Product In Action

Design of Experiments Software Helps Clear Wafer Transport Traffic Jams

Despite today's faster-is-better perception in chipmaking, speed isn't always the primary goal to seek. When costly and fragile devices are the output, wafer transport through the manufacturing processes must be deftly planned and carefully carried out. Product flows should be designed for punctuality in such a way that yield is not negatively affected. What's needed is a traffic solution that optimizes manufacturing speed without sacrificing product integrity.

 

Response surface methods were used to evaluate two effects for increasing material-handling speeds: increased velocity across the entire line (Factor A) and increased velocity in designated speed zones (Factor B). The floating contour grid indicates the best conditions beyond which no further improvement is predicted. (Graph courtesy of Stat-Ease)

PRI Automation (Billerica, MA) has been meeting traffic challenges like these by using designed experiments software. The firm designed and built overhead automated material handling systems (AMHSs) that transport wafers through the manufacturing lines of Intel, Motorola, and other global semiconductor giants. As a supplier of advanced factory automation systems, PRI has succeeded partly by discovering how rapidly transport vehicles ought to travel in order to minimize transit time between process steps while increasing AMHS throughput.

"There's a lot of traffic and turns in the transporting process," says John Rayter, senior industrial engineer at PRI's Mesa, AZ, office. "The acceleration limit is critical and is limited to 0.1 g because of the fragile nature of silicon. Too much jostling from rapid acceleration and deceleration can damage the wafers." Since there are millions of dollars worth of chips being transported by PRI-designed vehicles at a given time, any loss caused by damage is costly.

Monorail vehicles typically drop off and pick up semiconductor wafer lots from automated storage and retrieval systems for movement along transportation routes. When designing AMHSs, one of the tools Rayter and his colleagues use is design of experiments (DOE) software called Design-Expert, made by Stat-Ease (Minneapolis).

Every proposed AMHS design is simulated to predict average delivery time, which is defined as the average time to move a requested product to its destination once a request to move the product is initiated. To do this, explains Rayter, "it's necessary to simultaneously determine optimal speeds in congested and noncongested track areas." Using DOE software, he can determine three critical parameters when designing system-speed configurations:

• Diminished returns from excessive speed.
• Simultaneous evaluation of the effects of speeds in congested and noncongested areas.
• Emerging trends among contrasting factory systems.

Design of experiments—also known as designed experiments, experimental design, and DOX—comprehensively simplifies the experimentation process. The traditional one-factor-at-a-time (OFAT) method of analyzing variables requires time-consuming, cost-prohibitive experiments. Perhaps most significantly, the OFAT approach does not reveal factor interactions. Researchers and engineers can no longer afford to experiment in this trial-and-error manner.

DOE software helps experimenters to simultaneously evaluate all factors and factor interactions. Negative factors can be minimized and beneficial factors optimized. Factor interaction information discovered using DOE reveals how a total system works, so by fitting response data to mathematical equations within the software, DOE shows how interconnected factors will respond over a wide range of values without having to test all possible values directly. Collectively, these equations serve as models that predict outcomes for any given combination of values.

Using DOE software, PRI engineers optimize critical transport responses and find the best combinations of values, often resulting in positive synergistic reactions. Design-Expert's two-level factorial and response surface optimization tools produce robust parallel testing schemes. Its powerful statistical analysis uncovers dominant effects in an efficient manner that can save experimenters money and time.

Factors involved in AMHS designs include move volumes, product mix, equipment integration, scheduling logistics, and reentrant flow production patterns. These variables are best analyzed and predicted using simulation. However, simulation by itself is insufficient for estimating the interaction impact of AMHS equipment. To give a boost to the findings, DOE response surface methods (RSM) are applied to analyze and predict the impact of the speed of automated transport vehicles. RSM quantifies relationships among one or more responses and a number of input factors. It provides sophisticated maps to help identify peak performance, which is an attractive alternative when the nature of the product discourages product testing under actual field conditions.

"DOE software has the functionality to let you build complicated designs and evaluate comparative models to arrive at one that makes the most sense," says Rayter. "These designs have to be unbalanced in that the uncongested track speed always has to be greater than or equal to the speed in the congested areas. The software allows us to alter point selection strategies that minimize this effort, while providing easy-to-use diagnostics to ensure that the design points selected make for a valid design."

In developing one design, Rayter used the software to maximize efficiency ratings through trial-and-error design-point selections. This minimized the manual effort required to run simulations. "Once we had a design that met user-defined criteria and had an efficiency over 85%, which is recommended, we were in business," explains Rayter. "From a statistical perspective, you really don't buy any more improvement past a certain speed."

His findings are supported by DOE designs showing excellent "fit qualities," that is, empirical, mathematical data that are fit to the product test data. "This discovery is important," according to Rayter, "because it tells the product developers that there is no real need to spend extra time and money to have transport vehicles scoot around at the speed of sound." With this knowledge, product developers have the ability to discover outcomes simply by plugging in any two uncongested and congested area speeds. For example, if there is an upper bound on the uncongested speed, they can use the DOE software to statistically test for a more desirable design by fine-tuning congested area speeds.

"The four or five disparate facilities that we did this on all showed approximately the same behavior," continues Rayter, "indicating that we could generalize our conclusions somewhat regarding the upper limit. We could enhance speeds and still show significant benefits."

Using the DOE approach, PRI is developing a vehicle that autonomously accelerates to the upper bound speed as well as automatically slowing down for corners and stopping at stockers. "While all this is happening [in the software model]," Rayter says, "we can observe acceleration constraints on the wafer payload. It makes for shorter transit times, reduced vehicle counts, and higher throughput."

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