Nov. 6, 2008 – Chipmakers hope that forging new, more open alliances with process tool vendors will enable them to run much more productive, efficient wafer fabs in the future. Tension between chip companies and tool vendors developed with the off-again, on-again transition from 200mm to 300mm wafers. After hundreds of millions were spent on an aborted attempt to make the transition, the actual insertion of 300mm wafers came at a new node, requiring additional costly tool redesigns. Tool vendors traditionally have been wary of letting users understand what goes on inside their equipment. Internal workings of many individual tools have been cloaked in secrecy through the use of operating software, sometimes in binary or assembly code, that is hard or impossible for a user to fathom. This made it difficult for fabs to develop process control to improve processing and maintenance procedures.

Next-Generation Factory (NGF) concepts under the International SEMATECH Manufacturing Initiative (ISMI) call for new standards and more accessible tool operating data so that chipmakers and tool suppliers can collaborate on improving tool utilization and productivity. Many presentations at the recent ISMI Symposium on Manufacturing Effectiveness (Oct. 22-23) probed different aspects of this emerging collaboration.

Traditional barriers must be broken down for this more open approach to take hold. Informal exchanges at past conferences revealed some of the attitudes that must change. Litho engineers from a major chipmaker complained bitterly, for example, that they learned their exposure tool supplier was gathering data from its tool about lens imperfections to use in improving future lens designs. The toolmaker, however, refused to share this information with the customer. An engineer for a major toolmaker explained that a tool’s internal operating instructions should not be accessible to users because any changes could create hazardous conditions.

While chipmakers do want access to tool parameters and some metrics, this can be done without making potentially hazardous changes to internal controls. But there is still concern about built-in sensors revealing too much about a tool’s internal operations, according to Paul McGuire of Alan Weber & Associates. Control data may be the source for customer complaints, and intellectual property may “leak.” Also, service and spares revenue may be reduced if maintenance is stretched out using a condition-based maintenance (CbM) regime. McGuire showed an example where periodic maintenance had been replaced by CbM, so it was done only when needed (see Fig. 1).

Fig. 1: Example of condition-based maintenance replacing periodic maintenance.
CLICK HERE to view larger image

McGuire addressed both of these issues. sometimes maintenance may actually be done sooner rather than delayed. in any event, warrantees may have to be modified for flexible equipment maintenance, with charges for extra support in return for less maintenance and longer MTBFs.

Equipment-Performance Indicators (EPIs) are a way to shield IP, he explained. A single EPI value may be derived using data from multiple internal sensors, but only the EPI will be externally accessible (see Fig. 2). An internal tool model can be used to synthesize an EPI value, and EPIs can provide a status hierarchy for the process and equipment. This aggregates a measure of the health of lower-level modules in the hierarchy, he said. (see Fig. 3)

Fig. 3: EPI summary status values by equipment module or subsystem, can be defined hierarchically to aggregate health of lower level modules into higher ones.

These internal metrics will be needed to move beyond preventative maintenance to the predictive maintenance and anticipatory prognostics envisioned in the NGF, according to McGuire. If the EPI approach is adopted, tool designers will have to find a way to recalibrate internal sensors that are not externally accessible to ensure that readings remain accurate over the various sensors’ ranges.

Fig. 2: EPIs can be hierarchically related, matching the equipment structure hierarchy.
CLICK HERE to view larger image

Eliminating delays in processing and dealing with large amounts of data acquisition as process complexity increases was discussed by Rebecca Cooper of Adventa Control Technologies. Adventa works with Lam Research on “300 Prime” projects, which aim toward NGF goals with classic 300mm technologies.

Automated maintenance procedures can achieve a 31% reduction in data handling, she suggested (see Fig. 4) Maintenance data can be stored on a tool but sent to another site for data analysis and crunching, she said. Modular equipment design with smart wafer scheduling can also shift work to other modules on a cluster tool.

Fig. 4: Example of positive impact of automated procedures/maintenance on mean-time-to-clean (MTTC).
CLICK HERE to view larger image

Cooper said that targets include reducing first-wafer delay, recipe-to-recipe delay, lot-to-lot changeover, and job preparation times. The goal is seamless cascading of lots with different processing requirements (see Fig. 5)

Fig. 5: Cascading host & tool-level recipe tuning for run-to-run control.
CLICK HERE to view larger image

One example she cited was gathering lots of data during the 10-15 sec wafer transfer time. This might require a second port just for gathering data and supplemental communication ports. A Multi-GEM port is one without remote operations capability, she explained (see Fig. 6). Harvey Wohlwend of SEMATECH, the session moderator, added that such a port might handle 10k parameters/sec.

Fig. 6: Data collection using Multi-GEM ports.
CLICK HERE to view larger image

Recipe and parameter (RaP) management is a set of advanced factory capabilities now being devised that could be deployed as part of the NGF transition, according to Lance Rist of ISMI. It will provide tighter control of recipes stored on equipment and also provide for parameterization and on-demand visualization, according to Rist, who is both a chemical engineer and software specialist.

Seventeen device-maker members of ISMI have been working with Selete and other groups on early conceptualization and standards for RaP. The objective, Rist said, is to give chipmakers what they need while keeping requirements reasonable for toolmakers. Early prototyping has already begun with multiple versions ready to go as an alternative to SECS transport using an XML/SOAP protocol now in place in SEMI E139.3. To avoid excessive XML coding, Rist explained, recipes are put into ZIP files for transfer.

One advantage of RaP is that it thus eliminates recipe downloading during a process step, cutting operating costs. Costly “wrong recipe” mistakes caused by unmanaged recipe changes are eliminated by check-sum coding. Gathering use data makes it quicker to trace scrap due to recipe- and parameter-related errors, Rist said. On-demand access to recipe parameters as run-time changeable variable parameters can also speed APC implementation. He added that although RaP does not define how parameters are changed, it does show limit bands in XML format.

An RaP reference implementation (RRI), an executable emulator, has been devised as a tool for developers and users, according to Rist. This provides a learning aide showing how the program works, and it is now available from ISMI.

Implementing the NGF vision will require complete new approaches to enhance chip manufacturing capabilities, Rist stated, and RaP is one example of how this can be accomplished. — B.H.