Wafer front quantity detection method and semiconductor process equipment

Abstract

The application provides a wafer front quantity detection method and a semiconductor process equipment; in the wafer front quantity detection method, first, a feature vector set of a wafer is acquired; then, the feature vector set is input to a pre-trained prediction model, so that the prediction model outputs a prediction information set according to the feature vector set; a global feature vector of the wafer is generated according to the prediction information set, wherein the global feature vector includes a film thickness prediction mean value, a variation coefficient, a center-to-edge ratio, an azimuth non-uniformity parameter and an abnormal point coefficient; finally, a front quantity detection result of the wafer is generated according to the global feature vector and a preset feature threshold value, so that fast and lossless front quantity detection of the wafer is realized, a situation that a subsequent process is incorrectly operated due to wafer abnormalities is avoided, and process stability is improved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, and in particular to pre-fabrication methods for wafers and semiconductor process equipment. Background Technology

[0002] CMP (Chemical Mechanical Planarization) is a widely used process in IC (Integrated Circuit) manufacturing, primarily for the global planarization of metallic materials such as tungsten (W), copper (Cu), cobalt (Co), and aluminum (Al). Current CMP processes mainly consist of a primary polishing stage and a secondary polishing stage, equipped with an in-situ endpoint detection system to monitor the material removal rate in real time. Furthermore, to ensure the effectiveness of the CMP process, pre-measurement is required before CMP, i.e., measuring the thickness of the wafer's metal layer before CMP. This allows for the control of CMP process parameters and operation based on the measured film thickness.

[0003] In related technical solutions, pre-measurement often requires the combination of XRF (X-ray Fluorescence Spectrometer) technology or RS (Resistance). However, XRF technology has limited measurement penetration and is difficult to reflect changes in film thickness under multilayer structures. At the same time, due to the time-consuming measurement, it is not suitable for mass production. RS, on the other hand, is a contact test and will cause unexpected damage or contamination. Therefore, how to achieve rapid and non-destructive pre-measurement is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a pre-fabrication inspection method and semiconductor process equipment for wafers to alleviate the above-mentioned technical problems, thereby realizing rapid and non-destructive pre-fabrication inspection of wafers, avoiding the situation where subsequent processes run incorrectly due to wafer abnormalities, and thus improving process stability.

[0005] In a first aspect, embodiments of the present invention provide a pre-fabrication inspection method for wafers, the method comprising:

[0006] Obtain a feature vector set of a wafer; wherein the feature vector set includes feature vectors corresponding to multiple sampling points on the wafer, and each feature vector is determined according to the detection signal of the corresponding sampling point, including at least one of the following: time-domain feature parameters, frequency-domain feature parameters, and statistical feature parameters;

[0007] The feature vector set is input into a pre-trained prediction model so that the prediction model outputs a prediction information set based on the feature vector set; wherein, the prediction information set includes prediction information corresponding to multiple sampling points, and the prediction information includes the film thickness prediction value and uncertainty;

[0008] A global feature vector for the wafer is generated based on the predicted information set, wherein the global feature vector includes: the predicted mean film thickness, the coefficient of variation, the center-to-edge ratio, the orientation non-uniformity parameter, and the outlier coefficient.

[0009] The pre-detection results of the wafer are generated based on the global feature vector and the preset feature threshold; wherein, the pre-detection results include: pre-detection pass result, pre-detection fail result, and pre-detection conditionally pass result.

[0010] Optionally, the step of generating a global feature vector of the wafer based on the predicted information set includes: determining multiple effective sampling points based on the predicted information of the sampling points; and generating the global feature vector based on the film thickness prediction values ​​of the multiple effective sampling points.

[0011] Optionally, the plurality of effective sampling points include a first effective sampling point located at the center of the wafer and a plurality of second effective sampling points located at the edge of the wafer; the step of generating the global feature vector based on the film thickness prediction values ​​of the plurality of effective sampling points includes: calculating the film thickness prediction mean and film thickness prediction standard deviation based on the film thickness prediction values ​​of the plurality of effective sampling points, and determining the coefficient of variation based on the film thickness prediction mean and film thickness prediction standard deviation; calculating the center-to-edge ratio and radial gradient based on the film thickness prediction values ​​of the first effective sampling point and the film thickness prediction values ​​of the plurality of second effective sampling points, and determining the orientation non-uniformity parameter based on the center-to-edge ratio and the radial gradient; identifying abnormal sampling points among the plurality of effective sampling points, and determining the abnormal point coefficient based on the number of abnormal sampling points and the number of effective sampling points.

[0012] Optionally, the step of determining outlier sampling points among the multiple valid sampling points includes: calculating the outlier value of each valid sampling point based on the predicted film thickness value of each valid sampling point; and designating valid sampling points with outlier values ​​greater than a preset outlier threshold as outlier sampling points.

[0013] Optionally, the step of determining multiple valid sampling points based on the prediction information of the sampling points includes: calculating the confidence level of each sampling point based on the predicted film thickness and uncertainty of each sampling point; and selecting sampling points with a confidence level not less than a preset confidence threshold as the valid sampling points.

[0014] Optionally, the preset feature threshold includes: a preset target film thickness and a preset error; the step of generating the pre-measurement detection result of the wafer based on the global feature vector and the preset feature threshold includes: generating a film thickness detection result based on the predicted average film thickness, the preset target film thickness, and the preset error; wherein, the film thickness detection result includes: a film thickness detection qualified result and a film thickness detection unqualified result; if the film thickness detection result is the film thickness detection unqualified result, the pre-measurement detection result is determined to be the pre-measurement detection failure result.

