JP5206021B2 - Manufacturing method of semiconductor device - Google Patents

Description

  The present invention relates to a method for manufacturing a semiconductor device, and more particularly to a method for manufacturing a semiconductor device that determines a process condition of a machining process based on trend data in a multi-product production line.

  In a multi-product production line, in addition to restraining manufacturing costs by improving yield and productivity, it is important to provide products to the market without delay in accordance with customer requirements. For this purpose, it is required to improve the processing accuracy and the apparatus operating rate.

  In a production line with fine processing typified by semiconductor device manufacturing, the processing shape obtained may vary even when processing is performed under the same processing conditions due to changes in the processing device status and the processing results of the previous process. is there. It is necessary to quickly find the optimum processing conditions for each type of product to be produced.

  In order to improve the processing accuracy, the introduction of APC (advanced process control) technology that optimizes the processing conditions of the target process based on the processing results of the previous process and the subsequent process is being promoted.

  Japanese Patent Laid-Open No. 2007-220842 uses a polynomial model including a second-order or higher-order term of a pre-processing state quantity when the number of samples for parameter estimation is insufficient in a wafer polishing process. It is proposed to estimate the parameters by the least square method so that the error is minimized.

  There is an increasing demand for optimization of the maintenance period and reduction of test processing other than product processing in order to improve the device operation rate. For this reason, an FDC (fault detection and classification) technique for monitoring a device state in detail and predicting the occurrence of a failure in advance has been emphasized as a management technique in a production line in recent years.

  Japanese Patent Laid-Open No. 2006-146459 points out that there is a low frequency component and a high frequency component in the change of log information such as physical parameters of the manufacturing apparatus, obtains a normal distribution of the high frequency component excluding the low frequency component, We propose to set upper and lower limits based on the standard deviation and detect abnormal values that exceed the upper and lower limits.

  In APC and FDC, it is generally performed to record single or plural kinds of data for process data representing an apparatus state and a processing result, and monitor trend data indicating the change. The flow of fluctuation of the apparatus state and the processing result is grasped as a process drift, and the process data for the next processing is predicted based on the tendency of the process drift. One form of APC is to optimize the processing conditions of the next lot using the predicted values obtained. In the FDC, it is possible to avoid a failure in advance by comparing and determining process data obtained with respect to a preset management value and using a predicted value of process drift.

  The process data may change due to factors other than the apparatus state and the processing result. The main cause is varietal differences. If the difference between varieties is ignored, it may be determined that there is an abnormality even in the normal apparatus state or processing result. For example, in the etching process, the etching characteristics such as the etching rate change depending on the resist opening ratio. In the polishing process, the difference in the occupied area ratio of different materials in the wafer affects the polishing rate. Such a difference between varieties greatly affects the management of process data in a multi-product production line.

  In order to monitor the process drift with high accuracy, it is desirable that the process data is recorded continuously without being lost depending on the product type. If there is a period in which process data is missing, if process drift occurs during that period, the trend data will change abruptly, and prediction of process data cannot follow the change. However, in a multi-product production line, it is practically impossible to obtain process data for each product continuously.

  FIG. 7 shows an example of a trend chart of process data obtained when two types of products A and B are processed under the same processing conditions. The horizontal axis indicates the date, and the vertical axis indicates the process data. The value of the process data for the product type B is clearly smaller than the value of the process data for the product type A. As described above, it is not desirable to use the process data including the difference between types as it is. Process data may be managed separately for each product type, but in this case, a period in which process data is missing may occur for each product type. It is desirable to remove the effects of differences between varieties in some way.

  Japanese Patent Application Laid-Open No. 2002-202806 proposes that the relationship between processing parameters such as a film formation time and a film formation rate and a characteristic value such as a film thickness obtained as a function is collectively managed.

  Japanese Patent Laid-Open No. 2005-347585 obtains a relational expression between an etching time and a groove depth in trench etching, acquires past groove depth measurement information and etching time information, an offset variation amount due to the type from the acquired information, It is proposed to obtain the time series fluctuation amount due to the change with time, calculate the etching time to correct these fluctuations, perform the etching process, and collect the etching time data and groove depth measurement data after the process.

JP 2007-220842 A JP 2006-146458 A Japanese Patent Laid-Open No. 2002-202806 JP 2005-347585 A

  In a multi-product production line, there is no established method for managing process data, creating trend data, and reflecting it in the next processing.

According to one aspect of an embodiment of the present invention,
In the production line of multi-variety semiconductor devices, create trend data of standard products and difference data between products,
If the product type to be produced is a standard product type, the trend data is used as it is; if the product type to be produced is a product other than the standard product type, the trend data is corrected using the inter-product type difference data. To obtain the predicted value of processing,
Determine the optimal processing conditions based on the predicted values,
Processing is performed under the processing conditions obtained,
Process data is obtained by measurement after processing,
If the produced product is a standard product, the process data is used as it is. If the produced product is a product other than the standard product, the process data is corrected for the product difference, and the product is used as a standard product. Create trend data corresponding to
The processing schedule of each type is arranged in time series, and the change of the standard type is determined according to the processing amount of the produced type .
A method for manufacturing a semiconductor device is provided.

