Fault early warning method of semiconductor equipment

Abstract

A fault early warning method for semiconductor equipment comprises the steps that a plurality of sets of historical equipment operation data are provided, and each set of historical equipment operation data comprises equipment operation environment parameters and whether the semiconductor equipment breaks down or not; inputting the plurality of groups of training equipment operation data into the neural network model; inputting equipment operation environment parameters in the plurality of groups of test equipment operation data into a neural network model, and outputting a result of whether the semiconductor equipment has a fault or not by the neural network model; when the test accuracy of the neural network model reaches a preset threshold value, completing the model test; and inputting the operation environment parameters of the to-be-detected equipment into the neural network model to predict whether the semiconductor equipment has a fault or not. Whether the semiconductor equipment breaks down or not can be predicted through the neural network model, the risk that the semiconductor equipment breaks down can be reduced by overhauling or adjusting the operation parameters of the semiconductor equipment in advance, the operation time of the semiconductor equipment is guaranteed, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor manufacturing technology, and in particular to a fault early warning method for semiconductor equipment. Background Art

[0002] With the rapid development of semiconductor technology, the operating status of equipment in semiconductor manufacturing companies has a significant impact on the production capacity of the company, so equipment failure and downtime will cause huge economic losses to the company. Therefore, real-time monitoring of semiconductor equipment and effective fault alarm in production are the development trend of modern electronic manufacturing industry, which requires modern semiconductor equipment to not only be able to respond to the changing external environment in a timely manner, but also be able to quickly reconfigure various internal resources and issue fault alarms to achieve the purpose of optimizing production.

[0003] However, there are still many problems with the fault alarm of semiconductor devices in the prior art. Summary of the invention

[0004] The technical problem solved by the present invention is to provide a fault early warning method for semiconductor equipment to ensure the operating time of the semiconductor equipment and improve production efficiency.

[0005] To solve the above problems, the present invention provides a fault warning method for semiconductor equipment, comprising: providing several groups of historical equipment operation data, each group of the historical equipment operation data includes equipment operation environment parameters, and whether the semiconductor equipment fails under the equipment operation environment parameters; dividing the several groups of the historical equipment operation data into several groups of training equipment operation data and several groups of test equipment operation data; inputting the several groups of the training equipment operation data into a neural network model for model training; after the model training, inputting the equipment operation environment parameters in the several groups of the test equipment operation data into the trained neural network model, and the neural network model outputs several groups of corresponding results of whether the semiconductor equipment will fail; comparing the several groups of corresponding results of whether the semiconductor equipment will fail output by the neural network model with the results of whether the semiconductor equipment will fail in the several groups of the test equipment operation data, until the test accuracy of the neural network model reaches a preset threshold, and completing the model test of the neural network model; providing the equipment operation environment parameters to be detected of the semiconductor equipment; inputting the equipment operation environment parameters to be detected into the neural network model that has completed the model test, and predicting whether the semiconductor equipment will fail by the neural network model.

[0006] Optionally, the semiconductor equipment includes: etching equipment, growth equipment, deposition equipment and annealing equipment.

[0007] Optionally, when the semiconductor device is a physical vapor deposition device, the corresponding device operating environment parameters include: AC deflection current, DC deflection voltage, electrostatic suction cup clamping voltage, gas flow in the third chamber, heating plate pressure, heating plate temperature, process pressure, power generator current, power generator power, and power generator voltage.

[0008] Optionally, the neural network model adopts a back-propagation neural network model.

[0009] Optionally, the neural network model includes: an input layer, a hidden layer and an output layer, and the number of nodes h of the hidden layer is determined by the number of nodes m of the input layer and the number of nodes m of the output layer, that is:

[0010]

[0011] Among them, the number of nodes m in the input layer is the number of types of the equipment operating environment parameters; the number of nodes n in the output layer is the number of results output by the neural network model.

[0012] Optionally, the method of training the neural network model until the test accuracy of the neural network model reaches a preset threshold includes: when the test accuracy of the neural network model does not reach the preset threshold, increasing the number of model training times of the neural network model until the test accuracy of the neural network model reaches the preset threshold.

[0013] Optionally, the number of groups of the training device operating data is greater than the number of groups of the test device operating data.

