each measurement point after the wafer enters the subsequent deposition chamber during the early stage of new material layer growth. The curvature change rate is extracted and fused with the interface degradation feature matrix to obtain the interface bonding force attenuation factor for each measurement point, thus constructing the interface bonding force distribution field. The curvature time series refers to the sequence of local curvatures of the measurement point arranged according to the sampling sequence number within the acquisition window after the initial deposition of the new material layer. The curvature change rate refers to the slope of the linear fit of this sequence within the acquisition window. The interface bonding force attenuation factor is a dimensionless quantity characterizing the degree of interface bonding force retention at the measurement point; a value close to 1 indicates good interface bonding force retention, while a value close to 0 indicates severe interface bonding force attenuation. The interface bonding force distribution field refers to the full wafer distribution formed by assembling the interface bonding force attenuation factors point by point according to the measurement point number.

[0038] Specifically, after the wafer enters the next deposition cavity from the inter-cavity transfer channel and completes chuck adsorption, a multi-beam laser curvature meter within the next deposition cavity scans the measurement point array, which is registered point-to-point with the measurement points. The multi-beam laser curvature meter is an intracavity measurement component that emits a parallel laser array to irradiate the wafer surface and calculates the local curvature from the relative changes in the spacing of the reflected laser spots. Its laser spot array spacing is set to 3mm to 4mm, the curvature sampling period is set to 40ms to 50ms, and the acquisition window is set to 0s to 6s after the start of new material layer deposition. During scanning, the measurement point index i strictly follows the polar coordinate registration method, ensuring a one-to-one correspondence between curvature measurement points and ellipsoidal measurement points in the wafer coordinate system, thereby obtaining the curvature time sequence of each measurement point. The unit is m⁻¹, and k is the sampling point number, ranging from 1 to M.

[0039] Understandably, this registration process causes the interface degradation characteristics acquired outside the cavity and the mechanical response acquired inside the cavity to fall on the same measurement point index. The correlation between external causes and intracavitary manifestations is thus established, which is the key focus of this method for sensing external interface degradation.

[0040] Specifically, a first-order linear least-squares fit is performed on the curvature time series of each measuring point within the acquisition window, and the slope of the fitted line is taken as the rate of curvature change at that measuring point. The unit is m⁻¹·s⁻¹. Before fitting, initial sampling points caused by chuck adsorption jitter are removed; the first two sampling points are discarded. For measurement points where the standard deviation of the fitting residual exceeds the first residual threshold, the mean of the curvature change rate of adjacent measurement points is used to fill in the gap. The first residual threshold refers to the boundary of the standard deviation of the curvature residual used to determine whether the curvature fitting is effective; it is taken from 0.018 m⁻¹ to 0.022 m⁻¹. The reason for choosing this range is that this magnitude is comparable to the single-point curvature repeatability of a multi-beam laser curvature meter.

[0041] Understandably, the rate of curvature change compresses the noisy curvature time series into a single slope quantity. When the interface is well-bonded, stress is effectively transferred across the interface, and the curvature increases linearly with the film thickness; when a degradation layer exists at the interface, stress transfer is hindered, and the rate of curvature increase is suppressed. This slope quantity carries direct mechanical evidence of the interface bonding state.

[0042] Specifically, measurement points with equivalent optical thickness less than a first thickness threshold and in-plane coverage less than a first coverage threshold are selected from the interface degradation feature matrix to form a subset of benchmark measurement points. The first thickness threshold refers to the equivalent optical thickness limit for determining whether the thickness of the interface degradation layer is negligible, ranging from 0.28 nm to 0.34 nm. The reason for choosing this range is that degradation layers below this thickness are less than one oxide molecule layer, and their weakening effect on interfacial bonding is negligible. The first coverage threshold refers to the in-plane coverage limit for determining whether the interface degradation material exhibits a sparse point distribution, ranging from 0.16 to 0.20. The reason for choosing this range is that when the coverage is below this value, the degradation material exhibits a sparse point distribution, and the main interface remains in clean contact. The arithmetic mean of the curvature change rates of all measurement points within the subset of benchmark measurement points is taken as the non-degradation benchmark curvature change rate. The unit is m⁻¹·s⁻¹. When the number of measurement points in the reference measurement point subset is less than 1 / 8 of N, the historical calibration value of the previous stack of the same material combination of wafers in this batch is used as the non-degradation reference curvature change rate.

