Middleware layers could lead to plug-and-play automation
11/01/2004
The emergence of equipment data acquisition (EDA) standards, such as the new Interface A standard, opens up the possibility of plug-and-play manufacturing tools with greater levels of automation in semiconductor fabs. While additional standards are needed, device manufacturers are considering flexible and adaptive software layers to act as "wrappers" or "middleware" for homogeneous ports on tools for data, monitoring, and control functions.
Statistical process control (SPC), advanced process control (APC), and run-to-run control have all contributed some value to advanced automation in semiconductor fabs, but they are only as useful as the data provided to them. To keep pace with industry requirements for ongoing cost reduction at high yields, automation technology will need to deliver faster access to useful information from the entire processing line. This is only possible with hardware and software connectivity systems that "plug and play."
Historically, proprietary SECS/GEMS interfaces made development of plug-and-play connectivity systems inconceivable. Implemented more than 20 years ago, the ubiquitous SECS/GEM interface established a standard protocol for data exchange, but it lacked standards for how data should appear. So although data collected from different tools shared the same exchange protocol, it embodied different, and often proprietary, forms of data. For two decades, the end result was factories full of tools unable to communicate without a human "translator."
The emerging EDA Interface A standard driven by Semi represents the first step toward fabs full of tools that all speak the same data language. By standardizing how data is actually communicated from tool to tool, EDA Interface A's object model provides a universal translator that can accommodate complex tools such as cluster tools, diffusion systems, and linked lithography. The model also enables aggregation of standard mechanical-interface robots and automatic message-handling interfaces to ease their integration. It is a natural fit for the object-oriented factory systems in use today, and it is scalable for the more advanced systems yet to be deployed.
EDA interface standards are a significant advance toward full-factory automation; however, they still only allow tools to share passive data. This limitation was actually Semi's intent, since EDA was specifically developed as a passive way to collect data without interfering with the active controls of each tool. Nevertheless, until fabs have a standard way to manage the active data and control commands on their tools, they will have achieved only half their goal for true plug-and-play fab-wide connectivity and automation.
Semi has launched a number of initiatives to address this issue, including a recipe and parameters management task force (RaP) that attempts to define how recipes are described and applied. There is also the process control system (PCS) task force, which has the goal of defining standards that enable systems to rapidly detect, classify, and predict problems to control processes and equipment, as well as keep variability at a minimum. A third work group, labeled the common equipment model (CEM) task force, strives to design a common approach for the description of physical equipment structures.
There is no shortage of efforts to innovate and implement solutions for these complex problems. But one thing is certain: All these initiatives pursue data and the industry's pressing need for large amounts of it in more consistent and rapidly available forms (Fig. 1).
Factory middleware
While semiconductor equipment OEMs wait for new standards in data and control, today's innovative device manufacturers are considering flexible and adaptable software layers that offer the ability to "wrap" dissimilar tools with standard controls and interfaces. Considered as a sort of factory middleware, these software wrappers promise homogeneous ports for data, monitoring, and control, regardless of the underlying tool differences. The result is a distributed equipment environment (DEE) in which data streams from active tools and equipment are interconnected and interactive, at the operational event level as well as the end-user control level.
The effect of the DEE not only increases the transparency of processes, it also provides additional flexibility. Many OEMs on both the frontend and backend of IC manufacturing do not integrate metrology components into their tools, a fact which has motivated many chipmakers to buy from independent suppliers and perform the integration themselves. But this adds yet another disparate output at the tool, meaning the production system and the integrated metrology subsystem may be sending data independently. Fortunately, a new category of connectivity software that evolved from the CEM and the EDA Interface A standards enables chipmakers to merge these two sources into one model. This software is commonly referred to as an equipment information bridge (EIB). By creating a virtual model of what each tool is doing, EIB software unifies both data and controls in a single interface (Fig. 2).
 Figure 2. An equipment information bridge (EIB) folds communication data at the tool level with function definitions to provide a common equipment model. |
EIB software can achieve this information unification by enabling communication data at the tool level and then folding data and function definitions into a common equipment model. In other words, the software generates a virtual model of what each tool is doing. Adjusting the model parameters activates a change in the actual tools. Operators can then use open protocols to communicate and control individual tools or the overall process. An added benefit is that factories can integrate commercial software based on Semi standards to implement multiclient, simultaneous access from many factory points.
The CEM and EDA Interface A standards are prime enablers for these new software layers, and many subscribe to open-architecture computing models utilizing protocols such as .NET, CORBA, SOAP/XML, and others. This provides automation engineers with real-time visibility into critical processes and the data necessary to automate control systems, reduce process variations, and improve overall product quality — all from a standard and open foundation.
