Critical issues in gas delivery for advanced semiconductor processing
By Hubert Dinh, Mohamed Saleem, Ph.D. and Sowmya Krishnan, Ph.D., Ultra Clean Technology
The residual moisture within the gas panel can react with a halogen-based gas, leading to corrosion that can gradually affect the performance of critical components. The ability to predict these types of failures will help minimize the costs associated with scrapped wafers and tool downtime. A graphical user interface integrated within the gas panel can help monitor the health of components and the gas panel as a whole.
Repeatable gas delivery to process chambers is of paramount importance to advanced semiconductor processing. It is equally important to eliminate any potential problems with these systems, such as corrosion in the wetted areas of gas delivery components. As semiconductor processing enters extreme, deep submicron regimes, the requirement for precision gas delivery combined with good repeatability and reliability becomes even more critical, especially for a variety of processes such as etch, atomic layer deposition (ALD) and chemical vapor deposition (CVD). Many gas delivery components, such as mass flow controllers (MFCs), and pneumatic and ALD valves, are pushed to their limits for device geometries extending below 60 nm. It is essential to understand and predict the state of these components prior to running process recipes in the processing tool.
The repeatability of specialty gas flow to the process chamber can be dictated by the health of these critical components. For example, a reaction between halogen gases and residual moisture when present in the gas delivery system can trigger an onset of corrosion in the sensor tubes of the mass flow controllers, which if not monitored can lead to catastrophic failures. Similarly, any failures of the high-speed ALD valves (including corrosion/contamination and failure of the valve diaphragms and seats) can affect the dosage of the expensive ALD precursors leading to erroneous process results, scrapping of wafers and tool downtime.
Mass flow controllers (MFCs) are the most critical components in a gas delivery system, and process repeatability largely depends on the performance of MFCs from process to process. While current flow technology is adequate to meet existing geometries, the next generation devices will require enhanced flow performance.
Despite the recent advances incorporated in thermal-based MFC technology such as digital communication, improved accuracy specification, multigas/multirange capabilities, and pressure-insensitive technology, the sensor tube technology for the MFCs has essentially remained unchanged. Contamination and corrosion-related problems with sensor tubes in the MFCs most often contribute to the repeatability issues.
For example, sensor tubes can contain surface irregularities that can trap moisture on the surface. As a result, some failures are due to a chemical reaction between residual moisture in the gas delivery system and halogen gases such as chlorine, hydrogen chloride, hydrogen bromide, and reactive gases such as silane, dichlorosilane, and others. The sensor zero-drift that occurs over a period of time can gradually affect the sensor linearity and eventually its accuracy and repeatability. It should be noted that the corrosion of sensor tubes is a slow and gradual process and, very often, monitoring the repeatability of flow can help eliminate catastrophic failures and expensive replacement of components in the gas delivery system and the process chamber.
MFC technology has improved in recent years to address sensor zero-drift concerns, but this should still be constantly monitored prior to running processes. The surface defects that initiate corrosion and contamination can also be present on the wetted areas of other components in the gas delivery system, such as pneumatic valves, pressure regulators and filters. Such defects could result from imperfect electropolishing processes. Furthermore, the welded areas and heat-affected zones in tubing and components are especially vulnerable to corrosion.1
Unfortunately, today there is only anecdotal information available in the literature as to the critical moisture threshold below which there will be no, or minimal, reaction with corrosive and reactive gases. This threshold level needs to be characterized, but nonetheless, there are still a couple of ways the risks associated with MFC corrosion can be minimized.
The first method is to improve the surface finish characteristics of all critical wetted surfaces to improve their moisture dry-down characteristics. The second is by revisiting the purge methodology for moisture dry-down for the specific gas delivery systems in question.
The majority of existing purge methodologies for gas delivery systems have been in place for well over a decade and were originally instituted for submicron technology, which could tolerate variances in process gas flow repeatability. By carefully studying the purge efficacy and improving designs such that all dead-legs are eliminated, it is possible to obtain orders-of-magnitude improvements in moisture dry-down performance.
For some of the newer semiconductor processes, such as ALD used in high-k dielectric gates, valves with fast response are required to pulse the precursors to the wafer during the deposition processes. These valves typically see 5 million to 10 million cycles per year and are operated at high temperatures, typically exceeding 175°C. It is, therefore, imperative that they have high life cycles, zero particle shedding and high reliability.
