Wafer production condition abnormality identification method, device and equipment based on hash algorithm

Abstract

The application provides a wafer production condition abnormality identification method, device and equipment based on a hash algorithm, belongs to the wafer analysis field, and specifically comprises the following steps: receiving current wafer processing data in real time; adopting feature hashing to perform hash mapping on the current processing value of each working parameter item to obtain a current feature hash value; adopting record hashing to perform combined hash mapping on each current wafer processing data to obtain a current record hash value; according to a current timestamp, acquiring the record frequency of the current feature hash value and the current record hash value mapped to a hash bucket, and evaluating the abnormality of the current wafer processing data in the time dimension. The processing scheme of the application can efficiently and quickly perform abnormality analysis on the streaming data of wafer production, and the analysis accuracy is high, and the wafer production conditions subjected to manual rechecking are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wafer analysis, in particular to a wafer production working condition abnormality identification method, device and equipment based on a hash algorithm. BACKGROUND

[0002] In the wafer processing process, the sensors of the machine table can obtain a large amount of data. In the existing FDC system, the commonly used method for controlling the process parameters is to set a control threshold value. For example, a maximum upper limit value is set for the temperature parameter. If the current data of the temperature exceeds the threshold value, the system will issue an alarm and notify the relevant engineers. According to the system setting level, more serious cases will directly shut down. Engineers generally need to set different SPEC values (computer performance evaluation values) for EQP (machine table), Chamber (chamber), Recipe (process recipe), and Parameter (parameter). When the working condition changes, the value needs to be adjusted in time. For engineers, once the working condition changes, the temporary workload of these operations is very large, and there may be omissions in the operation process.

[0003] In addition, in the wafer processing process, various working conditions are very complex, and the machine table generates a large amount of data per second. The SPEC value is often a fixed value set by the user. The fixed value cannot cover all abnormal situations in the streaming data scenario, so some abnormal situations cannot be identified, or normal situations are judged as abnormal situations. These abnormal situations also need additional manual review and analysis, so the existing method has the problems of low efficiency, waste of resources, etc. in abnormality analysis. SUMMARY

[0004] Therefore, in order to overcome the shortcomings of the prior art, the present application provides a wafer production working condition abnormality identification method, device and equipment based on a hash algorithm, which can efficiently and quickly perform abnormality analysis on streaming data of wafer production, and has high analysis accuracy and reduces manual review.

[0005] In order to achieve the above purpose, the present application provides a wafer production working condition abnormality identification method based on a hash algorithm, comprising: receiving current wafer processing data in real time, the current wafer processing data comprising a current timestamp, a working parameter item of wafer processing and a corresponding current processing value; performing hash mapping on the current processing value of each working parameter item using a feature hash to obtain a current feature hash value; performing combined hash mapping on each current wafer processing data using a record hash to obtain a current record hash value; according to the current timestamp, obtaining the record frequency of the current feature hash value and the current record hash value mapped to a hash bucket, and evaluating the abnormality of the current wafer processing data in the time dimension, the number of the hash bucket being obtained by hash mapping of historical wafer processing data using the feature hash and the record hash.

[0006] In one of the embodiments, the current processing value of each of the work parameter items is hashed to obtain a current feature hash value by using feature hashing, including: performing logarithmic transformation on the current processing value of each of the work parameter items; obtaining a processing value range corresponding to the work parameter item; normalizing the current processing value after logarithmic transformation according to the processing value range; and mapping the current feature hash value according to the normalized current processing value.

[0007] In one of the embodiments, each of the current wafer processing data is combined and hashed to obtain a current record hash value by using record hashing, including: generating a random vector of a dimension number of the current wafer processing data, each element of the random vector being randomly sampled from a standard normal distribution N(0, 1); converting the current wafer processing data into a first vector based on the dimension of the current wafer processing data; calculating the dot product of the first vector and each random vector respectively, and converting the dot product into a binary to obtain a binary string corresponding to the first vector; and converting the binary string into a decimal integer to obtain the current record hash value.

[0008] In one of the embodiments, the abnormality of the current wafer processing data in the time dimension is evaluated according to the current timestamp, the record frequency of the current feature hash value and the current record hash value mapped to the hash bucket, including: respectively counting the current total record frequency of the feature hash value and the record hash value mapped to the hash bucket according to the current timestamp; respectively determining the abnormality score of the current feature hash value and the current record hash value according to the current total record frequency; and evaluating the abnormality of the current wafer processing data in the time dimension according to the abnormality score.

[0009] In one of the embodiments, the abnormality score of the current feature hash value is calculated by using the formula , wherein, represents the abnormality score of the current feature hash value,

[0010] The abnormality score of the current record hash value is calculated by using the formula , wherein, represents the abnormality score of the current record hash value, r i is the i-th current wafer processing data, r ij is the current processing value of the j-th work parameter item in the i-th current wafer processing data, is the number of occurrences of the hash value corresponding to the current wafer processing data r i in the current timestamp t, is the current wafer processing data r icurrent total record frequency of the corresponding hash value in the hash bucket, current processing value r in current timestamp ij occurrence number of the corresponding hash value, current processing value r in current timestamp t ij current total record frequency of the corresponding hash value in the hash bucket, current timestamp.

