New tools and fabs demand 300mm automation optimization
06/01/2001
Tony Bonora, David Feindel, Asyst Technologies, Fremont, California
overview
Automation for 300mm fabs absolutely requires comprehensive attention to the interrelationships between the four levels of automation: FOUPs and wafer-handling automation, tool frontend automation, intrabay automation, and interbay automation. The complexity of the manufacturing sequence continues to grow at a rapid pace, meaning more layers, more vertical circuit integration, and the need for faster feedback from metrology measurement to provide real-time process feed-forward and feed-backward process control.
The fact that 300mm wafer carriers are 2.5x the weight and volume of their 200mm predecessors (Fig. 1, Table 1) adds substantially to demands on wafer handling, transport, and automation systems. However, the impact of the migration to 300mm on automation goes much deeper. Sorting out the issues raised in integrating these many building blocks into a working 300mm fab has proven to be a difficult task. Optimizing each individual function does not necessarily produce an optimized automation system.
 Figure 1. 300mm FOUPs have added substantially to demands on wafer handling, transport, and automation systems. |
Changes in fab automation philosophies brought about by the change to 300mm can be seen in all four levels of automation in a fab:
- FOUPs and wafer-handling automation;
- tool frontend automation;
- intrabay automation (material movement within a bay); and
- interbay automation (material movement between bays).
Interrelationships between these levels of automation have proven to be of critical importance. For example, changes in wafer-handling automation, such as the change in container architecture to a front-opening unified pod (FOUP) impacts all three subsequent levels.
IC manufacturers who best understand these interrelationships will recognize larger economic rewards from superior fab efficiency and effectiveness. Tool suppliers who have products cutting across all of these levels are in the best position to help chipmakers improve the use of automation.
FOUPs and automation
Because they allow a single standard interface to metrology and process tools, FOUPs (the 200mm-wafer carrier equivalent for 300mm wafers) are one of the key enabling technologies for automation. Much discussion in the industry has centered on the FOUP and its required characteristics. Even today, after four years of work, changes are still being made to the standard.
One key decision was the number of wafers in the FOUP: the industry has standardized on 25. This has several implications for other automation levels in a 300mm fab. For example, it allows a fairly high density of wafers/unit of volume, lessening the size of stockers. It also lessens the number of moves/hr an automated material-handling system (AMHS) has to perform.
The trade-off, however, is that wafer lot cycle times lengthen, as all 25 wafers have to finish processing on a tool before the FOUP can move to the next tool. It also forces mandatory automated tool loading; FOUPs designed for perhaps 10-13 wafers would be light enough for operators to handle manually, whereas 25-wafer FOUPs are not.
 |
The emergence of 300mm wafers has promoted the use of sophisticated robotics, with accuracies on the order of 0.1mm, just to retrieve wafers from a FOUP. This accuracy requirement is particularly challenging because in many cases 300mm wafers demand the use of edge-grip end effectors (Fig. 2). Unlike the backsides of 200mm wafers that have a matte finish, 300mm wafers are polished on both sides to eliminate a source of contamination, facilitate particle detection, and ensure a flat wafer surface during the lithography process. An edge-grip end effector grips the edges of the wafer, avoiding contact with the wafer backside. Wafer-end effectors are thicker than previous vacuum-style end effectors, limiting available free space between wafers and making accurate motion critical. Currently, as much as 25% of 300mm wafer-processing tools use edge-grip end effectors.
Tool frontend automation
One of the most obvious changes in 300mm fabs is the ubiquitous presence of tool frontends, often called factory interfaces. Typically, equipment manufacturers need to combine a variety of operations, including atmospheric wafer handling, prealignment, wafer buffering, a local minienvironment, wafer-mapping capability, auto identification, optical character recognition (OCR), and FOUP-opening mechanisms into one system. This serves as the tool's interface to the fab's AMHS.