[0043] Understandably, the rate of curvature change without degradation is established in-situ on the same wafer and in the same deposition cycle as a reference for the rate of curvature growth that should be expected when the interface is without degradation. Subsequent calculations of adhesion degradation are thus unaffected by batch-to-batch deposition rate drift and cavity temperature differences; the degradation is attributed to the interface degradation itself.

[0044] Specifically, the attenuation of interfacial bonding strength originates from two independent pathways. The first is the physical barrier of the degradation layer itself; the thicker and wider the degradation layer, the smaller the clean contact area for chemical bonding at the interface, and the attenuation of bonding strength due to thickness is exponential. The second is the mechanical performance in the early stages of new material layer deposition; the drop in the rate of curvature change relative to the rate of curvature change of the non-degradable reference layer reflects a decrease in stress transfer efficiency across the interface. The bonding loss ratio is calculated by multiplying the in-plane coverage and the equivalent optical thickness relative to the bonding barrier feature thickness by the exponential attenuation rate. Taking the barrier pathway power of the residual bonding area ratio obtained by subtracting this bonding loss ratio from 1, and the mechanical pathway power of the ratio of the rate of curvature change to the rate of curvature change of the non-degradable reference layer, the two are multiplied and calculated using the following formula: ; in, Let be the attenuation factor of the interfacial bonding force at the i-th measuring point, which is dimensionless and ranges from 0 to 1; The characteristic thickness of the bonding barrier is measured in nm and ranges from 0.52 nm to 0.62 nm. The reason for choosing this range is that the interaction range of interfacial chemical bonding is about 1 to 2 atomic layers. When the thickness of the degradation layer reaches this order of magnitude, the bonding is basically blocked. The characteristic length of the measured bonding strength decay is concentrated around 0.55 nm. The barrier pathway weight index is dimensionless and ranges from 1.15 to 1.35. The reason for choosing this range is that physical barriers weaken the binding force slightly more than linearly, and the measured fitting index is concentrated around 1.25. The mechanical pathway weighting index is dimensionless and ranges from 0.68 to 0.82. This range is chosen because the rate of curvature change is also affected by the growth pattern of the new material layer itself, and its indicative effect on bonding strength is weaker than linear. The measured fitting index is concentrated around 0.75. The core of this formula is that the exponential attenuation rate in the first product term is the attenuation rate of the degradation layer thickness on the bonding capacity per unit area. When the thickness approaches zero, this attenuation rate is zero; when the thickness is much greater than the bonding barrier feature thickness, this attenuation rate approaches 1. Multiplying this by the in-plane coverage yields the bonding loss ratio over the entire measurement area. Subtracting 1 gives the residual bonding area ratio. Taking the power of the barrier pathway power reflects the superlinear attenuation of the barrier. The second product term is the normalized ratio of the rate of curvature change relative to the rate of curvature change of the non-degradation baseline. Taking the power of the mechanical pathway power reflects the sublinear indicative strength of the mechanical pathway. When the thickness of the degradation layer approaches zero and the rate of curvature change is equal to the rate of curvature change of the reference curvature without degradation, the interfacial bonding force attenuation factor is equal to 1, and the interfacial bonding force does not attenuate, which is consistent with physical intuition.

[0045] Understandably, the interface bonding attenuation factor integrates the geometric features of interface degradation acquired from outside the cavity with the mechanical response acquired from inside the cavity into a single dimensionless quantity. This quantity is independent of both macroscopic film thickness and a single optical signal. The implicit interface degradation, which originally did not change the macroscopic film thickness and was not covered by the acquisition time sequence, thus obtains a quantifiable and comparable characterization form, filling the gap in the perception of external interface degradation in existing schemes.