[0011] In one embodiment, the formula further includes a time decay coefficient a, and the specific values of and are multiplied by a and then substituted into the formula for calculation; the specific values of and are multiplied by a and then substituted into the formula for calculation.

[0012] In one embodiment, after the current wafer processing data is received in real time, the method further includes: obtaining historical wafer processing data; identifying the current work parameter item of the current wafer processing data and the historical work parameter item carried by the historical wafer processing data respectively; when it is determined that the current work parameter item and the historical work parameter item are consistent, using feature hashing to hash map the current processing value of each work parameter item.

[0013] A wafer production condition abnormality identification device based on a hash algorithm, the device comprising: a receiving module for receiving current wafer processing data in real time, the current wafer processing data including a current timestamp, a work parameter item of wafer processing, and a corresponding current processing value; a feature hashing processing module for using feature hashing to hash map the current processing value of each work parameter item to obtain a current feature hash value; a record hashing processing module for using record hashing to combine hash map each current wafer processing data to obtain a current record hash value; an evaluation module for obtaining the record frequency of the current feature hash value and the current record hash value mapped to a hash bucket according to the current timestamp, and evaluating the abnormality of the current wafer processing data in the time dimension, the number of the hash bucket being obtained by using the feature hashing and the record hashing to hash map historical wafer processing data.

[0014] In one of the embodiments, the record hash processing module comprises: a random vector generating unit configured to generate a number of random vectors according to a dimension number of the current wafer processing data, each element of the random vector being randomly sampled from a standard normal distribution N(0, 1); a first vector converting unit configured to convert the current wafer processing data into a first vector based on the dimension of the current wafer processing data; a dot product conversion unit configured to calculate a dot product of the first vector and each random vector respectively, and convert the dot product into a binary to obtain a binary string corresponding to the first vector; and a record hash value generating unit configured to convert the binary string into a decimal integer to obtain a current record hash value.

[0015] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0016] Compared with the prior art, the advantages of the present application are that a large amount of wafer processing stream data generated by a machine per second in a wafer processing process can be efficiently and quickly detected for abnormality, and real-time dynamic identification of abnormal conditions of wafer production conditions is realized. The present application also realizes processing of each record in a real-time data stream with constant time complexity and constant space complexity through feature hashing and record hashing, and captures the correlation of data in different dimensions through a local sensitive hashing (LSH) function and a time scoring mechanism, so that the consumed computing resources and computing time are greatly reduced, the accuracy of abnormality judgment is improved, the frequency of manual review is reduced, product yield is improved, and economic benefits are further created. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 is a flowchart of a wafer production condition abnormality identification method based on a hash algorithm in an embodiment of the present application;

[0019] Figure 2 is a flowchart of a feature hash mapping step in an embodiment of the present application

[0020] Figure 3 is a flowchart of a record hash mapping step in an embodiment of the present application;

[0021] Figure 4A structure block diagram of a wafer production condition abnormality recognition device based on a hash algorithm in an embodiment of the present application is shown in the figure.

[0022] Figure 5 An internal structure diagram of a computer device in an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0023] The embodiments of the present application will be described in detail below with reference to the drawings.

[0024] The above and other aspects of the present application will become more apparent by describing in detail the embodiments thereof with reference to the attached drawings in which:

[0025] It is to be understood that the following description is merely illustrative of the aspects of the present application and that numerous and various embodiments can be derived from the teachings described herein without departing from the spirit of the application. It is therefore contemplated to append claims to be intensive to cover all such modifications and embodiments of the application that fall within the true scope of the application. Accordingly, the application is not to be construed as limited to the precise examples described herein.

[0026] It is also to be understood that the following description is only illustrative of the aspects of the present application and that numerous and various embodiments can be derived from the teachings described herein without departing from the spirit of the application. It is therefore contemplated to append claims to be intensive to cover all such modifications and embodiments of the application that fall within the true scope of the application. Accordingly, the application is not to be construed as limited to the precise examples described herein.

[0027] In addition, in the following description, specific details are provided to thoroughly understand examples. However, it will be understood by one of ordinary skill in the art that the described aspects can be practiced without these specific details.

[0028] This application provides a method for identifying abnormal wafer production conditions based on a hash algorithm. This method can be applied to a server or terminal. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable smart devices. The server can be a standalone server or a server cluster composed of multiple servers. Figure 1 As shown, the method is applied to a server as an example, and includes the following steps:

[0029] Step 101: Receive the current wafer processing data in real time. The current wafer processing data includes the current timestamp, wafer processing working parameters, and the corresponding current processing values.