[0015] Optionally, the preset feature thresholds further include: a preset coefficient of variation threshold, a preset center-to-edge ratio threshold, a preset orientation non-uniformity threshold, and a preset outlier coefficient threshold; the step of generating the pre-measurement detection result of the wafer based on the global feature vector and the preset feature thresholds includes: if the film thickness detection result is a qualified film thickness detection result, generating a uniformity detection result based on the coefficient of variation and the preset coefficient of variation threshold; generating a radial distribution detection result based on the center-to-edge ratio and the preset center-to-edge ratio threshold; generating a circumferential distribution detection result based on the orientation non-uniformity parameter and the preset orientation non-uniformity threshold; generating an outlier detection result based on the outlier coefficient and the preset outlier coefficient threshold; and determining the pre-measurement detection result based on the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the outlier detection result.

[0016] Preferably, the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly detection result all include both qualified and unqualified results; the step of determining the preliminary quantity detection result based on the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly detection result includes: if there is only one unqualified result among the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly detection result, the preliminary quantity detection result is determined to be a conditionally passed result; or, if there is more than one unqualified result among the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly detection result, the preliminary quantity detection result is determined to be a failed result; or, if none of the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly detection result contain any unqualified results, the preliminary quantity detection result is determined to be a passed result.

[0017] Preferably, the method further includes: if the pre-quantitative detection result is a failure result, generating an alarm message and controlling the wafer to stop entering the next process for processing.

[0018] In a second aspect, embodiments of the present invention also provide a semiconductor process apparatus, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0019] The embodiments of the present invention bring the following beneficial effects:

[0020] This invention provides a pre-evaluation method and semiconductor process equipment for wafers. First, a feature vector set of the wafer is obtained; wherein the feature vector set includes feature vectors corresponding to multiple sampling points on the wafer. Then, the feature vector set is input into a pre-trained prediction model, so that the prediction model outputs a prediction information set based on the feature vector set; wherein the prediction information set includes prediction information corresponding to multiple sampling points, and the prediction information includes predicted film thickness and uncertainty. A global feature vector of the wafer is generated based on the prediction information set, wherein the global feature vector includes: the mean predicted film thickness, coefficient of variation, center-to-edge ratio, orientation non-uniformity parameter, and outlier coefficient. Finally, the pre-evaluation result of the wafer is generated based on the global feature vector and a preset feature threshold, thereby achieving rapid and non-destructive pre-evaluation of the wafer, avoiding subsequent process errors due to wafer anomalies, and thus improving process stability.

[0021] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart of a pre-fabrication inspection method for wafers provided in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of an eddy current system provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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.

[0027] As semiconductor manufacturing processes evolve towards more advanced nodes such as 28nm, 14nm, and even 7nm / 5nm, CMP technology has become increasingly crucial, especially in BEOL (Backend of Line) processes. CMP has become an indispensable key step, its core principle being the synergistic effect of chemical etching and mechanical polishing to eliminate step differences, depressions, and redundant metal layers on the wafer surface caused by previous processes. In practical applications, CMP processes need to be precisely coordinated with photolithography, etching, and other processes to ensure that the flatness error of the dozens of interconnect layers is controlled at the sub-nanometer level.

[0028] Furthermore, to ensure the effectiveness of the CMP process, pre-calculation testing is required before CMP. Pre-calculation testing measures the wafer thin film thickness (also known as the thickness of the front-end film) before CMP to prevent the preceding CMP processes from being run multiple times or not at all due to special reasons (such as thin film deposition). In actual testing, pre-calculation testing mainly measures the thickness of the wafer metal layers, often requiring the use of XRF or RS technologies. However, XRF technology has limited penetration capabilities and is difficult to reflect changes in film thickness under multilayer structures; moreover, due to its time-consuming measurement, it is not suitable for mass production. RS, being a contact test, can cause unintended damage or contamination.

[0029] Furthermore, the relevant technical solutions also employ eddy current systems for endpoint detection in the CMP process. Eddy current systems are non-contact measurement technologies suitable for multilayer metal structures. Utilizing the principle of electromagnetic induction, these systems can accurately measure the thickness of metal layers, such as Cu films, with a detection range from 1500 nm to less than 30 nm. However, eddy current systems cannot provide accurate film thickness information before polishing to guide CMP process parameter settings, meaning they cannot be applied to pre-CMP measurement scenarios. Additionally, since eddy current systems only provide film thickness measurement data, they cannot automatically determine whether to proceed with the CMP process or which parameters need adjustment based on pre-measurement data. They also cannot predict potential problems during CMP, such as incorrect process formulations, over-polishing, or under-polishing. Moreover, the lack of calibration for varying product types and different film thickness ranges leads to errors in film thickness measurement data, thus affecting the CMP process effectiveness.

[0030] Based on this, embodiments of the present invention provide a wafer pre-evaluation method and semiconductor process equipment. First, a feature vector set of the wafer is obtained; wherein, the feature vector set includes feature vectors corresponding to multiple sampling points on the wafer; then, the feature vector set is input into a pre-trained prediction model, so that the prediction model outputs a prediction information set based on the feature vector set; wherein, the prediction information set includes prediction information corresponding to multiple sampling points, and the prediction information includes film thickness prediction values ​​and uncertainties; a global feature vector of the wafer is generated based on the prediction information set, wherein the global feature vector includes: film thickness prediction mean, coefficient of variation, center-to-edge ratio, orientation non-uniformity parameter, and outlier coefficient; finally, the pre-evaluation result of the wafer is generated based on the global feature vector and a preset feature threshold, thereby achieving rapid and non-destructive pre-evaluation of the wafer, avoiding subsequent process errors due to wafer anomalies, and thus improving process stability.

