D. Rex Wright, PhD, Asyst Technologies, Inc.
Semiconductor manufacturing and cleanroom use have been tightly coupled since the earliest days; today, semiconductor cleanrooms continue to drive trends and technology.
Why semiconductor factories are different
It’s easy to understand why semiconductor factories (“fabs”) are put in cleanrooms, but why have semiconductor cleanrooms taken the particular structure they have today, and what are the trends guiding them? Let’s start by studying how semiconductor fabs are different from other manufacturing facilities.
If you compare a fab to an automobile factory, you will notice that, while the latter has a distinct linear flow (lines feeding an assembly line), no such line exists in a semiconductor cleanroom. That’s because what most differentiates one chip from another is feature size, which directly drives the number of functions per chip. Thus, process control has always been the dominant driver in semiconductor fabs.
Individual process steps are performed in highly optimized individual tools; because they are optimized for their own steps, they are not optimized to balance process times across multiple steps. The high expense of these tools means that they have to perform multiple (nonsequential) steps to keep their utilizations high, and thus the factory has a lot of re-entrant flow. Given this re-entrant “job-shop” routing among tools of differing process times, factory automation and design must deal with complex routing and intermediate work in progress (WIP), while keeping the wafer environments clean. How have cleanroom designs provided this flexibility over the years?
History
The early fabs (through approximately the 150 mm-wafer generation) were similar in design. Tools were aligned in rows along a wall in a “bay and chase” formation, with the wafers and the front (loading) portions of the tools in the very clean “bay” portion, and the maintenance and operation portions of the tools in the “chase” behind the wall-still a cleanroom, but of a significantly lower grade.
There was typically a lot of human interaction with the wafers (humans using tweezers and microscopes to perform quality control such as checking the “color” of the oxide coating under a microscope to confirm correct thickness). Human interaction, coupled with the need for numerous storage racks in front of the tools for WIP, meant bays had to be very clean and very wide (at least by today’s standards).
Wafer lots were stored in open cassettes, which were typically placed inside simple boxes for (human) transport between tools. Box exteriors were often adorned with multiple pages of cleanroom paper for “travelers” of instructions and hand-written notes. Automation was very rare or non-existent.
In general, the entire (bay) environment was kept very clean (ISO Class 3 or 4 [Class 10 or Class 1]). Laminar flow was typically provided by full-ceiling coverage with full-floor perforations.
As wafer sizes grew to 200 mm and linewidth features on the wafers shrank, the size of fabs grew enormously. Instead of using just open cassettes in boxes, fabs turned more to pods using standard mechanical interface (SMIF). Tools now needed to have “front ends” to enable cassette-to-cassette processing (see Fig. 1).
Figure 1. Typical 200 mm SMIF cleanroom. Note (clear) wall between bay, chase remained. |
The SMIF pods also tended to have “paper travelers” but as time passed, more automated forms of passive identification appeared. Barcodes, scanned by the operator, or infrared tags communicated directly with tools and shelves. Some had small human-readable screens to replace paper notes. On the tools, communication standards, such as general equipment model/semiconductor equipment communication standard (GEM/SECS), allowed some monitoring of tool and lot processing by central fab controls, but the standards were still primitive.
Between steps there was still much human interaction. Typically pods were taken to “clean areas” (hoods) for inspection, but more and more automation began to appear on inspection tools as well. As time passed, some interbay automated material-handling systems (AMHS) moved boxes (pods) between stockers at the end of each bay. Humans were still loading and unloading pods onto tools (intrabay), and stocking occurred mostly on passive shelves.
As users became more comfortable with isolation technologies, fabs using SMIF began to lessen their cleanroom classification levels; full-ceiling coverage of laminar flow now dropped-sometimes to half or less (see Fig. 2).
Figure 2. As automation spreads, semiconductor cleanroom requirements actually lessen outside of tools. |
null
Modern history: The rise of 300 mm
The transition to 300 mm took many years of false starts before gaining traction. The chief innovation was the universal use of pods, now chosen to be side-opening with the cassette and box as one piece, and thus called front-opening unified pods (FOUPs). Extensive standards defined not only the FOUPs, but also the load ports on the tools. Some unintended side effects of the standards activities included:
- The front-opening pod locked in a single size for transport carrier at 25 wafers. True, alternatives (13-wafer and 1-wafer) were demanded from time to time, but these disappeared quickly (despite considerable development costs for the tool and automation infrastructure). This large carrier size would have implications later for those seeking small lot sizes.
- The large numbers of suppliers and the complexity of the interface standards led to subtle interoperability issues as the systems became more complex.
- Process tools had to add equipment front-end modules (EFEMs) to handle the automation interface. In the past, humans or pod loaders handled some of this functionality. Note that even in regions where low labor costs have kept fabs human-transported, end users have continued to pay for this additional functionality (up to several percent of a process-tool cost for some tools).
FOUP identification went from passive (barcode and IR) to active (RFID). These small “tags” embedded in the carriers (and readers in load ports and stockers) allowed both read and write functionality. Coupled with “slot control” (always returning wafers to the same slot of the same carrier on output from a process tool), process tracking became more powerful.
On the process tools themselves, far more powerful control-automation standards allowed process jobs and control jobs to be tracked and controlled more effectively.
The tools themselves were now more often in a “ballroom” style room, where the walls between the bay and chase sides disappeared (essentially the whole fab became a chase). However, the tight linearity requirement of the intrabay AMHS systems (usually overhead hoist vehicles [OHT]) still forced the tool front ends into tight, straight lines. The tools themselves remained highly optimized for process, with multiple (nonsequential) processes and varying process times.