[0030] Current wafer processing data is real-time data collected from various wafer processing machines. This data is streaming data, meaning it is generated, transmitted, and processed in real-time as a continuous data stream. Current wafer processing data includes the current timestamp, wafer processing parameters, and the corresponding current processing values. Historical wafer processing data includes historical timestamps, wafer processing parameters, and the corresponding historical processing values. The processing parameters describe the working environment of the wafer processing machine, such as temperature, humidity, vacuum level, sputtering current, processing gas pressure, and deposition time. The processing parameters for current and historical wafer processing data are consistent for the same processing machine, and therefore, their dimensions are also consistent. The processing parameters for current and historical wafer processing data may differ between different processing machines.

[0031] The server receives the current wafer processing data in real time. When multiple processing machines are processing the same type of chip simultaneously and the processing parameters are consistent, the server can simultaneously receive the current wafer processing data from multiple machines in real time for subsequent simultaneous processing.

[0032] Step 102: Use feature hashing to perform hash mapping on the current processing value of each working parameter item to obtain the current feature hash value.

[0033] Feature hashing maps original features to a fixed-length vector space using a hash function. Each feature value is mapped to a fixed index position. If different feature values ​​hash to the same index position, they are considered to be the same feature. The feature hashing process can be expressed by the formula h(x) = (x1w1 + x2w2 + ... + x...). d w dThe expression ) mod m represents the hash function, where x is the original feature vector (the current processing value in this embodiment), d is the dimension of the original feature vector, w is the weight of the hash function, and m is the number of hash buckets. The hash function can be MurmurHash, farmhash, etc. The server uses feature hashing to perform hash mapping on the current processing value of each working parameter item to obtain the current feature hash value.

[0034] Step 103: Use record hashing to perform combined hash mapping on each current wafer processing data to obtain the current record hash value.

[0035] Record hashing refers to performing a hash operation on the current processing value in the current wafer processing data to generate a hash value. Record hashing performs specific hash processing on the current wafer processing data to reduce the feature dimensionality of the current wafer processing data.

[0036] The server uses record hashing to perform a combined hash mapping on each current wafer processing data entry, obtaining the current record hash value. The server can convert the current processing value into a format suitable for hash function processing, typically a string or byte sequence; then, it uses the selected hash function to perform hash calculations on the preprocessed data, generating the current record hash value. Alternatively, the server can hash each current processing value separately and then combine these hash values ​​to generate the final hash value. The hash functions selected in steps 102 and 103 can be the same or different. When the hash functions selected in the two steps are the same, the hash function maps the current wafer processing data to a fixed number of buckets, making it more likely that similar records will be mapped to the same bucket.

[0037] Step 104: Based on the current timestamp, obtain the current feature hash value and the record frequency of the current record hash value mapped to the hash bucket, and evaluate the anomalies of the current wafer processing data in the time dimension. The number of hash buckets is obtained by hash mapping the historical wafer processing data using feature hash and record hash.

[0038] The number of hash buckets is obtained by hash mapping historical wafer processing data using feature hashing and record hashing. The server can perform hash mapping on the historical wafer processing data in advance using steps 101 to 103 to determine the range of hash bucket values.

[0039] The server retrieves the current characteristic hash value mapped to the hash bucket and the record frequency of the current record hash value based on the current timestamp, and evaluates any anomalies in the current wafer processing data over time. The server can determine whether the current wafer processing data is abnormal by evaluating the record frequency of the current timestamp and historical timestamps. When the change in record frequency exceeds a preset range, the server determines the current wafer processing data as abnormal data and sends a warning message to the engineer's corresponding terminal account; when the change in record frequency does not exceed the preset range, the server determines the current wafer processing data as normal data and stores it as historical wafer processing data.

[0040] In one embodiment, when the server performs a hash mapping on the current wafer processing data and the hash value exceeds the range of the hash bucket, the server will first perform a program self-check. If the program self-check is fault-free, the server will create a new hash bucket and report the anomaly of the current wafer processing data.

[0041] The aforementioned method can efficiently and rapidly detect anomalies in the massive amounts of streaming data generated per second during wafer fabrication, and achieve real-time dynamic identification of abnormal conditions in wafer production. This application also utilizes feature hashing and record hashing to process each record in the real-time data stream with constant time and space complexity. Furthermore, by employing Locality Sensitive Hash (LSH) and a time-based scoring mechanism, it captures the correlation of data across different dimensions. This significantly reduces computational resources and time consumption, improves the accuracy of anomaly detection, reduces the frequency of manual review, and ultimately increases product yield, thereby generating further economic benefits.

[0042] In one embodiment, such as Figure 2 As shown, feature hashing is used to hash the current processing value of each working parameter item to obtain the current feature hash value, including the following steps:

[0043] Step 201: Perform a logarithmic transformation on the current processing value of each working parameter item.