[0031] To facilitate understanding of this embodiment, the embodiments of the present invention will be described in detail below.

[0032] Example 1

[0033] This invention provides a pre-fabrication inspection method for wafers. Pre-fabrication inspection refers to the process of performing anomaly detection on a wafer after the previous process is completed, before it enters the CMP process. For example... Figure 1 As shown, the method includes the following steps:

[0034] Step S102: Obtain the feature vector set of the wafer.

[0035] The feature vector set includes feature vectors corresponding to multiple sampling points on the wafer. Each feature vector is determined based on the detection signal of the corresponding sampling point and includes at least one of the following: time-domain feature parameters, frequency-domain feature parameters, and statistical feature parameters. In practical applications, time-domain feature parameters include amplitude features and shape features. Amplitude features include, but are not limited to, maximum value, mean value, peak value, and RMS (Root Mean Square) value, while shape features include kurtosis, skewness, and waveform factor. Frequency-domain feature parameters include, but are not limited to, the main frequency amplitude, phase, harmonic ratio, and power spectral density, while statistical feature parameters include, but are not limited to, signal stability, signal-to-noise ratio, and phase fluctuation. The specific time-domain feature parameters, frequency-domain feature parameters, and statistical feature parameters can be set according to the actual situation.

[0036] Furthermore, the detection signal at each sampling point can be an eddy current signal, meaning that each sampling point on the wafer is detected using an eddy current system. Specifically, such as... Figure 2 As shown, the eddy current system emits a high-frequency electromagnetic field by driving the electromagnet 22 to excite eddy currents 23 in the metal film layer 11 of the wafer 10. The secondary magnetic field generated by the eddy currents 23 is received by the eddy current sensor 21. After signal processing, the eddy current signal can be obtained, and the wafer film thickness can be determined from the eddy current signal. Therefore, by detecting each sampling point on the wafer using the eddy current system, the eddy current signal for each sample can be obtained, thus achieving rapid and non-destructive detection of the wafer film layer. Furthermore, this detection method reduces the detection time to 12 seconds per wafer, which is 5-10 times faster than traditional offline detection, further improving the pre-fabrication efficiency of wafers.

[0037] Therefore, after the wafer is loaded onto the CMP polishing pad, multiple sampling points are first set at the center and edges of the wafer; the edges can be areas at a preset distance from the wafer center, which can be set according to actual conditions. Then, each sampling point on the wafer is detected by an eddy current system, such as acquiring a 200ms eddy current signal for each sampling point, achieving rapid coverage of the metal layer; finally, the characteristic signal of each sampling point is determined based on the eddy current signal of each sampling point.

[0038] For example, for the eddy current signal S(i,t) of the i-th sampling point on the wafer within time t, t∈[0,T], where T represents the detection period, digital filtering is first applied to eliminate high-frequency noise, resulting in the processed eddy current signal S. f (i,t)=αS f (i,t)+(1-α)S f (i,t), where α represents the smoothing factor, and then, based on the processed eddy current signal S... f(i,t) determines the feature vector of the i-th sampling point. Here, the feature vector includes the average amplitude, phase information, and waveform factor. The average amplitude A(i) is calculated using the mean function, i.e., A(i) = mean(|Si,t). f (i,t)|), phase information Using formula The waveform factor K(i) is determined using the formula K(i) = max(|S f (i,t)|) / mean(|S f (i,t)|) is determined, and thus the feature vector of the i-th sampling point can be determined. This leads to the characteristic vector set of the wafer.

[0039] It should be noted that in the process of determining each feature vector, the detection signal can also be any signal other than the eddy current signal, and the feature vector can be generated based on the other signals. The specific method can be adaptively adjusted according to the actual situation, and will not be described in detail in this embodiment of the invention.

[0040] Step S104: Input the feature vector set into the pre-trained prediction model so that the prediction model outputs a prediction information set based on the feature vector set.

[0041] After obtaining the feature vector set of the wafer, the feature vector set is input into a pre-trained prediction model so that the prediction model outputs a prediction information set based on the feature vector set. The prediction information set includes prediction information corresponding to multiple sampling points, and each prediction information includes the film thickness prediction value and uncertainty. For example, for any sampling point i, the prediction information of that sampling point can be obtained. Here, the prediction information includes the film thickness prediction value T(i) and uncertainty U(i) of that point. Thus, through the prediction model and the feature vector of each sampling point, the prediction information of that sampling point can be calculated so that the wafer can be pre-measured based on the prediction information of each sampling point. Compared with the method of detecting the overall prediction information of the wafer, this detection method further improves the accuracy of the wafer pre-measured detection.

[0042] Step S106: Generate the global feature vector of the wafer based on the prediction information set.

[0043] The global feature vector includes: the predicted mean film thickness, the coefficient of variation, the center-to-edge ratio, the orientation non-uniformity parameter, and the outlier coefficient. Based on the global feature vector, pre-measurement inspection of the wafer is achieved. Compared with the method of pre-measurement inspection using single detection information, this method of pre-measurement inspection using multiple features not only further improves the accuracy of pre-measurement inspection, but also realizes precise process control of the wafer and ensures the process quality of the wafer.

[0044] Step S108: Generate the wafer pre-detection results based on the global feature vector and the preset feature threshold.

