Yield rate evaluation method and yield rate evaluation device

Through the information entropy calculator and the ultimate gradient enhancement model, the complexity and flexibility of yield evaluation in the semiconductor industry are solved, and more efficient packaging strategy and cost control are achieved, which improves shipment volume and reduces production costs.

CN118941099APending Publication Date: 2024-11-12WINBOND ELECTRONICS CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310759648.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-11
Filing Date
2023-06-26
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The yield evaluation method in the existing semiconductor industry relies on complex machine learning models, which makes the results difficult to explain and the model application is inflexible, and the model needs to be retrained when data changes, which increases the cost and inefficiency problems.

Method used

Using the information entropy calculator, the information entropy value of each parameter is independently calculated through the extreme gradient boosting (XGBOOST) model, the product information entropy calculator is established, and the complex fitting model is replaced, and the evaluation is used using fixed variables is simplified to model application and Data processing.

Benefits of technology

It improves model interpretability and application flexibility, reduces data volume requirements, reduces training costs, enhances the controllability of packaging strategies, improves shipment volume and reduces production costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941099A_ABST
    Figure CN118941099A_ABST
Patent Text Reader

Abstract

The invention discloses a yield evaluation method and a yield evaluation device. The method comprises the following steps of: collecting wafer manufacturing data, front-section wafer test data and rear-section product yield information in a semiconductor product manufacturing process, and selecting a plurality of parameters related to yield from the wafer manufacturing data, the front-section wafer test data and the rear-section product yield information; the relative information entropy of defective products in the multiple samples manufactured by using the parameters relative to global fixed probability defective products is calculated to establish a product information entropy calculator, and the global fixed probability defective products represent the defective products with the yield not changing along with the parameters; and collecting wafer manufacturing data of the current product and front wafer test data, and substituting the data into the product information entropy calculator to evaluate the yield of the current product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to an evaluation method and a device, and in particular to a yield evaluation method and a yield evaluation device. Background Art

[0002] In the semiconductor industry, packaging and testing costs account for a large part of the chip manufacturer's overall production costs. Among the shipped products, there will be great differences in packaging and testing costs due to the different specifications of the shipped products. For example, for memory chip products of the same density, the packaging and testing costs of automotive specifications will be nearly twice as much as those of general commercial specifications. Therefore, if the yield of high-specification products is not high, its packaging and testing will cause a considerable loss of profits. Therefore, after the chip completes the front-end manufacturing and testing, the person in charge of product engineering must carefully classify the finished products into appropriate specifications in order to properly control the packaging and testing costs. For example, chips with lower yields in the back-end are sent to packaging and testing with lower specifications, and products with higher yields in the back-end are packaged and tested with high specifications. In this way, a larger shipment volume can be achieved and costs can be controlled.

[0003] Figure 1 The basic process for determining the packaging specifications of existing products includes chip product manufacturing 12, front-end chip testing 14, back-end yield evaluation 16, and packaging testing 18. The packaging test 18 can be divided into high-specification packaging test 182 and low-specification packaging test 184. The back-end yield evaluation 16 determines whether the product is subjected to high-specification packaging test 182 or low-specification packaging test 184 based on the manufacturing data and test data obtained from chip product manufacturing 12 and front-end chip testing 14. At this stage, the early method is to intervene through experts 162 in the field and make judgments on the product based on their professional knowledge. In recent years, many companies have introduced machine learning auxiliary systems 164, such as deep neural networks (DNN), random forests, etc., to make auxiliary judgments to increase efficiency and improve the defects that may be caused by human judgment. Although machine learning can process large amounts of complex data, due to its complex mathematical model, there are often some unavoidable shortcomings.

[0004] For example, neural networks can synthesize complex yield functions based on nonlinear functional superposition in linear algebra. Because they need to use multi-layer and deep network models to achieve the best fit, the result is often a complex data combination that is difficult to explain, which greatly reduces its flexibility in use. For example, when users collect data in the future, if there is information missing or test changes, users cannot deconstruct the model and select reasonable data to bring into the model. Faced with this situation, users often have to retrain the model. Summary of the invention

[0005] The present invention provides a yield evaluation method and device, which can increase product shipments and reduce production costs by establishing an information entropy calculator and using it to evaluate product yield and packaging strategies.