[0046] Specifically, the interfacial bonding force attenuation factor of each measuring point is assembled one by one according to the measuring point number i, along with the polar coordinates of the measuring point and the material layer number to which the current interface belongs, forming a full-wafer interfacial bonding force distribution field. The material layer number to which the current interface belongs refers to the layer number of the interface formed by this transfer gap in the stacked structure. This distribution field is used for subsequent fusion with the mismatch stress between the old and new layers and for calculating the peeling instability tendency index. The material layer number is referenced when subsequently annotating the causal interface layer number.

[0047] This method obtains the interlayer mismatch stress at each measurement point based on the curvature time sequence. This interlayer mismatch stress is then fused with the interfacial bonding force distribution field to obtain the peeling instability tendency index for each measurement point, thus constructing a full-wafer peeling instability tendency distribution. The interlayer mismatch stress refers to the in-plane stress generated at the interface between the new material layer and the underlying deposited film due to differences in elastic constants and thermal expansion. The peeling instability tendency index is a dimensionless quantity characterizing the tendency of an interface to peel at that measurement point; a larger value indicates a greater tendency for interface peeling at that point. The full-wafer peeling instability tendency distribution refers to the distribution formed by assembling the peeling instability tendency indices point by point according to the measurement point number.

[0048] Specifically, using the material layer number associated with the interfacial bonding force distribution field, the elastic modulus and Poisson's ratio of the new material layer, the elastic modulus and Poisson's ratio of the underlying deposited film, and the deposition thickness of the new material layer at the end of the acquisition window are obtained from the material constant table of the deposition formula. The difference between the curvature value of the curvature time series at the end of the acquisition window and the curvature value of the same series at the beginning of the acquisition window is taken as the curvature increment introduced by the new material layer. This is converted into the average stress within the new material layer according to the Stony relation, and then distributed to the interface according to the elastic mismatch ratio between the new material layer and the underlying deposited film to obtain the interlayer mismatch stress at that measurement point. The unit is Pa. For measuring points whose polar coordinates fall within the wafer edge exclusion zone, the average value of the interlayer mismatch stress between the old and new layers at the remaining measuring points within the same radius zone is used to make up the difference. The wafer edge exclusion zone refers to the annular area within 2.7 mm to 3.3 mm from the wafer edge. The reason for choosing this range is that its width covers the edge action range of a conventional chuck clamping structure.

[0049] Understandably, the mismatch stress between the old and new layers provides the mechanical load driving the peeling at each measuring point. This load, along with the interfacial bearing capacity given by the interfacial bonding attenuation factor, constitutes two necessary aspects of the peeling criterion.

[0050] Specifically, the interfacial bonding force attenuation factors at each measuring point within the interfacial bonding force distribution field are sorted from largest to smallest, and the arithmetic mean of the measuring points with the highest values ​​after sorting is taken as the benchmark bonding force attenuation factor. Dimensionless. The high-value segment refers to the top 1 / 4 of the measurement points after sorting. The reason for taking the high-value segment instead of the overall average is that this segment represents the bonding strength level achieved by the region with the slightest interface degradation on the wafer. Using this as a reference, the attenuation of each measurement point is normalized to the relative loss relative to the best interface of the wafer, thereby eliminating the intrinsic bonding strength differences between different material combinations.

