Zilog Wafer Module III: A New Age Wafer Manufacturing Facility

		At a Glance
	An insider's look at how a modern wafer fab was built and equipped, and what impacted the many decisions that were made along the way, including the decision to use minienvironments.

Widely publicized "conventional wisdom" has it that a modern wafer fab must cost one or more billion dollars and is a massive site/footprint undertaking. We have never subscribed to that conventional wisdom and have always believed that one could build a small- to medium-sized fab for much less than a billion dollars and without the necessity of air conditioning a football field. Of course, it is one thing to believe that  this is possible, and it is something else to prove it in a real situation.

The proof lies in the Zilog wafer fab Module III, which has been shown to have an ultimate revenue capacity of ~$400 million. The investment in the fab to date has been $175 million. Our projections are that another $75 million will bring the facility to full capacity.

As a bit of historical perspective, Zilog's first fab, at its Nampa, Idaho, site, was a 4 in. NMOS facility, constructed and commissioned in 1979. With great originality, it was christened Module I. A second fab on that site, a 5 in. CMOS fab called Module II, was developed in subsequent years. These two fabs in combination provided Zilog with a revenue generation capacity of ~$200 million. In 1992, the company recognized that its revenue would grow beyond the capability of these original fabs, and planning began for the facility we will describe in this article.

Having been associated with the design and operation of ~20 semiconductor fabs going back as far as 1959, the writer has developed some strong opinions about how to create a reliable, predictable integrated circuit wafer manufacturing operation. Among his beliefs is that the wafers need to be kept as far away from the people as possible, the manufacturing process should require minimum interaction between the manufacturing personnel and the process applied to the wafer, and the environment in the fab should be as comfortable and convenient for the operating personnel as the state-of-the-art permits.

Fig. 1. Zilog used this building, which was originally constructed for the manufacture of electrofluoretic display panels but never completed, to house Mod III.

Fig. 2. A typical minienvironment installation is shown.

At the time consideration was being given to the development of the new plant, Zilog was using several foundries, including the Kawasaki fab at Utsunomiya, Japan; the company was very impressed with the yields being achieved in that facility. Accordingly, we spent time at the Utsunomiya fab, studying the concepts that had been incorporated there and considering whether those concepts should be incorporated in the development of the new Zilog fab. Among the features that were noted in the Kawasaki fab were that the wafers were transferred from bay to bay by a track system rather than by operating personnel, and the control of the equipment was done from a remote site rather than in the fab room itself. Other Japanese fabs we had visited in the period used on-floor robots to move the wafers. It was clear that the Japanese had as their objective the minimization of personnel traffic in the cleanroom.

In parallel with the consideration of the fab design concept, it was also necessary to make certain decisions about the building to house the facility. When the employees of Zilog and Warburg, Pincus Capital Corp., purchased Zilog from Exxon in 1989, an option was taken on another Exxon building located on an adjacent site. That building (Fig. 1) had been constructed for the manufacture of electroflouretic display panels and had never been completed. Zilog felt that the building might be attractive for future fab expansion because it had a basement with high ceilings to accommodate under-fab facilities as well as a good combination of on-slab and basement-under floor space. The company exercised its option to purchase the building in 1990 and held it "in inventory" against the possibility that it eventually would house a new manufacturing facility.

From the beginning of the project, the fab design team was managed by the people who would ultimately run the plant. Our first project leader was Rich Morley, who had been a manager of another of our fabs and who had many years of wafer manufacturing experience prior to that assignment. He selected a small group of associates to work with him in the development of the project concept. He was soon assisted by John Frank, who had been active in the SEMATECH operation and had a very good feel for the state-of-the-art 200 mm equipment.

As the Mod III project plan began to unfold in 1993, the planning group had to deal with several challenges. Outside architects concluded that the existing building might not be suitable for fab operation because of its location adjacent to a main east-west interstate highway; they proposed a new, ground-up construction on the site. As the budget for that alternative began to come together, however, it was clear that it would exceed the guidelines that we had set for the overall project.