[0045] The pre-quantitative inspection results include: pre-quantitative inspection pass result, pre-quantitative inspection fail result, and pre-quantitative inspection conditional pass result. In practical applications, a pre-quantitative inspection pass result indicates that the wafer is normal and can proceed to the next CMP process; a pre-quantitative inspection fail result indicates that the wafer is abnormal and cannot proceed to the next CMP process, thus avoiding increased equipment maintenance costs due to abnormal wafers entering the CMP process; a pre-quantitative inspection conditional pass result indicates that the wafer has an abnormality, but can be restored to normal and proceed to the next CMP process after adjustments. Therefore, by performing pre-quantitative inspection on the wafer, abnormal wafers are prevented from entering the next CMP process, thereby improving the process control accuracy of the wafer and ensuring the process quality of the wafer.

[0046] The wafer pre-fabrication inspection method provided in this invention enables rapid and non-destructive pre-fabrication inspection of wafers, avoiding subsequent process errors caused by wafer abnormalities, thereby improving process stability.

[0047] In one implementation, the step of generating a global feature vector of a wafer based on a set of prediction information includes the following sub-steps: (1) determining multiple effective sampling points based on the prediction information of the sampling points; and (2) generating a global feature vector based on the film thickness prediction values ​​of the multiple effective sampling points.

[0048] In sub-step (1), the process of determining multiple valid sampling points based on the predicted information of the sampling points is as follows: First, the confidence level of each sampling point is calculated based on the predicted film thickness and uncertainty of each sampling point; for example, for any sampling point i, the confidence level C(i) = exp(-U(i)) is calculated based on its predicted film thickness T(i) and uncertainty U(i). 2 / T(i) 2 Then, set the confidence level to be no less than the preset confidence threshold C. threshold The sampling points are taken as valid sampling points, that is, when C(i)≥C threshold When C(i) < C, the corresponding sampling point is taken as the valid sampling point; similarly, when C(i) < C threshold When this happens, the corresponding sampling point is considered an invalid sampling point.

[0049] Therefore, by filtering multiple sampling points to obtain multiple effective sampling points, and generating a global feature vector based on the film thickness prediction values ​​of multiple effective sampling points, the accuracy of the global feature vector is avoided due to invalid sampling points, thereby further improving the accuracy of wafer pre-measurement.

[0050] Furthermore, for the sake of clarity, for multiple valid sampling points, we assume that the total number of valid sampling points is n, and the corresponding set of film thickness prediction values ​​T is T = {T(1), T(2), ..., T(n)}. At the same time, the multiple valid sampling points are divided into the first valid sampling point located at the center of the wafer and the multiple second valid sampling points located at the edge of the wafer, that is, the second valid sampling points are the multiple valid sampling points located in the edge region.

[0051] Specifically, in sub-step (2), the step of generating a global feature vector based on the film thickness prediction values ​​of multiple effective sampling points includes: ① calculating the film thickness prediction mean and film thickness prediction standard deviation based on the film thickness prediction values ​​of multiple effective sampling points, and determining the coefficient of variation based on the film thickness prediction mean and film thickness prediction standard deviation; ② calculating the center-to-edge ratio and radial gradient based on the film thickness prediction values ​​of the first effective sampling point and multiple second effective sampling points, and determining the orientation non-uniformity parameter based on the center-to-edge ratio and radial gradient; ③ identifying the abnormal sampling points among the multiple effective sampling points, and determining the abnormal point coefficient based on the number of abnormal sampling points and the number of effective sampling points.

[0052] In section ①, the predicted mean μ of the film thickness is calculated using μ = mean(T); similarly, the predicted standard deviation σ of the film thickness is calculated using the std(standard deviation) function, i.e., σ = std(T); and the coefficient of variation CV is calculated using CV = σ / μ. Specifically, in some scenarios, the predicted maximum value Tmax = max(T), the predicted minimum value Tmin = min(T), and the predicted range R = Tmax - Tmin can be determined based on the set of predicted film thickness values ​​T. These can be set according to the actual situation.

[0053] Furthermore, in section ②, for ease of explanation, the predicted film thickness at the first effective sampling point is referred to as Tcenter, and the predicted film thickness at the second effective sampling point is referred to as Tedge. The center-to-edge ratio (CER) can then be calculated using CER = Tcenter / mean(Tedge), where mean(Tedge) represents the average predicted film thickness at multiple second effective sampling points. Similarly, the radial gradient Gr can be calculated using Gr = (mean(Tedge) - Tcenter) / D, where D represents the average radius of multiple second effective sampling points. That is, the total radius is obtained based on the distance from each second effective sampling point to the center of the circle, and D is determined based on the total radius and the number of second effective sampling points. For the azimuth inhomogeneity parameter U... θ Then it can be based on U θ = std(Tedge) / mean(Tedge) is calculated.

[0054] Furthermore, in section ③, outlier sampling points among multiple valid sampling points are first identified. The specific identification process is as follows: The outlier value of each valid sampling point is calculated based on its predicted membrane thickness value; valid sampling points with outlier values ​​greater than a preset outlier threshold are designated as outlier sampling points. Specifically, the improved GESD (Generalized Extreme Studentized Deviate) algorithm can be used to calculate the outlier value, and the outlier value of each valid sampling point is calculated based on Z(i) = 0.6745(T(i) - median(T)) / MAD, where median(T) represents the median of the membrane thickness prediction set T, MAD (Median absolute deviation) represents the absolute deviation of the median, and |Z(i)| is used as the outlier value of each valid sampling point. At this point, if the outlier value is greater than the preset outlier threshold (preferably 3.5, which can be adaptively adjusted according to the actual situation), i.e., |Z(i)|>3.5, then the corresponding valid sampling point is regarded as an outlier sampling point; conversely, if the outlier value is not greater than the preset outlier threshold, i.e., |Z(i)|≤3.5, then the corresponding valid sampling point is regarded as a non-outlier sampling point. Therefore, outlier sampling points among valid sampling points can be determined based on the outlier value of each valid sampling point.