[0051] Specifically, the occurrence of peeling instability is determined by two conditions. First, the degree of overload of the interlayer mismatch stress on the interface relative to the remaining bearing capacity at that point; second, the degree of relative loss of the bonding force at that point relative to the optimal interface of the substrate. The overload ratio is calculated as the quotient of the product of the interlayer mismatch stress, the interface bonding force attenuation factor, and the intrinsic critical peeling stress of the interface. This overload ratio is assigned an overload path weight. The normalized bonding force loss is calculated as the sharpened power of the ratio of the interface bonding force attenuation factor to the reference bonding force attenuation factor (1 minus the normalized power). This normalized bonding force loss is assigned a relative loss path weight. The two weighted terms are then added together and calculated using the following formula: ; in, Let be the peeling instability tendency index of the i-th measuring point, which is dimensionless; The intrinsic critical peel stress at the interface is given in Pa, and is taken as 1.6 × 10⁻⁶. 8 Pa up to 1.9 × 10 8 Pa, the reason for choosing this range is that the critical peel stress level of the interface between common metal and dielectric layers in the non-degradation state is 100 MPa, and its value is determined by the offline pull-out test of the material combination. The relative loss sharpening index is dimensionless and ranges from 2.2 to 2.6. The reason for choosing this range is that the relative loss of binding force contributes very little to the peeling when it is close to the optimal level of this film, and the contribution increases sharply when it deviates significantly. The measured fitting index is concentrated around 2.4. The overload path weight is dimensionless and ranges from 0.55 to 0.65. The relative loss path weight is dimensionless and ranges from 0.35 to 0.45. The sum of the two weights is 1. The reason why the overload path weight is higher than the relative loss path weight is that the overload of the interlayer mismatch stress is the direct driving force of peeling, while the relative loss of bonding force is the enabling condition. The core of this formula is that the numerator of the first term is the interlayer mismatch stress borne by the measuring point, and the denominator is the product of the interface bonding force attenuation factor and the intrinsic critical peeling stress of the interface, that is, the actual remaining critical peeling stress after degradation at the measuring point. Dividing the two gives the dimensionless overload ratio. The smaller the attenuation factor, the lower the remaining critical peeling stress and the larger the overload ratio. The second term, within square brackets, is the normalized loss of bonding force relative to the best interface of the film. Taking the power of the relative loss sharpening exponent reflects the nonlinear amplification of loss after deviating from the baseline. When there is no attenuation of the interface bonding force and the interlayer mismatch stress is much lower than the intrinsic critical peeling stress of the interface, the first term approaches zero, the second term is exactly zero, and the peeling instability tendency exponent approaches zero, which is consistent with physical intuition.

[0052] Understandably, the peeling instability tendency index unifies the load-bearing capacity loss caused by the degradation of the external cavity interface with the mechanical load applied by the deposition of the new material layer inside the cavity into a single dimensionless criterion. A type of latent instability, where the film thickness is normal, the mismatch stress between the old and new layers does not exceed the conventional criteria, but the interface is weakened by an extremely thin degradation layer and is on the verge of peeling, can thus be captured by a single index.

[0053] Specifically, the peeling instability tendency index of each measuring point is assembled one by one according to the measuring point number i, along with the polar coordinates of the measuring point and the material layer number of the location of the interface bonding force distribution, forming a full-wafer peeling instability tendency distribution. This distribution is used for subsequent comparison with the dynamic judgment threshold, screening of measuring points exceeding the threshold, and clustering to locate hidden instability regions. The polar coordinates of the measuring points are referenced when outputting the coordinates of the instability location, and the material layer number is referenced when annotating the interface layer number of the inducing factor.

[0054] This method generates a dynamic judgment threshold based on the distribution of peel instability tendency across the entire wafer. It then filters out over-threshold measurement points whose peel instability tendency index exceeds this dynamic judgment threshold. These over-threshold measurement points are spatially clustered to delineate latent instability regions. The method outputs the latent instability detection results, including the location coordinates of the latent instability region, the index of the inducing interface layer, and the confidence level. The dynamic judgment threshold is an adaptively generated judgment boundary based on the statistical characteristics of the peel instability tendency index across the entire wafer. Over-threshold measurement points are those whose peel instability tendency index exceeds the dynamic judgment threshold. Latent instability regions are clusters formed by merging over-threshold measurement points through spatial clustering.

