Attention Fab Managers: Defeat the dreaded yield bust with “MY GOODNESS!™”, S-Cubed’s all new, AI powered subsystem that combines machine learning (ML) and intelligent sensing to save the day!
Understanding Yield Bust and Its Effects on Semiconductor Production
The yield bust in the semiconductor manufacturing industry occurs when a high volume wafer fab consumes a huge amount of resources (“like an elephant”), yet only produces a small fraction of optimal results (“like a canary”), due to an unknown/uncontrolled process parameter. Even when optimal results are produced, it cannot be replicated easily, leading to large amounts of wasted resources and capital. In summary: it’s when we make the “good ones” by accident, rather than on purpose.
The solution to avoid yield bust is simpler than one may expect: gaining awareness of as many process parameters as possible during processing (keeping in mind low cost and complexity).
So how do we go about achieving this awareness? Well, just as in inspection routines where we use a known good wafer (aka “golden wafer”), the same concept can be applied to process parameter monitoring. By intelligently observing through ML as many process parameters as possible with the “golden wafer”, one can establish what constitutes acceptable statistical variation and what constitutes being out of range. With this new data, any out of range deviations that occur can be detected quickly, and the tool can be alarmed accordingly to enable the cognizant engineer to determine the root cause and ensure all is in order before proceeding.
Improving Wafer Processing Through AI and Effective Data Collection
For photolithographic wafer processing, which is often characterized by many process steps involving multiple parameters, one must guard against unwanted variations in the following parameters (this is for just starters):
- Enclosure (ambient) temperature
- Enclosure humidity
- Particulate concentration throughout enclosure and the process
- Exhaust flow rate and pressure
- Consumption(flow) rates of solvents
- Consumption (flow) rates of polymers (resist)
- Spin speed
- Acceleration
- Bake temperatures and length of bake
- Chill temperatures and length of chill
- Overall electrical current flow to the tool
With this information in hand, the MY GOODNESS!!™ system logs histograms of all the parameters listed above (and more as needed) and develops with ML what constitutes nominal (good) operation of the tool. When any deviation occurs the tool alarms, and the engineer can download the appropriate histograms of “nominal” and deviant tool behavior. This data helps the engineer find the root cause of an issue and decide upon the appropriate next course of action.
Do you think the MY GOODNESS!!™ system might be right for you? Reach out to our experts to learn more and/or get a quote.