[0044] The server's current processing value for each working parameter item j Perform a logarithmic transformation. The server... Perform a logarithmic transformation to obtain the transformed value The formula is: .

[0045] Step 202: Obtain the processing value range corresponding to the working parameter item.

[0046] The server retrieves the processing value range corresponding to each processing parameter item. Each processing parameter item in the current wafer processing data corresponds to multiple processing values; therefore, the server can update the maximum processing value (max) corresponding to processing parameter item j in real time based on the current wafer processing data. j and processing minimum value min j The processing numerical range [min] is obtained. j , max j ].

[0047] Step 203: Normalize the current processing value after logarithmic transformation according to the processing value range.

[0048] The server normalizes the current processing value after logarithmic transformation based on the range of processing values, obtaining the normalized value. The normalization formula can be .

[0049] Step 204: Map the current feature hash value based on the normalized current processing value.

[0050] The server maps the current feature hash value to the normalized current processed value. The number of hash buckets is b, and the current feature hash value is... mod represents the modulo operation.

[0051] The above method uses feature hashing to hash individual working parameter items, compressing high-dimensional features into lower dimensions and reducing storage and computation costs.

[0052] like Figure 3 As shown, in one embodiment, a combined hash mapping is performed on each current wafer processing data using a record hash to obtain the current record hash value, including the following steps:

[0053] Step 301: Based on the number of dimensions of the current wafer processing data, generate a random vector of the number of dimensions. The elements of each random vector are randomly sampled from the standard normal distribution N(0,1).

[0054] The server generates a random vector of a certain number of dimensions based on the number of dimensions in the current wafer processing data. The current wafer processing data r i It contains d working parameters (dimensions) and randomly generates K vectors. Each vector a has a dimension equal to d. The elements of each random vector a are randomly sampled from a standard normal distribution N(0,1).

[0055] Step 302: Based on the dimension of the current wafer processing data, convert the current wafer processing data into a first vector.

[0056] The server transforms the current wafer processing data into a first vector based on the dimension of the current wafer processing data. Current wafer processing data r i The first vector is .

[0057] Step 303: Calculate the dot product of the first vector and each random vector respectively, and convert the dot product into binary to obtain the binary string corresponding to the first vector.

[0058] The server calculates the dot product between the first vector and each random vector, converts the dot product to binary, and obtains the binary string corresponding to the first vector. The server calculates the dot product between each random vector 'a' and the first vector. The dot product. The server converts the dot product to binary, denoted as bits, and the server can use the formula... Convert the dot product to binary, where, This represents the dot product operation. The server then concatenates the bit results of each vector into a binary string.

[0059] Step 304: Convert the binary string to a decimal integer to obtain the hash value of the current record.

[0060] The server converts the binary string into a decimal integer. Get the hash value of the current record. .

[0061] The above method uses record hashing to perform combined hash mapping on the entire record, compressing the high-dimensional feature space into a lower dimension and reducing storage and computation costs.

[0062] In one embodiment, based on the current timestamp, the record frequency of the current feature hash value and the current record hash value mapped to the hash bucket is obtained, and the anomaly of the current wafer processing data in the time dimension is evaluated, including: calculating the current total record frequency of the feature hash value and the record hash value mapped to the hash bucket based on the current timestamp; determining the anomaly score of the current feature hash value and the current record hash value based on the current total record frequency; and evaluating the anomaly of the current wafer processing data in the time dimension based on the anomaly score.

[0063] The server calculates the current total record frequency of feature hash values ​​and record hash values ​​mapped to hash buckets based on the current timestamp. The current wafer processing data r in the current timestamp t i The number of times the corresponding hash value appears. As of the current timestamp t, the current wafer processing data r i The current total record frequency of the corresponding hash value in the hash bucket.

[0064] r iFor the i-th current wafer processing data, r ij This represents the current processing value of the j-th working parameter item in the i-th current wafer processing data.

[0065] The current processing value r in the current timestamp ij The number of times the corresponding hash value appears.

[0066] The current processing value r up to the current timestamp t. ij The current total record frequency of the corresponding hash value in the hash bucket.

[0067] This is the current timestamp.

[0068] The server determines the anomaly score of the current feature hash value and the current record hash value based on the current total record frequency.

[0069] The server assesses the anomalies in the current wafer processing data over time based on anomaly scores. The server can obtain an anomaly threshold M; if the anomaly score is greater than M, the data is determined to be anomalous.

[0070] In one embodiment, the anomaly score of the current feature hash value is calculated using the formula... Calculate, where, This indicates the anomaly score of the current feature hash value.

[0071] The anomaly score of the current record's hash value is calculated using the formula. Calculate, where, This indicates the abnormal score of the current record's hash value.

[0072] r i For the i-th current wafer processing data, r ij This represents the current processing value of the j-th working parameter item in the i-th current wafer processing data.