These functions were previously performed by a combination of solutions supplied by tool manufacturers and OEMs. For example, many 200mm vacuum cluster tools have load ports where a human operator places an open cassette onto an in situ cassette elevator that is part of the process tool. An integrated robot then retrieves the wafers and places them on processing stations within the tool, and returns the wafers to the cassette when they are finished.
In contrast, 300mm-wafer transfer requires a dedicated atmospheric wafer pick-and-place robot that loads and unloads the FOUP and places wafers in the process tool or otherwise makes them available to the process tool (Fig. 3). The number of functions that have to be performed on the tool frontend has grown compared to 200mm wafer processing because of the FOUP architecture. Unlike a 200mm SMIF pod from which the entire cassette can be removed and placed inside a process tool, the FOUP's integrated construction prevents removal of the "cassette" portion of the FOUP.
Many factors in 300mm automation have combined to drive the transformation of tool frontends:
- wafer extraction and loading;
- flexible robotic handling;
- multiple automated interfaces;
- numerous Semi Standards covering loading heights and permissible loading zones; and
- wafer container architecture changes.
Human operators are very adaptive in addressing tool-loading tasks and compensating for tool utilization and operability issues. At 300mm, however, the skills and capabilities that people have traditionally brought to this job are being replaced by automated mechanisms and control software.
Integrating all of these components — robots, 300mm FOUP-opening devices, prealigners, wafer handlers, minienvironments, necessary software, receiving stations for wafers, and software interfaces — into a single tool frontend ensures optimal matching among individual components to meet the specific requirements of the 300mm FOUP design and the fab automation system.
Tool buffering
Tool buffering (i.e., the storage of extra product lots at a tool) is a concept long recognized for enhancing process tool utilization. An International Sematech study revealed that 15-20% of tool time can be spent waiting for an operator or the arrival of product lots. Considering capital investment in a 600-tool factory, this translates into tens of millions of dollars/year in lost opportunity. Placing appropriate quantities of work in process (WIP) in buffers close to the tool can improve tool productivity by reducing or eliminating this wait time.
From the perspective of the tool manufacturer, the important performance metric is throughput, measured in wafers/hr (wph). A key target is to minimize idle time on the tool while waiting for one wafer lot to exit and another to arrive. This is one major reason why all process and most metrology tools have multiple loadports (i.e., so a tool can process one FOUP while waiting for another to be removed and replaced with a new lot).
With human handling, when a tool is finished processing, the operator is alerted to remove one lot and replace it with another. This material change may take 30-90 sec if an operator and material are immediately available. For example, throughput for an average tool may be 100-150wph, exchanging a half-dozen product lots/hr. With a reserve capacity of three to four FOUPs available, plus associated empty locations to store completed FOUPs, tool utilization can be improved by as much as 5-10% over conventional means of tool loading while reducing the peak time demands on the automated transport system.
But it is proving hard to implement buffering in 300mm fabs in an economical fashion. Early 300mm guidelines from International Sematech had each process tool responsible for providing a buffer of one hour of WIP for the tool. This required larger tool footprints, a trade-off that many equipment manufacturers and chipmakers were unwilling to make. A second problem with this strategy was that if a tool went down, all lots were trapped in its buffer, significantly reducing operating flexibility. The alternative concept was to provide buffering at the stocker. This provides the best level of operating flexibility, but the drawback to this strategy is that it takes time — as much as 10-15 min in some cases — to get a new FOUP to the tool. This can leave a tool starved for lots.
A newer approach to the problem is to separate the function of transporting lots from the tool-loading storage function. This combines the best of both worlds: rapid delivery of lots to the tool and the storage of lots in the vicinity of the tool in a space-efficient manner, providing operating flexibility.
To take advantage of buffering wafers at or adjacent to the frontend of the tool, Asyst's FastLoad (Fig. 4) combines an x-z robotic mechanism to provide buffering and tool-loading functions. The system may be integrated with the tool structure or the facility structure. It is comprised of a gripping mechanism mounted on a vertical linear slide that can pick up a FOUP. Lateral movement from the overhead transport system continues down to the 900mm loading position at the front of the tool. Positional accuracy within one millimeter is an achievable target with this mechanically deterministic system.