Impact on 300mm processing
CEM, EDA, RaP, and other emerging standards will certainly help APC technology reduce lot-to-lot variation by ensuring that individual tools produce materials within thresholds. But the golden ring for the semiconductor industry is less focused on tool-centric control than on wafer-to-wafer control for 300mm processes — or on advanced 200mm processes, for that matter. Operators should be allowed to focus their resources on producing good yields, not on unproductive attempts to decipher arcane tool data.
Wafer-to-wafer process control is very desirable, and by definition it offers two things. First, it provides visibility into the processing of individual wafers, allowing earlier detection of variations that affect quality. Second, it enables systems to deduce corrective measures for one wafer while continuing to process other wafers, instead of waiting for the entire lot to complete its run. A few fabs have achieved this goal, to some degree, for processes in which wafer-to-wafer nonuniformity is consistent. But in order to become more dynamic and flexible, the technology needs to detect the distribution of the variables that affect process performance and either fix the problem's upstream source or compensate for it at the wafer level downstream.
While this may be a great scenario, the bottom line for fabs is that advanced analysis and decision-making support requires data. Just as important, these systems require data quickly if critical adjustments are to correct process variations before they spread.
The benefits are not limited to one area. Coherent, comprehensive data can streamline or eliminate some existing run-to-run operations. In common practice, the disjointed sources of process data in older fabs caused many manufacturers to readjust or recalibrate tools after each lot rather than trust process parameters from one run to the next. This overbearing approach to building confidence into the process can be costly, and often is analogous to fixing something not broken. Reducing these setup cycles will result in instant gratification in semiconductor throughput calculation. Clarifying and consolidating product health indicators would provide operators with the confidence to keep running from lot to lot, eliminate separate setups, decrease cycle times, and improve overall throughput.
Impact on TAP automation
As frontend wafer fabs enhance their efficiency through automation and hyper-volume manufacturing methods, the backend test, assembly, and packaging (TAP) sector is only now beginning to adopt automation. Indeed, the contrasting level of automation between fabs and TAP facilities has become increasingly apparent in the component cost that each contributes to products. Once heavily weighted toward the fabs, the cost scale has increasingly shifted toward the relatively nonautomated TAP side.
Independently operating tools in many facilities have made TAP processes unnecessarily complex and unreliable (Fig. 3). Fortunately, the TAP sector can leverage technological advances designed to minimize variations in wafer fabs. One strategy is to connect tools to line controllers using EIB software to speed tool setups, manage recipe selections, and collect data (Fig. 4). Leveraging CEM, EDA, RaP, and PCS standards available to fabs, the TAP segment has been able to link tools and automate entire processes, resulting in less maintenance, lower cost-of-ownership (CoO), and higher return-on-investment (ROI). Early adopters in the TAP sector have targeted specific applications, such as tracking of probe cards and wafers, as well as automated prober setup and data collection. The benefits of these initiatives include increased prober and tester equipment productivity, and reduction of human error.
Some leading integrated device manufacturers have already started linking their process and metrology tools to station controllers, MES, SPC, and dispatch servers to implement automated data collection, recipe selection, and automated material validation. This helps them improve equipment productivity and reduce production cycle time. Such integrated functions represent the next generation of automation systems based on CEM, EDA, and other new standards, making it possible to implement advanced technologies without strict dependence on SECS/GEM interfaces on all equipment. Free to move forward toward better solutions, facilities can gain the benefits of advanced automation at a fraction of the deployment time and cost that would have been required to implement this level of automation just a few years ago.
Conclusion
With the prospect of readily available and open standard methods for communicating and controlling factory tools, automation engineers are now rethinking how they can design the new execution and control systems of tomorrow. Middleware — like EIB, which can bridge the gap between the many different evolutions of manufacturing equipment that must be automated — will make building flexible plug-and-play architectures a reality. Considering the many tangible benefits of plug-and-play software architectures, along with the semiconductor industry's growing leverage of the types of open standards currently enjoyed by other technology segments, it is reasonable to conclude that the days of hardware and software disconnects may finally be numbered.
Acknowledgments
Asyst EIB and NexEDA are trademarks of Asyst Technologies Inc.
Robert Foy is director of professional services and strategic accounts in the Connectivity Systems Group at Asyst Technologies Inc., 48761 Kato Rd., Fremont, CA 94538; ph 510/661-5000, e-mail [email protected].