Using SMIF technology to help retrofit a high-volume fab

John Ort and Roy Hunter, Texas Instruments; and Chris Humphreys, Asyst Technologies

The integration of SMIF components with 200-mm tools enabled a fab to move to a new process quickly with minimal disruption of its production capacity.

The rapidly increasing demand for digital signal processors has presented the semiconductor industry with the challenge of increasing fab capabilities without disrupting current production capacity. One solution is to retrofit an existing facility so that it can handle larger substrates and smaller device geometries but to keep parts of the fab running during the upgrading project. In the case of Texas Instruments, for example, which has had an accelerated technology development schedule in place for five years, one area where efforts have been focused is converting the firm's 150-mm wafer capacity to 200 mm. Achieving faster ramp-ups and better equipment utilization are complementary goals.1 As part of this program, the DMOS 4 fab in Dallas was recently retooled to run a sub-half-micron process on 200-mm wafers.

To minimize the cost and impact of the project, a novel approach was required. After careful consideration of the options, TI chose to use standard mechanical interface (SMIF) technology as the means to upgrade the fab's facilities, install additional tools, and qualify new processes without sacrificing fab output. This article discusses the reasons why SMIF technology was selected, the company's expectations regarding the project's outcome, and the steps taken to minimize any adverse effects related to the adoption of SMIF systems. The installation and qualification processes are also described, along with project results.

Project Background

Before the retrofit, the DMOS 4 fab was a Class 5 facility running a high volume of 150-mm (6-in.) wafers. Built in the mid-1980s, the 60,000-sq-ft building was designed as a combination of bay/chase and room/chase areas. Process area flooring was raised on a waffle slab, and there were numerous small fan towers on the fab periphery to recirculate the air. No stockers, automated material-handling systems (AMHS), or sorters were used; all wafers were stored and transported in carriers. Bar codes were used for lot ID in conjunction with tool controllers that were based on Autoshell and routed through an in-house protocol converter system and the in-house manufacturing execution system (MES).

The goal of the retooling project was to qualify a leading-edge process on 200-mm (8-in.) substrates within 9 months, while keeping some existing production tools operational. Select tools would be removed, other tools rearranged, and 200-mm pilot tools added. In addition, an AMHS would be installed and brought on-line, and facilities services would be upgraded. Once the new process was qualified, more 200-mm tools would be added and additional 150-mm tools would be removed. Ramp-up and ramp-down would be staged in parallel in order to minimize their effect on total fab capacity.

To initiate the project, management assembled a team comprising representatives from equipment engineering, process engineering, manufacturing, training, facilities, purchasing, and automation. The team reviewed various options and assessed them in terms of the dual project constraints of holding down costs and minimizing the impact on production. Equipment vendors were brought in for discussions of the options, and Anam Semiconductor, with which TI has a close relationship, was consulted regarding the benefits of and issues involved in running a SMIF-equipped fab. Company representatives also studied SMIF fabs in Germany, Oregon, and Colorado. In addition, a return-on-investment (ROI) analysis was done on the implementation of SMIF technology. This analysis factored in the costs associated with the addition of SMIF input/output (I/O) modules as well as the capital and operating cost savings that would accrue from reduced heating, ventilation, and air-conditioning (HVAC) requirements. As expected, the ROI analysis found that payback was closely tied to electrical yield, with payback time being inversely proportional to electrical yield improvement (see Figure 1).

 

Figure 1: ROI payback time versus die yield improvement for the implementation of SMIF technology. The arrow represents the targeted yield increase for the DMOS 4 fab.