[0055] After identifying the abnormal sampling points, the number of abnormal sampling points, m, can be determined. Then, based on the number of abnormal sampling points m and the number of valid sampling points n (i.e., the total number of valid sampling points), the abnormal point coefficient Q = m / n can be determined.

[0056] Therefore, a global feature vector G for the wafer can be generated, where G = [μ, CV, CER, U]. θ In some scenarios, the global feature vector G can also be G=[μ,σ,CV,CER,Gr,U]. θ The global feature vector G provides comprehensive and detailed information on the wafer film thickness distribution. Furthermore, for ease of explanation, this embodiment uses G = [μ, CV, CER, U]. θ The process of generating the pre-quantitative detection results of the wafer is illustrated using [Q] as an example. The global feature vector G in other cases can be adaptively adjusted with reference to this case. The embodiments of the present invention will not be described in detail here.

[0057] In one implementation, the preset feature threshold includes: a preset target film thickness T. target Preset error δ, preset coefficient of variation threshold CV limit Preset center-to-edge ratio threshold CER limit Preset orientation non-uniformity threshold U θlimit and preset outlier coefficient threshold Q limitThe pre-detection results of the wafer are generated based on the global feature vector G and the preset feature threshold, including the following scenarios:

[0058] (1) Generate film thickness detection results based on the predicted average film thickness, the preset target film thickness and the preset error; wherein, the film thickness detection results include: film thickness detection qualified results and film thickness detection unqualified results; if the film thickness detection result is film thickness detection unqualified results, determine the previous quantity detection result as the previous quantity detection unqualified results.

[0059] Specifically, the absolute value of the difference between the predicted mean film thickness and the preset target film thickness is calculated, and the film thickness detection result is generated based on the absolute value of the difference and the preset error; wherein, when the absolute value of the difference is not greater than the preset error, i.e., |μ-T t arg et When |≤δ, the film thickness test result is a qualified result; conversely, when the absolute value of the difference is greater than the preset error, i.e., |μ-T t arg et When |>δ, the film thickness detection result is a film thickness detection failure result, thereby realizing the average film thickness detection of the wafer based on the predicted average film thickness, the preset target film thickness and the preset error.

[0060] Furthermore, when the film thickness detection result is unqualified, the previous measurement detection result is directly determined to be unqualified. This means that the wafer is abnormal and cannot proceed to the next CMP process. The cause of the abnormality is recorded, and alarm information is generated. This avoids the wafer abnormality causing the CMP process to fail, thereby reducing the waste of CMP consumables (polishing pads, polishing fluid).

[0061] (2) If the film thickness detection result is a qualified film thickness detection result, generate a uniformity detection result based on the coefficient of variation and a preset coefficient of variation threshold; generate a radial distribution detection result based on the center-to-edge ratio and a preset center-to-edge ratio threshold; generate a circumferential distribution detection result based on the azimuth non-uniformity parameter and a preset azimuth non-uniformity threshold; generate anomaly detection results based on the anomaly coefficient and a preset anomaly coefficient threshold; wherein, the uniformity detection result, radial distribution detection result, circumferential distribution detection result and anomaly detection result all include qualified detection results and unqualified detection results; determine the preliminary quantity detection result based on the uniformity detection result, radial distribution detection result, circumferential distribution detection result and anomaly detection result.

[0062] Specifically, when the film thickness detection result is acceptable, the remaining feature parameters in the global feature vector G are then detected. For example, if the coefficient of variation is not greater than a preset coefficient of variation threshold, i.e., CV ≤ CV... limit If the uniformity test result is positive, the result is considered acceptable; otherwise, if the coefficient of variation is greater than the preset coefficient of variation threshold (CV > CV), the result is considered unacceptable. limit If the uniformity test result is not met, then the test result is unqualified.

[0063] Similarly, if the center-to-edge ratio CER and the preset center-to-edge ratio threshold CER limit satisfy |CER - 1| ≤ CER limit , then the radial distribution detection result is a qualified detection result. On the contrary, if the center-to-edge ratio CER and the preset center-to-edge ratio threshold CER limit satisfy |CER - 1| > CER limit , then the radial distribution detection result is an unqualified detection result. If the azimuthal non-uniformity parameter is not greater than the preset azimuthal non-uniformity threshold, i.e., U θ ≤ U θlim it , then the circumferential distribution detection result is a qualified detection result. On the contrary, if the azimuthal non-uniformity parameter is greater than the preset azimuthal non-uniformity threshold, i.e., U θ > U θlimit , then the circumferential distribution detection result is an unqualified detection result. If the anomaly point coefficient is not greater than the preset anomaly point coefficient threshold, i.e., Q ≤ Q limit , then the anomaly point detection result is a qualified detection result. On the contrary, if the anomaly point coefficient is greater than the preset anomaly point coefficient threshold, i.e., Q > Q limit , then the anomaly point detection result is an unqualified detection result.