[0014] Optionally, the ratio of the number of groups of the training device operating data to the number of groups of the test device operating data is 6:5 to 9:5.

[0015] Correspondingly, the technical solution of the present invention also provides a storage medium on which computer instructions are stored, and when the computer instructions are executed, the steps of any of the above methods are executed.

[0016] Correspondingly, the technical solution of the present invention also provides a terminal, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and the processor executes the steps of any of the above methods when executing the computer instructions.

[0017] Compared with the prior art, the technical solution of the present invention has the following advantages:

[0018] In the fault warning method of semiconductor equipment of the technical solution of the present invention, by performing model training and model testing on the neural network model, in the model testing stage, when the test accuracy of the neural network model reaches a preset threshold, the neural network model can predict whether the semiconductor equipment will fail, and the risk of failure of the semiconductor equipment can be reduced by early maintenance or adjustment of the operating parameters of the semiconductor equipment, thereby ensuring the operating time of the semiconductor equipment and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of a fault warning method for semiconductor equipment according to an embodiment of the present invention;

[0020] Figures 2 to 3 It is a schematic diagram of the structure of each step of the fault warning method for semiconductor equipment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] As described in the background art, there are still many problems with fault alarms of semiconductor devices in the prior art, which will be described in detail below.

[0022] Currently, the focus of work within the factory is mainly on repairing semiconductor equipment after it has failed, but there is no complete mechanism for fault warning of semiconductor equipment, which in turn leads to problems such as reduced available time of semiconductor equipment, low production efficiency, and heavy workload for personnel.

[0023] On this basis, the present invention provides a fault warning method for semiconductor equipment. By performing model training and model testing on a neural network model, in the model testing stage, when the test accuracy of the neural network model reaches a preset threshold, the neural network model can predict whether the semiconductor equipment will fail. The risk of failure of the semiconductor equipment can be reduced by overhauling or adjusting the operating parameters of the semiconductor equipment in advance, thereby ensuring the operating time of the semiconductor equipment and improving production efficiency.

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 The present invention is a flowchart of a method for early warning of a semiconductor device failure according to an embodiment of the present invention, comprising:

[0026] Step S101, providing several groups of historical equipment operation data, each group of the historical equipment operation data including equipment operation environment parameters, and whether the semiconductor equipment fails under the equipment operation environment parameters;

[0027] Step S102, dividing the plurality of groups of historical equipment operation data into a plurality of groups of training equipment operation data and a plurality of groups of testing equipment operation data;

[0028] Step S103, inputting several groups of training device operation data into a neural network model for model training;

[0029] Step S104, after model training, inputting the device operating environment parameters in the several groups of the test device operating data into the trained neural network model, and the neural network model outputs the results of whether several groups of corresponding semiconductor devices will fail;

[0030] Step S105, comparing the results of whether the semiconductor device will fail in several groups corresponding to the output of the neural network model with the results of whether the semiconductor device will fail in several groups of the test device operation data, until the test accuracy of the neural network model reaches a preset threshold, and the model test of the neural network model is completed;

[0031] Step S106, providing operating environment parameters of the semiconductor device to be tested;

[0032] Step S107, inputting the operating environment parameters of the device to be tested into the neural network model that has completed the model test, and using the neural network model to predict whether the semiconductor device will fail.

[0033] The steps of the de-embedding method of the transmission line are described in detail below with reference to the accompanying drawings.

[0034] Figures 2 to 3 It is a schematic diagram of the structure of each step of the fault warning method for semiconductor equipment according to an embodiment of the present invention.

[0035] Please refer to Figure 2 , providing several groups of historical equipment operation data, each group of the historical equipment operation data includes equipment operation environment parameters, and whether the semiconductor equipment fails under the equipment operation environment parameters.

[0036] In this embodiment, the historical equipment operation data is the equipment operation data corresponding to the processed wafers during the preparation process.

[0037] In this embodiment, the several groups of historical device operation data include device operation data of the semiconductor device under normal operation and device operation data of the semiconductor device under fault condition.

[0038] Please continue to refer to Figure 2 , dividing the several groups of historical equipment operation data into several groups of training equipment operation data and several groups of testing equipment operation data.

[0039] The semiconductor equipment includes: etching equipment, growth equipment, deposition equipment and annealing equipment.