[0006] An embodiment of the present invention provides a yield evaluation method, which is applicable to an electronic device with a processor. The method includes the following steps: collecting chip manufacturing data, front-end chip test data, and back-end product yield information in the semiconductor product manufacturing process, and selecting multiple parameters related to the yield; calculating the relative information entropy of defective products in multiple samples manufactured using each parameter relative to the global fixed probability defective products, so as to establish a product information entropy calculator, wherein the global fixed probability defective products represent defective products whose yield does not change with the parameters; and collecting the chip manufacturing data and front-end chip test data of the current product and substituting them into the product information entropy calculator to evaluate the yield of the current product.

[0007] An embodiment of the present invention provides a yield evaluation device, which includes a connection device, a storage device and a processor. The connection device is used to connect multiple machines related to semiconductor product manufacturing and testing. The storage device is used to store computer programs. The processor is coupled to the connection device and the storage device, and is configured to load and execute a computer program to use the connection device to collect chip manufacturing data, front-end chip test data and back-end product yield information from the machine in the semiconductor product manufacturing process, select multiple parameters related to the yield, and calculate the relative information entropy of defective products in multiple samples manufactured using each parameter relative to the global fixed probability defective products, so as to establish a product information entropy calculator, wherein the global fixed probability defective product represents a defective product whose yield does not change with the parameters, and use the connection device to collect the chip manufacturing data and front-end chip test data of the current product from the machine and substitute them into the product information entropy calculator to evaluate the yield of the current product.

[0008] Based on the above, the yield evaluation method and device of the present invention replaces the original complex model with the constructed product information entropy calculator, and replaces the arbitrary variable part required for the original fitting with fixed variables. Therefore, when using the method, the user can more clearly understand the model and the results it produces, and use the results to formulate packaging strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The basic process for determining product packaging specifications for existing products;

[0010] Figure 2 is a comparison diagram of a yield evaluation method according to an embodiment of the present invention;

[0011] Figure 3is a block diagram of a yield evaluation device according to an embodiment of the present invention;

[0012] Figure 4 is a flow chart of a yield evaluation method according to an embodiment of the present invention;

[0013] Figure 5 is a schematic diagram of establishing an information entropy calculator according to an embodiment of the present invention;

[0014] FIG. 6A to FIG. 6C is an example of calculating information entropy according to an embodiment of the present invention;

[0015] Figure 7 is a flow chart of a method for establishing an information entropy calculator according to an embodiment of the present invention;

[0016] Fig. 8A and Figure 8B is an example of calculating weights according to an embodiment of the present invention;

[0017] Fig.9A and Fig. 9B is a weight table shown according to an embodiment of the present invention.

[0018] Description of Figure Numbers

[0019] 12: Chip product manufacturing

[0020] 14: Front-end chip testing

[0021] 16: Back-end yield evaluation

[0022] 162: Expert

[0023] 164: Machine Learning Assisted Systems

[0024] 18: Packaging test

[0025] 182: High specification packaging test (high cost, high price)

[0026] 184: Low specification packaging test (low cost, low price)

[0027] 20: Parameters

[0028] 22: Neural Networks

[0029] 24: Information Entropy Calculator

[0030] 30: Yield evaluation device

[0031] 32: Connecting device

[0032] 34: Storage device

[0033] 36: Processor

[0034] 52: The first group (characteristic parameters)

[0035] 522: Wafer manufacturing data

[0036] 524: Front-end chip test data

[0037] 54: The second group (label parameters)

[0038] 542: Back-end product yield information

[0039] 56: Data cleaning and pre-processing

[0040] 58: Information entropy calculator established

[0041] 62, 64: Relationship diagram

[0042] 82: Distribution map

[0043] 84: Change relationship diagram

[0044] 92: Weight table

[0045] 94: Sample weight table

[0046] Ep1~Epn: relative information entropy

[0047] p1~pn:area

[0048] S402~S406, S702~S718: Steps DETAILED DESCRIPTION

[0049] The embodiments of the present invention improve the main shortcomings of the traditional yield assessment method using machine learning. By using the definition of information entropy and the Extreme Gradient Boost (XGBOOST) model, a more interpretable result can be obtained, which can give users more flexibility in application. For example, the independence between each variable can give users more choices and variability when collecting data in the future, and can also improve model inspection or debugging to a certain extent.

