Managing
05/01/2001
Paul Niemyski, Texas Instruments, Dallas, Texas
overview
Texas Instruments is nearly on track with its goals of being the "second or third IC manufacturer to ramp up to 300mm-wafer fabrication." Its methodology in dealing with the unique demands of the jump from 200mm to 300mm - for example the inability to copy exact, the limited supply of 300mm wafers for process evaluation, and the need to push supplier's development of 300mm tools - is an interesting lesson in fab management for the industry.
Texas Instruments is equipping its first 300mm-wafer fab, DMOS6, in Dallas. Originally designed to be a 200mm ballroom-style fab, the shell was constructed in 1996. This is a "megafab" with a 118,000 ft2 "waffle table" (i.e., the area in the main ballroom comprising the cleanroom) and 25,000 ft2 of cleanroom off the waffle table. The fab's original design was three floors, open ballroom, and Class 1. This original design concept was derived when the plan for DMOS6 was for 200mm equipment using open cassettes and manufacturing devices down to 0.25µm design rules. The three-floor wafer fab design became the standard for Texas Instruments in the 1990s, a period when the company was building wafer fabs for memory ICs overseas and fabs for logic in the US.
During the construction of DMOS6, the semiconductor industry began to discuss and plan a move from 200mm to 300mm wafers. The was during the sweet spot in the semiconductor cycle and we were all looking at ways to increase production capacity while continuing to improve productivity and reduce cost. During this period, the industry began to struggle with the concept of the 300mm-wafer transition; the operative question was "who would bear the cost of this wafer size conversion?"
 Copper metal module at DMOS6 facility at Texas Instruments in Dallas, TX. |
Previously, Intel led the conversion to 150mm and IBM led the 200mm conversion both with the associated financial and technical pains. TI was one of late adopters of 200mm technology, building many of the last 150mm facilities in the world. The company did not want to be in the same situation with 200mm, but the decision was made near the end of the construction of DMOS6 to wait for 300mm. At the time of that decision, 300mm technology was targeted for introduction in late 1997 at 0.25µm aluminum-interconnect technology. We waited to see who would be first; seemingly along with the rest of the industry, our strategy was to be the second or third manufacturer to transition to 300mm. Along with 13 other major IDMs and foundries, our participation in the International 300mm Initiative (I300I) and cooperation with other IC manufacturers were significant parts of our early work.
300mm-development team
TI's 300mm-development team was initially formed as a centralized equipment organization called the Manufacturing Science and Technology Center (MSTC). MSTC had the responsibility for equipment evaluation, specification, and selection. This team eventually transitioned to a manufacturing oriented team; a group of about 16 equipment, process, and manufacturing engineers came together to complete the equipment selection activities started by MSTC that were now specific to DMOS6.
Process equipment evaluations and eventual selection was the primary task of this group of engineers. The team was also responsible for modeling the factory, supporting interactions with I300I, and participating on standards committees. Key concerns for the team, besides equipment maturity or readiness, included understanding and planning for operations in a front opening unified pod (FOUP) wafer carrier environment, nonproduct-wafer reduction or elimination, and factory capacity modeling.
The team had the responsibility for selecting the equipment for processing 180nm aluminum and 130nm copper interconnect technology. The team planned on obtaining much of the data for 300mm equipment selection from I300I equipment evaluations. I300I had very extensive test methodology, but there were some issues that led us to question whether we could count on I300I completely for equipment evaluation data. First, the data being collected did not address the process technology that TI would be using to start up DMOS6. In addition, I300I only planned to evaluate a limited range of equipment. Finally, with the semiconductor industry downturn in 1997-1998, some equipment suppliers significantly reduced or even stopped 300mm development efforts. So, I300I had a very difficult time just getting suppliers to complete committed evaluations, much less sign up for new ones.
At TI, our 300mm-evaluation team was challenged to collect all the data I300I did not, and faced challenges similar to those that I300I faced in terms of gaining supplier commitment to do evaluations for the team.
Equipment demonstration methodology
When we started our evaluations, the 300mm team's strategy, addressing the limited resources outlined above, concentrated on getting equipment suppliers to focus on demonstrating their tool sets by doing short loop integrations of our existing process flows at 200mm; this included processing wafers between suppliers with TI engineers managing all the logistics of wafer transport. We did not directly evaluate unit processes (i.e., a discrete process within a flow, such as gate polysilicon etch, for example), but we did evaluate how supplier-developed unit processes matched our integration needs. We did this with an internal DMOS6 project called Virtual Wafer Fab (VWF). By using our access to International Sematech's Advanced Technology Demonstration Facility (ATDF) and our supplier's application labs, we were able to demonstrate short flow integration of our existing 200mm processes in terms of mechanical operations and electrical results.
