Ansys IDEAS Digital Forum Banner 1
WP_Term Object
(
    [term_id] => 386
    [name] => Semiconductor Services
    [slug] => semiconductor-services
    [term_group] => 0
    [term_taxonomy_id] => 386
    [taxonomy] => category
    [description] => 
    [parent] => 0
    [count] => 584
    [filter] => raw
    [cat_ID] => 386
    [category_count] => 584
    [category_description] => 
    [cat_name] => Semiconductor Services
    [category_nicename] => semiconductor-services
    [category_parent] => 0
)

Exensio: Big Data in the Fab

Exensio: Big Data in the Fab
by Paul McLellan on 03-03-2015 at 7:00 am

 For 20 years PDF Solutions have been working with fabs on yield enhancement. Today, they announced their Exensio Platform for big data manufacturing environments. They haven’t really been keeping it a secret and have been talking about it at events since late last year, but it has basically been in stealth mode for the last 3 years. The primary focus of Exensio is to help fabs ramp to production volumes and then keep yield up once they are there.

There are 700-800 steps in a typical production process today. However, inspection is only done every 40 or 50 processing steps. This makes it hard to identify exactly what a problem is caused by. What is really wanted is to catch the problem when it happens and identify the tool (or material) causing the issue, a huge Fault Detection and Classification (FDC) system. But there are tens of millions of datapoints per second and so to do this would require hooking up every tool in the fab to a huge central database, capturing data within milliseconds so as to be able to catch even small excursions, uploading terabytes of data every day, and analyzing it all in real-time. Ideally the system would be able to react before the wafer was even finished processing.

And that is just what Exensio does. For example, every 300mm tool at TSMC is hooked up to Extensio. They started with beta customers such as Sony four years ago in 2011.


I talked last week to John Kibarian, CEO of PDF, about the Exensio announcement. He said that a lot of this is being driven by the explosive growth in data availability across the manufacturing flow (not just in the fab but in packaging, assembly and test too). There are more types of data collected, including data that has not really been used before. Larger factories and larger wafers generate more data. Plus there is a need for fast decisions so that corrective action can be taken immediately, as well as the slower analysis of longer term trends such as the day of the week, the specific operator, the specific tool and so on.

Most customers can and do conduct some of this analysis themselves. But typically this takes days or weeks to complete as opposed to minutes or hours. At some level, it is all about driving variability reduction which is the key to ramping a process to high volume manufacturing (HVM) and gradually driving yields even higher once that is achieved. With yield now sometimes driven by single layer atomic variation, the analysis can be very complex.


Exensio is built on top of the Cassandra database for scalability (as, by the way, is Facebook who have perhaps the biggest scalability problem of anybody). One top of this are:

  • Exensio-yield (dataPOWER)
  • Exensio-control (maestra)
  • Exensio Yield Analysis Services


PDF solutions have more process yield ramps to HVM than any other company, with over 60 below 90nm. They are the established leader with connections to all the major foundries such as TSMC, GF, ST, On, TowerJazz and more, along with relationships with the major fabless companies such as Qualcomm, NXP, Silicon Image, CSR and many more who work closely with their foundries as “virtual IDMs”. Exensio is the next step in the evolution of PDF, successfully leveraging big data architecture and technology. 80% of their business is leading edge with 28/22/20nm now just a niche and most work going on at 16/14/10nm.

PDF Solutions website is here.


Comments

There are no comments yet.

You must register or log in to view/post comments.