Manufacturers have designed valves with special diaphragms and seat materials with life cycles greater than 25 million cycles.2 However, all of the life cycle data is based on operation in inert gases such as nitrogen. Many of the precursors used in ALD processes, on the other hand, are highly reactive and require high temperatures to maintain vapor phase. It is not known how these valves will perform under these specific process conditions. As ALD processes become mainstream, and more and more new precursors are identified that would impart specific properties to the gate dielectric, understanding the reliability of the components used for pulsing the precursors will become critical.
Advanced monitoring of gas panel health
Since the inception of e-diagnostics in early 2000, multiple teams within the semiconductor industry, including chip makers and OEMs alike, have been actively engaged in e-diagnostic activities to support the industrywide implementation of advanced process control (APC), and fault detection and classification (FDC) techniques. APC refers to passive and active processes, and wafer data mining and tool-to-tool process adjustments. FDC is a technique focused on detecting real-time tool and process deviations and isolating the root cause.3 The implementation of these techniques has historically faced several challenges such as the lack of a consensus strategy for integration of various systems from different suppliers. This leads either to poor-quality parametric data transfer or to unavailability of critical data when requested.4
Today, with the availability of more and more in-situ metrology capabilities and FDC on their tools, OEMs have begun to integrate the resultant data streams into a centralized fab data system. However, large amounts of this sensor and tool data are either unsynchronized or not applicable for use in APC protocols.5 Even though there are rich pools of sensor and tool data available, maintenance alerts are still most often provided only after a failure has occurred. The time spent on trial-and-error design of experiments isolating the root cause may take weeks, forcing extended tool downtime, the cost of which can typically be in excess of $100,000 per hour.6 It is, therefore, imperative that any process-critical parameters deviating from specifications be detected in advance at the gas delivery system level. This will allow for the rapid implementation of corrective action.
The techniques used in APC and FDC can be applied to monitor the state of health of a gas delivery system. This can be accomplished through the use of a graphical user interface (GUI) tool integrated with the gas panel. The GUI can be capable of monitoring the entire gas system with the touch of single key (see Fig. 1). In principle, this software may also work across industry-common communication protocols via proprietary algorithms. Using such a tool, the user can quickly scan the performance of a gas panel by accessing digital information from various components. By processing the information, an assessment of the health of the components can be made for the entire gas panel. This ability to monitor, analyze and diagnose real-time transient values, gives users the opportunity to troubleshoot flow excursions prior to running processes.
Process gas flow repeatability is a highly critical element in the deep submicron regimes of advanced semiconductor processing. Typical repeatability problems stem from corrosion or contamination issues in critical wetted surfaces, resulting in performance degradation of active components such as MFCs. For example, MFC sensor drifts can result from contamination build-up or corrosion occurring gradually over time. Process repeatability issues can also be caused by imperfections in the wetted areas of other critical components such as pneumatic valves, pressure regulators and filters that may trigger an onset of corrosion. Some corrosion-related problems can be addressed by revisiting and fine-tuning existing purge methodologies to improve the moisture dry-down characteristics. In addition, it is important to predict problems related to gas flow using techniques established for APC and FDC. The ability to predict a problem with a gas delivery component will help minimize the costs associated with scrapped wafers and tool downtime. A graphical user interface integrated at the gas-panel level will help predict the health of critical gas delivery components and gas panel as a whole.
Hubert Dinh is a technology development engineer at Ultra Clean Technology (UCT; Menlo Park, Calif.).
Mohamed Saleem, Ph.D., is director of technology at UCT.
Sowmya Krishnan, Ph.D., is vice president and chief technology officer at UCT. Dr. Krishnan is also a member of the CleanRooms Editorial Advisory Board.
1. Glime, William and Tom Seidel. “Fast-Switching Valves for High Productivity ALD,” Semiconductor International, September 2005.
2. Saleem, M., S. Krishnan, D. Beery, and A. Kabansky. “Examining Corrosion Resistance of Chromium-passivated Stainless Steel,” Micro, 2, 43-51 (1998).
3. Wohlwend, Harvey. “An E-factory Vision: Building On e-Diagnostics,” Future Fab Intl. Volume 11, June 2001.
4. Scanlan, John. “Knowledge-Based Process Control for Fault Detection and Classification,” Semiconductor Manufacturing Magazine, October 2003.
5. Smith, John A., K.C. Lin, Matt Richter and Uzi LevAmi. “Practical, Real-Time Multivariate FDC,” Semiconductor International, December 2004.
6. Kot, Vincent, Manjunath Yedatore. “The Next Step in E-diagnostics: Mining the Tool Sensors,” Semiconductor International, October 2003.