Semiconductor FDCSemiconductor FDC
With the Maestria FDC solution, the fault detection and classification functions are referred to as global process control (GPC). GPC detects anomalies using Hotelling's T2 statistic, which represents the degree of divergence from a cluster median, taking into account the cross correlation between variables. It also employs unique algorithms to classify an anomaly.The following illustrates the GPC modeling process.
In order to be able to implement each step in the analytical procedure, it is essential to possess familiarity with and information on the process, manufacturing recipe, and equipment behavior. It should be pointed out that a very important and difficult problem concerns what process variables should be used to generate a model when there are 30~50 to choose from.Moreover, in creating a model it is important to retrace detailed process behavior, going over the SPC charts and time-series charts for each variable. Familiarity with the process and equipment is absolutely essential if one is to achieve a proper interpretation of the behavior.
After using this analysis to obtain a general idea of the cause of the failure, various types of experimentation – such as changing the recipe or intentionally adjusting the settings – can be employed to produce yet more detailed information that will help to pinpoint the root cause.
Once the cause of a failure is determined, one will look for solutions to the problem, taking such actions as upgrading the equipment or modifying the process recipe. Even in cases where a full solution cannot be found, by executing the fault identification model online, one can hope for useful results such as early detection of abnormal wafers, preventing them from proceeding further into the manufacturing process and thus reducing scrap as well as equipment downtime.
Thanks to these analytical processes, Maestria makes it possible to gain an understanding of process and equipment behavior; we can thus strive to continuously improve operational stability. This has made it possible in recent years to build stable manufacturing systems that support small-lot production and shorter process lifecycles. However, for a device manufacturer to effectively employ such an FDC solution would require the deployment of a large number of trained engineers to look after all of the production lines and manufacturing equipment. When one considers such problems as the cost of hiring and training such specialized staff, it must be said that it would be difficult for all device manufacturers to do this unaided.
Through Yamatake's partnership with Si Automation, we are gathering information on equipment and processes being used in Japan and overseas. We also offer a variety of statistical analytical packages, including Yamatake's own modeling technologies. Drawing on these, we are developing an “analysis service”: when commissioned by a device manufacturer, we can perform data analysis and work with the customer to upgrade both processes and equipment.
At Yamatake, in order to refine this analysis service we are engaged in (1) conducting technical R&D aimed at developing more effective analytical algorithms, as well as realizing an efficient online data analysis environment; (2) developing ways in which to apply technologies that we already possess for use in other fields to semiconductor manufacturing; and (3) accumulating and sharing expertise related to equipment and processes – in other words, focusing on human resources and the knowledge possessed by experienced engineers. To illustrate this, let us look at an example of (2).
In order to meet the required specifications, when setting the conditions for a semiconductor process one inevitably has to make a trade-off between various parameters. It is true that there are process engineers working for device manufacturers who actually take an empirical approach to this task, experimenting with different parameters before deciding on the optimum conditions. But it is not unusual for such settings to be based on an engineer's intuition and experience.
RSM-S, a unique multipurpose optimization technology developed by Yamatake, facilitates optimum condition setting using statistical models based on small-set experimental data. It is expected that this can play a useful role in improving semiconductor-manufacturing recipes and reducing the time taken to get new production equipment up and running.
<<Measurement data processing