[0050] Please refer to Figure 2At present, the mainstream machine learning method in the industry is to use a neural network 22, which mainly uses the method of linear algebra to input various parameters 20 (including parameters F1~Fn) into the activator for non-linear transformation, and then optimizes the conditions of the optimizer, and finally superimposes the functionals composed of the original parameters into a complex yield function Y. In the process, it is necessary to set the coefficients bi and bj, and use a large number of random variables for weights Wi and Wj, and achieve the best fit result through the continuous optimization process.

[0051] Although the neural network 22 can achieve the purpose of fitting a complex yield function by generating a large number of random variables and optimizing the process, it also introduces too many and complex variables, making it difficult for users to truly interpret and understand the final results.

[0052] To address this problem, the embodiment of the present invention replaces the original model method using complex structures such as neural network 22 with an information entropy calculator 24 constructed by information entropy in data science to replace the original complex model that was originally optimized to achieve yield function fitting. The arbitrary variable part originally required for fitting is replaced by fixed variables. This method allows users to more clearly understand the model and the results it produces when using the method, and use this result to set a periodic packaging strategy.

[0053] When calculating parameters and sample yield performance, the information entropy calculator 24 of this embodiment calculates the information entropy value E of each parameter (such as Fn) relative to the yield performance separately and independently. Therefore, there is a lot more flexibility in use. If the parameters of the new product are different from the parameters used in the model construction, a simple filter can be used to process the input parameters without re-constructing the model.

[0054] In addition, the amount of data required for training can be greatly reduced by using the information entropy calculator 24. In the case of a small sample, a screening model can still be constructed by the change relationship between the information entropy of the parameters.

[0055] Please refer to Figure 3The yield evaluation device 30 of this embodiment is, for example, an electronic device such as a file server, a database server, an application server, a workstation or a personal computer with computing capabilities. In some embodiments, the yield evaluation device 30 is connected to a machine related to semiconductor product manufacturing and testing via a connection device 32. In other embodiments, the yield evaluation device 30 can be integrated into the above-mentioned machine to perform yield evaluation, which is not limited here. The yield evaluation device 30 includes a connection device 32, a storage device 34 and a processor 36. The functions of each component are described as follows:

[0056] The connection device 32 is, for example, any wired or wireless interface device used to connect to and transmit data with external devices such as semiconductor product manufacturing machines and test machines. For wired methods, the connection device can be a universal serial bus (USB), RS232, universal asynchronous receiver and transmitter (UART), internal integrated circuit (I2C) or serial peripheral interface (SPI), but is not limited to this. For wireless methods, the connection device can be a device that supports communication protocols such as wireless fidelity (Wi-Fi), RFID, Bluetooth, infrared, near field communication (NFC) or device to device (D2D), but is not limited to this. In some embodiments, the connection device 32 is, for example, a network card that supports wired network connections such as Ethernet or a wireless network card that supports wireless communication standards such as IEEE 802.11n / b / g of the Institute of Electrical and Electronics Engineers, which can connect to the network through wired or wireless methods and connect to the above-mentioned machines to obtain data.

[0057] The storage device 34 is, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk or other recording media, and is used to store computer programs executable by the processor 36 and data obtained by the connection device 32. In some embodiments, the storage device 34 can store relevant data (such as a weight table) of the product information entropy calculator established by the processor 36, but the present embodiment is not limited thereto.

[0058] The processor 36 is, for example, a central processing unit, or other programmable general-purpose or special-purpose microprocessor, microcontroller, digital signal processor, programmable controller, application-specific integrated circuit, programmable logic device or other similar device or combination of these devices, and the present invention is not limited thereto. In this embodiment, the processor 36 can load a computer program from the storage device 34 to execute the yield evaluation method of the embodiment of the present invention.

[0059] Please also refer to Figure 3 and Figure 4 The method of this embodiment is applicable to Figure 3 The yield evaluation device 30 is shown.

[0060] In step S402, the processor 36 of the yield evaluation device 30 uses the connection device 32 to connect to multiple machines related to semiconductor product manufacturing and testing to collect chip manufacturing data, front-end chip testing data and back-end product yield information in the semiconductor product manufacturing process from the machines, and select multiple parameters related to the yield.

[0061] The wafer manufacturing data is, for example, wafer acceptance test (WAT) data, or inline process data such as device critical dimension, exposure shifting, etc. The front-end wafer test data is, for example, chip probe results including yield and failure rate of each item, or direct current (DC) test data. This embodiment does not limit the types and numbers of the above data.