[0073] The current wafer processing data in the current timestamp t The number of times the corresponding hash value appears.

[0074] Records up to the current timestamp t. The current total record frequency of the corresponding hash value in the hash bucket.

[0075] Features in the current timestamp The number of times the corresponding hash value appears.

[0076] For the current timestamp t, the feature The current total record frequency of the corresponding hash value in the hash bucket.

[0077] This is the current timestamp.

[0078] The above method can calculate the anomaly scores of the current feature hash value and the current record hash value, thereby facilitating further data analysis.

[0079] In one embodiment, α is the time decay coefficient. and Multiply the specific value by α and then substitute it into the formula for calculation; and The specific value is multiplied by α and then substituted into the formula for calculation.

[0080] Specifically represented as .

[0081] Taking α as 0.9 as an example, at timestamp t=1, record the generation of r1, map it to bucket 3, and update the counter of bucket 3. When substituting into the formula, =1*α=1*0.9=0.9, =1*α=1*0.9=0.9.

[0082] At timestamp t=2, record the generation of r1, map it to bucket 3, and update the counter of bucket 3. The total counter for bucket 3 When substituting into the formula, =(1+α )*α=(1+0.9)*0.9=1.71, =0.9(1+α) )*α=(1+0.9)*0.9=1.71.

[0083] At timestamp t=3, record the generation of r1, map it to bucket 3, and update the counter of bucket 3. The total counter for bucket 3 When substituting into the formula, =(1+α )*α=(1+1.71)*0.9=2.43, =0.9(1+α) )*α=(1+1.71)*0.9=2.43.

[0084] The above method reduces the impact of past records on the current score by using a time decay coefficient.

[0085] In one embodiment, after receiving the current wafer processing data in real time, the method further includes: acquiring historical wafer processing data; identifying the current working parameter items in the current wafer processing data and the historical working parameter items carried in the historical wafer processing data; and when it is determined that the current working parameter items and the historical working parameter items are consistent, performing a hash mapping on the current processing value of each working parameter item using feature hashing.

[0086] The server retrieves historical wafer processing data. This historical wafer processing data can be stored in a database. The server identifies the current working parameters in the current wafer processing data and the historical working parameters carried in the historical wafer processing data. When the current working parameters and historical working parameters are consistent, the server uses a feature hash to perform a hash mapping on the current processing value of each working parameter. When the current working parameters and historical working parameters are inconsistent, the server directly issues an anomaly warning for the current wafer processing data.

[0087] Example 1

[0088] The following explanation uses a Failure Data Collection (FDC) system as an example to illustrate the Physical Vapor Deposition (PVD) process. In the semiconductor industry, an FDC system is also known as a Failure Data Collection system, used to collect, store, and analyze data on faults or defective products detected during the manufacturing process.

[0089] Step 101: Receive the current wafer processing data in real time. The current wafer processing data includes the current timestamp, wafer processing working parameters, and the corresponding current processing values.

[0090] Anomaly detection was performed on the working parameters of wafer processing, namely vacuum level (Pv), temperature (T), sputtering current (lsp), gas pressure (pg) and deposition time (T).

[0091] The server receives real-time wafer processing data S as shown in the table. For ease of description, the working parameter items are represented by English abbreviations.

[0092]

[0093] Step 102: Use feature hashing to perform hash mapping on the current processing value of each working parameter item to obtain the current feature hash value.

[0094] Taking the first wafer processing data in the table as an example, the value of the feature Pv is 6.9. Applying the logarithmic transformation: log(1+6.9)=0.89. As of now, the maximum value corresponding to this working parameter is 9, and the minimum value is 6, so the processing value range is (6,9). The normalized value is (0.89-log(6)) / (log(9)-log(6))=0.7. b (hash bucket)=10, mapping: (0.7*10)mod 10=7, that is, the Pv hash value of the first record is 7.

[0095] The server uses feature hashing to perform a hash mapping on the current processing value of each working parameter item to obtain the current feature hash value.

[0096] Step 103: Use record hashing to perform combined hash mapping on each current wafer processing data to obtain the current record hash value.

[0097] Five random vectors are generated, each with five dimensions (consistent with the dimensions of the first wafer processing data). The elements of each random vector are typically sampled randomly from a standard normal distribution N(0,1).

[0098]

[0099] The first record has 5 dimensions, which can be converted into a vector: r=[6.9,190,15,5,260]

[0100] The server calculates the dot product of r with each random vector.

[0101]

[0102] .

[0103] The server concatenates the binary string as

[01101] , which is 13 in decimal. The hash bucket b=10, and 13 mod 10=3, meaning the hash value of this wafer processing data is 3.

[0104] Step 104: Based on the current timestamp, obtain the current feature hash value and the record frequency of the current record hash value mapped to the hash bucket, and evaluate the anomalies of the current wafer processing data in the time dimension. The number of hash buckets is obtained by hash mapping the historical wafer processing data using feature hash and record hash.