Fig. 3. The layout of the fab, as originally designed.

Another dilemma had to do with the movement of the wafers interbay and  intrabay. We had been impressed with the conveyor concepts being used in the Japanese fab, but found that if we chose an American conveyor company to work with, we would be a pilot site. We did not believe we should take that risk.

Finally, there was a debate about 150 mm wafers vs. 200 mm wafers. The Japanese fabs were still adhering pretty much to the 150 mm standard at that time, but we saw the equipment evolution moving rapidly toward 200 mm. After evaluating and resolving these and other issues, the planning group did the following:

• Confirmed the selection of 200 mm as the wafer size
• Determined that the existing building could be reinforced at a relatively low cost, such that when appropriate vibration damping facilities were added, it would meet the vibration requirements down to 0.25 µm (250 nm) and below
• Selected an excellent set of outside consultants and subcontractors
• Most significantly, concluded that this fab would be among the first in ground-up construction in North America where the fundamental concept was "minienvironments"

By 1993, the industry had been experimenting for quite some time with the concept of maintaining the wafer cassettes in sealed enclosures (often designated as SMIF boxes) in the act of transport from one tool to another. The minienvironment concept goes one step further and houses each piece of equipment in its own temperature-controlled chamber, such that the wafers never see the environment of the greater fab area at all.

One of the reasons that we were enthusiastic about the minienvironment concept, even though there was not, at the time, a great deal of experience with it, was that it notionally eliminated the requirement for cleanroom gowning. Getting into head-to-toe, full cleanroom garb takes time -- and getting out of it takes time. Communication between people whose faces are hidden and whose voices are muffled is impeded. An engineer who might visit the fab floor and quickly help solve a problem, being human, thinks twice before he goes through the full gowning process. If we could eliminate the need for full gowning, we felt that we could greatly reduce processing errors, improve communications between the operations people and the engineering people and generally create an environment in which it was much more pleasant to work. The minienvironment concept seemed the answer to that desire.

The site

The Zilog Nampa, Idaho, facilities are situated on a 52 acre site, located ~15 miles west of downtown Boise.

One consideration in the planning for the project was due to allowance for construction limitations during inclement weather. The election to go with the existing building was biased, in part, by the desire to have the structure weather-tight and ready for internal installation as quickly as possible.

There are no serious ground water or drainage problems on the site, and the city water supply is acceptable. The only infrastructure problem we have experienced has been with the electrical power supply, which, in earlier years, was somewhat unreliable. Recently, with the addition of auxiliary feeds, an on-site power substation and uninterruptable power supply facilities, power supply continuity has become more acceptable.

The one major site challenge that had to be overcome was that of vibration from the adjacent highway. Some very good engineering work by "owner's advocate" Paul Oliphant of Applied Dynamics resulted in a redesign of the floor and foundation so that vibration-sensitive equipment sat on very stable pads. The concrete floors were decoupled from the rest of the structure, and "sensitive" sections of the floor were decoupled from other sections. Measurements taken after these modifications indicated that the fab could manufacture down to 0.25 µm dimensions and perhaps below. In addition, all sensitive equipment had additional shock-absorbing measures provided at the footprint.

Fig. 4. The layout as the fab expanded to its final capacity of 2500 wafers out per week.

The building itself was reinforced by raising the roof ~6 ft and installing clear span trusses. This not only strengthened the structure in the lateral direction but also provided more headroom for certain air handling equipment that is located on a balcony at the rear of the building. The building reinforcement process took approximately three months and was  completed by May 1994; internal building construction was under way by June 1994.

Fab concept: minienvironments

The writer has been an advocate of "peopleless" wafer manufacturing for many years. The idea of keeping people out of the wafer manufacturing process does not come from any lack of appreciation for the human being, but rather from the realization that high-yield wafer manufacturing is naturally benefited by total automation and that no matter how clean the surrounding air, human beings tend to deposit defects on the wafer whenever they are in close proximity.

