A quality-cost model of in-line inspections for excursion detection and reduction

Abstract

A methodology for optimized defect excursion (EXR) monitoring is proposed using an economic statistical process control (SPC) model for defect limited yield. The cost of in-line defect inspections is increasing at an exponential rate, particularly for a 300mm fabrication facility. Therefore, optimized in-line inspection schemes have become more critical for controlling the costs of semiconductor manufacturing. In order to minimize the inspection costs while maintaining acceptable yield, a cost function which incorporates the power of the inspection, the interval between inspections, and the yield impact (cost) is optimized for all inspection locations in a given process flow with a fixed sampling budget. This methodology can be used to allocate inspections based upon the risk of yield excursions at defect limited process layers. This model can also be used to establish quantitative estimates of return on investment (ROI) of inspections to inform decisions regarding purchase of in-line monitoring (ILM) tools or sampling adjustment. A quality-cost model has been derived using the theory of economic SPC, and has been implemented in a high volume CMOS fabrication facility with a high degree of success.

Optimized management of excursions in semiconductor manufacturing

“…The risk of adding or removing an inspection per time unit at a specific point in the flow depends on the information gained per inspection (Hall et al 2008). This information can be the reduction of risk in terms of material at risk (uninspected lots since the last-known-good inspection) or the yield loss due to an excursion (when a process or a tool shifts out of specification).…”

“…And the data required to support failure analysis is quite diverse, ranging from immediate needs such as test datalogs and layout databases to less frequently accessed data such as lot related information. (Hall et al 2008) provide a quality-cost model for excursion detection and reduction. Their results indicate that inspections can be allocated based upon the risk of yield excursions at defect limited process layers.…”