[0105] The hash value of record r1 is 3, which means it is mapped to bucket 3.

[0106] The hash value of record r2 is 3, which means it is mapped to bucket 3.

[0107] The hash value of record r3 is 5, which means it is mapped to bucket 5.

[0108] At timestamp t=1, record the generation of r1, map it to bucket 3, and update the counter of bucket 3. .

[0109] At timestamp t=2, record the generation of r1, map it to bucket 3, and update the counter of bucket 3. The total counter for bucket 3 .

[0110] At timestamp t=2, calculate the anomaly score for record r2:

[0111] .

[0112] .

[0113] The anomaly threshold is M. If the anomaly score is greater than M, the record is considered an anomaly.

[0114] It should be understood that, although Figures 1-3 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 1-3 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0115] In one embodiment, such as Figure 4 As shown, a wafer production condition anomaly identification device based on a hash algorithm is provided. The device includes a receiving module 401, a feature hash processing module 402, a record hash processing module 403, and an evaluation module 404.

[0116] The receiving module 401 is used to receive the current wafer processing data in real time. The current wafer processing data includes the current timestamp, wafer processing working parameters, and the corresponding current processing values.

[0117] The feature hash processing module 402 is used to perform a hash mapping on the current processing value of each working parameter item using feature hashing to obtain the current feature hash value.

[0118] The record hash processing module 403 is used to perform a combined hash mapping on each current wafer processing data using the record hash to obtain the current record hash value.

[0119] Evaluation module 404 is used to obtain the current feature hash value and the record frequency of the current record hash value mapped to the hash bucket based on the current timestamp, and to evaluate the anomalies of the current wafer processing data in the time dimension. The number of hash buckets is obtained by hash mapping the historical wafer processing data using feature hash and record hash.

[0120] In one embodiment, the feature hash processing module includes:

[0121] The logarithmic transformation unit is used to perform logarithmic transformation on the current processing value of each working parameter item.

[0122] The numerical range acquisition unit is used to acquire the processing numerical range corresponding to the working parameter item.

[0123] The normalization unit is used to normalize the current processing value after logarithmic transformation according to the range of processing values.

[0124] The feature hash value mapping unit is used to map the current feature hash value based on the normalized current processing value.

[0125] In one embodiment, the hash processing module includes:

[0126] The random vector generation unit is used to generate a random vector of a certain number of dimensions based on the number of dimensions of the current wafer processing data. The elements of each random vector are randomly sampled from the standard normal distribution N(0,1).

[0127] The first vector transformation unit is used to transform the current wafer processing data into a first vector based on the dimension of the current wafer processing data.

[0128] The dot product conversion unit is used to calculate the dot product of the first vector and each random vector respectively, and convert the dot product into binary to obtain the binary string corresponding to the first vector.

[0129] The record hash value generation unit is used to convert a binary string into a decimal integer to obtain the hash value of the current record.

[0130] In one embodiment, the evaluation module includes:

[0131] The statistics unit is used to calculate the current total record frequency of feature hash values ​​and record hash values ​​mapped to hash buckets based on the current timestamp.

[0132] The score calculation unit is used to determine the abnormal scores of the current feature hash value and the current record hash value based on the current total record frequency.

[0133] The anomaly detection unit is used to assess the anomalies in the current wafer processing data in the time dimension based on the anomaly score.

[0134] In one embodiment, the anomaly detection device further includes:

[0135] The acquisition module is used to acquire historical wafer processing data.

[0136] The parameter identification module is used to identify the current working parameter items in the current wafer processing data and the historical working parameter items carried in the historical wafer processing data.

[0137] The determination module is used to perform a hash mapping on the current processing value of each working parameter item when the current working parameter item and the historical working parameter item are consistent.

[0138] Specific limitations regarding the hash algorithm-based wafer manufacturing process anomaly identification device can be found in the above-mentioned limitations of the hash algorithm-based wafer manufacturing process anomaly identification method, and will not be repeated here. Each module in the aforementioned wafer manufacturing process anomaly identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0139] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores current wafer processing data, historical wafer processing data, etc. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for identifying anomalies in wafer production conditions.

[0140] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0141] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: receiving current wafer processing data in real time, the current wafer processing data including a current timestamp, wafer processing working parameters, and corresponding current processing values; performing a hash mapping on the current processing value of each working parameter using feature hashing to obtain a current feature hash value; performing a combined hash mapping on each piece of current wafer processing data using record hashing to obtain a current record hash value; and, based on the current timestamp, obtaining the record frequency of the current feature hash value and the current record hash value mapped to the hash bucket, and evaluating the anomalies of the current wafer processing data in the time dimension, wherein the number of hash buckets is obtained by performing a hash mapping on historical wafer processing data using feature hashing and record hashing.

