Line Yield
Line yield is the fraction of wafers that successfully complete the entire manufacturing process flow from start to finish, measuring how many wafers physically survive the journey through hundreds of process steps rather than measuring anything about the die on those wafers. A wafer can be lost to breakage, pulled for engineering analysis, scrapped after a misprocessing event, or consumed as a test wafer — any of these removes that wafer from the output count entirely, regardless of how good the die on it might otherwise have been.
A wafer removed from the manufacturing flow because of breakage represents an entirely avoidable cost if it traces back to a specific handling or equipment issue, which is exactly why line-yield tracking by process step matters: breakage events that repeatedly cluster around one specific tool or one specific handling transfer point point directly at a fixable root cause rather than at random bad luck. Physical wafer breakage — during a robotic handoff, inside a process chamber, during a measurement step — removes a wafer from the line permanently regardless of how the die on it were performing up to that point, which means every single breakage event is a complete, total loss of whatever processing investment that wafer had already absorbed.
Process scrap, triggered when a wafer is misprocessed in a way that makes continuing to run it through the remaining flow pointless — wrong recipe applied, contamination detected, an out-of-spec measurement that can't be corrected downstream — removes wafers for reasons distinct from physical breakage, even though both show up identically in the line-yield denominator. Scrap decisions are deliberate: rather than continuing to invest additional process time and material into a wafer that engineering judges unrecoverable, the fab cuts the loss immediately, converting what would otherwise be a wasted continued investment into a bounded, already-known loss instead.
Engineering pulls — wafers deliberately removed from the normal flow for destructive or time-consuming analysis, such as cross-sectioning to verify a specific process step's actual result — are a necessary, planned category of line-yield loss rather than an unwanted defect, existing specifically to support the process-control and yield-improvement work that keeps the rest of the line yielding well. These wafers aren't lost to any problem; they're intentionally sacrificed in exchange for process-verification data that protects the yield of every other wafer in that lot and every subsequent lot, which is why planned engineering-pull rates are usually budgeted in advance rather than treated as an unplanned loss to chase down.
Line yield is measured and tracked at the level of individual process steps as well as across the entire flow, because a single mini line-yield loss rate at one specific step — even a small one — compounds multiplicatively across the hundreds of steps a modern process flow contains, and per-step tracking is what identifies exactly which step is contributing the most to total loss. A process flow with 500 individual steps, each losing even a tiny fraction of a percent of wafers, can compound into a materially lower overall line yield than the per-step numbers alone would suggest — which is exactly why root-cause analysis for line-yield loss starts by breaking the overall number down by individual step rather than treating it as one undifferentiated total.
Line yield interacts directly with [cumulative yield](cumulative-yield) as one of several multiplicative stages — wafer survival through the fab, die functionality on the wafers that survive, package survival, and final test pass rate — each of which represents a genuinely separate loss mechanism measured at a different point in the overall manufacturing flow. Because line yield measures only whether a wafer survives the fab process flow itself, and says nothing whatsoever about how good the die on a surviving wafer actually turn out to be, a fab can simultaneously run excellent line yield and poor die yield, or the reverse — which is exactly why cumulative yield analysis has to track each stage separately rather than collapsing everything into one undifferentiated number.