APC: The next frontier in wafer-level contamination control
05/01/2003
Many believe this highly touted discipline will be the ultimate yield-improving tool. Be prepared to dedicate the internal effort necessary to refine your process
By Alan Weber
The demand for advanced process control (APC) solutions continues to increase as the world's semiconductor manufacturing companies strive to improve process control, enhance yield and improve the efficiency of equipment utilization.
Measurement equipment for process and wafer-level monitoring, data acquisition, equipment-level data transfer as well as host communication, feedback and feed-forward control are some of the major issues being worked on by the world's leading chip manufacturers—many of which are teaming with hardened competitors to work together on ways to increase the speed to factory-wide APC.
As many semiconductor manufacturing managers are learning the hard way, a successful long-term APC strategy must not only meet a host of difficult technical requirements, it must also constitute commercially viable business for the suppliers involved.
Definitions are critical
APC can mean a lot of different things to different people, so a few definitions are in order:
- Fault Detection (FD)—Monitoring and analyzing variations in tool and/or process data to detect anomalies. Fault detection includes both univariate and multivariate statistical techniques.
- Fault Classification (FC)—Determining the cause of a fault once it has been detected.
- Fault Detection and Classification (FDC)—Combination of FD & FC.
- Fault Prediction (or Prognosis) (FP)—Monitoring and analyzing variations in process data to predict anomalies.
- Statistical Process Control (SPC)—The use of statistical methods to analyze process or product metrics for taking appropriate actions to achieve and maintain a state of statistical control and continuously improve the process capability.
- Run-to-Run Control (R2R)—Modifying recipe parameters or the selection of control parameters between runs to improve processing performance. A "run" can be a batch, lot or an individual wafer.
- Process Control System (PCS)—A system capable of performing Process Control, which includes one or more of R2R Control, FD, FC, FP, SPC, or any future Process Control functionality defined in this standard for a PCS functional group.
- Advanced Process Control (APC)—The manufacturing discipline for applying control strategies and/or employing analysis and computation mechanisms to recommend optimized machine settings, and to detect faults and determine their cause.
APC will benefit from the industry's current attention
There's a bewildering set of standard practice information being written, and an overwhelming amount of work for industry standards folks. But the bottom line is that the industry is paying more attention to the importance of more accurate and timely data out of the equipment and putting it into the hands of those who can interpret it.
These new standards, and the embedded control/integration approaches that will be required to implement them, will directly support the APC applications that apply statistical techniques to volumes of tool data (Fault Detection, SPC, etc.) for drawing their conclusions.
Even for run-to-run control applications, incorporating key information from the process tool about its behavior during a particular run can be used to enrich the control models, improving their fidelity and the quality of the resulting control behavior.
Payback on well-chosen APC applications is high
Yield and productivity improvement are the principal operational metrics that provide the ROI justification for most APC investments. These have now been analyzed and summarized across the industry in a number of papers/presentations (notably Tim Stanley, International SEMATECH, Sept. 2002), and include specifics on Cpk, rework and scrap rates, use of non-product wafers, tool productivity, cycle time and consumables cost. Moreover, in some cases, the typical expected results can be broken down by process area.
null
Leading implementers of APC have discovered that after the key sources of systematic variation in the most critical process areas have been identified and comprehended in a control scheme, the need for control in closely-related upstream and downstream processes is reduced or eliminated altogether.
This seems obvious enough when one views the wafer/substrate manufacturing process as a tightly-coupled set of interdependent steps. Few companies, however, actually develop their operational practices from this perspective (nor could they, when SPC was the only fab-level monitoring/control technology available).
Open door for other direct benefits
Once an APC system is in place, the data management infrastructure that results (collection, validation, storage, distribution, access, and transformation) can be used for a wide variety of purposes beyond process control. Because of the wide separation in time and space of the data required by many advanced control algorithms, it is likely that much of this data has never been easily accessible for use in other advanced manufacturing decision-support applications.
A robust APC infrastructure can change all that. A partial list of applications that can be enabled with the infrastructure includes production scheduling, equipment maintenance planning, process window analysis and design, yield learning, and other product life-cycle management tools.
APC systems are becoming more complex
In the late 1990s, the prevalent form of production APC systems consisted of: lot-based, run-to-run control that used post-process product feature metrology data; the "inversion" of an experimentally-determined process model, or transfer function; and whatever information was available about the incoming material to recommend recipe adjustments for a specific combination of lot, tool and process.
Some systems also monitored various tool/process conditions, generating events and alarms for use by human operators or complementary systems for notification, interdiction, or further analysis.
Architecturally speaking, these systems consisted of network-based application and database servers that collected/stored the data and responded to requests from tool and/or factory system (MES) interfaces to perform and return recipe adjustment calculations. The complexity in this first generation of systems arose not from the control mathematics involved but rather from the bookkeeping required to maintain the thousands of parameter sets in a high-product mix setting, and the mission-critical nature of automated control in a production environment.
As Moore's Law continues to drive the economics of computing, data storage and communications technologies, the number and type of available "sockets" for implementing APC functionality have increased enormously. Moreover, the levels of automation envisioned for 300-mm factories and the advent of Integrated Metrology will drive the application of APC on a wafer-to-wafer basis, which will be implemented in the process tool's embedded control system (see Figure 1).
As a result, multiple versions and instances of these applications may be provided by a number of suppliers (including the internal IT groups) in a given fab.
Yet none of these systems can stand alone. Most will depend on one another for some of their data, or assume that a fab-wide process data repository exists. As a result, the end user's systems engineering task to pull all this together has gotten far more complicated.