AMHS systems employed lots of stocking, initially in the form of large central stockers, and as time passed, with more dispersed, smaller stockers.
Emerging trends at 300 mm
So what are the trends seen in today’s newest 300 mm fabs? Let’s consider each of the key drivers for automation, wafer environment control, fab layout, and cleanroom structures.
Automation
We see the gradual removal of humans becoming more widespread as software systems do a better job of choosing and running lots. One key trend here is the rise of tool information standards using the equipment data acquisition (EDA) standards, which allow client subscribers to monitor tool and chamber status to keep tool utilization high. Another trend is the growing amount of process and metrology data tracked among process steps for immediate feedback (feed forward) using advanced process control (APC).
We also see new importance given to sorter automation. Tasks formerly done by humans with wands, such as adding test wafers, performing splits, and re-ordering wafers within a lot, now require automation. Modern sorters must combine high throughput with complex software capabilities to avoid slowing fab throughput (see Fig. 3).
Figure 3. At 300 mm, sorters (such as the Asyst Spartan Sorter) are needed to do test wafers, splits, and other activities performed by humans in earlier generations. |
Likewise, the front-end automation of process tools now must do more tracking of wafers. We see the increased importance of various single-wafer ID tracking marks on wafers, as well as huge data loads for the tools and factories as characteristics are tracked at the wafer level instead of the lot level. Wafer tracking will become more important as the percentage of smaller lots increases, especially in high-mix fabs (such as foundries).
Process
New materials are pushing the envelope and moving away from stable and well-understood materials to more exotic blends. Fabs no longer use just silicon, silicon dioxide, and aluminum; we see copper, low-dielectric-constant materials, heavy-metal oxides in gate dielectrics, and many others. This explosion in the number of materials in the line increases the chance of interaction, reliability problems, etc. While today we already see segregation of “copper” and “pre-copper” sections of the fab, we may start to see more isolation needed to prevent interaction (ironically, some of those walls may reappear in the ballrooms).
At the wafer level, wafers now need environments that are not only free from dust, but also well characterized atmospherically. While lithography engineers have been able to replace many of the exotic water-hating deep-ultraviolet (DUV) photoresists with more stable ones, other new materials are coming with their own requirements. In the front end of the process flow, certain thin-film steps (e.g., nitridation, oxidation) require either short time between steps or purging to avoid water monolayers on the wafer. In the back end (interconnect formation), the use of copper along with exotic low-dielectric materials and alkaline polish mixes for chemical mechanical polish (CMP) can leave tiny areas of copper lines exposed to corrosive residues. Here, avoidance of not only water but also oxygen is required in some cases. Cleanroom design must arrange for inert purging features without endangering humans (who still need the oxygen, after all). Special FOUPs, and select load ports and stocker shelves, are designed with purging features at these key steps.
Economic drivers of fab layout
The staggering multibillion-dollar investment in large fabs is driving the need for reuse of buildings, or at least alternate layouts such as multifloor and multibuilding fabs. The shorter ceilings of reused 200 mm fabs provide special challenges for the AMHS systems to find room to move and store lots. The cleanrooms themselves now also have additional interfaces, not just room-to-room, but floor-to-floor and building-to-building. Standards are developing for automated-door front-opening shipping boxes (a-FOSBs) to replace the formerly human-opened wafer-shipping containers. These new standards allow entry to buildings to be more predictable as they keep wafers safer from human contamination.
In parallel, while Moore’s law (i.e., the increase in number of functions per die every year) is still pushing the extreme performance end, as more and more consumer uses arise (e.g., communications, automotive, sensors, medical) we are seeing a clear rise in the need for faster cycle time and lower cost. While subsets of customers will always want the most powerful chips, many more customers are now emerging who simply want acceptable performance at less cost and with shorter lead times.
What this means for fab and cleanroom design in the near term is a push for more efficient manufacturing with less bulk storage (less WIP), more clever approaches to buffering, and more dependable scheduling and dispatch approaches. Automation (hardware and software) becomes more important as predictability of lot transport and processing become more important than flexibility in keeping the fabs running.
In the longer term, the desire to run smaller lots (for smaller runs of custom chips) may drive redesign of tool front ends (loading and unloading from chambers) and work to balance process times in order to allow faster small-lot cycle time through the fab without a loss in overall fab productivity (total output per unit time).
Conclusions
Cleanroom design has been a central part of semiconductor manufacturing over several generations of factories. Over time, it has changed from bay and chase, with high cleanliness and high human interaction, to ballrooms, with moderate cleanliness and almost no direct human interaction with the wafers. The trend now is toward larger fabs in more complicated shapes (multifloor and multibuilding) with product moving by ever more complex automation between “islands” of high cleanliness inside processing tools and inside clean (and possibly even purged) carriers. The rise of consumer applications wanting “good enough” performance with very fast lead times means that factory flow will become more important, with significant focus on reducing WIP levels and cycle time through more clever monitoring of tools, use of buffering and storage, and more effective automation.
Dr. Rex Wright is director of automation technology at Asyst Technologies, Inc. He studies AMHS systems engineering as well as the interoperability of various automation subsystems (FOUPs, load ports, AMHS cars, conveyors, stockers). Prior to joining Asyst, Dr. Wright worked at SEMATECH and at Applied Materials, managing projects on hardware and process-tool design and applications. He holds a PhD in Applied Physics from Stanford University.