The SMIF box solved the problem of keeping the wafers clean in transportation from machine to machine, but the equipment itself, circa 1993, was still largely open to the environment. Wafers coming out of the SMIF box would soon become dirty from the impurities that had collected on the machines and were subsequently deposited on the wafer.

The minienvironment concept, which had been promoted by vendors such as Meissner & Wurst and Asyst, neatly solves the problem of keeping the wafer clean as it passes through the equipment. In the minienvironment fab, each piece of processing equipment is enclosed in a floor-to-ceiling plastic, glass and metal chamber that is fed with highly filtered air at the top and exhausted from the bottom. Where precise temperature and humidity control is required, that shroud can also become an environmental chamber. Air locks on the side of the shroud permit the SMIF pod to dock with the minienvironment so that the wafers are never exposed to the general fab environment. A typical minienvironment installation is shown in Figure 2.

In the summer of 1993 when the minienvironment decision was being made, there were not too many working examples of the concept that we could visit and review. A domestic minienvironment installation at that time was the NCR fab at Colorado Springs, Colo. However, this fab was a conversion where the minienvironments had been used to upgrade the fab rather than a ground-up construction, which was our objective. Other installations using minienvironments at the time were IBM in East Fishkill, N.Y., and TSMC in Taiwan.

The hard concerns that had to be dealt with were the following:

• Would the minienvironments, when shrouding equipment that had not been specifically designed for their use, create any unanticipated problems?
• Could the temperature, humidity and particle count objectives be achieved for each tool in a real working environment?
• Did the minienvironments reduce the utilization of the floor space?
• Was the software available to permit successful integration of the automated load/unload robots with the processing tools?

There was also the "soft" concern. By going to the minienvironment concept when the rest of the industry still thought in terms of air conditioning football fields, were we being too "politically incorrect"? By that, we mean would our customers or process engineers be frightened by what might seem to them to be a radically new concept?

Fortunately, we were permitted to visit both the NCR and the TSMC facilities and were encouraged by what we saw and heard. While the application of minienvironments in the NCR facility had been constrained by the fact that it was a conversion rather than a ground-up construction, those close to the facility reported success. We also had data on single tool tests of the concept that was very encouraging. The decisive event, at least insofar as this writer is concerned, was a minienvironment demonstration in the middle of a very active construction area with a sensitive particle count monitor with two sensors on it, one inside the minienvironment chamber and one outside. The sensor inside was totally quiet, whereas the sensor on the outside was merrily ticking away.

After examination of the alternatives, Zilog selected Asyst Corp. to manufacture and install the minienvironments in the new module. Asyst set up a shop in a portion of the building adjacent to the fab itself and created the chambers to fit the equipment that had been selected for the fab. The air supply to the general area of the fab has been shown to meet Federal Standard 209E at Class 10, with temperature held at 71°F at better than ±1.5°F (action limits) and humidity controlled at 40% ±4% (action limits). The specifications on the conditions inside the minienvironment chambers with the tools operating have been measured to be better than Federal Standard 209E, Class 0.1, with the most sensitive equipment having temperature control at 22.5°C ±0.1°C and humidity control at 40% ±1.0%.

