Advanced Process Control will become an integral part of next-generation, factory-wide planning as chip makers look to sharpen overall contamination control
By James Moyne
As the semiconductor manufacturing industry becomes increasingly competitive under the current economic conditions, chip manufacturers continue to look for new ways to maintain a sharp, competitive edge. The industry is rapidly reaching consensus on the fact that advanced process control (APC), implemented factory-wide, is becoming a required, integral part of current and next-generation semiconductor facilities.
SEMI defines APC as: “The manufacturing discipline for applying control strategies and/or employing analysis and computation mechanisms to recommend optimized machine settings and detect faults and determine their cause.”1 Simply put, APC involves observing the processes and the wafers, and then allows the ability to automatically make changes to the processes to improve them.
The APC concept evolved from a method called Statistical Process Control (SPC) in which process or product data, such as metrology readings, are collected and analyzed statistically to see if the process is deviating from statistical norms. This approach, also called 6-Sigma analysis, is still widely used in many industries; however, it is somewhat limiting because it can only detect a small class of process deviations, and generally doesn't tell you what caused the deviations or how to correct for them.
These deficiencies are precisely what APC addresses.
APC's main components
The two major APC components are run-to-run (R2R) control and Fault Detection and Classification (FDC). R2R control, which includes wafer-to-wafer and batch-to-batch control, is a type of discrete process control in which the product recipe is modified between “runs” to minimize process drift, shift and variability.2
An example of R2R control would be adjusting the process time on your CMP tool before each wafer is processed to better hit your removal target. In minimizing process drift, shift and variability, process capability is increased and product scrap is reduced. More importantly, product specifications (such as CD) can be tightened, leading to higher performance devices.
Fault Detection and Classification (FDC) is a form of tool, process and/or product health monitoring and diagnosis, where large amounts of data from the process and wafer are analyzed—usually over the course of a process or process step, and compared against established limits to determine if a process or tool “fault” has occurred and what might have caused it. By detecting these faults, fewer wafers are scrapped and, by determining the cause of the faults, the tool mean-time-to-repair is reduced.
Improving process capability
R2R control first appeared in the early 1990s when Sema-tech-funded project results illustrated that thickness of wafers in a CMP process could be controlled by modeling the process and then making adjustments at each run.
Today, R2R control is used throughout the factory in such process areas as CMP, lithography, etch, CVD, diffusion and EPI. Additionally, these solutions have been extended to support “interprocess control,” or control across multiple process types.
The R2R control solution can be broken down into three levels—processing, measurement and control. At the processing level, just about all front-end processes are candidates for R2R control. The control is often dictated by the measurement capability. Traditionally, univariate control of parameters, such as thickness (THK) or CD, is attempted first; this is often expanded to include uniformity (UNIF) control, resulting in a multivariate solution. At the control layer, standard R2R control solutions are process-centric and can utilize feedforward (FF) and feedback (FB) measurement data. These can be augmented with interprocess control solutions, such as CD control between lithography and etch. At the highest level, factory-level control systems can tie control to factory-wide metrics, such as yield, throughput and device characteristics.
Figure 1 (above) provides an illustration of a typical R2R control implementation in a factory. Nearly all R2R controllers operate by maintaining dynamic or adjustable models of the processes they are controlling. Initial models are usually developed by running a set of experiments, called a Design of Experiments (DOE), to determine which parameters on the tool should be adjusted to correct for process drift and shift.
The DOE process can be costly, thus R2R control suppliers often maintain model libraries for common process types. Once the models are developed, they must be continuously tuned to account for changing process conditions. This requires some form of downstream (post) metrology to evaluate the controller performance. The controllers can also use incoming (pre) metrology to compensate for incoming wafer variations.2
For R2R control to be effective as an automated solution in the factory, it must be able to operate in an environment where process and integration conditions are far from ideal. This has been the major obstacle to the rapid acceptance of R2R control. Over the past few years, however, R2R control suppliers have provided a number of solution capabilities that have overcome this obstacle.
These capabilities, which should be required of all R2R control solutions, include:
- Accommodating missing, delayed, out-of-order, or “bad” metrology data;
- Providing for settable limits on R2R control adjustments;
- Providing model adjustments to accommodate PM events at a tool;
- Providing model adjustments to accommodate product changes and automatically calculating models for newly introduced products;
- Providing mechanisms to evaluate the health and effectiveness of the R2R controller;
- Providing interface specifications for integration of R2R control at a tool or factory-wide level.
With respect to the last point, integration at the factory level is generally preferred because it provides a better environment for combining control elements and tying control to factory objectives, such as CD, throughput and yield.
With R2R control, CMP process variability and cycle time are reduced, and process accuracy is improved.
An example of what can be achieved with R2R control is shown in Figure 2 (above), which illustrates that with R2R control, CMP process variability and cycle time are reduced, and process accuracy is improved.3
In its simplest form, Fault Detection and Classification (FDC) can consist of monitoring various sensors on a tool, detecting out-of-norm conditions for the sensor values, and relating these conditions to problems with the tool. But there are a large number of enhancements to this simple form that are usually delivered with FDC systems today.
Monitoring variations via FDC
Enhancements include multivariate analysis techniques, where the analysis incorporates relative contributions from all variables and accommodates correlations between these variables. Also, “Golden Run” analysis can be applied, where the setup of the FDC system is tied to the pattern of a good “golden” run or set of runs.