After a thorough investigation, the team decided to adopt SMIF technology for the fab upgrade for a number of reasons related to the system's wafer isolation capability. It was thought that by using isolation the 200-mm process could be brought on-line very quickly because tool qualification could begin immediately following installation regardless of the room conditions. It was also assumed that because wafers would be isolated from the cleanroom ambient air, the fab's HVAC system could be modified and some components removed, allowing more routing options for the facility's infrastructure improvements. Similarly, the chase wall could be removed, allowing more layout flexibility. Another potential benefit of wafer isolation was that process and electrical yields would be expected to improve. Finally, the adoption of SMIF technology would provide a testing ground for many of the automation and isolation strategies that would be standard in future 300-mm fabs. Similar reasons for using SMIF technology have been reported by LSI Logic, which provided postinstallation data showing there was as much as a 10% yield improvement in a SMIF fab compared with a non-SMIF facility.²

Project Strategy

Once the team had decided on the approach, the more difficult tasks of planning and implementation began. While SMIF technology was being used in some TI facilities, the DMOS 4 fab would be the first to become a fully automated fab, employing all aspects of SMIF technology. Wafers would be moved and stored in SMIF pods, and process tools would all be equipped with SMIF I/O modules and enclosures with internal environments that are better than Class 1 (see Figure 2). Pods would be transported via the AMHS from stocker to stocker, and cassette-to-cassette wafer transfers would be done in sorters. To optimize lot movement, the AMHS and tools would be interfaced with the fab's MES. At each tool, the auto-ID tags attached to the SMIF wafer pods would be read, the product verified, and the recipe downloaded; the process would then begin automatically, requiring little interaction from the manufacturing associates. The single- and multireticle pods (containing up to six reticles) that would be used to transport and load reticles into the steppers also would carry auto-ID tags.

 

Figure 2: Comparative schematics of process tools with and without SMIF systems; the Class 1 isolation capabilities of SMIF components enable the use of lower-rated cleanrooms. (Arrows represent airflows.)

Several SMIF vendors were invited to a series of meetings with the project team to discuss their products' performance specifications. After an internal review, Asyst Technologies was chosen as the SMIF provider and the first OEM purchase agreement of the project was signed. The vendor was then brought in as part of the project team and worked on developing specifications for the enclosures, the SMIF I/O modules, and the tool automation systems. In anticipation of further upgrades, the specifications also included requirements for 300-mm bridge tools.

During the specification development process, a common terminology was agreed on to define process tools in terms of their SMIF system integration and automation levels. Although the vast majority of semiconductor process tools are readily available with integrated SMIF components or as SMIF-ready, SMIF technology has not been widely deployed in the north Texas area. Therefore, a common lexicon had to be established so that the local equipment vendor community understood what level of equipment integration was being discussed, and what the responsibilities of the various parties involved were in regards to system integration.

Process Tools. Approximately 90% of the tools required for the project were available from OEMs with integrated enclosures, and adaptive enclosures were ordered from the SMIF system provider for the other 10%. In all cases, the enclosures had integrated filter fan units and PTFE filters, which made them relatively independent of the facility's HVAC system. With regard to loadports, the tools fell into two basic categories: integrated and adaptive. Integrated tools were available from the OEM with an integral SMIF I/O module, whereas adaptive tools did not contain such a module but were purchased SMIF-ready; that is, they were equipped with appropriate interface software and modifications had been made to the frame and electronics to accommodate the attachment of an I/O module on-site. Approximately 55% of the tools required for the project were available with integrated SMIF I/O systems.

To work out details of the SMIF interface and automation strategies, the project team engaged in technical negotiations with the tool OEMs. For the adaptive tools, drawings were provided by the SMIF system supplier to clarify the interface requirements. In addition, for each first-of-a-kind tool, a fit-up of the SMIF loadport to the tool was performed at the tool vendor's site to ensure that there were no problems with the hardware or software interfaces between the module and the tool.

A specification regarding the semiconductor equipment communication standard (SECS) messages that had to be supported through the tools was part of the information discussed with the tool vendors. Compliance to this message standard was critical if the host automation system was to work efficiently with the equipment interface (EI) and auto-ID systems. The tool vendors were also queried with regard to their tools' means of communications: Could auto-ID tag information be accessed through the tool or would it require a separate communications port? Could the SMIF I/O module be controlled through a serial port on the host tool (single-wire EI) or would it require a separate control port (multiwire EI)? Based on the information supplied by the vendors, the equipment interfaces were written in accordance with the capabilities of the tools and the requirements of the fab. The majority of tools were classified as multiwire A (see Figure 3), a designation that indicates the SMIF interface is controlled by the host tool, but auto-ID communications are conducted through a separate wire linked to the fab's MES.