[0064] When the film thickness detection result is a qualified film thickness detection result, at this time, it is judged whether the number of unqualified detection results in the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly point detection result is only one. If so, it is determined that the previous inspection result is a conditional pass result of the previous inspection, that is, the average film thickness detection is qualified but there is one unqualified distribution feature, and the parameters corresponding to the unqualified distribution feature need to be adjusted and then enter the next CMP process. For example, if the radial distribution detection result is an unqualified detection result, the chamber pressure distribution is adjusted to improve the radial uniformity of the wafer; or, if the circumferential distribution detection result is an unqualified detection result, the rotation speed is adjusted to improve the circumferential uniformity of the wafer, thus realizing the automatic adjustment of the parameters of some distribution features according to the previous inspection result to ensure the safe execution of the CMP process.

[0065] (3) When the film thickness detection result is a qualified film thickness detection result, if the number of unqualified detection results in the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly point detection result is greater than one, at this time, it is determined that the previous inspection result is a non-pass result of the previous inspection, control the wafer not to enter the next CMP process, record the reason for the anomaly, and generate an alarm message at the same time to prompt the operator to perform a manual inspection on the semiconductor process equipment; at the same time, it avoids the failure of the CMP process caused by the wafer anomaly, thereby reducing the waste of CMP consumables (polishing pads, polishing liquids) and the contamination caused by the incorrect operation of the wafers that have not undergone the previous inspection.

[0066] (4) When the film thickness detection result is a qualified film thickness detection result, if the uniformity detection result, radial distribution detection result, circumferential distribution detection result and abnormal point detection result do not contain any unqualified detection results, that is, each feature parameter in the global feature vector G is qualified, the number of unqualified items is 0, then the pre-quantity detection result is determined to be the pre-quantity detection pass result, and the wafer is controlled to proceed to the next CMP process.

[0067] Therefore, by performing pre-measurement inspection on the wafer using a global feature vector G, and controlling the semiconductor process equipment based on the pre-measurement inspection results, if the pre-measurement inspection result is a pass, the wafer is controlled to directly enter the CMP process; if the pre-measurement inspection result is a conditional pass, the parameters of the non-conforming distribution characteristics are adjusted before entering the CMP process. This not only avoids the waste of CMP consumables (polishing pads, polishing fluid) caused by wafer anomalies, but also reduces equipment maintenance costs due to wafer anomalies, such as reducing CMP process costs by about 10% to 12%. It also avoids running incorrect process recipes when directly entering CMP due to abnormal feature parameters, thus improving CMP process stability. If the pre-measurement inspection result is a fail, the wafer is directly controlled not to enter the next CMP process, and the cause of the anomaly is recorded, along with alarm information to prompt operators to manually inspect the semiconductor process equipment. This not only achieves comprehensive wafer inspection and improves quality control capabilities, but also eliminates the need to transfer wafers to dedicated metrology equipment, saving logistics time and labor costs, and significantly improving inspection efficiency.

[0068] In one implementation, the pre-trained prediction model in this embodiment of the invention is constructed and trained as follows:

[0069] (S1) Construct a training set of feature vectors.

[0070] Specifically, multiple baseline wafers with standard film thicknesses are selected, each with a different metal layer. Furthermore, the thickness range of each baseline wafer covers ±30% of the previous metal layer thickness. The thickness distribution should include at least five sampling points of different thicknesses, evenly distributed at the center and edges of the baseline wafers. Additionally, the true film thickness of the baseline wafers can be pre-verified using high-precision offline measurement tools (such as XRF).

[0071] For any baseline film wafer, multiple sampling points are detected, ensuring that the center, mid-diameter, and edge regions are all detected. Simultaneously, each sampling point is detected k times to obtain the signal set S(x,j,k) of the baseline film wafer. Here, x represents the identification information of the baseline film wafer among multiple baseline film wafers, such as sorting information, j represents the j-th sampling point of the baseline film wafer, and j∈[1,y], y is the total number of sampling points. For example, for the second sampling point out of 17 sampling points on the first baseline film wafer, sampling is performed 3 times, and the signal set is S(1,2,3).

[0072] Then, for the signal set S(x,j,k) of the baseline film wafer, calculate the average signal value AvgS(x,j)=∑S(x,j,k) / k for each sampling point. At the same time, the signal standard deviation stdS(x,j)=sqrt(∑S(x,j,k)-AvgS(x,j)) for each sampling point can also be calculated. 2 If stdS(x,j) / AvgS(x,j)>0.5, then sampling point j is marked as an unstable point and needs to be re-detected until each sampling point satisfies: stdS(x,j) / AvgS(x,j)≤0.5. Finally, multiple stable point data pairs are obtained. Each data pair includes the average signal value and standard film thickness of the stable point, i.e., {AvgS(x,j), T(x,j)}. Here, AvgS(x,j) represents the eddy current signal of the j-th sampling point in the x-th baseline film wafer, and T(x,j) represents the standard film thickness of the j-th sampling point in the x-th baseline film wafer.

[0073] Therefore, training data pairs for a baseline film wafer can be obtained from data pairs of multiple sampling points on the same baseline film wafer. Training data pairs from multiple baseline film wafers can form a training data pair set, so as to generate a feature vector training set based on the training data pair set.

[0074] For the eddy current signal at any sampling point, feature extraction can be performed. This feature extraction includes, but is not limited to, time-domain feature extraction, frequency-domain feature extraction, and statistical feature extraction. Specifically, for the time-domain feature parameter F... time This includes amplitude features and shape features. Amplitude features include, but are not limited to, maximum value, mean, peak value, RMS value, etc., while shape features include kurtosis, skewness, and waveform factor, etc. Similarly, for frequency domain feature extraction (such as through FFT (Fast Fourier Transform)), the frequency domain feature parameters F... freq Including but not limited to the main frequency amplitude, phase, harmonic ratio, and power spectral density; for statistical feature extraction, the extracted statistical feature parameter F fastThis includes, but is not limited to, signal stability, signal-to-noise ratio, and phase fluctuations. Therefore, a feature vector F = [F] can be generated for this sampling point. time F freq F fast ].

