Maximize Yields Zone by Zone

Laura Peters, Senior Editor -- Semiconductor International, 9/1/1999

Part 8of Series

Computerized wafer zone yield analysis addresses today's dynamic manufacturing environment, which requires yield ramps of 5%-6% per month and an increasing number of single-wafer processing steps. Nick Atchison and Ron Ross of Silicon Systems (Santa Cruz, Calif.) developed a wafer-zone-based methodology that helps determine which pieces of process equipment contribute to yield loss. Summarized in this eighth article of our Integrated Yield Management series (see www.semiconductor.net for full series), the analysis uses artificial intelligence programs to monitor zone-related wafer yields.

Yield zones are designed to separate zones on the wafer that have distinctly different yields. This is accomplished by mathematically analyzing yield patterns, viewing large numbers of wafer probe maps or using exploratory geometric analysis. Certain software database programs can automatically stack wafer probe-bin maps for large numbers of wafers, while also facilitating the tracking of bin data by x-y location.

Distinct yield patterns in different regional zones of the wafer simplify the tracking of yield problems to specific proccess tools.

Having an idea of the typical zone variations, engineers can use computer programs to summarize and plot yields by zone. The zones could be individual stepper fields, radial zones, quadrants or combinations thereof. The engineers have used both RS1 programs and compiled programs on VAX computers, the latter of which provide speed advantages. One zone map that has worked particularly well for the authors (see Yield Management, SI August 1999, p. 46.) has three radial zones and four quadrants in the outer zone.

With identified zones and computer-based yield tracking by zone, other analyses (parametric, V₀D₀, etc.) can correlate zone-specific yield differences to specific pieces of process equipment. Engineers can perform zone-based random defect analysis using KLA defect maps and bin map overlays to attribute regional yield loss to certain parametric sensitivities or to random defects of specific types.

It also is necessary to determine the statistical significance of yield differences among zones. If detection of a 1% yield difference at a 95% confidence level is desired, and if the yield standard deviation is 10%, a 0.3125 standard error of the mean is required. To achieve this error factor, with an average of six process tool units per process step, data from about 6000 wafers is needed (10/[(6000/6)^0.5] = 0.3125).

Zone-based yield tracking can show: yield variations in a single zone due to a process problem limited to that region; yield variations occurring in unison in different zones; inversely related variation of one or more zones due to mutually exclusive process interactions (Figure); yield swings in a zone, possibly indicating test program variation if both products follow the same process flow; or zone variations occurring in unison, indicating simultaneous process problems occurring in all zones.

After locating low-yield zones, statistically verifying the data, relating to specific process equipment and analyzing for random vs. systematic losses, subsequent analysis often requires SEM cross-sectioning or internal electrical probing. Atchison and Ross encourage monitoring corrective efforts with the same methods used to find problems.

An example illustrates the utility of zone-based yield analysis. In this case, zone 1A was yielding 10%-20% lower than the outer zones. Probe-bin failure maps indicated the rate of failure for I_cc stand-by leakage was excessively high in the zone, while leakage values also increased dramatically near the wafer edge. Wright etching and SEM analysis showed a 100X increase in the density of dislocation faults in the silicon substrate relative to other regions. Further analysis led to a variation in oxygen content of supplied silicon wafers, remedied by tightening the spread of O₂ content in the wafers.