 

Figure 3: SMIF and auto-ID integration strategies. Most tools in this project are classified multiwire A.

The Auto-ID System. The project team evaluated two auto-ID systems, a radio-frequency (RF) tag-and-antenna system and a short-range infrared technology. The RF ID system, which has a 128-byte memory, has plates that are read and cross-referenced to a database housed on the fab's MES. The infrared system employs the Asyst SmartTag 8400, which uses an open SECS architecture and has a 192-Kb memory. Each tag includes an LCD that can be configured for up to eight lines of text. This latter capability was among the reasons TI choose the infrared system, believing that it would be easier for manufacturing personnel to switch from one type of visual lot indicator (a lot card) to another (the LCD screen). The system was also selected because of its ability to act as an interactive interface between manufacturing personnel and the fab's MES.

The chosen auto-ID tags attach to the SMIF pods via a tag bracket, in an easy-on, hard-off attachment, as seen in Figure 4. Each SMIF I/O module is equipped with an IR probe that enables the host automation system to upload information from the tags and to transmit data to them, including display, lot information, and pod information files. Display files are used to provide human-readable text on the LCD screen; lot information files contain machine-readable data such as lot ID, last process, and next process; and pod files hold information on pod segregation level, last pod clean, and next pod clean.

 

Figure 4: A 200-mm wafer SMIF pod with the auto-ID tag attached.

When a wafer lot is loaded from a pod into a tool, the lot information stored on the pod tag can be uploaded to the MES (directly or through the host tool) and an "empty pod" message sent to the tag. Then, when the lot is reloaded into a pod, the lot information is resent to that pod's tag. This retrieve-and-resend capability makes it unnecessary to dedicate each cassette to a specific pod; any cassette coming out of a tool can be loaded into any available pod. In practice, however, this feature has not been fully utilized and pod and cassette integrity have been closely maintained at the DMOS 4 fab. When lot information is retransmitted to a tag at the end of a process, the host tool must keep the reloaded pod locked to the I/O module plate until the information is received. After implementing the system, it was discovered that several tools could not support waiting for a pod-unlock command from MES and automatically unlocked a pod once the cassette was replaced and the pod was closed.

Project Management. Throughout the project, the core team members, who included representatives from the SMIF system vendor, met regularly to monitor such concerns as schedules, costs, barriers to success, and technology transfer. Risk analysis meetings were conducted to identify potential risks to the project and take measures to mitigate them. The risks were categorized and assigned a priority for resolution based on potential impact to the project's success. Any deviations from the "copy-exact" philosophy of specification compliance were scrutinized carefully. Training sessions and interactive discussions with engineering and manufacturing personnel were also held to define the new terminology and to prepare them for the upcoming changes.

One of the most challenging aspects of the project was understanding how the use of SMIF pods, enclosures, and auto-ID would affect manufacturing and engineering functions. Manufacturing personnel held meetings to work through the process module by module and determine what changes would be necessary. They then presented a list of questions regarding manufacturing strategies to the project team, which used information from an existing SMIF fab and the SMIF system provider, along with other benchmark data, to work out answers. Some of the concerns addressed were how to manage nonproduct wafers (calibration, test, dummy, and conditioning wafers), various types of product wafers (pilots, split lots, qualification wafers), segregation and pod cleaning, tool maintenance, and sorter use. A major difference between non-SMIF and SMIF fabs is the need for wafer sorters in a SMIF fab. Because pods are used, sorters are needed to insert and remove test wafers, pilot wafers, conditioning wafers, and so on, as well as for handling engineering splits and recombinations. When solutions to the manufacturing issues had been agreed on, the revised procedures were submitted to the computer-integrated manufacturing (CIM) department to develop the necessary automation strategies.

