Defect characterisation and reduction on a multi-chamber inter layer dielectric (ILD) tool

Abstract

Due to the decreasing geometry and increased production complexity in the Semiconductor industry, particle defects have become more critical in regard to wafer yield parameters. Current cleanroom technology success leads to contamination being attributed mainly to equipment operation. One of the key mechanisms is wearout of equipment parts. This paper discusses the techniques applied at Fab 10, Intel's European Semiconductor Manufacturing site and the success of correlating machine components wearout to metrology measurements. In 1 micron technology days, one could visually inspect machine components during preventive maintenance and make a "judgment call" on replacement. However, risks are too great in sub-micron technology where a flake from a worn-out part can cost several die and contaminate downstream tools. In an effort to put more science into the changeout frequency of parts, removing subjectivity, a Noran 'Voyager 2100' Microanalysis system on an Amray 2030L FESEM was used to collect and analyze the X-rays from both defects on blank oxide wafers and machine parts. A background baseline of new multi-chamber ILD Chemical Vapour Deposition (CVD) machines was initially performed, which corresponded to normal station monitor defect counts. When defect monitors on a particular machine trend upwards, further monitors were performed. In tandem, a basic description of machine components was obtained from the vendor for approved materials. Where we did not have specific composition of suspect parts, the suspect part would be changed out and another Energy Dispersive X-ray Spectroscopy (EDX) monitor performed. We were able to obtain spectra of submicron defects on blanket oxide wafers which were held in a catalog and compared to analysis results of machine parts such as o-ring wearout of chamber doors, showerhead wearout and quality of graphite components. The information allowed us to set realistic changeout frequency of parts before they became a problem. The EDX analysis was backed up by inline metrology measurements on product. The success of this led to EDX on a new inline system being the standard test before the ILD process tool was vented up for troubleshooting. From the spectra information library, a response flow checklist (RFC) was written, which is now used as a standard method of troubleshooting Out Of Control (OOCs) particle situations. This is also the standard test performed when upgrades/new parts are being qualified in our tools.