DeepFab

Wafer level yield prediction

inProbe provides automated yield prediction to recognize, monitor and report yield loses before probing. The model is used to determine whether any of the wafers have higher yield than a wafer yield threshold and can be skipped from the next testing.

Minimal total cost

The yield threshold is optimized to minimize the total cost of Unit Probe + Assembly and to significantly save required Unit Probe capacity. The total cost optimization is based on our unique solution.

MES integration

Real time prediction integrated in MES or batch mode predictions.

How it works?

See our Semiconductor Digest article for more information.

Estimate the inProbe cost savings for your product and fab:

Die level performance prediction

Using our cutting-edge technology, we deploy die level performance prediction based. The model produces failure probability for each die.

Process feedback

Yield prediction can be used to troubleshoot process equipment immediately after defect inspection.

Reliability risk under control

Dice that passed Wafer Sort but have a failure probability higher than a selected probability threshold are filtered out. The corresponding Wafer Bin Map with pass/fail dice created after Wafer Sort is then downgraded. The probability threshold can be set either on the basis of a cost model or according to the required failure rate target.

Data

inLab allows to integrate any kind of data into the single platform and to add, merge and combine data easily within the existing framework. Most importantly, the inLab approach means to curate data at the beginning considering analysis requirements at the end.

Analytical edge

inLab creates analytical edge through a well-thought-out combination of machine learning tools with customer's domain expertise. In a multi-dimensional manufacturing space with thousands of parameters, this approach makes it possible to identify a yield impacting issue with an unprecedented precision and speed.

Defectivity insight

inLab allows to link defect inspection maps to probe maps so that probe maps can be used as the response for the die and defect level analysis. At the die level, inLab Bayesian tools can be employed to evaluate detected defects potency and estimate yield loss explained. At the defect level sensitive die areas can be identified using inLab die defect impact heatmaps.

Universal issues

Pattern recognition of universal issues such as scratches, probe tips, stepper field and mask on Wafer Bin Maps with high accuracy.

Online mode

Automated pattern recognition allows to track universal issues online across a large product portfolio.

Historical data analysis

It is also possible to label historical data for analysis that would otherwise be infeasible.