[0062] In addition, the machine is, for example, a manufacturing machine or a testing machine for semiconductor products, or a machine that stores a database of wafer manufacturing data, front-end wafer testing data, and back-end product yield information. This embodiment does not limit the type and number of machines.

[0063] In detail, the yield evaluation method of the embodiment of the present invention is mainly divided into two stages, namely, the stage of establishing an information entropy calculator and the production process after the selection strategy method is introduced. In the early stage of establishing the information entropy calculator, product information of a reference cycle is selected and collected, which includes wafer manufacturing data, front-end wafer test data and back-end product yield information, and based on these data, the information entropy calculator of the product is constructed.

[0064] Please refer to Figure 5 , this embodiment collects data first. The collected data is divided into two groups. The first group of data 52 includes model parameters (as feature parameters) of chip manufacturing data 522 and front-end chip test data 524; the second group of data 54 is product yield information 542 (as label parameters) of back-end testing. Then, the collected data is first cleaned (Data Clean) and pre-processed (Pre-treatment) 56, which includes making some basic assumptions based on the product information provided by field experts, and based on these assumptions, the collected data is projected into the assumed mathematical space to facilitate the subsequent operation of the information entropy calculator. After all the data is processed and projected into the assumed mathematical space, the establishment of the information entropy calculator 58 is launched.

[0065] Back to Figure 4 In step S404, the processor 36 calculates the relative information entropy of defective products in multiple samples manufactured using various parameters relative to the global fixed probability defective products, so as to establish a product information entropy calculator. The global fixed probability defective products represent defective products whose yield does not change with the parameters.

[0066] FIG. 6A to FIG. 6C This is an example of calculating information entropy according to an embodiment of the present invention. Fig. 6A First, a certain parameter is selected and a relationship diagram 62 of the yield rate versus the above parameter is drawn, and then it is divided into n regions (numbered p1 to pn) in order of size, and finally a distinction standard p0 is selected to distinguish good and bad products, and the product above the standard is called a good product, and the product below the standard is a bad product. In actual operation, m standards can be selected to calculate the information entropy, where m is a positive integer.

[0067] Please refer to Figure 6B , this embodiment further defines a situation where the yield does not change with a certain parameter, and draws a relationship diagram 64 of the yield versus the above-mentioned parameter.

[0068] After defining "good and bad products", this embodiment calculates the relative information entropy value of each "partition bad product" to the "global fixed probability bad product", for example Figure 6C As shown, the relative information entropy value of the defective products in area p1 in the relationship diagram 62 to the global fixed probability defective products in area p1 in the relationship diagram 64 is calculated. Thus, the relative information entropy value of the yield performance of the parameter in each area can be obtained. This relative information entropy value can indicate "the strength of the instability (uncertainty) or the degree of surprise (surprise) of the parameter on the yield performance". The calculation formula of the relative information entropy value Entropy is as follows:

[0069] Entropy = -(pn - p0)log(p0).

[0070] Among them, pn is the above-mentioned "partition defect", and p0 is the above-mentioned "global fixed probability defect". The larger the relative information entropy value Entropy is, the less the distribution of the sample is like "invariant proportion behavior". In other words, the more it looks like invariant proportion behavior, the more surprising it is.

[0071] In coordination, this embodiment can give the parameters a relative weight based on the assumptions of the product information on the data to describe the relationship between the parameters and the yield. Finally, following the above procedure, the relationship between all parameters and the yield performance is calculated in sequence and recorded in the storage device 34, thus completing the establishment of the information entropy calculator of the product.

[0072] Please refer to Figure 7 This embodiment illustrates Figure 4Detailed process of step S404.

[0073] In step S702 , the processor 36 selects a parameter from a plurality of parameters related to yield.

[0074] In step S704, the processor 36 selects a distinction standard from a plurality of distinction standards. In some embodiments, the processor 36 sets a distinction standard for a predetermined value or ratio for the yield range, and selects a distinction standard for calculating the relative information entropy. For example, the processor 36 may select 20%, 40%, 60%, and 80% of the maximum yield value as the distinction standard.

[0075] In step S706, the processor 36 divides the yield distribution of the samples manufactured using the parameters into a plurality of regions. The processor 36 divides the parameter range in the yield distribution into a plurality of regions (e.g., Fig. 6A shown).

[0076] In step S708, the processor 36 classifies the samples in each area into good products and bad products according to the selected classification standard, and calculates the relative information entropy of the bad products relative to the global fixed probability bad products.

