Process control through integrated metrology
01/01/1999
In situ sensors
Process control through integrated metrology
Stephani Watts Butler , Texas Instruments, Dallas, Texas
In the past few years, several semiconductor manufacturers have attempted to use in situ sensors to improve productivity and control. In addition, several companies have formed to meet the needs of the industry by providing unique sensor-based solutions and applications assistance. The applications have varied in their level of success. An examination reveals that four elements are required for the development of a sensor-based controller:
meeting of required business factors,
development of all needed components,
integration of components, and
form that meets goal.
The term "integrated metrology" is proposed to signify the integration of all these components into a particular form that meets a business need. An example of a required business factor is an identified business need with a feasible and profitable solution. Having a strong industry-wide business need, such as the drive to shrink the die size, provides the critical mass of effort required to spur development of needed components. Table 1 lists currently available controllers and their corresponding drivers.
 |
Another driver for sensor-based control is the expense of traditional maintenance schedules for equipment. Without in situ metrology to reflect the true state of the tool, expensive maintenance procedures are often overspecified to ensure confidence in tool status. In addition, the wafer state itself may alter the process: if the equipment processes more than one device or more than one step in the flow (even with the same recipe) then the equipment may function differently. As control needs increase and market opportunities emerge, process equipment vendors will increase the level of sensor-based control offered on equipment.
There are several types of measurement tools. To contrast sensors with stand-alone metrology, we can define where the metrology is physically located and what it may measure, as opposed to what categories of data are taken (Table 2).
 |
Decision-making
To make a processing decision using in situ sensors, raw data must first be converted into information, and this information can then influence an action (see figure). Examples of data are an optical emission spectrum from an etcher, or the lamp power trace from a rapid thermal processor. Examples of information are "the etch rate is 2 ?/sec less than desired," or "the temperature is unusually high." Examples of actions are "increase the time by 2 sec," or "fix the power supply to lamp number 2."
It is critical that software be set up to suggest actions. If the sensor-based controller simply signals a process failure without indicating likely corrective action, then the engineer or operator is likely to view the controller as an enemy. Frequent alarms with little assistance increase the complexity of the engineer`s or operator`s job. The final result will be that people will "fix" their problems by turning off the controller.
Due to high data volumes, extracting useful information is challenging. In addition to sheer volume, sensor data may be confounded by the aging of sensor components such as viewport windows, or influenced by factors unrelated to the process. The result is difficulty in creating high-quality information that can direct action.
Methods. These include the mathematics, models, and techniques used. Models are needed for both the steady-state behavior and the aging of the process. In addition to models, algorithms must be generated to create information from the data and to suggest action. Algorithms must not generate too many false positive (Type I error) and false negative (Type II error) messages, however. If the false positive rate is too high, then manufacturing personnel will become frustrated trying to fix what is not broken and will turn off the controller. Conversely, if the false negative rate is too high, then yield loss will go uncorrected and the controller will be considered useless. While the false positives are the most visible, it is the long-term false negative rate that determines the improvement that the controller will achieve. Thus, the right method, and the tuning of that method, to achieve acceptable Type I and Type II errors is a very important component of any control system.
The 3Rs of sensors. Other important aspects are the repeatability, reproducibility, and resolution of sensors. As reproducibility and repeatability become worse, the controller must be detuned to prevent sensor noise from influencing decisions. A detuned controller is less able to recognize and respond to small changes, and is slower to respond to big changes. Resolution is the smallest difference that the measurement device can detect. The new Semiconductor Industry Association (SIA) National Technology Roadmap for Semiconductors stresses the emerging importance of resolution in determining how tightly a system can be controlled. As device dimensions continue to shrink, measurement tools must have not only less noise, but better resolution.
 |
Steps in decision-making
Integrated metrology forms
To find correlation among different data sources and to identify root causes of faults, events and data sources must be time-aligned to one another. To achieve this alignment, a single clock must be used for every piece of equipment and data collection in the fab. Wafer positional tracking is also needed, for both batch and single-wafer cluster tools.
Due to the lack of randomization during processing, wafers are likely to remain in the same position through many cluster tools, thereby confounding the identification of the machine or position that caused faults. For example, low-yielding wafers often occur in the same run position across many lots on several different machines. Tracked randomization, either within the tool itself or when placing wafers in the boat, allows a low-yielding wafer to be more easily associated with a single position in one machine.
To send all gathered data to the manufacturing execution system (MES) for analysis would swamp both data connection lines and the MES host computer. Thus, data must be analyzed at the equipment and only summaries sent to the MES. More data should be sent if there is a failure, and less data if the process appears to be normal. Similarly, data collection should be event- and test-driven, so that more data is collected during production ramps or after a test failure.
One of the basic performance issues with current equipment is the lack of self-checking during idle time. For example, vacuum problems on Chamber C of a cluster tool could be detected by trying to pull a vacuum when the machine has no wafer queue. Today, however, a vacuum problem would be discovered only when the first wafer arrived at Chamber C (often after processing through A and B). Merely performing an "are you there?" electronic ping of an idle tool would detect many hard fails.
Though it is easiest for the process equipment vendors to do the integration of sensors, space (and possibly viewports) must be present. To simplify chamber designs, however, viewports have decreased in size and quantity. Retrofit kits need to be commercially available for the physical adding of sensors. For example, kits are available from CMP vendors for adding on-line thickness sensors, and from etch vendors for adding full-wafer interferometers.
The most important requirement is that data from all sources (from the many different process tools to the MES) be available to the controllers. A control system will still exist at the MES level, but it will focus more on module control and work in synergy with the distributed equipment control systems by specifying requirements and targets. The MES control system will transform from a "wrapper" on equipment that is unaware of the wrapper to an integrated system working in harmony with the many equipment control systems. Ultimately, Integrated Metrology will result in equipment that can monitor changing processing capability, adjust to achieve desired results, and perform diagnosis and prognosis to identify possible causes of failures.
STEPHANIE WATTS BUTLER is the manager of the Advanced Technology Branch in Silicon Technology Development at Texas Instruments. Previously, she managed the Advanced Process Control and Metrology Branch. Texas Instruments, PO Box 650311/MS 3701, Dallas, TX 75265, e-mail [email protected].