[0040] In this embodiment, taking the semiconductor device as a physical vapor deposition device as an example, the corresponding device operating environment parameters include: AC deflection current, DC deflection voltage, electrostatic suction cup clamping voltage, gas flow in the third chamber, heating plate pressure, heating plate temperature, process pressure, power generator current, power generator power, and power generator voltage.

[0041] In this embodiment, the purpose of dividing several groups of the historical device operation data into several groups of training device operation data and several groups of test device operation data is to use several groups of the training device operation data to train the subsequent neural network model. In the model training stage of the neural network model, the conditions and results in each group of the training device operation data need to be simultaneously informed to the neural network model for deep learning of the neural network model. The neural network model after model training is tested by several groups of the test device operation data, that is, the neural network system outputs the results through the input conditions, and the results output by the neural network model are compared with the actual results until the test accuracy of the neural network model reaches the preset threshold. At this time, it is considered that the neural network model has a relatively accurate prediction ability, and it can be put into practical applications later.

[0042] The number of groups of the training device operation data is greater than the number of groups of the test device operation data; the ratio of the number of groups of the training device operation data to the number of groups of the test device operation data is 6:5 to 9:5.

[0043] In this embodiment, taking 500 groups of historical device operation data as an example, 300 groups are divided into the training device operation data, and the remaining 200 groups are divided into the test device operation data. The ratio of the number of groups of the training device operation data to the number of groups of the test device operation data is 3:2.

[0044] Please refer to Figure 3 , inputting several groups of the training device operation data into the neural network model 100 for model training.

[0045] In this embodiment, during the model training stage, the device operating environment parameters in each group of the training device operating data and whether the semiconductor device fails under the device operating environment parameters are simultaneously input into the neural network model 100, and the neural network model 100 learns and establishes a mapping relationship internally.

[0046] In this embodiment, the neural network model 100 adopts a back propagation neural network model 100 (Back Propagation Neural Network, BPNN).

[0047] In this embodiment, the neural network model 100 includes: an input layer 100a, a hidden layer 100b and an output layer 100c, and the number of nodes h of the hidden layer 100b is determined by the number of nodes m of the input layer 100a and the number of nodes m of the output layer 100c, that is:

[0048]

[0049] Among them, the number of nodes m in the input layer 100a is the number of types of the device operating environment parameters; the number of nodes n in the output layer 100c is the number of results output by the neural network model 100.

[0050] In this embodiment, the number of types of the device operating environment parameters is 10, so the number of nodes m of the input layer 100a is 10.

[0051] In this embodiment, the number of results output by the neural network model 100 is only one, that is, the semiconductor device fails or the semiconductor device does not fail, so the number of nodes n of the output layer 100c is 1.

[0052] By substituting the number of nodes m of the input layer 100a as 10 and the number of nodes n of the output layer 100c as 1 into the formula, we get:

[0053]

[0054] After substitution, h is calculated to be 6.07, and h is rounded off to 6, that is, the number of nodes h in the hidden layer 100b is determined to be 6.

[0055] Please continue to refer to Figure 3 After the model is trained, several groups of equipment operating environment parameters in the test equipment operating data are input into the trained neural network model 100, and the neural network model 100 outputs several groups of corresponding results on whether the semiconductor equipment will fail.

[0056] After the model training phase is completed, it is necessary to verify whether the neural network model 100 can make accurate predictions. Therefore, the device operating environment parameters in several groups of the test device operating data are taken as input, and the neural network model 100 performs predictive output through the mapping relationship established in the training phase.

[0057] Please continue to refer to Figure 3, compare the results of several groups of whether the semiconductor device fails corresponding to the output of the neural network model 100 with the results of several groups of whether the semiconductor device fails in the test equipment operation data, until the test accuracy of the neural network model 100 reaches a preset threshold, and the model test of the neural network model 100 is completed.

[0058] In this embodiment, the method for training the neural network model 100 until the test accuracy rate reaches a preset threshold includes: when the test accuracy rate of the neural network model 100 does not reach the preset threshold rate, increasing the number of model training times of the neural network model 100 until the test accuracy rate of the neural network model 100 reaches the preset threshold rate.

[0059] At this point, the model training phase and the model testing phase of the neural network model 100 are completed, and the neural network model 100 will be subsequently applied in an actual process.

[0060] Providing operating environment parameters of the semiconductor device to be tested.