[0055] Specifically, the peeling instability tendency index of most measurement points across the entire wafer falls within the same low-value substrate band, while latent instability measurement points appear as a few outliers with high values. To allow the threshold to adaptively fluctuate with the substrate level of this wafer, a dynamic judgment threshold is generated using the following formula: ; in, The threshold is dynamically determined and is dimensionless. Let be the arithmetic mean of the stripping instability tendency index of all N measurement points, which is dimensionless; The maximum value of the stripping instability tendency index among all N measurement points is dimensionless. The dispersion factor is dimensionless and ranges from 2.4 to 2.8. The reason for choosing this range is that the exponential distribution within the base zone is approximately normal, and taking 2.5 times the standard deviation will reduce the misjudgment rate of the base measurement points to the order of one percent. The outlier stretching factor is dimensionless and ranges from 0.22 to 0.28. This range is chosen because when extreme outlier values ​​appear on the wafer, the outlier increases the standard deviation, causing the threshold to be dragged up and leading to missed detections of the second-highest measurement points. The threshold is adjusted upwards to a limited extent based on the deviation of the maximum value from the mean. A coefficient around 0.25 balances both missed and false detections. The core of this formula is that the first term on the right-hand side is the base center level, the second term is the conventional judgment margin obtained by multiplying the base dispersion and the dispersion multiplier, and the third term (within parentheses) is the dimensionless stretching factor constructed based on the deviation of the maximum value from the mean. When the entire field exponent is perfectly uniform, the standard deviation is zero, and the dynamic judgment threshold equals the mean; when the entire field dispersion is large, the dynamic judgment threshold automatically increases.

[0056] Understandably, the dynamic judgment threshold is adaptively generated based on the material combination, deposition formula, and cavity state of the wafer. This avoids the situation where a fixed threshold can lead to both missed and false judgments under different stack-up combinations.

[0057] Specifically, the peeling instability tendency index of each measuring point is compared with the dynamic judgment threshold. Points exceeding the threshold are selected and added to the exceeding-threshold measuring point set, along with their polar coordinates. This set and its accompanying polar coordinates are used for subsequent spatial clustering.

[0058] Specifically, the polar coordinates of each measuring point within the set of out-of-threshold measuring points are converted to rectangular coordinates, and density-based clustering is used to delineate the latent instability regions. Let the neighborhood radius be the first cluster radius, and the minimum number of measuring points in the region be the first cluster cardinality. The first cluster radius refers to the spatial distance limit for determining whether two out-of-threshold measuring points belong to the same cluster, and is set to 11mm to 13mm. The reason for choosing this range is that the measuring point division ensures that the distance between adjacent measuring points is approximately on the order of magnitude; a radius slightly larger than the adjacent distance ensures that physically connected degenerate regions belong to the same cluster without crossing irrelevant regions. The first cluster cardinality refers to the minimum number of out-of-threshold measuring points required to form a cluster, and is set to 2 to 3. The reason for choosing this range is that a single isolated out-of-threshold measuring point is more likely to be due to local measurement jitter, and at least 2 to 3 adjacent out-of-threshold measuring points are needed to constitute an instability region with engineering significance. During execution, an unclassified out-of-threshold measurement point is randomly selected. The remaining out-of-threshold measurement points within its first cluster radius are retrieved. When this number reaches the first cluster cardinality, the selected out-of-threshold measurement point is merged with its neighboring measurement points into a single cluster, and the neighborhood is recursively expanded until no new measurement points are added. Isolated out-of-threshold measurement points that do not reach the first cluster cardinality are removed. The resulting clusters represent the regions within the latent instability region.

[0059] Specifically, for each latent instability region, the centroid of the converted rectangular coordinates from the polar coordinates of all the exceeding-threshold measurement points within it is taken as the instability location coordinates of that region, and the material layer number associated with each exceeding-threshold measurement point is taken as the inducing interface layer number. Material layer numbers are consistent within the same region, indicating that the measurement points in that batch reside at the interface formed by the same transfer interval. The region confidence level is determined by the excess magnitude of each measurement point's index relative to the dynamic judgment threshold and the number of measurement points in the region. For each measurement point within the region, the excess ratio is calculated. This excess ratio is the quotient of the difference between the stripping instability tendency index and the dynamic judgment threshold, plus the dynamic judgment threshold. The arithmetic mean of the excess ratios within the region is multiplied by the scaling factor obtained by dividing the number of measurement points in the region by the first clustering cardinality, and then compressed to the interval of 0 to 1 using the hyperbolic tangent function to obtain the region confidence level. The instability location coordinates, inducing interface layer numbers, and confidence levels of each latent instability region are arranged in descending order of confidence level to form the latent instability detection results, which are then output to the process terminal.