Table 1. Original Mod III Tool Set
Equipment	Application
Canon 2500 I2	Step align
Canon 3000 I4	Step align
DNS Track 80B	Interface photo
DNS Track 80B (pix)	Pix track
Fusion PCU200 DUV	Photo stabilizer
KLA 5011 overlay	Alignment
Hitachi SEM 8840	CDs
Hitachi SEM 6780H	CDs
Hitachi SEM 7280H	CDs
Lam 4428 poly/nit	Etch poly/nit
Lam 4528I	Etch oxide
Lam 9608SE	Etch metal
SemiTool SST (solvent strip)	Strip photoresist
Fusion MCU 200	Strip photoresist
Fusion ACU 200	Strip photoresist
FSI Mercury MP (resist strip)	Strip photoresist
TEL vertical diffusion 858S	Drives/oxides anneal
DNS prediffusion WD WS820	Wet clean
DNS nitride strip WD WS820	Wet strip
FSI Mercury MP (cleans)	Cleans all
MDC CV plotter 1094	Measure diff
SDI SPV 1020	Measure diff
Rudolph ellipsometer FE IV	Measure diff
Eaton implant GSD	Implant
Eaton implant HE	High energy
Thermawave TP 400	Measure implant
Prometrics NC 110	Measure implant
AST RTP 2800	Rapid thermal anneal
Novellus Concept One	Dielectric dep
DNS 80A-SOG	SOG spin/bake
AME Endural metal dep	Al/tin dep
DNS premetal WD WS820	Metal preclean
Tencor UV1050	Measure thin film
Tencor SM300	Measure thin film
FTIR Biorad QS 408M	Measure dopant
KLA 2131 defect analysis	Defect density
HP 4062 - ETEST	PEVAL

The layout

An objective for the floor plan of Mod III was that one be able to stand in one corner and see all the way through the fab to the opposite end. That clearly ruled out the conventional aisle and chase concept and was another factor in favor of the minienvironment approach. Because of the high ceilings in the facility, the air supply to the minienvironments, as well as the fab itself, could come from above. However, the electricity, gases, chemicals and effluents needed to be routed below. For the part of the fab that was over the full basement, this did not present a problem; but, for the portion of the fab that was over concrete slab, it was necessary to create subfloor pathways under a raised floor to carry the ingress and egress facilities.

Figure 3 shows the layout of the fab as originally designed, and Figure 4 shows the layout as the fab being expanded to its final capacity of 2500 wafers out per week. The corridors marked "viewing" are designed for visitor viewing and have large windows into the fab. The corridor marked "equipment corridor" is de-signed for moving equipment in and out; that corridor also has large windows into the fab facility. We did not quite accomplish the objective of being able to see from one corner of the fab to the other because of certain high-profile tools that get in the way, but within the various sections, visibility is excellent.

Equipment selection

Selection of the tool set for a fab is usually biased by a combination of factors:

• Wafer size
• Process flexibility
• Previous experience with vendor
• Maturity of the equipment being selected
• Footprint
• Compatibility with the company's other fabs

In the case of Zilog Mod III, the decision had been made to go to 200 mm, which meant that we were looking for a tool set that had some degree of maturity at that wafer size. Zilog is an applications specific standard product manufacturer and specializes in the superintegration concept. That is, our designs are created by computer association of previously proven cores and cells. Since we do not want to go back and redo our entire core and cell library with each new generation of technology, all our products must be made in process flows that are subsets or supersets of a standard set of unit processes. Since we also import cores and cells under license from many sources, our subsets and supersets are designed to be as "industry standard" as possible. While "microns" are important to Zilog, they are often not the sole determinant of either the cost or the success of our business.

Those with fab experience know how process regimes tend to multiply. Historically, too many processes in a fab confuse the manufacturing personnel and impede their learning curve. Fortunately, we are heading for the day when the processing personnel have much less to do with the on-floor specification of the unit processes than was the case in the past, and the issue of people learning curves will be diminished over time by automatic recipe download, but that is another article.

In making our equipment selection, we had in mind a module startup with a single process at 0.8 µm (80 nm), with a conversion immediately thereafter to 0.6 µm (60 nm) and then future conversions to 0.35 µm (350 nm) and 0.25 µm. We also had in mind no more than two or three flows in the module at any point in time. No surprise, but what actually happened has differed a bit from those original objectives.

There was one other factor in the equipment selection for the module that is typical of our approach to fabs. We always prefer to put in the floor space and support facilities for the ultimate production capacity, give the tools vendors conditional letters of intent for the complete build-out and release equipment only as the demand for the capacity evolves. That approach accomplishes several objectives. Even within a tool generation, improvements are being made serial number to serial number so that we have the advantage of the most recent improvement as we release the next tool in a series. Further, we are not carrying the depreciation on unloaded tools too far ahead of the production/revenue demand. Finally, if it turns out from experience with the first set of tools that we have made a particular mistake, we are not too deeply committed.