  Continuous trend data regarding the standard product is obtained, and the standard product is changed at an appropriate timing.

  By setting a function for a plurality of varieties and an offset fluctuation amount based on a difference between varieties, it becomes possible to share trend data among a plurality of varieties. However, the varietal differences themselves can also vary. The amount of offset variation based on the function itself targeting multiple varieties and the difference between varieties changes.

  The present inventor considered a method of determining a reference product type when correcting a difference between product types in the process data and using the product type difference with respect to the reference product type for other types. It is not necessary to correct the difference between products for the standard product. In the production of the standard product, it can be said that the trend data is more sensitive to the apparatus status and the processing result than the other product that corrects the difference between products. As a standard product type, it is desirable to select a product type in which many lots are regularly processed.

  FIG. 1 shows a flowchart of creation and management of trend data according to an embodiment of the present invention. In step S1, process data for each target product type is acquired, and in step S2, an average value of the process data is obtained for each product type. Assume that the product type A is set as the reference product type. After obtaining the process data, the reference product type may be selected as product type A.

  For example, as shown in FIG. 2A, using the same processing apparatus, product type A, product type B, product type A, product type B, product type C are processed in time series, and the respective process data are measured. These process data are acquired, and average values Da, Db, and Dc are calculated for the products A, B, and C, respectively.

  In step S3, a correction value between product types is calculated. By subtracting the process data value Db of the product B from the process data value Da of the reference product A, the difference BA of the product A with respect to the product B is calculated. Similarly, Da−Dc = CA is obtained. In addition, if it is limited to a short time, the difference BC between the varieties C with respect to the varieties B is BC = BA−BC. When the equation is modified, Db + BA = Da, Dc + CA = Da, and when the inter-product differences BA and CA are added to the process data of the products B and C, it can be converted into process data corresponding to the product A. In step S4, the process data of the product other than the standard product is corrected (converted) to process data corresponding to the standard product. In step S5, process data corresponding to the reference product is collected to create trend data.

  FIG. 2B shows a state in which trend data is created by correcting the process data of the types B and C to process data corresponding to the type A in this way. Note that the process data of the standard type does not include correction of the difference between types, and the reliability is high accordingly. The process data of a product having a large correction value for the difference between products is less reliable. This difference in reliability can be expressed by weighting process data for each product type. If the weight of the process data of the standard type is 1.0, the weight of the process data of other types is less than 1.0.

  In the multi-product production line, new varieties are added according to the demands of the market, and usually a cycle of increasing, decreasing, and disappearing is performed. Depending on the time, the frequency of processing for each variety may also vary. It is not preferable to fix the standard variety for a long time. It is desirable to appropriately change the standard product type according to the processing amount of each product type.

  In step S10, the reference type is changed as necessary. Along with the change of the reference product type, the trend data is converted into the product type that has become the new reference product type in step S20.

  FIG. 3A is a flowchart showing details of the reference product change process. In step S11, order status and / or sales plan information for each product type is acquired. In step S12, the processing schedule for each product type is arranged in time series.

  FIG. 3B shows an example of a graph in which the number of lots scheduled for production is arranged in time series. The varieties A, B, A, B, and C are sequentially arranged in the production schedule. In practice, multiple varieties often coexist when switching.

  In step S13, the processing result of the previous process is referred to. It is possible to consider a lot that has been canceled after receiving an order or has been discontinued due to an abnormality in the machining process. It is also possible to make adjustments between different varieties with different numbers of processes and orders from delivery to delivery. It is also possible to estimate the arrival time for each product type by using dispatch information for managing the lot processing schedule. However, step S13 can be bypassed.

  In step S14, it is determined whether or not the reference product should be changed based on steps S12 and S13. A plurality of determination criteria can be prepared.

  The case where the standard type is A and the type to be produced is changed from A to B is taken as an example. Even if the production of the variety B is started, it is too early to immediately change the standard variety to B. The production record of the kind A is stored, and the kind A as the standard kind is highly reliable. There is little or no processing result of the type B, and the reliability of the type B as the standard type is still low. The time for changing the reference varieties from A to B is when the production amount of the varieties B will increase in the future, and the production results of the varieties B can be achieved to some extent. You may consider the past results and the stability of the machining process.

  FIG. 4 shows an example of determination of change of a reference product type in a plurality of determination criteria. The horizontal axis indicates the date, and the vertical axis indicates the process data. The top column shows the case where the largest number of trend data to be collected is set. Since the production amount of product A increases again before changing the standard product to B, the change of the standard product is not necessary. I will not. This is applied when the process is stable and the reliability of the processing process of the product type A is high. The bottom column shows the standard varieties when the production of the varieties B has progressed to some extent, but the production of the varieties A is still scheduled, but there are more production varieties B than the production plans of the varieties A. Is changed from A to B. It is applied when the stability of the processing process is still insufficient and it is important to focus on varieties for which a larger number of production is planned. The middle two columns are examples in the middle of these two cases. After the production of the product A is finished and only the product B is produced, the production result of the product B is improved to some extent, and then the reference product A is selected. Change from B to B.