[0075] Furthermore, the generated eigenvector F can be dimensionality reduced using PCA (Principal Components Analysis). For example, the eigencovariance matrix of the eigenvector F can be calculated, and the eigenvalues ​​and eigenvectors of the eigencovariance matrix can be obtained. The eigenvectors corresponding to the L largest eigenvalues ​​can be selected to form a projection matrix P, thereby obtaining the dimensionality-reduced eigenvector F. reduced =P*F.

[0076] Therefore, by reducing the dimensionality of the feature vector at each sampling point, a feature vector training set can be generated. This training set includes the data training pairs corresponding to each sampling point, and each data training pair includes the dimensionality-reduced feature vector F of that sampling point. reduced And standard film thickness. It should be noted that the above-mentioned dimensionality reduction and feature extraction can refer to existing technologies, and the embodiments of the present invention will not be described in detail here.

[0077] After generating the aforementioned feature vector training set, a prediction model is constructed. This prediction model employs an Enhanced Multi-Feature Fusion Calibration Model (EMFCM) to address the complex nonlinear mapping relationship between eddy current signals and film thickness. The EMFCM model is a piecewise model; different piecewise models are determined for different film thicknesses W. For example, when W < 100 nm, the piecewise model is an SVR (Support Vector Regression) model; when 100 nm ≤ W < 500 nm, the piecewise model is an XGBoost (eXtreme Gradient Boosting) model; and when W ≥ 500 nm, the piecewise model is a multinomial regression model. The specific piecewise model and film thickness range can be adaptively adjusted according to actual conditions.

[0078] After the segmented model is determined, each segmented model is trained using the feature vector training set. During the training process, grid search can be used to determine the optimal hyperparameters, and K-fold cross-validation can be used to avoid overfitting. For details, please refer to the existing technology. The embodiments of this invention will not be described in detail here.

[0079] Furthermore, for the trained segmented model, an error compensation value C(F) is pre-set to ensure model accuracy. This error compensation value C(F) is a fixed value and can be adjusted according to actual conditions. Therefore, for the trained prediction model, its prediction function is T. final Ensemble(F) = C(F) + C(F), where C(F) represents the error compensation value, and Ensemble(F) represents the piecewise model. For example, when the film thickness W < 100 nm, Ensemble(F) is an SVR model; when 100 nm ≤ W < 500 nm, Ensemble(F) is an XGBoost model; and when W ≥ 500 nm, Ensemble(F) is a multinomial regression model. The specific piecewise model can be found in existing technologies, and will not be detailed here. Furthermore, actual production data can be collected periodically to update the model, ultimately obtaining a well-trained prediction model that meets the requirements. It should be noted that the specific model training process can be found in existing technologies, and will not be detailed here.

[0080] Therefore, the embodiments of the present invention can quickly output the prediction information set of the wafer based on the feature vector set of the wafer through the prediction model, thereby further improving the pre-detection efficiency of the wafer.

[0081] Example 2

[0082] To facilitate understanding, the pre-evaluation process of a 300mm wafer with a preset Cu film thickness of 300nm is used as an example. First, a feature vector set is generated based on the eddy current signals from multiple sampling points on the wafer. This feature vector set is then input into a pre-trained prediction model to output a prediction information set. Finally, a global feature vector G = [μ, σ, CV, CER, Gr, U] is generated based on this prediction information set. θ [,Q], where μ is 585nm, and |μ-T] target |>δ indicates that the film thickness reading is significantly higher than expected, meaning the film thickness test result is unqualified.

[0083] Similarly, σ is 35.1 nm, which is greater than the corresponding threshold, indicating that the standard deviation is too large, and the detection result is unqualified; CV = 6.0%, which is greater than the CV. limit (e.g., 3%) indicates a high coefficient of variation, resulting in a non-compliant uniformity test result; CER = 1.15, |CER-1| > 0.1 indicates a significantly thicker center than edge, resulting in a non-compliant radial distribution test result; Gr = 0.05 nm / mm indicates a large radial gradient; U θ It is 3.5%, which is greater than U. θlimit(2%) indicates high azimuth non-uniformity, and the circumferential distribution test result is unqualified; the outlier coefficient Q = 0.22, which is greater than Q. limit (e.g., 5%), the abnormal point detection result is a test failure result. Therefore, the previous quantity detection result is determined to be a previous quantity detection failure result. At this time, the semiconductor process equipment needs to control the wafer to prevent it from entering the next CMP process, record the cause of the abnormality, and generate alarm information to prompt the operator to manually inspect the semiconductor process equipment. For example, based on the feedback of abnormal data, the upstream deposition process finds a fault in the airflow control valve, thereby achieving timely repair.

[0084] Furthermore, embodiments of the present invention also provide a semiconductor process apparatus, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method embodiments.

[0085] The semiconductor process equipment provided in this embodiment of the invention has the same technical features as the wafer pre-measurement method provided in the above embodiment, so it can also solve the same technical problems and achieve the same technical effects.

[0086] This embodiment also provides a machine-readable storage medium storing machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions cause the processor to implement the above-described wafer pre-detection method.