[0060] Understandably, this detection result will address interface degradation occurring outside the deposition chamber acquisition sequence at the transfer gap, translating into an instability determination that is located at wafer coordinates, traced back to the specific interface layer, and ranked by severity. Latent exfoliation instability, which would otherwise be in a detection blind zone due to no change in macroscopic film thickness and lack of in-situ signal coverage, is now identified and intercepted before subsequent layer deposition is completed. The limitation of existing solutions in their inability to detect instability induced by interface degradation outside the chamber is resolved here.

[0061] The present invention also discloses a wafer-level multi-material thin film deposition instability detection system applied to the above method, including an interface exposure acquisition module, an interface degradation inversion module, an adhesion attenuation estimation module, a peeling tendency fusion module, and an instability determination output module.

[0062] The interface exposure acquisition module is used to arrange elliptic characterization units and oxygen partial pressure probes in the intercavity transfer channel, discretize the wafer surface into several measurement points, and collect the exposure duration, time-averaged effective oxygen partial pressure and elliptic reflection phase difference of each measurement point to construct the interface exposure matrix of the transfer gap. The exposure duration is the length of the exposure period of the measurement point from the opening of the gate valve of the previous deposition cavity to the closing of the gate valve of the next deposition cavity.

[0063] The interface degradation inversion module is used to obtain the estimated thickness of the degradation layer at each measurement point based on the exposure time and the time-averaged effective oxygen partial pressure in the interface exposure matrix of the transfer gap. Based on the elliptic reflection phase difference in the matrix and the estimated thickness of the degradation layer, the equivalent optical thickness and in-plane coverage of each measurement point are obtained, and the interface degradation feature matrix is ​​constructed.

[0064] The bonding force attenuation estimation module is used to collect the curvature time sequence of each measuring point after the wafer enters the next deposition chamber during the early stage of new material layer growth, extract the curvature change rate, and fuse the curvature change rate with the interface degradation feature matrix to obtain the interface bonding force attenuation factor of each measuring point, and construct the interface bonding force distribution field.

[0065] The peeling tendency fusion module is used to obtain the interlayer mismatch stress between the old and new layers at each measurement point based on the curvature time sequence, and to fuse the interlayer mismatch stress between the old and new layers with the interface bonding force distribution field to obtain the peeling instability tendency index at each measurement point, and to construct the whole wafer peeling instability tendency distribution.

[0066] The instability determination output module is used to generate a dynamic determination threshold based on the distribution of the instability tendency of the entire wafer peeling, to screen out the over-threshold test points whose peeling instability tendency index is greater than the dynamic determination threshold, to perform spatial clustering on the over-threshold test points, to delineate the hidden instability region, and to output the hidden instability detection result containing the location coordinates of the hidden instability region, the index of the inducing interface layer and the confidence level.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A wafer-level multi-material thin film deposition instability detection method, characterized in that, include: Elliptic characterization units and oxygen partial pressure probes are arranged in the intercavity transfer channel. The wafer surface is discretized into several measurement points. The exposure time, time-averaged effective oxygen partial pressure and elliptic reflection phase difference of each measurement point are collected to construct the transfer gap interface exposure matrix. The exposure time is the length of the exposure period of the measurement point from the opening of the gate valve of the previous deposition cavity to the closing of the gate valve of the next deposition cavity. Based on the exposure time and average effective oxygen partial pressure in the transfer gap interface exposure matrix, the estimated thickness of the degradation layer at each measurement point is obtained. Based on the elliptic reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of the degradation layer, the equivalent optical thickness and in-plane coverage of each measurement point are obtained, and an interface degradation feature matrix is ​​constructed. The curvature time sequence of each of the measurement points after the wafer enters the next deposition chamber is collected during the early stage of the growth of the new material layer. The curvature change rate is extracted and fused with the interface degradation feature matrix to obtain the interface bonding force attenuation factor of each measurement point, and the interface bonding force distribution field is constructed. Based on the curvature time sequence, the new and old interlayer mismatch stresses at each measurement point are obtained. The new and old interlayer mismatch stresses are fused with the interface bonding force distribution field to obtain the peeling instability tendency index at each measurement point, and the whole wafer peeling instability tendency distribution is constructed. A dynamic judgment threshold is generated based on the distribution of the whole wafer peeling instability tendency. Exceeding threshold measurement points with peeling instability tendency index greater than the dynamic judgment threshold are screened out. Spatial clustering is performed on the exceeding threshold measurement points to delineate the hidden instability region. The hidden instability detection result containing the location coordinates of the hidden instability region, the index of the inducing interface layer and the confidence level is output.

