Under Control

SILICON IN CONTROL

 

Recently this columnist attended a workshop on semiconductor manufacturing held in San Antonio, home of the Alamo and the Rough Riders. Next to the hotel was an Imax theater showing a special Alamo film and "Jurassic park," three stories high. Guess which one your token nerd saw.

This workshop - one in a series looking at manufacturing problems in the semiconductor industry- focused specially on wafer fabrication problems.

Anyway, I had to listen to that most common of all laments "My control problems are special." Years of experience taught me that control problems are similar across many industries. The secret is to locate the true differences that make the problems' special. Issues involving materials handling, software, yield (quality), and diagnostics are common across the control spectrum.

What's special about the semiconductor industry can be summarized as follows:

It pleases me to see there was considerable interest in learning how to better maintain control of processes battered by unforeseen events. This led to many hallway discussions about chaos and emergent systems, subjects close to my heart.

A wafer production line must meet process specifications in the face of the extreme swings of multiple variables. Any control system needs to be insensitive to stimulation. But in wafer manufacturing, this requirement is crucial. Quick time-to-market and high throughput require use of the latest in modeling and feed forward control.

Imagine a spread sheet of control solutions. The horizontal axis presents an increasing level of technical complexity, from direct feedback control involving a few sensors and effectors to plant wide kilosensor / effector systems.

In the vertical axis, we consider the sophistication of control type - of solution, not problem - from direct one-to-one control-to-sensor fusion to full-model control with dead reckoning. In other words, a spread sheet of problem "size" vs. the solution technology.

For example, early PLC's controlled via direct binary feedback based around "limit switches." PLCs today are used in large kiloplant systems.

Personal and minicomputers were used in the early days for intelligent control, but lacked the power to handle fast, large multipoint systems. Combining the PC and PLC is what's done today.

Model-based control will be next. Modeling is best explained by an example. A wrist watch does not tell time. It models time. It is, at best, an approximation of real-world time.

Like a submarine, the watch navigates by dead reckoning, you reset it once a month and that's good enough. The submarine uses a model of the currents, and awareness of its approximate direction and speeds to 'guesstimate' its current position. Course correction is done by surfacing and using a sextant.

The control system of the future will be a model of the process running faster than real time. Feedback from the modeled process will impact the real process. At appropriate intervals, the process model will be re calibrated for further use.

Parameters for wafer fab, for example, can be determined using feedback based on predicted performance - a control scheme resident in the database, with update from sensors. Sensor busses need to accommodate large quantities of dumb sensors and/or small quantities of smart sensors.

Updating of the real processes' I/O data will be done asynchronously from the model. Process parameters will feed forward into the real manufacturing process from the model runs.

Wafer fabricators can only use, not drive, existing control offerings. Many fabricators are beginning to concentrate on product, and have forgotten about reinventing the control industry. This growing maturity will lead to a robust, agile process. Working with the technology providers will lead to a broad spectrum control techniques. The industry is off and running. Congratulations, and keep on trucking.

As appeared in Manufacturing Systems Magazine November 1994 Page 16
http://www.manufacturingsystems.com




Homepage
Manufacturing Systems Magazine Articles
References - Table of Contents

Send mail to rmi.info@barn.org for more information.
Please send mail to webmaster@barn.org regarding web site structure.
Copyright © 1996-2002 R.Morley Inc. All Rights Reserved

R. Morley Incorporated
614 Nashua Street, Suite 56
Milford, NH 03055-4992 USA
Tel: 603-878-4365 FAX: 603-878-4385