Historically, the negative to this approach has been that the installation of additional tools in the fab during ongoing production damages the fab environment and interferes with the production flow. Mod III, by design, alleviated that problem. First of all, since we were using minienvironments, new tools could be moved in and installed without impacting the manufacturing environment, since each tool is encapsulated in its own individual chamber. In addition, a tool preparation/decontamination room was set up outside the fab at the end of the wide access corridor to permit us to move the tool from the preparation room to its place in the fab very quickly, without significant interference with the existing fab operations.

The strategy for tool installation in the fab provided an initial capacity of 350 200 mm wafers out per week, which, in phases, would be expanded to 1550 wafers out per week. It was later determined that with certain alterations to the fab architecture (now already in place), the ultimate capacity of the installation can be 2500 wafers out per week. Table 1 shows the initial tool set for Mod III (unit quantities not given, since, as noted, the number of tools of a particular type varied over time). Additional equipment that was installed in the process of bringing the facility to 0.35 µm and adding mixed signal capabilities includes tools for CMP oxide, CMP W, CVD W (Novellus), CVD TiN (Novellus) and WSi_x deposition.

Fig. 5. The wafer output buildup achieved by the Mod III start-up team.

The relationship between the semiconductor manufacturing industry and the semiconductor equipment industry is extremely close: "The tool is the process." The writer is often amused by those outside the industry who go around wringing their hands over the question, "How is company so-and-so going to get to such-and-such microns?" Those in the industry know the obvious answer, which is, "Just purchase such-and-such micron tools."

The following is not meant to be a commercial for the manufacturers of the particular equipment selected for Mod III, since we have no way of knowing that some other selection might not have been more optimum. However, we can say that contrary to our experience with certain previous fabs, we have had very few regrets. The most serious challenges in bringing the tool set to full operation had more to do with integration of the software than the reliability and performance of the hardware.

In contracting for the equipment set, we did what are probably now considered to be "all the usual things." The vendors were required to first demonstrate a certain level of performance with the tool on their floor and then a certain level of performance on our floor after it was installed. Complete spare parts kits (both consumables and non-consumables) had to be maintained on our site. Gas, chemical and environmental conditions were specified in detail.

The timeline

Table 2 shows the timeline for the Mod III project from initiation to the level of 1000 production outs per week. Only 12 months elapsed from the time when the internal building construction was initiated and the pilot production began. While this may not be a record, it was evidence of the fact that the planning and execution on the project was very good.

Table 2. Mod III Fab Timeline
Planning group Formed Dec 1992	Visits to Japanese fab Feb 1993	Preliminary concepts,"New building" Mar 1993	Refinement of concept to include minienvironments Aug 1993	Board approval of appropriation request Sep 1993	Detailed project planning with contractors Dec 1993
Reinforcement of building completed May 1994	Startup process more specifically documented Jun 1994	Internal building construction under way Jun 1994	Major building facilities being installed Jun 1994	Equipment begins to arrive Oct 1994	Equipment installation begins in the fab Nov 1994
At least one of each equipment set installed May 1995	Pilot production begins Jun 1995	Module dedication and press event Jun 1995	Full production begins Jul 1995	Output exceeds 1000 wafers/week May 1996
Management
Rich Morley 12-7-92 to 4-13-94 Ed McBain 6-26-95 to 11-12-96 John Frank 4-13-94 to 7-16-95 Rick White 11-12-96 to present

In retrospect, we wish we had had full production from Mod III six months earlier than it occurred, since, as some will recall, 1995 was a year of capacity shortage in the industry. We probably lost three months in the project in 1993 when we transitioned through the false starts noted earlier in the article.

The major contractors

Fig. 6. An internal view of Mod III.