[0077] In step S710, the processor 36 assigns a weight to the selected parameter according to the calculated relative information entropy of each region. In detail, regarding the assignment of weights, this embodiment only assigns higher weights to "parameters with directional change behaviors". For example, if the parameter value is larger, the information entropy value is smaller, it can be determined that the parameter has directional change behavior and a weight is assigned; if the parameter value is smaller, the information entropy value is smaller, it can be determined that the parameter has directional change behavior and a weight is assigned. It should be noted that if the information entropy value is negative, it will be set to zero and not considered because it is meaningless by definition.

[0078] For example, Fig. 8A and Figure 8B is an example of calculating weights according to an embodiment of the present invention. Fig. 8A Shown via FIG. 6A to FIG. 6C A distribution diagram 82 of the relative information entropy calculated by the example of FIG. 8 includes relative information entropies Ep1 to Epn calculated for n regions. Figure 8B A change relationship diagram 84 of the relative information entropy is shown. In this embodiment, weights are given to the parameters according to the change relationship diagram 84.

[0079] With respect to the relative information entropies Ep1-Epn in the distribution diagram 82, this embodiment first sets the relative information entropies with negative values ​​to zero, and at this time no weight is given to the parameters.

[0080] In view of the situation where the information entropy value decreases as the parameter increases, this embodiment will compare the relative information entropies Ep(n-1) and Ep(n) of adjacent regions. If the relative information entropy Ep(n) of the current region is greater than the relative information entropy Ep(n-1) of the previous region, the relative information entropy Ep(n) of the current region is set to the relative information entropy Ep(n-1) of the previous region, that is, Ep(n) = Ep(n-1), and the weight given to the parameter is increased; and if the relative information entropy Ep(n) of the current region is not greater than the relative information entropy Ep(n-1) of the previous region, the relative information entropy Ep(n) of the current region is maintained.

[0081] On the other hand, in view of the situation where the information entropy value increases as the parameter decreases, the present embodiment also compares the relative information entropies Ep(n-1) and Ep(n) of adjacent regions. If the relative information entropy Ep(n) of the current region is less than the relative information entropy Ep(n-1) of the previous region, the relative information entropy Ep(n) of the current region is set to the relative information entropy Ep(n-1) of the previous region, that is, Ep(n)=Ep(n-1), and the weight given to the parameter is increased; and if the relative information entropy Ep(n) of the current region is not less than the relative information entropy Ep(n-1) of the previous region, the relative information entropy Ep(n) of the current region is maintained.

[0082] Through the above method, the relative information entropy calculated for the selected parameters using the selected distinction criteria and the weight given to the parameters can be obtained.

[0083] Next, in step S712, the processor 36 determines whether all the classification criteria have been used. If there are still classification criteria that have not been used, the process returns to step S704, reselects the classification criteria, and repeats steps S706 to S710 to obtain the relative information entropy calculated using different classification criteria and the weights given to the parameters.

[0084] In step S712, if it is determined that all the classification criteria have been used, the process proceeds to step S714, where the processor 36 records the calculated relative information entropy and weight of each parameter in the storage device 34. The processor 36 records the relative information entropy and weight in a table, for example.

[0085] In step S714, the processor 36 determines whether the relative information entropy and weights of all parameters have been calculated. If there are still parameters that have not been calculated, the process returns to step S702, reselects parameters, and repeats steps S704 to S714 to obtain weights given to different parameters.

[0086] In step S716, if it is determined that all parameters have been calculated, the process proceeds to step S718, where the processor 36 uses the calculated relative information entropy and weight of each parameter to establish a product information entropy calculator. The processor 36, for example, superimposes the weights calculated for each parameter using different distinction criteria as the weights given to the parameters.

[0087] Back to Figure 4 In the process, after the product information entropy calculator is established, in step S406, the processor 36 collects the wafer manufacturing data and the front-end wafer test data of the current product and substitutes them into the product information entropy calculator to evaluate the yield of the current product. Among them, the processor 36, for example, calculates the sum of the weights corresponding to the multiple parameters used to manufacture the current product as the score for evaluating the yield of the current product. Afterwards, the processor 36 can determine the packaging and testing specifications applicable to the current product according to the yield score.

[0088] Specifically, at this stage, the present embodiment collects the manufacturing data of the target batch of products and the previous test data, and substitutes them into the previously established product information entropy calculator to calculate the corresponding weight score of each product and output the selection strategy of the product packaging and testing specifications. Based on the selection strategy, the packaging and testing cost control can be carried out.

