Relative Control and Management Philosophy

Silicon wafers for the semiconductor industry are extremely complex materials with characteristics such as high purity levels, crystallographic perfection, precise mechanical tolerances, complicating efforts to effectively monitor process stability and control quality for individual product types. Material of silicon wafer can be doped with more than 12 kinds of dopants, such as B, C, N, Al, Si, Sb and others. Currently, the sizes of the firm’s products are 4-, 5-, 6-, 8- and 12inch. Considering dopants and sizes, and each kind of product has different attributes according to which, 7~12 minutes are required to slice a piece of wafer. About 2 minutes are required to inspect the quality of a piece of wafer. A wafer can be easily broken during inspection, because of its thinness and brittleness (Lin et al., 2002). Moreover, slicing is a kind of cutting technique that has difficulty in yielding the required precision. Three scenarios will incur damage on the work piece: (1) inaccurately estimating the precision of the slicing machine, (2) engineers set parameters and change the type of material and (3) inconsistently controlling the wafer quality by applying the sampling method owing to the small batch size of wafer slices in the production model. Consequently, given unstable yields of synchronously multiple quality characteristics are unstable or drifting accuracy of wire saw machines, inspectors must consider employing machine control and monitoring measures. Five synchronously occurring precision quality characteristics, namely thickness (THK), bow, warp, total thickness variation (TTV), center thickness (CTRTHK) and total indicator reading (TIR) must be simultaneously inspected using automatic testing equipment (ASTM F534, 1995; ASTM F657, 1995; Takeshi, 1998). Those multiple quality characteristics destabilize the slicing. The case firm used quantitative methods, such as process capability indices (PCIs) and statistical process control (SPC) charts, are severely limited in monitoring slicing problems (Lin et al., 2002).