2. The method according to claim 1, characterized in that, Discretizing the wafer surface into several measurement points specifically includes: The elliptic characterization unit and the oxygen partial pressure probe are arranged above the robotic arm's travel path along the intercavity transfer channel. The elliptic characterization unit uses a single-wavelength laser source, and the incident angle is set in the Brewster angle range of the metal and dielectric stack. As the wafer passes beneath the elliptic characterization unit at a constant speed with the robotic arm, the wafer surface is discretized into several measurement points according to the equal arc length and equal radius ring zone division method, and each measurement point is assigned a measurement point polar coordinate composed of the radius and the azimuth angle relative to the travel direction. The oxygen partial pressure probe is installed on the inner wall of the intercavity transfer channel, and the polar coordinates of the measuring points are used for the calculation of the exposure time of each measuring point and the acquisition of the ellipsoidal reflection phase difference.

3. The method according to claim 2, characterized in that, Assign polar coordinates to each of the aforementioned measuring points, consisting of the radius and the azimuth angle relative to the direction of travel, specifically including: The exposure time of each measuring point is synthesized from three factors: the fixed reference time for the opening and closing of the gate valve and the start and stop of the robot arm, the ratio of the geometric projection position of each measuring point in the direction of travel to the linear velocity of the robot arm, and the edge stagnation correction factor linearly amplified according to the normalized radius of the measuring point. According to the exposure start and end time window of each measurement point, the channel oxygen partial pressure time sequence signal continuously collected by the oxygen partial pressure probe is extracted, and the arithmetic mean within the window is taken as the time-averaged effective oxygen partial pressure of that measurement point. The p-polarized reflected light intensity and s-polarized reflected light intensity are collected at each measurement point by the elliptic characterization unit. The elliptic reflection phase difference at the measurement point is solved by the analyzer rotation method. The elliptic reflection phase difference contains an integer number of phase order determined by the phase continuity expansion of adjacent measurement points.

4. The method according to claim 1, characterized in that, Based on the ellipsoidal reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of the degradation layer, the equivalent optical thickness and in-plane coverage of each measurement point are obtained, specifically including: According to the self-limiting growth law, the estimated thickness of the degradation layer is synthesized by the self-limiting saturation thickness, the reference oxidation rate constant, the power ratio of the exposure time and the time-averaged effective oxygen partial pressure relative to the calibrated reference oxygen partial pressure. Using the estimated thickness of the degradation layer as the initial value for iteration, the elliptic reflection phase difference is substituted into the thin-film dielectric phase delay model. The equivalent optical thickness is solved according to the refractive index of the degradation layer, the refractive index of the channel atmosphere, the refractive index of the substrate of the underlying deposited film, and the elliptic incident angle. The iteration convergence criterion is set to the thickness change between two adjacent iterations being less than a given convergence amount. The ratio of the equivalent optical thickness to the estimated thickness of the degradation layer is taken as the in-plane coverage, and this ratio is limited to the range of zero to one.

5. The method according to claim 1, characterized in that, The curvature time sequence of each measurement point after the wafer enters the next deposition chamber is collected during the early stage of new material layer growth, and the curvature change rate is extracted, specifically including: The multi-beam laser curvature meter in the subsequent deposition cavity scans the array of measurement points that are registered one by one with the measurement points point by point, and calculates the local curvature from the relative change of the spacing between the reflected light spots to obtain the curvature time sequence of each measurement point. A first-order linear least squares fit is performed on the curvature time sequence of each measurement point within the acquisition window. The initial sampling point caused by chuck adsorption jitter is removed, and the slope of the fitted straight line is taken as the curvature change rate of the measurement point. For measurement points whose standard deviation of the fitted residual exceeds the first residual threshold, the curvature change rate is made up by the average curvature change rate of the adjacent measurement points.