Clicking on a data point in the bull's-eye view lets the user investigate details of the fault, plot data history and overlay data.
Interestingly enough, improvements in FDC analysis that have occurred over the past five years have, in many ways, hindered its adoption. Today's FDC solutions are often too complex and too time-consuming to be understood or maintained by process engineers; thus, an important capability that should be considered when implementing any FDC solution is ease-of-use. For example, as shown in Figure 3 (page 18), Web-based front-ends for FDC solutions can provide a snapshot of the tool health with a single plot or chart, but also support drill-down into details of the data through clicking on individual data points.
This example of wizard-like approaches for development of FDC models shows how an etch endpoint detector can be used as a virtual particle sensor and as a monitor for detecting false endpoints.
Another important aspect of ease-of-use in FDC systems is wizard-like approaches for the development of FDC models. An example of what can be achieved with FDC is shown in Figure 4 (at right). Here we see how an etch endpoint detector can be used as a virtual particle sensor utilizing a moving average technique, and can also be used as a monitor for detecting false endpoints.
Because of its initial focus on individual processes, such as CMP, APC technology development has generally occurred from the ground up, with solutions developed as “islands of APC.”
The vast majority of APC's benefits, however, rest in the deployment of factory-wide integrated solutions. For example, R2R control systems can target improvement of factor-wide objectives, such as CD control, device electrical characteristics, die yield and throughput, rather than just process-centric goals, such as CMP process capability. FDC systems can provide higher-level analysis capabilities, such as tool-matching and process qualification, with faster deployment. More importantly, APC becomes an integral part of factory operations that can be incorporated into process and product design.
When selecting a factory-wide APC solution, you need to consider a number of important requirements in evaluating whether the R2R control and FDC solutions will support cost-effective, factory-wide deployment. One requirement is that the system must support the required number of R2R and FDC implementations in a consolidated fashion.
For example, the user interface should provide various factory views that support specific tools or groups of tools for APC analysis. This user interface should be similar for all tools so that the training effort is reduced and consolidated.
Another important requirement of an APC solution is use of a central repository for data so that higher-level analysis can be conducted. The solution should also keep pace with emerging SEMI standards for integration. At this time, the Process Control Systems (PCS) task force within SEMI is developing a comprehensive suite of APC integration standards.
Finally, any pre-integration with other factory systems like the Manufacturing Execution System (MES), such as user authentication, process flow management and context management, reduces the training and software maintenance effort.
Return on investment
Improvements due to APC implementation have been reported since the mid-1990s. In the last few years, these reports have also begun to include return on investment (ROI) testimonials. As an example, Sem-atech conducted a study of technical papers on APC implementations and produced a report on the ROI of R2R control.
The report concludes benefits of over $300M/year for a 20K wafer start/month logic fab.4 It also cites ROI resulting from increased process capability and accuracy, and reduced rework, process time, downtime, consumables cost, operator intervention, scrapped wafers and send-ahead wafers.
Significant ROI can also be achieved with FDC, generally resulting from reduced scrapped wafers, reduced downtime, improved mean-time-to-repair and faster process qualification.
Although APC has already demonstrated significant ROI, we have only begun to tap its true power. Once factory-wide APC systems are in place, individual APC solutions can be consolidated and tuned to achieve factory-wide, rather than process-centric objectives.
For example, factory-wide R2R control systems can coordinate lower level R2R control to achieve yield, throughput and device performance objectives.
FDC systems can actually be used to monitor the health of R2R control systems, and R2R systems could help in the selection of FDC models. Other equipment engineering capabilities, such as e-diagnostics, inventory control and maintenance management, could be integrated with and take advantage of APC. Finally, the comprehensive APC system could be fully integrated into the factory MES and other components to optimize scheduling, WIP and eventually even impact device design specifications.
It's clear that APC is here to stay and will become an integral part of future factory-wide fab operation. The key with implementing these solutions today is to make sure that they can be extended to the factory-wide, multi-component APC solutions that will be required in the future.
JAMES MOYNE, Ph.D., is director of EES technology, MES Business Unit at Brooks Automation, Inc. He can be reached at: (734) 516-5572.
- SEMI Draft Doc. #3527: “Provisional Specification for Automated Process Control Systems Interface,” Semiconductor Equipment and Materials International (September 2003).
- J. Moyne, E. del Castillo and A. Hurwitz, Run-to-run Control In Semiconductor Manufacturing, CRC Press, (2000).
- J. Moyne, “AEC/APC Vision: A Research and Suppliers Point of View, ” 3rd European AEC/APC Conference, Dresden, Germany, (April 2002).
- T. Stanley, et.al. “Cost and Revenue Impact of Advanced Process Control (APC) with an Emphasis on Run-to-Run Control (R2R),” Sematech AEC/APC Symposium XIV Proceedings, Snowbird, Utah, (September 2002)
Editor's Note: We introduced the first technical feature on the subject of APC back in May, when e-diagnostic specialist Alan Weber gave CleanRooms readers a healthy overview of the terms that define APC—the next frontier in contamination control—as well as an estimation of the time, money and personnel necessary to make it work. This month, Brooks Automation's James Moyne delves into factory-wide strategies.