Published: 01 January 1989
| ||Format||Pages||Price|| |
|PDF (248K)||12||$25||  ADD TO CART|
|Complete Source PDF (6.8M)||467||$60||  ADD TO CART|
Cite this document
For on-line statistical process control to be effective, it is necessary to correctly specify the distribution of the process generating the defects. Manufacturing and assembly operations have traditionally assumed measurement data to follow the normal distribution and defect data to follow the Poisson distribution. This paper will discuss the Integrated Circuit manufacturing industry, where the Poisson distribution assumption is known to be false. Suggestions will be made on how to modify existing control charts to allow for the clustering effects known to occur in IC manufacturing. By modifying control charts which assume the Poisson distribution, the same procedures and interpretations may be used in many cases: only the sample size or control limits would change. With other control charts however, significant clustering precludes their use.
Quality control, clustered defects
Member of Technical Staff, AT&T Bell Labs, Murray Hill, NJ