Boring Label, Good Stuff

Interface A: It is undoubtedly one of the most uninspiring labels for anything ever, but it is sure to have a profound impact on the semiconductor industry for years to come. It's basically a communication standard for equipment that has just recently become available (see "The Standard Pieces of SEMI's Interface A ").

Why is that so important? Because, increasingly, success in semiconductor manufacturing is ultimately less about new materials and new device structures and more about manipulating data. It's about collecting and analyzing data to improve efficiency and productivity, yield, quality, turnaround time, ramp time and about every other metric by which a manufacturing operation is judged. It's also about better integrating data from many various sources: design data with manufacturing and test data, metrology and equipment data with process, modeling and simulation data with empirical data. This all becomes more important, of course, as processes become more complex, more expensive and with an exploding number of variables that are harder to understand and control.

Today, the semiconductor industry is focused on the collection and analysis of data in many different disciplines. Design-for-manufacturing, yield and test (DFM, DFY and DFT), advanced process and equipment control (APC/AEC), modeling and simulation, e-diagnostics, metrology, yield analysis and just basic automation. This month's issue has several examples of work in these areas.

For example, as Carl Fiorletta of Adventa Control Technologies notes in this month's cover story, "How to Improve Fab Productivity ," automation can be used to tune processes to keep tools in spec, online and producing revenue wafers; collect data and automate process optimization to minimize hands-on process tweaking by process engineers and operators; and allow machines to be taken offline for maintenance only when recipes cannot be tuned to keep the machine within specification.

Alexander Braun's article, "Process Complexity Fuels Integrated Metrology ," notes that metrology providers must make their systems more than just "data geysers" and indicate what action should be taken, what process marginality is coming down the line, and the seriousness of the drift.

In "Modeling Interconnect Variation — Optional or Mandatory?" author Stephen Fisher of Praesagus notes how trying to quantify systematic vs. random variation is a key challenge. "Is the variation within die, within wafer, wafer-to-wafer, lot-to-lot or fab-to-fab? Answering these questions is a daunting task today and a nightmare for future processes." Part of the solution to this nightmare is modeling.

Underlying all of this activity in these disparate areas is data collection, analysis and distribution. Challenges abound: Desirable data might not even be available to collect (imagine you'd like to measure the temperature in a certain part of the tool; no sensor, no data), or an even more common problem is there might be too much data. There is usually not very good correlation between different sets of data. This is a notable problem in the DFM world, where what's printed on the wafer can be quite different than what the designer intended, compounded by the use of resolution enhancement techniques. And, perhaps not surprisingly, there is often not enough processing power to crunch the data. To analyze a complex design, it may require 100 processors running for days.

The key to making all of this work is simply the infrastructure: Much of the data is in different formats or protocols, moving on very different types of buses. Interface A, and pending sister standards Interface B and Interface C (which facilitate data sharing between applications) simply make data handling easier. Ultimately, the industry will probably need Interfaces D through Z, but for now we should embrace and love the not so glamorous, but critical, Interface A.