System Automatically Detects &nand Quantifies Wafer Surface Defects

Alexander E. Braun, Associate Editor -- Semiconductor International, 9/1/1998

A s wafer sizes increase and device dimensions decrease, it is important to remove the often limited human factor from the defect detection and classification loop. Automated detection and classification of crystalline defects on micro-grade silicon wafers are important yield impactors because of their wider defect classifying capabilities, and because they eliminate the increased contamination risk human operators present.

A technique to detect and classify wafer crystalline defects is being studied and tested in production environments. Developed by ADE Optical Systems Corp. (Charlotte, N.C.), the optical system consists of a conventional laser scanning surface inspection system (SSIS) to quantify light-scattering events caused by contaminants on the wafer surface and a quad cell (QC) silicon photodetector for light channel (LC) detection. The LC detects and classifies material defects by analyzing information from this scattered and reflected light. Compared to SSISs currently in use, the technique provides higher sensitivity for small defect detection and more signatures for defect classification.

The system can detect and classify epi protrusions (mounds and spikes), slip dislocations, pits, epi stacking faults (ESF) and other defects capable of catastrophically degrading a device's performance. Identifying these killer defects and their source has a direct impact on yield.

The enhanced SSIS LC uses a multi-channel silicon photodetector as the detection source. The QC consists of four quadrants parallel and perpendicular to the scan line. The QC's additional channels allow surface data to be collected, enabling the specular beam's directional shifts to be detected and quantified. The QC also can detect small deviations within the wafer surface plane. Small slope changes alter the position of the reflected laser beam's focal spot on the QC. Biasing between the QC's four quadrants allows slope change data to be gathered from one or any combnnation of quadrants. Beam intensity attenuation also can be detected.

Without LC capability for defect inspection, an SSIS has two disadvantages. First, sometimes distinguishing between different defects is as important as detecting them. Epi mounds and surface particles can cause similar dark channel (DC) light responses, but epi mounds are killer defects, while particles can be cleaned. Second, some defects have smooth and shallow surface profiles and do not scatter light. As device features shrink, these defects affect yield increasingly. The enhanced SSIS, using information from both LC and DC, overcomes these difficulties and detects and classifies these defects.

Conventional SSIS is used for wafer defect inspection, but because it lacks defect classification capabilities, visual inspect remains important. A conventional SSIS that collects scattered, or DC light from surface flaws cannot provide flaw image and classification information. ADE's SSIS system's QC LC monitors the motion and intensity of the laser beam reflected from the wafer surface.

Data collected by the system are run past an analysis and defect classification algorithm based on statistical modeling using real defect samples and mathematical modeling. The algorithm analyzes each event's DC and LC signatures and identifies them as a specific defect type. Analysis results are displayed as bin sorting results. Table 1 lists the characteristic LC signatures for typical killer defects in the 4 to 10 µm range.