Similar to many IDMs in the industry, TI uses a "copy exact" manufacturing strategy for moving processes from development into production facilities. Copy exact posed a particular challenge to the TI 300mm team since it was technically impossible to copy exact to 300mm tools with development being done at 200mm. To minimize the risk, we went to our equipment suppliers and asked them to copy our 200mm processes onto their 300mm platforms. While we could not copy the 200mm processes and equipment "exactly," we could at least copy the process parameters themselves or "process copy exact." With process copy exact we were copying materials, methods, and process parameters, everything except the exact piece of equipment.
Process copy exact brought with it several advantages. First, it allowed us to test at least two different suppliers at a given process operation. This gave us the ability to competitively bid for the equipment. We told suppliers that they had an equal chance for our business and that we would help make them viable for selection through VWF integration testing. Second, it gave us the insight into a given supplier's readiness by testing our existing 200mm process on the 300mm tools prior to purchase.
During the evaluation phase, the TI 300mm team had access to a very limited amount of quality 300mm silicon wafers. The quality and availability of silicon required to do the evaluations that we planned posed a unique challenge to the evaluation team. In addition, the evaluation team wanted to focus on integration and not on unit process development given the size of the team and the value of the data to be collected.
For example, the team spent a significant amount of time evaluating process tools maturity to maximize the probability of first or second pass success. To insure sufficient maturity, we evaluated equipment using a modified version of the Sematech equipment maturity assessment (EMA); we modified EMA so that it gave a numerical score (the Sematech version scores a system as alpha, beta, or gamma). Our modified EMA, which we dubbed TIMA, rated equipment with a score from 0-100 with 0-50 labeled "alpha," 50-75 "beta," and 75-100 "production capable."
We evaluated hardware, process, software maturity, and standards compliance. Our expectations dictated that all equipment needed to score as beta and greater than 80% needed to score production capable for us to begin our integration work. Our equipment set reached this maturity level in early 1999.
 Sputter system during installation at DMOS6. |
To minimize the use of silicon, we focused on using what silicon we had available only for integration tests. We encouraged suppliers to demonstrate baseline capability with their silicon or we would not commit our integration silicon to evaluating their process. Most all data was shared with the supplier from our integration work. Here we wanted to insure that suppliers had visibility into their process issues so they could remedy them prior to our selection process. For each major process step, we wanted two technically capable suppliers so we would have them in a competitive bidding situation when the equipment was finally selected.
Our team focused on four major process "loops" in the integration of a device: the isolation, the transistor, the gate salicide, and the interconnect loops.
With the interconnect loop, we focused solely on copper integration, copying a process from our copper development facility. The decision to run aluminum in DMOS6 came later and, closer to start-up, we also tested that loop.
As our work progressed, each loop was completed and electrically tested. However, we did not complete the transistor loop, which had the longest cycle time, because while it was in progress we were given the go-ahead to start up our 300mm fab. From this point on, other process development work took priority at our suppliers. The isolation loop could not be tested electrically, but both the salicide and the interconnect loops were, and we found that parametrics matched those with 200mm processing. This was critical data needed for the go-ahead decision for DMOS6.
Equipment selection methodology
Data from our VWF was critical in the decision-making process for selection of the tools for DMOS6. This data gave us the confidence we needed to proceed with the completion of the fab and the start-up of production.
 Overhead intrabay delivery track at DMOS6 |
To analyze the data from over 100 individual process and metrology tools, the 300mm team used an evaluation tool called the equipment capability comparison (ECC) matrix, which had been developed by MSTC. This matrix evaluates four critical areas of interest in equipment selection: process capability, equipment reliability and productivity, compliance with Semi and I300I standards and guidelines, and, finally, cost of ownership. The ECC matrix takes inputs - such as specifications and actual performance - and, using Taguchi loss functions, compares the results. The matrix gives a supplier's equipment a score from 1 to 10 for a particular parameter. Equipment suppliers with the highest score oftentimes won the business for DMOS6, but not always.