Installation, Qualification, and Ramp-Up

To make room for the first set of 200-mm tools, some 150-mm tools were removed and others rearranged. In addition, some administrative areas were converted to Class 100 space, including areas that were not on the waffle slab. Some recirculating air handlers also were removed and those production areas that would no longer contain open cassettes were downgraded from Class 5 to Class 100. In those Class 5 areas where 150-mm processing was continuing, however, construction protocols were followed closely to minimize any yield impact.

To help maintain the project's accelerated installation schedule and prevent any unnecessary delays, the SMIF vendor assigned a team to the fab and installed a cut shop so that minor modifications to the enclosures could be done on-site. This shop was also used to build some specialty items quickly. Prior to tool delivery, a testing lab was set up for the fab's CIM group to begin work on the automation scheme. The lab included SMIF I/O hardware that was typical of the new integrated and adapted process tools and configured for single-load and batch operation; examples of auto-ID tags, pods, and simulation software were also available. The CIM team worked with manufacturing and engineering to dissect each operation and "fingerprint" step-by-step requirements, which were then used to write the equipment interface codes.

The pilot 200-mm tools were moved into the designated areas as soon as they arrived, and assembly and installation began. For the integrated tools, installation was essentially the same as it would be in a non-SMIF fab, with the exception of some additional testing. Figure 5 shows a tool with both an integrated SMIF I/O module and an integrated enclosure. For the adapted tools, installation of the SMIF I/O modules was done in parallel with other vendor work, after the tools were mechanically functional. Figure 6 shows an adapted tool with adapted enclosures covering the interfaces between the SMIF I/O modules and the tool chamber. Because this tool has two loadports, there are two SMIF I/Os units, which are seen here attached to the machine but which can easily be unlatched and rolled away for maintenance. Figure 7 shows a third type of tool, which combines an adaptive SMIF I/O module and an integrated enclosure.

 

Figure 5: A process tool with an integrated SMIF I/O module and an integrated enclosure.

Figure 6: A process tool with two adapted SMIF I/O modules and adapted enclosures.

Figure 7: A process tool with an adapted SMIF I/O module and an integrated enclosure.

Earlier in the project, testing documents had been created that captured the particular requirements of the SMIF interfaces and enclosures. For example, tests had been designed to assess the ergonomic and safety requirements of the total system, the interlocks between the loadports and the tools, and the performance of the enclosures. Generally, conducting this testing and the resulting signoffs proceeded with little difficulty. However, it was found that the differential pressure specified for the enclosures was difficult or impossible to achieve on some of the tools while maintaining air velocities within specified limits, and the overall distribution of differential pressure by tool was centered about the lower specification limit (see Figure 8).

 

Figure 8: Airflow velocity and differential pressure performance of the tool enclosures.

There were also some problems with the adapted tools. The limited SMIF experience of the personnel involved led to long learning curves on several tools, and in some cases, the proper cabling, software, and configurations for tags, I/O modules, and tool communications had to be researched further before the tools could be released to production. While some of these difficulties had been discovered and corrected before the tools were shipped, many were not resolved until after installation because of incomplete testing at the vendor. For example, interfaces with the fab's host communications system could not be tested at the tool vendor sites.

Process qualifications started on the tools immediately on equipment release. Because of the isolation provided by the SMIF system, there was no need to wait for the tool bay or area to recover from the efforts of construction and installation activity or for cleanroom recertification. First-silicon runs were started as soon as the pilot tools were released. The first-pass qualification lots met all testing requirements and achieved yields comparable to the foundry benchmarks and higher than the assumptions made on the ROI model. Significantly, these high yields occurred in spite of the fact that room testing showed ambient particle levels in production areas had increased since construction had been initiated and defectivity levels of the non-SMIF-protected 150-mm product had risen somewhat in the same time frame. The electrical yield increases stemmed primarily from reduced levels of particles, corrosion, and scratches on wafer surfaces.