[0142] In one embodiment, when the processor executes a computer program, it implements a hash mapping of the current processing value of each working parameter item using feature hashing to obtain the current feature hash value, which includes: performing a logarithmic transformation on the current processing value of each working parameter item; obtaining the processing value range corresponding to the working parameter item; normalizing the current processing value after logarithmic transformation according to the processing value range; and mapping the current feature hash value according to the normalized current processing value.

[0143] In one embodiment, when the processor executes a computer program, it implements a combined hash mapping of each current wafer processing data using record hashing to obtain the current record hash value, including: generating a random vector of the number of dimensions of the current wafer processing data, wherein the elements of each random vector are randomly sampled from a standard normal distribution N(0,1); converting the current wafer processing data into a first vector based on the dimensions of the current wafer processing data; calculating the dot product of the first vector and each random vector respectively, and converting the dot product into binary to obtain a binary string corresponding to the first vector; and converting the binary string into a decimal integer to obtain the current record hash value.

[0144] In one embodiment, when the processor executes a computer program, it implements the following: based on the current timestamp, it obtains the current feature hash value and the current record hash value mapped to the hash bucket, and assesses the anomaly of the current wafer processing data in the time dimension. This includes: calculating the current total record frequency of the feature hash value and the record hash value mapped to the hash bucket based on the current timestamp; determining the anomaly score of the current feature hash value and the current record hash value based on the current total record frequency; and assessing the anomaly of the current wafer processing data in the time dimension based on the anomaly score.

[0145] In one embodiment, when the processor executes the computer program, it further performs the following steps: acquiring historical wafer processing data; identifying the current working parameter item in the current wafer processing data and the historical working parameter item carried in the historical wafer processing data respectively; when it is determined that the current working parameter item and the historical working parameter item are consistent, using feature hashing to perform hash mapping on the current processing value of each working parameter item.

[0146] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program performs the following steps: receiving current wafer processing data in real time, the current wafer processing data including a current timestamp, wafer processing working parameters, and corresponding current processing values; performing a hash mapping on the current processing value of each working parameter using feature hashing to obtain a current feature hash value; performing a combined hash mapping on each piece of current wafer processing data using record hashing to obtain a current record hash value; and, based on the current timestamp, obtaining the record frequency of the current feature hash value and the current record hash value mapped to the hash bucket, and evaluating the anomalies of the current wafer processing data in the time dimension, wherein the number of hash buckets is obtained by performing a hash mapping on historical wafer processing data using feature hashing and record hashing.

[0147] In one embodiment, when the computer program is executed by the processor, it implements a feature hashing method to hash the current processing value of each working parameter item to obtain the current feature hash value, including: performing a logarithmic transformation on the current processing value of each working parameter item; obtaining the processing value range corresponding to the working parameter item; normalizing the current processing value after logarithmic transformation according to the processing value range; and mapping the current feature hash value according to the normalized current processing value.

[0148] In one embodiment, when the computer program is executed by the processor, it implements a combined hash mapping of each current wafer processing data using record hashing to obtain the current record hash value, including: generating a random vector of the number of dimensions based on the number of dimensions of the current wafer processing data, wherein the elements of each random vector are randomly sampled from a standard normal distribution N(0,1); converting the current wafer processing data into a first vector based on the dimensions of the current wafer processing data; calculating the dot product of the first vector and each random vector respectively, and converting the dot product into binary to obtain a binary string corresponding to the first vector; and converting the binary string into a decimal integer to obtain the current record hash value.

[0149] In one embodiment, when a computer program is executed by a processor, it implements the following: obtaining the current feature hash value and the record frequency of the current record hash value mapped to the hash bucket based on the current timestamp, and evaluating the anomalies of the current wafer processing data in the time dimension. This includes: calculating the current total record frequency of the feature hash value and the record hash value mapped to the hash bucket based on the current timestamp; determining the anomaly score of the current feature hash value and the current record hash value based on the current total record frequency; and evaluating the anomalies of the current wafer processing data in the time dimension based on the anomaly score.

[0150] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: acquiring historical wafer processing data; identifying the current working parameter item in the current wafer processing data and the historical working parameter item carried in the historical wafer processing data respectively; when it is determined that the current working parameter item and the historical working parameter item are consistent, using feature hashing to perform hash mapping on the current processing value of each working parameter item.

