Emerging Requirements in Real-Time Data Analysis and Management for Improved Process Control
Arun Srivatsa and Scott Bushman
Two factors have dominated recent technology transitions for both logic and memory devices and increased the challenges for process control.
The first is the introduction of several new materials and more complex multilayer structures. For example, with the introduction of high-k metal gate in logic, the gate film-stack has gone from a single layer of poly around 100nm in thickness to about 5–6 layers of different metal films, most of which are only a few nanometers thick. Also, memory has seen a tremendous multiplication in the number of layers required to produce the device, most recently with V-NAND devices.
The second factor is the transition from planar structures to three-dimensional devices. While memory devices first saw increasing 3D complexity, logic devices have now joined the “party” with the introduction of FinFETs.
These changes in technology bring about significant process-control challenges for advanced manufacturing. The challenges include, for example, precise Angstrom level (in some cases sub-angstrom) process control for the deposition of most layers in the high-k metal gate film-stack. With logic FinFETs, one also has to also contend with sidewall conformality and uniformity of these thin layers. Ultrashallow dopant control on fin sidewalls is another challenge. Similarly, there are many increased challenges in process control for memory devices.
This increased complexity requires intelligent control systems that can maintain necessary shape morphology in addition to controlling thickness. To enable this level of control, the modern cluster tool has a plethora of sensors. A typical deposition chamber, for instance, may have 200–300 sensors to monitor gas flow rates, pressure, temperature, etc., at predetermined sampling frequencies.
Depending on the sensor and parameter being measured, the sampling frequency could range from a fraction of a hertz to several hertz. In accumulation, a tremendous of amount of data is being generated by these sensors at high rates. It is estimated that a typical 300mm cluster tool for metal gate deposition (as shown in figure 1) outputs data at a rate of nearly 3 million terabytes every year. This is roughly around 3 million bytes of data per minute.
Figure 1. Schematic of an Applied Materials Endura integrated metal gate stack PVD system.
The data being generated at these high rates must be analyzed in real time to enable the intelligent process control required for high yields in high-volume manufacturing. In turn, this increases requirements for both real-time computation and data storage. These large volumes of data also need to be archived for review and analysis. Clearly, modern high-volume manufacturing places increased demands on both real-time data analysis and data archival systems.
Traditional equipment engineering system (EES) data collection has been <20 terabytes, but because of increased data sampling rates, the number of sensors, and the length of data retention, this number is expected to grow 10-fold, such that a facility EES system would be 200 terabytes. Additionally, the kinds of data that customers want to use in analytics is not only sensor and statistical data, but data in the form of images, categorical data related to yield characteristics, and metrology data such as die-level measurements.
There is growing interest among semiconductor manufacturers for distributed data storage solutions that can handle transactional and non-transactional (unstructured) data, especially for EES solutions like Applied E3. As a result, Applied Materials is investing in solutions that include distributed and federated databases for use in offline data storage.
In addition to storing large volumes of data, an expedient method of retrieving and using this data for advanced process control is critical to the development of new applications built around data analysis. These applications, along with innovative methodologies, are critical to lowering the cost of manufacturing and improving factory yield as device geometry and materials increase in complexity.