ECC data was ultimately used in technical negotiations with suppliers to reach commitments on their part regarding overall performance of the equipment. Marketing or lab data collected during VWF development rolled into the ECC and then into technical negotiations. Since all suppliers were in a competitive situation for our business, data showing they could commit to specifications greatly improved commitments that we attained from them in negotiations.
Cost of ownership (COO) played a critical role in our decision-making process given the fact that TI, as well as most other IDMs, was making the move to 300mm for strictly productivity reasons.
If gains of 30% or more in productivity were not achieved, the move to 300mm would not be worth the effort, risk, or cost. The 300mm team used Two Cool COO software from Wright Williams and Kelly to calculate total COO for equipment. Model information came as direct inputs from equipment demonstrations out of VWF and I300I evaluations, supplier technical negotiations, and supplier commercial negotiations.
Copy exact was the final consideration in our selection of equipment, at least in terms of the supplier and tool design. All other factors being equal, if an incumbent 200mm supplier was not beaten in an ECC evaluation, a draw was decided in their favor. We felt this would help minimize some transfer issues given the fact that, simply stated, a number of the "knobs" on the tools were the same. The scaling of a process from 200mm to 300mm would be easier and problems could be addressed by our 300mm and 200mm process engineers.
DMOS6 start-up strategy
TI's 300mm team was given the go-ahead to start up the 300mm fab in late 1999. The plan then was to start with the current 180nm-aluminum process, targeting qualification in the third quarter of 2001. This would allow us to shake out the complete tool set on a proven technology node and give us needed capacity for product demands at that node. We also would begin a nearly concurrent start-up of copper qualification in the fab, following one to two quarters behind aluminum qualification, which is now targeted for qualification by the end of 2001.
Given the criticality to TI of a successful start-up of DMOS6, staffing for the facility has been a top priority. Starting with the best and brightest from within TI, we have added new college hires and experienced external candidates selected from across many fields, including manufacturing, equipment engineering, process engineering, product engineering, finance, training, automation, and HR. High performance team training and development have been major activities of the fab's leadership team.
Since DMOS6 is a production facility, it has been commissioned to run high volume from the start. All the actions taken by the 300mm team have been geared toward rapid qualification of the wafer fab based upon lessons learned with 200mm processing and by copying 200mm processes. The layout of the fab has been created with full production capability in mind; we did not begin with a pilot line layout with the idea of transitioning to a production layout. Systems infrastructure, process specification, specification formats, and control methodology are all copied intact if not nearly intact from existing TI facilities.
Also, to assist in rapid start-up and qualification of our 300mm-wafer fab, we again exercised our VWF concept to support production of qualification pilots for the start-up of equipment. In past 200mm start-ups, TI has had access to either joint venture or wholly owned 200mm wafer fabs to produce pilots necessary to start up and release production tools. This was not the case with 300mm. We had no other source of support to start up the 300mm fab other than from our equipment suppliers and Sematech, which supported our lithography development needs.
As part of our purchase agreements with suppliers, we required that they support the production of our qualification silicon for start up. In a typical 200mm start-up, this number can be in excess of 10,000 wafers. For DMOS6, this amounted to over 8000 wafers total for two technologies - aluminum and copper. Using this method, DMOS6 saved over three months in start-up time, not having to build the patterned and nonpatterned pilots in house as the tools were coming on line. As tools were released to process engineering, pilots required to install, release, and qualify tools came from supplier sites. Having to build these wafers in DMOS6 during the start-up would have required additional time. The tactic was to build the wafers at supplier sites and have them staged in front of a tool when required for tool qualification.
The fab is now in start-up mode and is in the process of qualifying the aluminum-process tool set. The fab will shortly begin qualifying the copper-process tool set. By the time you read this article, we will have evaluated our initial integration and yield data and will have started qualification lots with a target completion in late September.
Summary
The era of 300mm-wafer fabrication is now at TI. The fab is currently in the middle of process start-up. Significant engineering evaluations will continue throughout most of 2001 and production ramp of the facility will take place going forward past 2002. This fab will have TI's highest volume (in terms of die), lowest cost/die, and most advanced technology. TI's business strategy is for rapid movement of products to new technologies to address our continuous strive toward manufacturing and technology excellence. DMOS6 is another example of TI's commitment to one of its core competencies: manufacturing.
Paul Niemyski received his BS in materials science from North Carolina State University and MS in engineering science from RPI. He has worked at IBM and TI. Niemyski is process engineering manager for DMOS6 at Texas Instruments, P.O. Box 655303, M/S 3786, Dallas, TX 75265; email [email protected].