Unlike a cable-hoist system that, when positioned over a given tool location can only drop a FOUP vertically to a given level, this mechanism is able to access two or three tiers of stored FOUPs positioned above a process tool by moving laterally around the FOUP on the top row to extract product beneath. This allows chipmakers to store additional WIP in the bay, reducing or eliminating stockers at the head of the bay. Asyst has determined, through fab modeling, that a significant percentage of stockers in a fab can be eliminated using this strategy.
Interbay and intrabay automation
Conventional 200mm-factory automation has been primarily based on interbay transport systems consisting of overhead, car-based transport and large stocking locations. Stocking locations are usually large stockers located at the head of every bay, holding 150-250 boxes or pods. These stockers also serve as the human operator interface, with the WIP being retrieved by human operators and carried to process tools or workstations.
Interbay systems were first introduced around 1988. During the last five years, interbay transport has become an industry norm; its absence in the fab is a rarity. Interbay systems are typically characterized by their capacity, expressed in moves/hr, with each move being the transport of one container from a sending stocker to a destination stocker. Typical run rates are 250-350 product lot moves/hr, although a few large fabs are in the 700+ moves/hr range.
Intrabay transport was envisioned as an automated means for moving material from stockers at the end of bays to individual process tools and, in some manner, loading those tools. Compared to interbay transport, intrabay material-handling challenges are much more daunting, requiring interfacing to a wide variety of process and metrology tools. Lack of standard equipment interfaces and complexity has limited the application of intrabay automation in 200mm fabs to a few bays at the largest fabs.
Currently, most 300mm fabs are considering a technology called overhead hoist transport (OHT) — a vehicle carrying one FOUP suspended below a monorail track mounted just below the ceiling. Each vehicle can raise or lower a FOUP to a loadport (usually a vertical drop of about 2.5m).
 |
Interbay and intrabay transports have traditionally been implemented as two different functions, usually with two different systems, perhaps provided by two or more different manufacturers. The industry is starting to realize, though, that optimizing an interbay transport system for moving material between bays and optimizing a second intrabay transport system for moving material within a bay does not produce an optimal transport system for the fab. Routing WIP from a process tool to the stocker at the head of the bay and waiting for an interbay vehicle to arrive and transport it to the next stocker, where it again waits for another intrabay vehicle to arrive and deliver it to a tool, is too inefficient and time consuming. Having a common transport system, fab wide, for both interbay and intrabay transport, eliminates handoffs and wait times, substantially improving delivery times.
Safety is a major issue for transport systems (Table 2), especially for the tool-loading function. Chipmakers need to permit operators to load and unload lots during startup and pilot plant phases, while developing methodologies and operational procedures for automated loading. To further industry work in this area, International Sematech and Selete sponsored a workshop (held in Tokyo in March 2001) to clarify, if not resolve, these issues. The trend seems to be in favor of physical shields or barriers to separate operators from the automated tool-loading system, with interlocks to permit one or the other to load a tool, but not both simultaneously.
A second issue that has developed at 300mm pilot lines involves interoperability issues between the OHS and the frontend of the tools. Upon loading a FOUP, for example, when the overhead hoist lowers a FOUP onto a loadport, misalignment or timing errors have resulted in tool stoppage. One way to tighten interfaces and improve reliability is to integrate individual building blocks of the entire frontend of the tool.
A third issue that has surfaced is the increasing requirement for high material transfer capacity within a bay. Early estimates concluded that a high-traffic bay would require 125-175 moves/hr; current estimates recognize more than 300 moves/hr. If send-ahead wafers or test lots are added to the mix, this number grows even higher. This puts a considerable strain on OHT technology. Vehicle-based technologies inherently require wait times for vehicles to arrive at the required pickup or drop-off location. OHT vehicles also take 20-25 sec to place or retrieve a FOUP, blocking the track during this time. These built-in delays make it very difficult to achieve 300 moves/hr, even with large numbers of high-speed vehicles, bypass lanes (if they can be fitted into today's narrow bays), and advanced control algorithms.
