Integrated Auto Macro Defect Inspection was implemented onto all photolithography clusters at IBM’s 300mm wafer manufacturing line. Inspection is performed post photoresist develop step and has resulted in achieving 100% wafer inspection while minimizing tool downtime to less than 5%. This paper discusses the challenges encountered towards making the inspection process manufacturable. The challenges started as tool integration issues during implementation, but have evolved to include hardware and software issues as well.
The tool integration issues encountered at the outset pertained to the physical installation of a macro metrology tool into the photolithography tool cluster. Subsequently, the hardware challenges included ensuring that the macro inspection, recipe management, and disposition systems were designed properly in order to effectively receive, send, and store the inspection data.
Software challenges, on the other hand, are very similar to those encountered with the automated macro defect inspection standalone tools. With the existing software providing a high degree of defect detection capability the challenge has been more towards decreasing the high number of false fails while ensuring that the true fails do not end up passing inspection. False fails significantly impact manufacturing throughput of the photolithography sector. Ongoing software improvements in response to these false fails have resulted in improving inspection accuracy. False fails are further being addressed by moving towards automated recipe creation (utilizing the scanner/stepper file shot map offset parameters). Heightened sensitivity of non-optimized recipe parameters, also lead to trivial defect fails that are not necessarily reworked. This significantly impacts cycle time while causing tool downtime and thus needs to be avoided. On the other hand preventing true fails from passing inspection involves ensuring that the recipe parameters are adequately sensitized.
To further improve automated macro defect inspection several hardware and software improvements have also been pursued. Software improvements include accurately grouping the detected defects and performing automated defect classification into pre-taught categories. Hardware related improvements include improving defect coordinate repeatability and integrating a "smart" system that will provide real-time feedback, thus resulting in automated tool shut down and product rework when necessary. This paper will also discuss the importance of defect detection capability, defect classification, and KLARF file conversion in implementing automated disposition.