Gt data mining ai algorithm for fabs

Edith Ohri
Datalert (startup)
edith@datalert.co.il
FabHighQ
- a GT data mining application

Team member - Edith Ohri
• Edith Ohri - Industrial & Management Eng. from Israel
Technion and MSc from NY Polytech.
• The developer of GT data mining (big data); GT
received a research grant from SMU Singapore.
• Currently works on applications of GT.
• Previously initiated a project sponsored by Israel Chief
Scientist on “Procedure Systems Diagnostics”.
• Prior to that worked in Israel and the USA as an IE
Eng. And System Analyst.
7Sep-2018 Garage+ presentation by Datalert © GT analytics Slide 2

Fab-High-Q Value Proposition:
HIGHER YIELDS
• Fab-High-Q produces higher yields by finding
the root causes of Quality faults, and guiding
their detection and early stage prevention.
• The solution has a very high net value thanks
to using existing infrastructure, thus saving
the large part of investment, preparations,
integration, training and maintenance costs.
7Sep-2018 Garage+ presentation by Datalert © GT analytics 3

Further Benefits
• Shorter time to market for new products
• Flexible capacity planning
• Faster Learning
• More automation
• Reduced overhead

The product: Fab-High-Q
Fab-High-Q is a data analytics algorithm that
spots Quality risks, sends alerts and suggests
corrections, based on GT – a proprietary
universal solution.
Main strengths:
– Discovery of hidden patterns
– Early detection
– In-depth Machine Learning
– Robustness, very little pre requirements
– Supplied as a turn-key project.
7Sep-2018 Garage+ presentation by Datalert © GT analytics Slide 5

Options of implementation
• Per order project
• Licensing
• Collaboration.

Example – Fab Higher Yields
• The case: a Fab which specializes in small-batch
manufacturing, experienced a steep decrease
in yields.
• GT was called in “to find out which work
stations are responsible for the quality failure”.
• It found that the root cause was rather in the
wafer supply processes. The findings led to
new insights about prevention and
improvement of yields.

Example cont. Overcoming Bad Data
• The data input was challenging. It had many outliers,
noise, missing values, and disrupted records due to a
computer upgrade that took place in the critical period.
• GT overcame the bad data by applying cluster analysis.

Example cont. Finding a clue
Among the clusters there was an irregular tiny
yet significant pattern (marked in red) that
occurred in the beginning of the Quality
deterioration. (cont. next slide)

Example cont. The Root Cause
7Sep-2018
The Quality failure was indeed related to a special
event: a very large production order in the 2 lines under
study (red and blue).
The large order created a shortage in high quality
wafers. The “blue” line started to fail consequently until
brought to a halt.
Garage+ presentation by Datalert © GT analytics 10

Example cont. The Solution
• The answer to a raw-material shortage is
increased supply by acquiring additional
quantities from internal or external sources.
• The best way though to increase the supply
would be to improve the quality of the in-
house wafers, by learning with GT their “long
tail” quality distribution.

GT founds in the wafer long tails several sub-
groups. The high end sub-groups could be
expanded thereby improving the total ovens
turn out.
Example cont. Exploiting Long Tails

Example cont. Lessons to draw
• All data sources are equally good; the more
unsupervised the better it is.
• Presuming in big data analysis doesn’t help.
• Good AI creates more answers than asked-for.
• Focus first and most on root causes.
“Defining the source of the problem is half
the way to the solution” (a Hebrew proverb).

Thanks
Edith Ohri
Datalert (startup)
edith@datalert.co.il
* The pictures are from free sources on the internet

Gt data mining ai algorithm for fabs

More Related Content

What's hot (20)

Similar to Gt data mining ai algorithm for fabs (20)

Recently uploaded (20)

Gt data mining ai algorithm for fabs