Contrary to what may be the case with larger companies, Zilog does not maintain its own plant construction department. Rather, the company sets up a small (five to eight people) internal project management team and works through carefully selected architects, consultants and contractors. It would be impossible to list all of the vendors who contributed to Mod III. Some of the major contributors were the following:

• Applied Dynamics -- owner's advocate
• Keller & Gannon -- architectural and engineering
• Marshall Contractors -- construction
• Tri-State Electric -- electrical supply and distribution
• Praxair and CMPA -- specialty gas piping systems and tool hookup
• Scott -- mech  -anical piping and HVAC
• Flanders -- cleanroom grid and filters
• Systems Chemistry -- chemical distribution design
• Harder Mech-anical -- phase II wet and dry side
• Don Yeaman -- layout simulation

Again, our pur-pose here is not to write a commercial for these vendors but to indicate the major functional areas that were covered in the development of the facility.

The start-up and start-up team

Having been involved in the "start-up" of a number of fabs in his industrial lifetime, the writer has some well-formed opinions about the right and wrong way to do things. First, the start-up team should be small and, individually, highly experienced. The diffusion engineer should be among the best diffusion engineers that you have. The implant engineer should be among the best implant engineers that you have. The alignment tool engineer should be among the best alignment tool engineers that you have, and so forth.

Assuming that you are starting the fab up on a 24-hour-a-day, seven-day-a-week schedule, the four or five shift forepeople, who are the real floor management of the fab, should be well seasoned. The manufacturing manager (or general foreman, as we used to call the position) must be a pro. You do not want a lot of operators on the floor in the start-up phases, since too many to manage diverts the attention of the manufacturing and engineering personnel from the many issues that they will face. It helps if you can select five- and 10-year operatives with good track records to be part of the start-up cadre.

One area where you probably cannot overkill is in tool maintenance. It makes no difference how much the tool has been run in on the vendor's floor, and subsequently, on the fab floor, it is going to exhibit early life breakdown. Industry practice, of course, has been to require that the vendor have maintenance personnel on-site to train and troubleshoot, but in periods when there is a boom in fab startup activity, the skill and stability of vendor-supplied on-site personnel can be highly variable.

Figure 5 shows the wafer output buildup that the Zilog Mod III Start Up Team achieved. Well-trained and certified operatives were fed into the facility on a measured basis. Off-shift technical and engineering personnel were added, and the floor foreperson cadre was doubled in the process.

Of course, talented people can also create their own chaos if they are not properly managed. Obviously, the precise recipes used in the unit processes in the new fab will not be the same as was the experience of individual team members in their previous employment (whether at the same company or elsewhere), and stable high-yield chip manufacturing is, if nothing else, totally unforgiving of "instant ingenuity" (or "knob twiddling" as it has also been called).

Zilog has developed a fab lot processing management system we call the "Dispatch Bench." It is designed to help the fab operator avoid processing mistakes. In addition to having the usual on-floor process documentation, the operator receives a unit process run card for each step to be performed. When the operator selects the next lot indicated for processing in his or her area, a bar code swipe provides a ticket that tabulates in detail the specific steps to be performed. After the operator performs those steps, another bar code swipe brings questions on a computer screen that the operator must answer successfully in order that the lot be cleared to move to the next step. By appropriate use of this concept, the mistakes that can be fatal to high-yield wafer processing are very much reduced. Bottom line, the Dispatch Bench concept helped us bring Mod III up the experience curve with very few missteps.

Results

There are, of course, a variety of metrics for measuring the performance of a wafer fab. The meaning of these factors is somewhat dependent upon the fab mission. For example, the weighting of the performance factors of a fab producing one or a few products with a single process (a DRAM fab, for example) might be different from that of a fab such as Mod III, which is running a half a dozen processes and several dozen products. We provide the following data not so much to create a competitive comparison as to show that the concepts and execution inherent in the Zilog Mod III project resulted in very good performance.