  Returning to FIG. 3A, after it is determined in step S14 that the reference product has been changed, the reference product is changed in step S15. Along with the change of the reference product type, the correction value for the difference between product types is changed to a value corresponding to the product type that has newly become the reference product type in step S16, and the weighting is also changed as necessary. In step S20, the trend data is updated to a new target product. Thereafter, in step S30, trend data is created together with the results of the machining process.

  FIG. 5 is a block diagram showing a system configuration for manufacturing a semiconductor device according to an embodiment of the present invention. The plurality of apparatuses 10 are apparatuses such as film forming, polishing, exposure, etching, and ion implantation, and include a processing unit and a measurement unit. The computer 30 supplies the processing conditions of the processing process to each device 10 via the database 20, and each device performs processing processing under the given processing conditions, and processing data such as process data measured after the processing is stored in the database 20. To the computer 30.

  FIG. 6 is a flowchart of a semiconductor device manufacturing process using the system shown in FIG. In step S101, the value of the next processing is predicted from the trend data so far. This predicted value is for the standard variety. In step S102, the type of the target lot is determined. In step S103, the predicted value is converted according to the target product type using the difference between product types. In step S106, an optimum processing condition is calculated based on the predicted value according to the target product type. If necessary, the correlation between the process data and the processing result can be referred to in step S104 before step S106. For example, the etching time is determined by using the etching rate as trend data and referring to the correlation between the etching time and the etching depth. Moreover, step S105 etc. which refer the measured value of the previous process of an object lot as needed can also be performed.

  In the process of obtaining the optimum processing conditions, it is possible to determine an abnormality in trend data, an abnormality in the actual value of the previous process, and the like. If the calculated processing condition can be processed, the processing condition is transmitted to the processing apparatus in step S107, the processing process is performed in step S108, the result of the processing process is measured, and the processing data is transmitted in step S109. To do. If the calculated processing condition cannot be processed, an alarm is transmitted in step S110. Also in this case, the accuracy of abnormality determination can be improved by removing the influence of the product type from the trend data. In this way, the processing process proceeds.

  Although the present invention has been described with reference to the embodiments, the present invention is not limited thereto. It will be apparent to those skilled in the art that various modifications, substitutions, combinations, improvements, and the like can be made.

FIG. 1 is a flowchart of trend data creation management according to an embodiment of the present invention. FIG. 2A is a graph showing the process data of the product under processing by one processing device with respect to the date, and FIG. 2B is a process data corresponding to the product A by correcting the process data of the product other than the standard product, It is a graph which shows the state which produced data. FIG. 3A is a flowchart showing details of the standard product change process, and FIG. 3B is a graph in which the number of lots scheduled for production is arranged in time series. FIG. 4 is a graph showing an example of determining a change of a reference product type in a plurality of determination criteria. FIG. 5 is a block diagram showing a system configuration for manufacturing a semiconductor device according to an embodiment of the present invention. FIG. 6 is a flowchart of a semiconductor device manufacturing process using the system shown in FIG. FIG. 7 is a graph showing an example of a trend chart of process data obtained when two types of products are processed under the same processing conditions.

Explanation of symbols

S step,
D Average value of process data,
10 Equipment (working part, measuring part),
20 database,
30 Calculator.

Claims (5)

In the production line of multi-variety semiconductor devices, create trend data of standard products and difference data between products,
If the product type to be produced is a standard product type, the trend data is used as it is. If the product type to be produced is a product other than the standard product type, the trend data is corrected using the inter-product type difference data. To obtain the predicted value of processing,
Determine the optimal processing conditions based on the predicted values,
Processing is performed under the processing conditions obtained,
Process data is obtained by measurement after processing,
If the produced product is a standard product, the process data is used as it is. If the produced product is a product other than the standard product, the process data is corrected for the product difference, and the product is used as a standard product. Create trend data corresponding to
The processing schedule of each type is arranged in time series, and the change of the standard type is determined according to the processing amount of the produced type .
A method for manufacturing a semiconductor device. The method for manufacturing a semiconductor device according to claim 1, wherein the determination of the change of the reference product type is performed using at least one of an order reception status and a sales plan. The method for manufacturing a semiconductor device according to claim 2, wherein the determination of the change of the standard product type is performed with reference to the processing results of the process to be managed. In the determination of the change of the standard varieties, if the number of scheduled processing is large and the production already started is determined as the standard varieties, and if it is different from the current standard varieties, the standard varieties The method of manufacturing a semiconductor device according to claim 1, wherein trend data is updated by changing the data.   5. The method according to claim 1, wherein when determining an optimum processing condition based on the predicted value, at least one of a correlation between process data and a processing result and a measured value of a previous process of a target lot is referred to. A method for manufacturing a semiconductor device.

Images