Lucent Technologies, Bell Labs Innovations


S-Wafers Project

Overview

Each week, thousands of wafers enter production in Lucent Technologies' Microelectronics manufacturing facilities. In most cases, the path through the fab will involve hundreds of processing steps and can take several weeks. On each wafer devices are created by the hundreds, arranged in a regular grid. The vast majority of these wafers are thin disks of silicon and the devices are integrated circuits, from DSP's to the latest FPGA chips. In increasing numbers, the wafers are made of gallium arsenide and the devices are semiconductor lasers, again packed densely on each wafer. While the processing of lasers and IC's is quite different, a common feature is that the object under manufacture consists of a regular array of individual devices, each device a potential product to be sold. Another commonality is the complexity of the processing to which each wafer is subjected.


If you have a Web browser that supports Java applets, then in the accompanying applet on the left you should find a single lot of wafers probed in Madrid. The colored squares represent defective chips, while the white squares are good chips. As you move the scroll bar, the panels display the binary maps formed by coloring black those chips that failed the selected test. The (composite/binary) button toggles between displays of these binary maps and the composite or sitewise averaged maps, again calculated for the indicated failure mode. The focus of much of our work is on identifying common patterns of failed chips across lots of a given code or technology for the purpose of quality improvement and process monitoring.


Quality control á la Walter Shewhart consisted of simple charts tracking a single variable at a time. Effective process monitoring and quality improvement efforts in this context, however, need to address the spatial character of these data. Researchers in the Statistics and Information Analysis group, together with product, process and FMA engineers at each of the Lucent Microelectronics manufactring sites, have designed a powerful set of sofware tools and acompanying statistical techniques for efficiently analyzing the vast amounts of data collected during the fabrication process. The S-wafers software tools provide an environment for simple, but powerful, wafer visualization; improved process monitoring; the detection of defect signatures; and spatial analysis of variance to analyze data from final probe, parametric tests, particle scans, memory bitmap failures, and split-lot experiments. New yield models that directly exploit spatial patterning have also been implemented in the S-wafers environment. These tools are based in the S language and are available on UNIX workstations and personal computers (PC's). In addition, we have designed a point-and-click interface for personal computers running Microsoft Windows.

Impact

Our efforts begain with FMA (Failure Mode Analysis) engineers in Orlando, designing tools for monitoring data collected at final probe. The result of this work has been a collection of reports that provide a daily summary of the patterns of defective chips by technology and code. To make further contacts within the factories, we have led several extremely succesful courses on Spatial Data Analysis in Orlando , Madrid , Allentown , and Reading . As a result of these courses, we have expanded our tools to incorporate a variety of other data sources: bitmap data from the ORCA FPGA codes; test results from the numerous zone monitors run in Allentown and Orlando; and point-level particle data taken from the Tencor and KLA instruments. Recently, we were approached by the Wafer Mapping Team from Optoelectronics who want to use an expanded version of the S-wafers software as the analysis engine behind their process monitoring and quality improvement effort. This new direction is particularly exciting because this new data source poses a number of novel statistical problems.

As technology advances from 0.9 to 0.5 and 0.35 microns, engineers cannot depend on visual inspections and must increasingly rely on sophisticated data analysis and automatic statistical procedures for monitoring the manufacturing process. For the last three years this project has provided and will continue to provide the next generation of data analysis techniques and software tools to Lucent Microelectronics.

Further Information

For those of you outside of Lucent Technologies, feel free to contact one of the S-wafers team members. Within the Statistics and Information Analysis department, contact Lorraine Denby (requests or enhancement suggestions for the GUI), Mark Hansen (new statistical methodologies), or David James (requests or enhancement suggestions for the S-wafers software).