An excellent example of the kind of multi-level, cascaded system that may soon be prevalent was presented at this year's European AEC/APC Conference (Botros, Funk and Weber; March 2003). But this is just one example—many other configurations are possible.
APC technology, by its nature, will always be a moving target because the external tools and systems on which it depends are constantly changing. So, this increase in complexity is with us to stay.
Ironically, the fundamental process control technology on which today's recipe update decisions are made is probably the most stable component of this entire picture. And even this is subject to change as more work is done in the areas of "physics-based" or "first principles-based" simulation and control.
Process control engineers are still needed
Process control engineers possess a set of skills ideal for participating in process integration and improvement activities in semiconductor production processes.
But the process control engineer is still far from being a common job description in most companies, and is likely the by-product of the industry downturn rather than a commentary on the value that this expertise can bring to a process engineering organization.
Standards are important—and improving
The original "Ten things you need to know about APC" (Paul McGuire, Tony Mullins, Future Fab, 1997) article pointed out that the use of open, standards-based architectures would accelerate industry adoption of APC. Market forces being what they are, however, it's now clear that the industry will probably never share a common implementation architecture, as this strikes too close to a supplier's product design freedom.
But given the drastic improvements that have been made in system interface technologies (both the data representation and "plumbing" aspects of the problem), this is no longer necessary. The current need is for users and suppliers to agree on the standard mechanisms that will be used to connect independent APC products.
This is the precise goal of the SEMI Process Control Systems (PCS) Task Force, which was formed in late 2001 and just published its first formal ballot (SEMI Doc. #3527).
Moreover, the industry has learned from its first attempt to standardize APC integration approaches, and current initiative has the quality (production experience), breadth (geography and company type) and level of participation (core team of creators and active reviewers) to ensure a viable result.
One aspect of product marketing is making sure the problem and solution domains are in alignment, and that they are consistent with the customers' purchasing model of the world.
When one views APC from the various perspectives of the current and potential participants in this business, it's no wonder that the commercial APC market has been slow to mature—especially when there are so many different parties vying for their share of the overall value proposition.
Semiconductor manufacturers take the position that they are the process/applications experts and don't expect equipment suppliers to provide the total solution. What they really want is external visibility and control of tool behavior, and customizable standard product solutions that are cheaper than they could build and support for themselves.
Process equipment suppliers, on the other hand, feel strongly that they provide not just hardware but the entire process, including the control aspects that are becoming more important in achieving the required process specifications. As a result, they prefer to provide a complete solution to the user's APC needs.
Metrology equipment suppliers put a different spin on the problem, making the case that the quality of the product parametric data is the key to good control—and that buying one's APC system from the process tool vendors is akin to the fox guarding the henhouse.
Integrated metrology suppliers can fall on either side of this fence, but emphasize the importance of timely parameter measurements in minimizing process variability and product jeopardy. Sensor/subsystem suppliers, meanwhile, are asking, "Just where do you think all this data comes from, anyway?" Their objective is clearly to broaden their stake in this market without making exclusive supply agreements with the process tool OEMs.
The automation systems/software suppliers complete this chorus, suggesting that only they are the best source for truly independent solutions that span the fab while incorporating the robust information technologies necessary to support fully automated, high-volume operation.
Current economic conditions have effectively suspended this debate, but it is far from over.
A new breed of APC supplier is emerging
The synergy between off-line process analysis/yield management systems and on-line control applications has attracted the attention of a number of suppliers who possess deep domain knowledge about their particular process areas. These applications use much of the same information, and are often procured and maintained by the same end-user organizations.
The major challenge for suppliers will be to apply this knowledge in a way that delivers the performance and reliability required of on-line, mission-critical manufacturing systems. It requires a special set of system architecture skills and a healthy dose of real production experience.
A secondary issue will be making the transition from a single-process mentality to support a wider range of production operations in a consistent way. Despite these challenges, this group is a promising new entrant into the APC market.
A cycle of underinvestment
Throughout the downturn, end users who depend on leading-edge process technology for some of their competitive advantage have continued to invest in APC solutions. Much of this work, however, has been done on private or semi-private systems, so very little of this investment has trickled down into the supplier community.
In fact, only a couple of suppliers can truthfully claim to have any significant, high-volume production experience or market traction. And without it, one can't really predict how a software product will hold up when scaled up from its "proof of concept" incarnation to support an entire fab.
In this unclear market environment, it has been difficult for APC suppliers to justify major new developments in this space, especially for companies whose core business is process or metrology equipment. The result has been a cycle of serious under-investment in APC technology, which if not addressed soon will exacerbate the separation between user expectations and supplier offerings.
The silver lining in this otherwise dark cloud is that since no dominant supplier has captured much of the market, there is plenty of headroom for companies that have the insight, resources, patience, and the will to serve this important segment.
Internal effort needed
APC is now a mainstream semiconductor manufacturing technology, and is considered essential for the 130-nm, 300-mm technology nodes and beyond. Nevertheless, the approaches for applying this technology are all over the map, and many lead to missed expectations, cost overruns—or worse.
So, be prepared to dedicate the internal effort necessary to refine your requirements and engage your suppliers as genuine partners in the process.
Alan Weber is president of Alan Weber & Associates, Inc., a consulting company specializing in semiconductor Advanced Process Control, eDiagnostics, and other related manufacturing systems technologies. Before founding his own company, he was vice-president/general manager of the KLA-Tencor Control Solutions Division, acquired from ObjectSpace, Inc. in March 2000. He can be reached at [email protected]