6. The method according to claim 5, characterized in that, The curvature change rate is fused with the interface degradation feature matrix to obtain the interface bonding force attenuation factor for each measurement point, specifically including: In the interface degradation feature matrix, measurement points with equivalent optical thickness less than a first thickness threshold and in-plane coverage less than a first coverage threshold are selected to form a reference measurement point subset. The arithmetic mean of the curvature change rate of each measurement point in the reference measurement point subset is taken as the non-degradation reference curvature change rate. The bonding loss ratio is formed by the product of the in-plane coverage and the exponential attenuation rate of the equivalent optical thickness relative to the bonding barrier feature thickness, and the barrier path is raised to the power of the residual bonding area ratio obtained by subtracting the bonding loss ratio. The mechanical path power of the ratio of the curvature change rate to the non-degradable reference curvature change rate is taken, and the power is multiplied by the barrier path power of the residual bonding area ratio to obtain the interface bonding force attenuation factor.

7. The method according to claim 1, characterized in that, By fusing the interlayer mismatch stress with the interfacial bonding force distribution field, the peeling instability tendency index of each measuring point is obtained, specifically including: The interface bonding force attenuation factor of each measuring point in the interface bonding force distribution field is sorted from large to small, and the arithmetic mean of the high-value measuring points after sorting is taken as the benchmark bonding force attenuation factor. The quotient of the product of the mismatch stress between the old and new layers, the attenuation factor of the interfacial bonding force, and the intrinsic critical peel stress of the interface is taken as the overload ratio, and the overload ratio is assigned an overload path weight. The sharpening power of the ratio of the interface bonding force attenuation factor to the benchmark bonding force attenuation factor is taken as the normalized bonding force loss, and a relative loss path weight is assigned to the normalized bonding force loss. The weighted overload ratio is added to the weighted normalized binding force loss to obtain the stripping instability tendency index.

8. A wafer-level multi-material thin film deposition instability detection system, applied to the method described in any one of claims 1-7, characterized in that, include: The interface exposure acquisition module is used to arrange elliptic characterization units and oxygen partial pressure probes in the intercavity transfer channel, discretize the wafer surface into several measurement points, and acquire the exposure duration, time-averaged effective oxygen partial pressure and elliptic reflection phase difference of each measurement point to construct the interface exposure matrix of the transfer gap. The exposure duration is the length of the exposure period of the measurement point from the opening of the gate valve of the previous deposition cavity to the closing of the gate valve of the next deposition cavity. The interface degradation inversion module is used to obtain the estimated thickness of degradation layer growth at each measurement point based on the exposure time and time-averaged effective oxygen partial pressure in the transfer gap interface exposure matrix, and to obtain the equivalent optical thickness and in-plane coverage at each measurement point based on the elliptic reflection phase difference in the transfer gap interface exposure matrix and the estimated thickness of degradation layer growth, thereby constructing an interface degradation feature matrix. The bonding force attenuation estimation module is used to collect the curvature time sequence of each of the measurement points after the wafer enters the next deposition chamber during the early stage of new material layer growth, extract the curvature change rate, fuse the curvature change rate with the interface degradation feature matrix to obtain the interface bonding force attenuation factor of each of the measurement points, and construct the interface bonding force distribution field. The peeling tendency fusion module is used to obtain the interlayer mismatch stress of the new and old layers at each of the measurement points based on the curvature time sequence, fuse the interlayer mismatch stress of the new and old layers with the interface bonding force distribution field to obtain the peeling instability tendency index of each of the measurement points, and construct the whole wafer peeling instability tendency distribution. The instability determination output module is used to generate a dynamic determination threshold based on the full wafer peeling instability tendency distribution, filter out the over-threshold measurement points whose peeling instability tendency index is greater than the dynamic determination threshold, perform spatial clustering on the over-threshold measurement points, delineate the latent instability region, and output the latent instability detection result containing the location coordinates of the latent instability region, the index of the inducing interface layer, and the confidence level.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images