[0089] For example, for products with a higher yield score (e.g., exceeding a predetermined threshold), they can be sent to a high-specification plus packaging test, while for products with a lower yield score (e.g., below a predetermined threshold), they can be sent to a low-specification plus packaging test. In this way, the shipment volume of the products can be increased and the production cost can be reduced.

[0090] Fig.9A and Fig. 9B is a weight table according to an embodiment of the present invention. Fig.9A The weight table 92 shows the Figure 7 The weights of the parameters (including parameters 1 to n) calculated by the process are shown in the weight table 92, which represent the weights of the regions in the direction of increasing parameters from top to bottom. According to the weight table 92, the weights corresponding to the parameters of each sample can be found, and the Fig. 9B The sample weight table 94 is shown. The total weight can be obtained by summing up the weights of the parameters of each sample (including samples A to n), and the total weight can be used as a reference value for evaluating the rank of the sample.

[0091] In summary, in the yield evaluation method and device of the embodiment of the present invention, since each parameter is separately and independently calculated for its information entropy value relative to the yield performance when calculating the parameters and sample yield performance, there is a lot more flexibility in use. If the parameters of the new product are different from the parameters used when the model is constructed, a simple filter can be used to process the input parameters without the need to rebuild the model. By means of an information entropy calculator, the amount of data required to be collected during training can be greatly reduced. Through the screening action of the information entropy calculator, parameters with variations can be excluded and not included in the weight calculation, thereby obtaining a set of appropriate weight scores and selection strategies, thereby achieving the purpose of controlling packaging and testing costs.

[0092] Although the present disclosure has been disclosed by the above embodiments, the embodiments are not intended to limit the present disclosure. It will be apparent to those skilled in the art that various modifications and changes may be made to the structure of the present disclosure without departing from the scope or spirit of the present disclosure. Therefore, the protection scope of the present disclosure belongs to the appended claims.

Claims

1. A yield evaluation method, applicable to an electronic device having a processor, comprising the following steps: Collect wafer manufacturing data, front-end wafer testing data and back-end product yield information in the semiconductor product manufacturing process, and select multiple parameters related to the yield; Calculating the relative information entropy of defective products in a plurality of samples manufactured using each of the parameters relative to defective products with a global fixed probability, so as to establish a product information entropy calculator, wherein the defective products with a global fixed probability represent defective products whose yield does not change with the parameters; as well as The wafer manufacturing data and the front-end wafer testing data of the current product are collected and substituted into the product information entropy calculator to evaluate the yield of the current product.

2. The method according to claim 1, wherein the step of calculating the relative information entropy of defective products in a plurality of samples manufactured using each of the parameters relative to the global fixed probability defective products to establish the product information entropy calculator comprises: For each of the parameters, dividing the yield distribution of the sample manufactured using the parameter into a plurality of regions; Distinguishing the samples in each of the regions into good products and bad products according to at least one differentiation standard, and respectively calculating the relative information entropy of the bad products relative to the global fixed probability bad products; According to the calculated relative information entropy of each of the regions, a weight is given to the parameter; as well as The product information entropy calculator is established using the calculated relative information entropy and the weight of each parameter.

3. The method according to claim 2, wherein the step of giving the weight to the parameter according to the calculated relative information entropy of each of the regions comprises: The relative information entropy having a negative value is set to zero.

4. The method according to claim 3, wherein the step of giving the weight to the parameter according to the calculated relative information entropy of each of the regions further comprises: comparing the relative information entropies of adjacent regions; If the relative information entropy of the current area in the area is greater than the relative information entropy of the previous area, setting the relative information entropy of the current area to the relative information entropy of the previous area, and increasing the weight given to the parameter; as well as If the relative information entropy of the current region among the regions is not greater than the relative information entropy of the previous region, the relative information entropy of the current region is maintained.

5. The method according to claim 3, wherein the step of giving the weight to the parameter according to the calculated relative information entropy of each of the regions further comprises: comparing the relative information entropies of adjacent regions; If the relative information entropy of the current area in the area is less than the relative information entropy of the previous area, the relative information entropy of the current area is set to the relative information entropy of the previous area, and the weight given to the parameter is increased; as well as If the relative information entropy of the current region among the regions is not less than the relative information entropy of the previous region, the relative information entropy of the current region is maintained.

