Everyone likes a touchdown, right? Tom Brady's touchdown pass to Julian Edelman with just two minutes left on the clock in Super Bowl XLIX brought undiluted joy to millions of New England Patriots football fans. While many touchdowns in a single game are widely celebrated in football, a large number of “touchdowns” in semiconductor manufacturing would not be met with the same level of celebration. In fact, touchdowns can be a costly quality issue in the semiconductor manufacturing process, and if not for the analysis capabilities of big data solutions, they would be a very difficult problem to identify and contain.
Demand for high-quality semiconductors in popular consumer products like smartphones and safety-critical components such as automotive anti-lock braking systems is at an all-time high. The demand makes it difficult for leading integrated device manufacturers (IDMs) and fabless semiconductor companies to maintain high product yield levels while also maintaining an elite level of quality. The continual threat of expensive product recalls is a chief concern for these manufacturers. How can semiconductor companies reduce the number of bad devices delivered to manufacturers? In many ways, it starts with limiting the number of touchdowns in wafer sort.
In semiconductor testing, a “touchdown” occurs when the prober touches a bare die in wafer sort. Too many touchdowns or excessive test probing in semiconductor manufacturing can actually damage the bare die and result in a low-quality part being packaged up and sent into the supply chain. The part may not immediately fail, but too may touchdowns during wafer sort can result in a part that may fail prematurely, which is unacceptable for safety-critical automotive applications or for high-end smartphones.
Adding to this challenge is parallel testing of dice. Parallel testing is a common practice because it enables the prober to simultaneously descend on multiple dice, enabling two, four or eight dice to be tested at once, dramatically speeding up tester time. While this method increases throughput performance, it also can result in too many unnecessary touchdowns for previously-determined good dice. As the prober moves across the wafer and stops to retest a questionable die, it may also probe other seemingly reliable dice due to the automated parallel testing. Semiconductor operations are often not aware of the number of touchdowns that occur, and the resulting questionable dice they may inadvertently create. These suspect devices may not even be caught in final test if the parts are still functional, albeit with a diminished level of quality.
Fortunately, the majority of touchdown-related “escapes” (suspect semiconductor devices released to final assembly) can be prevented in wafer sort with the right big data analytics.
Big Data Solutions to the Rescue
By providing more detailed, near real-time information during manufacturing operations, big data solutions can now identify and avoid costly touchdown errors much earlier in the manufacturing flow. Using specific algorithms to track touchdowns, big data solutions can automatically “flag” dice with too many touchdowns in real time and remove them from the production flow prior to final assembly. Previously, these suspect devices would continue on undetected only to fail prematurely in the field. By creating a dynamic rule to detect excessive touchdowns, subcons can prevent the passage of questionable devices to final assembly, increasing overall quality.
In today's fast-paced manufacturing environment, where there is mounting pressure to achieve greater throughput while also improving quality, it is essential for semiconductor and electronics companies to streamline operations by taking advantage of dynamic data rules. This approach ensures a checks and balances system that enables superior product yield with higher quality and reliability. Limiting touchdowns is just one way to provide a better defense against costly return merchandise authorizations (RMAs) and product recalls.