[0087] The pre-fabrication inspection method for wafers and the computer program product for semiconductor process equipment provided in the embodiments of the present invention include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0089] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0090] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0092] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A pre-processing inspection method for wafers, used to perform anomaly detection on wafers after the previous process is completed before they enter the CMP process; characterized in that, The method includes: Obtain a feature vector set of a wafer; wherein the feature vector set includes feature vectors corresponding to multiple sampling points on the wafer, and each feature vector is determined according to the detection signal of the corresponding sampling point, including at least one of the following: time-domain feature parameters, frequency-domain feature parameters, and statistical feature parameters; The feature vector set is input into a pre-trained prediction model so that the prediction model outputs a prediction information set based on the feature vector set; wherein, the prediction information set includes prediction information corresponding to multiple sampling points, and the prediction information includes the film thickness prediction value and uncertainty; A global feature vector for the wafer is generated based on the predicted information set, wherein the global feature vector includes: the predicted mean film thickness, the coefficient of variation, the center-to-edge ratio, the orientation non-uniformity parameter, and the outlier coefficient. The pre-measurement detection results of the wafer are generated based on the global feature vector and the preset feature thresholds; wherein, the pre-measurement detection results include: pre-measurement detection pass result, pre-measurement detection fail result, and pre-measurement detection conditionally pass result; the preset feature thresholds include: preset target film thickness, preset error, preset coefficient of variation threshold, preset center-to-edge ratio threshold, preset orientation non-uniformity threshold, and preset anomaly point coefficient threshold; The step of generating a global feature vector for the wafer based on the predicted information set includes: determining multiple effective sampling points based on the predicted information of the sampling points; generating the global feature vector based on the film thickness prediction values ​​of the multiple effective sampling points; wherein the multiple effective sampling points include a first effective sampling point located at the center of the wafer and multiple second effective sampling points located at the edge of the wafer; calculating the film thickness prediction mean and film thickness prediction standard deviation based on the film thickness prediction values ​​of the multiple effective sampling points, and using the ratio of the film thickness prediction standard deviation to the film thickness prediction mean as the coefficient of variation; determining the film thickness corresponding to the multiple second effective sampling points based on the film thickness prediction values ​​of the multiple second effective sampling points. The predicted mean and standard deviation of film thickness are calculated. The ratio of the predicted film thickness value of the first effective sampling point to the predicted mean of film thickness of a plurality of second effective sampling points is used as the center-to-edge ratio. The difference between the predicted mean of film thickness of a plurality of second effective sampling points and the predicted film thickness value of the first effective sampling point is calculated. The radial gradient is determined based on the difference and the average radius of a plurality of second effective sampling points. The ratio of the predicted standard deviation of film thickness of a plurality of second effective sampling points to the predicted mean of film thickness of a plurality of second effective sampling points is used as the azimuth non-uniformity parameter. Abnormal sampling points are identified among a plurality of effective sampling points, and the ratio of the number of abnormal sampling points to the number of effective sampling points is used as the abnormal point coefficient.

2. The method according to claim 1, characterized in that, The step of identifying outlier sampling points among a plurality of valid sampling points includes: Calculate the outlier value of each effective sampling point based on the predicted film thickness value of each effective sampling point; The valid sampling points whose outliers are greater than a preset outlier threshold are designated as the abnormal sampling points.

3. The method according to claim 1, characterized in that, The step of determining multiple valid sampling points based on the prediction information of the sampling points includes: The confidence level of each sampling point is calculated based on the predicted film thickness and uncertainty of each sampling point. Sampling points with a confidence level not less than a preset confidence threshold are considered as valid sampling points.

4. The method according to claim 1, characterized in that, The step of generating the pre-detection result of the wafer based on the global feature vector and the preset feature threshold includes: A film thickness detection result is generated based on the predicted average film thickness, the preset target film thickness, and the preset error; wherein, the film thickness detection result includes: a film thickness detection qualified result and a film thickness detection unqualified result; If the film thickness detection result is a film thickness failure result, the pre-measurement detection result is determined to be a pre-measurement failure result.

5. The method according to claim 4, characterized in that, The step of generating the pre-detection result of the wafer based on the global feature vector and the preset feature threshold includes: If the film thickness detection result is a qualified film thickness detection result, a uniformity detection result is generated based on the coefficient of variation and the preset coefficient of variation threshold; a radial distribution detection result is generated based on the center-to-edge ratio and the preset center-to-edge ratio threshold; a circumferential distribution detection result is generated based on the azimuth non-uniformity parameter and the preset azimuth non-uniformity threshold; and an anomaly detection result is generated based on the anomaly coefficient and the preset anomaly coefficient threshold. The pre-measurement detection result is determined based on the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly point detection result.

6. The method according to claim 5, characterized in that, The uniformity detection results, the radial distribution detection results, the circumferential distribution detection results, and the anomaly point detection results all include qualified and unqualified results. The step of determining the pre-measurement detection result based on the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the outlier detection result includes: If there is only one unqualified result among the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the abnormal point detection result, the pre-quantity detection result is determined to be a conditionally passable result of the pre-quantity detection. or, If the number of non-compliant results among the uniformity detection result, the radial distribution detection result, the circumferential distribution detection result, and the anomaly point detection result is greater than one, then the preliminary quantity detection result is determined to be a preliminary quantity detection failure result; or... If none of the uniformity detection results, radial distribution detection results, circumferential distribution detection results, and abnormal point detection results contain the unqualified detection results, then the pre-quantity detection result is determined to be the pre-quantity detection pass result.

7. The method according to claim 1, characterized in that, The method further includes: If the pre-quantification detection result is a failure, an alarm message is generated, and the wafer is controlled to stop entering the next process for further processing.

8. A semiconductor process apparatus, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1-7.

Images