Through the four quarters of 1997, the average process yield (good wafers in a lot seriatum out over good wafers in those lots started) averaged 98% with individual weeks where the yield was 99% or better (in Zilog, an output wafer is counted as "good" only if it passes post-fab process evaluation on the scribe line test sites). The average cycle time (lot released to the fab to lot released from process evaluation) was 18 days. The yield model Y = Ke - da is commonly employed by the company to measure and predict good probe yield per wafer. In the 0.8 µm process, Mod III typically operates with K = 0.97 and d = 0.58. In the 0.6 µm process, typical values are K = 0.99 and d = 0.97. A recent company press release announced achievement of the first 100% yielding 200 mm wafer from the Mod III facility. This was significant because of the size of the die. Each of the 378 die measured 110,000 square mils and was free from particles or any other pattern defects.

We have consistently shown in this facility that having people in the module in normal work dress has absolutely no impact on fab yield because of the use of the minienvironments. Gowning is not necessary other than to the degree it provides an attitudinal bias. The maximum that is required in Mod III is a simple head covering, a lab coat and fab shoes. Personnel can enter and exit the facility without elaborate dress/undress activities, and since faces are open, communications are as convenient as they are in an office environment. Because of the open design of the module, the ceiling height and the elimination of the bay and chase concept, the room is quiet and free from some of the air flow noise that sometimes interferes with easy communication in conventional fabs. Overall, the fab personnel feel that the environment that has been created makes it "a very nice place to work." Figure 6 shows an internal view of Mod III.

Conclusion

We believe that the success of the Zilog Module III project demonstrated a number of things:

• Fabs do not have to cost a billion dollars or more
• A small project team working with skilled contractors can accomplish significant results in a reasonably short period
• Minienvironments work and work well
• The "drag" of complete cleanroom gowning will become a thing of the past
• With the use of minienvironments, the confining strictures of the bay and chase concept are no longer applicable
• The efficiency of the SMIF/minienvironment concept can significantly reduce the amount of labor required for the fab to achieve a particular output

In interviewing those who were active both in the development of Mod III and in its operation today, the writer has asked repeatedly, "What should we have done differently than we did?" The responses result in a remarkably short list, and the defects are not difficult to correct.

Fab personnel feel that not enough attention was given to places to stage work in process ahead of the tools. Of course, we could argue in return that not having places to build inventory is an asset with regard to throughput time. Fab maintenance personnel feel that the facilities pathways under the fab are becoming crowded, making it difficult to hook up and service new tools. We certainly agree with that and would suggest that these areas be larger (more people-friendly) in the future. On some tools, it is necessary to shut down the tool in order to change a point-of-use filter; that problem can be solved with a dual point-of-use filter arrangement and appropriate valving.

Personal environmental safety and orderly tool bring-down power backup was provided in the original design. We should have gone further and provided for a full-operating backup to the main power grid, and plans are under way to install the requisite additional generation facilities in the future.

We are told by those who have worldwide visibility in this area that there are a number of fabs that were brought on-line in the early 1990s that are now facing the need for upgrade, either from 150 mm to 200 mm wafer size or from one generation of processing capability to another. Given our Mod III experience, we would certainly recommend that the minienvironment concept be carefully evaluated in the process of considering these upgrades.

Acknowledgments

The list of people and organizations who contributed to the success of a project is very long. I am indebted to Richard Morley, now at Paradigm Technologies; John Frank, now at Meissner & Wurst; Jane Sinclair, director of accounting at Zilog; Mihir Parikh, CEO of Asyst, and his associates, Rick White, Mod III manager, and the Zilog Mod III crew; and Paul Oliphant of Applied Dynamics for patiently supplying me with its detailed recollections of the project. Mike Bradshaw, Zilog's senior vice president of operations, was the senior manager for the project and remains responsible for the module today, along with his other responsibilities. I also want to thank Theresa Hedger of my Zilog staff for assisting me in the preparation of this article.

Dr. Edgar A. Sack, CEO (retired) of Zilog Inc., is a veteran of the IC industry. He received a doctorate in electrical engineering from Carnegie Mellon University in 1954. E-mail: edsack@pacbell.net