6. The method according to claim 2, wherein the step of assigning the weight to the parameter according to the calculated relative information entropy of each of the regions comprises: The weights calculated using the different distinction criteria are superimposed as the weights given to the parameters.

7. The method according to claim 2, wherein for each of the parameters, the step of dividing the yield distribution of the samples manufactured using the parameters into a plurality of regions comprises: The yield distribution is divided into the plurality of regions according to predetermined values ​​or ratios.

8. The method according to claim 2, wherein the step of evaluating the yield of the current product comprises: For a plurality of parameters used in manufacturing the current product, a sum of the weights corresponding to the parameters is calculated as a score for evaluating the yield of the current product.

9. The method according to claim 1, wherein after the step of collecting wafer manufacturing data, front-end wafer testing data and back-end product yield information in the semiconductor product manufacturing process, it further comprises: The collected wafer manufacturing data, the front-end wafer test data and the back-end product yield information are cleaned and pre-processed, and the processed data are projected into a hypothetical mathematical space.

10. The method according to claim 1, wherein after the step of evaluating the yield of the current product, the method further comprises: The packaging and testing specifications applicable to the current product are determined based on the yield.

11. A yield evaluation device, comprising: Connecting devices that connect multiple machines related to semiconductor product manufacturing and testing; a storage device storing a computer program; as well as A processor, coupled to the connection device and the storage device, is configured to load and execute the computer program to: Using the connection device to collect wafer manufacturing data, front-end wafer testing data, and back-end product yield information during the semiconductor product manufacturing process from the machine, and selecting a plurality of parameters related to the yield; Calculating the relative information entropy of defective products in a plurality of samples manufactured using each of the parameters relative to defective products with a global fixed probability, so as to establish a product information entropy calculator, wherein the defective products with a global fixed probability represent defective products whose yield does not change with the parameters; as well as The connection device is used to collect the wafer manufacturing data and the front-end wafer testing data of the current product from the machine and substitute them into the product information entropy calculator to evaluate the yield of the current product.

12. The yield evaluation device according to claim 11, wherein the processor comprises: For each of the parameters, dividing the yield distribution of the sample manufactured using the parameter into a plurality of regions; Distinguishing the samples in each of the regions into good products and bad products according to at least one differentiation standard, and respectively calculating the relative information entropy of the bad products relative to the global fixed probability bad products; According to the calculated relative information entropy of each of the regions, a weight is given to the parameter; as well as The product information entropy calculator is established using the calculated relative information entropy and the weight of each parameter. 13 . The yield evaluation apparatus according to claim 12 , wherein the processor sets the relative information entropy having a negative value to zero.

14. The yield evaluation device according to claim 13, wherein the processor further compares the relative information entropy of the adjacent regions, wherein if the relative information entropy of the current region among the regions is greater than the relative information entropy of the previous region, the relative information entropy of the current region is set to the relative information entropy of the previous region, and the weight given to the parameter is increased, and if the relative information entropy of the current region among the regions is not greater than the relative information entropy of the previous region, the relative information entropy of the current region is maintained.

15. The yield evaluation device according to claim 13, wherein the processor further compares the relative information entropy of the adjacent regions, wherein if the relative information entropy of the current region among the regions is less than the relative information entropy of the previous region, the relative information entropy of the current region is set to the relative information entropy of the previous region, and the weight given to the parameter is increased, and if the relative information entropy of the current region among the regions is not less than the relative information entropy of the previous region, the relative information entropy of the current region is maintained. 16 . The yield evaluation apparatus according to claim 12 , wherein the processor comprises superimposing the weights calculated using different discrimination criteria as the weights given to the parameters. 17 . The yield evaluation apparatus according to claim 12 , wherein the processor is configured to divide the yield distribution into the plurality of regions according to a predetermined value or ratio. 18 . The yield evaluation apparatus according to claim 12 , wherein the processor calculates a sum of the weights corresponding to a plurality of parameters used in manufacturing the current product as a score for evaluating the yield of the current product.

19. The yield evaluation device according to claim 11, wherein the processor further performs data cleaning and pre-processing on the collected wafer manufacturing data, the front-end wafer test data and the back-end product yield information, and projects the processed data into a hypothetical mathematical space. 20 . The yield evaluation device according to claim 11 , wherein the processor further determines a packaging and testing specification applicable to the current product according to the yield.