[0061] It should be noted that, in this embodiment, the device operating environment parameters to be tested of the semiconductor device are device operating environment parameters that will change to predicted after a period of time (such as 24 hours) with reference to the current device operating environment parameter change trend of the semiconductor device.

[0062] Please continue to refer to Figure 3 , the operating environment parameters of the device to be tested are input into the neural network model 100 that completes the model test, and the neural network model 100 predicts whether the semiconductor device will fail.

[0063] In this embodiment, by performing model training and model testing on the neural network model 100, in the model testing stage, when the test accuracy of the neural network model 100 reaches a preset threshold, the neural network model 100 can predict whether the semiconductor device will fail, and can reduce the risk of failure of the semiconductor device by overhauling or adjusting the operating parameters of the semiconductor device in advance, thereby ensuring the operating time of the semiconductor device and improving production efficiency.

[0064] Correspondingly, an embodiment of the present invention further provides a storage medium on which computer instructions are stored, wherein the computer instructions execute the steps of the above-mentioned method when they are executed.

[0065] Correspondingly, an embodiment of the present invention further provides a terminal, including a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, and the processor executes the steps of the above-mentioned method when executing the computer instructions.

[0066] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

Claims

1. A semiconductor device fault warning method, characterized in that: include: Providing several groups of historical equipment operation data, each group of the historical equipment operation data includes equipment operation environment parameters, and whether the semiconductor equipment fails under the equipment operation environment parameters; dividing the plurality of groups of historical equipment operation data into a plurality of groups of training equipment operation data and a plurality of groups of testing equipment operation data; Inputting several groups of training equipment operation data into a neural network model for model training; After model training, inputting the device operating environment parameters in several groups of the test device operating data into the trained neural network model, and the neural network model outputs several groups of corresponding results of whether the semiconductor devices will fail; Comparing the results of whether the semiconductor device will fail in several groups corresponding to the outputs of the neural network model with the results of whether the semiconductor device will fail in several groups of the test device operation data until the test accuracy of the neural network model reaches a preset threshold, thereby completing the model test of the neural network model; Providing operating environment parameters of the semiconductor device to be tested; The operating environment parameters of the device to be tested are input into the neural network model that completes the model test, and the neural network model predicts whether the semiconductor device will fail.

2. The semiconductor device failure warning method according to claim 1, characterized in that: The semiconductor equipment includes: etching equipment, growth equipment, deposition equipment and annealing equipment.

3. The semiconductor device failure warning method according to claim 2, characterized in that: When the semiconductor equipment is a physical vapor deposition equipment, the corresponding equipment operating environment parameters include: AC deflection current, DC deflection voltage, electrostatic chuck clamping voltage, gas flow in the third chamber, heating plate pressure, heating plate temperature, process pressure, power generator current, power generator power, and power generator voltage.

4. The semiconductor device failure warning method according to claim 1, wherein: The neural network model adopts a back-propagation neural network model.

5. The semiconductor device failure warning method according to claim 4, characterized in that: The neural network model includes: an input layer, a hidden layer and an output layer. The number of nodes h in the hidden layer is determined by the number of nodes m in the input layer and the number of nodes m in the output layer, that is: h=√0.43mn+0.12n 2 +2.54m+0.77n+0.35+0.51; Among them, the number of nodes m in the input layer is the number of types of the equipment operating environment parameters; the number of nodes n in the output layer is the number of results output by the neural network model.

6. The semiconductor device failure warning method according to claim 1, wherein: The method for training the neural network model until the test accuracy of the neural network model reaches a preset threshold includes: when the test accuracy of the neural network model does not reach the preset threshold, increasing the number of model training times of the neural network model until the test accuracy of the neural network model reaches the preset threshold.

7. The semiconductor device failure warning method according to claim 1, characterized in that: The number of groups of the training device operation data is greater than the number of groups of the test device operation data.

8. The semiconductor device failure warning method according to claim 7, characterized in that: The ratio of the number of groups of the training device operation data to the number of groups of the test device operation data is 6:5 to 9:

5.

9. A storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed, the steps of the method according to any one of claims 1 to 8 are executed.

10. A terminal comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, characterized in that: When the processor runs the computer instructions, the steps of the method according to any one of claims 1 to 8 are performed.

Images