Outlier Detection incorporates Parts Average Testing (PAT) and GDBN and is a vital enabler for chip companies serving the automotive industry. Our previous version of Outlier Detection was successful in several companies but we have decided to rewrite it for scalability. We also have outlier detection in place for aerospace customers but the requirements are slightly different. The new upgraded module for Outlier Detection is for release very soon and is for high-volume wafer sort. Several companies needing automotive capabilities later this year have already signed up for our new module.
What is Outlier Detection?
“Outliers: parts whose parameters are statistically different from the typical part.“AEC
Outlier detection is a method of identifying a member in the group that deviates grossly from the norm. You test semiconductors across different parametrics, such as voltage, current and how it reacts to different stimuli. You calculate the average manually or let your Outlier Detection system crunch the numbers for you. Using an algorithm, outliers that deviate vastly from the norm are found. In other words, you calculate what is average, then exclude what isn’t.
Outlier detection is a major way to ensure quality by identifying and removing potentially defective chips.
When applied in yieldHUB, the system detects the anomalies using pre-defined recipes after the data has been uploaded to the cloud. This makes the solution highly scalable and puts the control of quality firmly in the hands of our customers instead of in the hands of the test houses.
What is an outlier?
In statistics, an outlier is a result that deviates from the normal values. In the semiconductor industry, an outlier is a chip that has passed all standard tests. But it differs from standard parameters in one or more categories. According to the AEC, these chips are more likely to fail earlier than standard chips, then they are in the device. For that reason, it’s important to remove them at wafer level.
Why is outlier detection important?
Outliers pass the normal test program as there is nothing inherently wrong with them when compared with the datasheet. However, would you want a semiconductor that is very different from the norm controlling the airbags of your car, or inserted into your high end smart-phone? Don’t think so. So while the die would pass normal testing, algorithms are applied to remove these units so they never see the light of day.
Outlier Detection: Key points
Large array of analysis for wafer sort, final test and WAT/PCMSPAT, DPAT, GDBNDPAT applied per wafer, per test, per site
Specialist tools also available or can be added
No need to download data first, just typically: “search->select->analyse”
Puts you in control of quality and reliability
Module is fully Compliant with AEC-Q001 Various input and output formats
Test limits for every site on every wafer recordedData can be stored indefinitely
What industries need outlier detection?
Outlier detection is essential to the automotive industry. Quality and uniformity are essential to ensure safety in the running of motor vehicles.
Increasingly we are seeing the need for some form of outlier detection to be applied to dice going into consumer goods such as high-end smart-phones.
What’s different about our implementation?
The outlier detection algorithms will be easy to set-up per test and per product. The actual DPAT limits used for any die will be available in an audit trail. Who changed algorithms and when will be recorded. The effect on yield will be recorded and massively scalable analysis will be available like many other analysis tools in yieldHUB. You will also be able to run simulations easily on our web-based system before deciding on which algorithms you would like to apply for a given test or bin.