Fab control automation
Five or six years ago, an IC process flow might have had 400 process and metrology steps, with each step requiring an operator to move material. Now, near-term future fabs anticipate 600 steps with an associated number of wafer lot moves into and out of tools. The sequence of steps follows what may appear to be a random path from process tools to a stocker, then to a metrology area, back to another stocker, then to another process tool. The best route is determined through computer monitoring and control, allowing for scheduled tool downtime and adjustment of process parameters. Ideally, factory automation maintains overall factory balance as more complexity is added to the process.
While most fabs have wafer-scheduling systems and predictive models that attempt to optimize the flow through the factory, real-time dispatching of wafer lots in an optimized fashion to different parts of the fab is just starting to gain momentum. This is a very complex task for a more than 600-step process. The scheduling-dispatching control software must analyze tremendous amounts of information in real or near-real time (process tool setup and current status, scheduled and unscheduled machine downtime, lot status and location, transport system performance, lot prioritization, future work schedules, etc.).
Today, a typical large 200mm factory has 4000-5000 wafer lots in process at any given time. This number will grow with process complexity, thus increasing the value of WIP inventory. For a 300mm factory, assuming that each lot has an average realizable value of $200,000, the fab will be carrying $800 million to $1 billion in WIP. The magnitude of this inventory will force higher-quality WIP management through better scheduling and dispatching of lots to process tools.
Looking ahead
In the era of 130nm linewidths and billion-dollar fabs, perhaps successful fabs can still be built employing human operators to transport FOUPs on carts. As the processing sequence becomes more complex and fabs ramp their volume, the challenge of managing the flow of material becomes challenging. The combination of more process steps and more process stages that require geographic isolation (CMP, copper, etc.) is converging with the 300mm-wafer era to require more efficient flow of material throughout the fab and, consequently, greater automation content.
The 300mm era may see the introduction of smaller lot sizes, driven by inventory costs and increased process complexity. Today, in large 200mm fabs, 4000-5000 wafer lots are being routed through the fabrication process at any given time. This number grows with increased process complexity and number of process steps. More inventory also means longer queue times, increased inventory expenses, greater risk of obsolescence, and longer learning cycles (i.e., it takes longer to catch yield errors, etc.). The use of smaller lot sizes offers leverage to address these issues; cycle time through the fab is decreased, average waiting time at a process tool or between process tools can be significantly reduced, etc. However, the impact on the factory's automation system is significant.
Ten years ago, industry experts were projecting a lights-out fab by the late 1990s, a large, dark factory with fewer human operators and virtually total automation. The reality is that human loading of wafers is still common. The complexity of the manufacturing sequence continues to grow rapidly, meaning more layers, more vertical circuit integration, and the need for faster feedback from metrology measurement to provide real-time process feed-forward and feed-backward process control. This puts additional demands on automation. More important, successful automation can accelerate this increased complexity in chip fabrication. By optimizing the factory's overall automation systems, chipmakers can achieve their technical and economic goals.
Acknowledgments
Asyst SMIF-Pod, FasTrack, FastLoad, and Plus Portal are trademarks of Asyst Technologies.
Anthony Bonora received his BS and MS in mechanical engineering from the University of California, Berkeley. He was a 1999 recipient of the Semi Award for North America for his contributions to the semiconductor industry. Bonora is executive VP and CTO at Asyst Technologies, 48761 Kato Rd., Fremont, CA 94538; ph 510/661-5000, fax 510/661-5387.
David Feindel received his BS in management science from Case Institute of Technology, and his MBA from Harvard. Feindel is a consultant for Asyst Technologies' Transport Products Group.