Following the process qualifications, the 200-mm run rate was ramped up while 150-mm production was ramped down and additional 150-mm tools were removed to free space for more 200-mm tools. The tool removals and installations were staged to allow the transition to proceed with as little impact on product output as possible (see Figure 9). No decrease from the first-pass 200-mm yields was seen during the tool movements.

Figure 9: Average production levels in terms of wafer and square-inch-of-silicon starts during the transition from 150- to 200-mm wafers.

Tool automation proved to be the most difficult aspect of the project. The configuration of the communications interfaces varied from tool manufacturer to tool manufacturer and even from model to model. In addition, the SmartTag 8400 was a relatively new product when the project began, and many of the SMIF I/O modules on the integrated tools did not have software that could communicate with the tags. Thus, software had to be upgraded on-site. The automation involved in the use of sorters also was more complicated than anticipated and the applications more numerous than originally planned. Not only is there no manual handling of wafers, product or not, with SMIF, but sorters are used to verify pod content via an optical character reader.

As the tool automation came on-line, a truly place-and-go philosophy was implemented. When a pod is placed on the tool loadport, a pod-in-place signal initiates a tag read by MES (directly or through the tool), the lot number is referenced with the correct process step, and the appropriate recipe is downloaded to the tool, along with a start command. To shorten the cycle time, tools are set up to preload while waiting for this command. The loadports have wafer-mapping capability, which is used to reconcile wafer counts at each process step. Because manufacturing personnel are no longer involved in recipe selection or lot verification, misprocessing has essentially been eliminated. The implementation of the AMHS and stockers also has contributed to improvements in productivity. The wafers remain in pods during stocker storage and transport. For reticles, a bare-reticle stocker was installed, which has a SMIF I/O module for combining from one to six reticles in a pod.

Conclusion

The success of the DMOS 4 fab upgrade project was dependent on several key strategies: the adoption of SMIF technology to achieve wafer isolation, early involvement of the SMIF system provider, well-written tool specifications, effective communication with tool vendors and affected fab personnel, and a well-rounded project team dedicated to smoothing the transition to 200-mm production. Overall, the introduction of SMIF technology at the fab has resulted in higher-than-expected device yields and proved to be a cost-effective means to increase capacity quickly with minimal facility modifications. The approach has now been adopted at Texas Instruments' fabs in Japan and Germany as well as at DMOS 6, the 300-mm fab in Dallas.

References

1. R Helms, "Fabwide Automation Is Critical to Microelectronics' Future," Solid State Technology 43, no. 1 (2000): 49­51.
2. M Gatov, "Why Our Fab Is a SMIF Facility," A²C² 3, no. 7 (2000): 19­25.

John Ort is an equipment engineer at the Texas Instruments DMOS 4 fab in Dallas with responsibility for the automated material-handling system and metrology equipment. Before joining TI, he worked as an equipment engineer for National Semiconductor. Ort earned a BS in mechanical engineering from Clemson University in Clemson, South Carolina. (Ort can be reached at 972/995-4290 or j-ort1@ti.com.)

Roy Hunter has been a computer engineer for TI for six years and is the factory automation section lead for material-handling systems and CMP at DMOS 6. Hunter also has held positions at the TI DMOS 4 and Twin Star facilities. He received a BS in computer engineering from the University of Arkansas (Little Rock). (Hunter can be reached at 972/927-3635 or rhunter@ti.com.)

Chris Humphreys is a program manager for Asyst Technologies (Fremont, CA) and is responsible for the company's projects at Texas Instruments worldwide. He is now stationed at DMOS 4. He has 18 years' experience in the semiconductor industry. Before joining Asyst in 1999, he managed cleanroom construction projects for CMPA, a Texas-based construction management firm. He has also spent 13 years with Motorola in engineering and management. Humphreys received a BS in chemical engineering from Arizona State University (Tempe), and serves as education vice president for the Institute of Environmental Sciences and Technology. (Humphreys can be reached at 972/995-1986 or chumphre@asyst.com).

MicroHome | Search | Current Issue | MicroArchives
Buyers Guide | Media Kit