[0151] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for identifying abnormal conditions in wafer manufacturing based on a hash algorithm, characterized in that, include: Receive current wafer processing data in real time. This current wafer processing data includes the current timestamp, wafer processing parameters, and the corresponding current processing values. The current processing value of each of the aforementioned working parameter items is hashed using a feature hash to obtain the current feature hash value; The current record hash value is obtained by performing a combined hash mapping on each of the current wafer processing data records using record hashing. Based on the current timestamp, obtain the current feature hash value and the record frequency of the current record hash value mapped to the hash bucket, and evaluate the anomalies of the current wafer processing data in the time dimension. The number of hash buckets is obtained by hash mapping the historical wafer processing data using the feature hash and the record hash. The anomaly score of the current feature hash value is calculated using the formula. Calculate, where, This indicates the anomaly score of the current feature hash value. The anomaly score of the current record's hash value is calculated using the formula. Calculate, where, This indicates the abnormal score of the current record's hash value. r i For the i-th current wafer processing data, r ij This represents the current processing value of the j-th working parameter item in the i-th current wafer processing data. The current wafer processing data r in the current timestamp t i The number of times the corresponding hash value appears. As of the current timestamp t, the current wafer processing data r i The current total record frequency of the corresponding hash value in the hash bucket. The current processing value r in the current timestamp ij The number of times the corresponding hash value appears. The current processing value r up to the current timestamp t. ij The current total record frequency of the corresponding hash value in the hash bucket. This is the current timestamp.

2. The wafer manufacturing anomaly identification method according to claim 1, characterized in that, The step of hashing the current processing value of each of the working parameter items using feature hashing to obtain the current feature hash value includes: Perform a logarithmic transformation on the current processing value of each of the aforementioned working parameter items; Obtain the processing value range corresponding to the aforementioned working parameter item; The current processing value after logarithmic transformation is normalized according to the range of processing values; The current feature hash value is obtained by mapping the current processed value after normalization.

3. The wafer manufacturing anomaly identification method according to claim 1, characterized in that, The step of performing a combined hash mapping on each of the current wafer processing data entries using record hashing to obtain the current record hash value includes: Based on the number of dimensions of the current wafer processing data, a random vector of the specified number of dimensions is generated, and the elements of each random vector are randomly sampled from a standard normal distribution N(0,1). Based on the dimensions of the current wafer processing data, the current wafer processing data is converted into a first vector; Calculate the dot product between the first vector and each random vector, and convert the dot product into binary to obtain a binary string corresponding to the first vector; Convert the binary string to a decimal integer to obtain the hash value of the current record.

4. The wafer manufacturing process anomaly identification method according to claim 1, characterized in that, The step of obtaining the current feature hash value mapped to the hash bucket and the record frequency of the current record hash value based on the current timestamp, and evaluating the anomalies of the current wafer processing data in the time dimension, includes: Calculate the current total record frequency of the feature hash value and the record hash value mapped to the hash bucket based on the current timestamp; Based on the current total record frequency, determine the anomaly score of the current feature hash value and the current record hash value, respectively. The anomaly score is used to assess the anomalies in the current wafer processing data in the time dimension.

5. The wafer manufacturing anomaly identification method according to claim 1, characterized in that, The formula also includes a time decay coefficient α. Will and Multiply the specific value by α and then substitute it into the formula for calculation; and The specific value is multiplied by α and then substituted into the formula for calculation.

6. The wafer manufacturing process anomaly identification method according to claim 1, characterized in that, After receiving the current wafer processing data in real time, the method further includes: Obtain historical wafer processing data; Identify the current working parameter items in the current wafer processing data and the historical working parameter items carried in the historical wafer processing data respectively; When it is determined that the current working parameter item and the historical working parameter item are consistent, the current processing value of each working parameter item is hashed using feature hashing.

7. A wafer manufacturing anomaly identification device based on a hash algorithm, characterized in that, The device includes: The receiving module is used to receive the current wafer processing data in real time. The current wafer processing data includes the current timestamp, wafer processing working parameters, and the corresponding current processing values. The feature hash processing module is used to perform a hash mapping on the current processing value of each of the work parameter items using feature hashing to obtain the current feature hash value; The record hash processing module is used to perform a combined hash mapping on each of the current wafer processing data using the record hash to obtain the current record hash value; The evaluation module is used to obtain the record frequency of the current feature hash value and the current record hash value mapped to the hash bucket based on the current timestamp, and to evaluate the anomalies of the current wafer processing data in the time dimension. The number of hash buckets is obtained by hash mapping the historical wafer processing data using the feature hash and the record hash. The anomaly score of the current feature hash value is calculated using the formula. Calculate, where, This indicates the anomaly score of the current feature hash value. The anomaly score of the current record's hash value is calculated using the formula. Calculate, where, This indicates the abnormal score of the current record's hash value. r i For the i-th current wafer processing data, r ij This represents the current processing value of the j-th working parameter item in the i-th current wafer processing data. The current wafer processing data r in the current timestamp t i The number of times the corresponding hash value appears. As of the current timestamp t, the current wafer processing data r i The current total record frequency of the corresponding hash value in the hash bucket. The current processing value r in the current timestamp ij The number of times the corresponding hash value appears. The current processing value r up to the current timestamp t. ij The current total record frequency of the corresponding hash value in the hash bucket. This is the current timestamp.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the wafer production condition anomaly identification method according to any one of claims 1 to 6.

Images