For JMP users: Save hours every time with yieldHUB

By John O’Donnell, Founder & CEO, yieldHUB

So you’re a JMP user and your company has just purchased yieldHUB. So you’re wondering, is this a good thing or a bad thing? You have invested years into JMP and although you find it always took a lot of time to get the right data into JMP, you were happy once the data was in there. JMP doesn’t do data management. So now that yieldHUB is available to you, does this make it easier or harder to get your data into JMP, if at all?

Well actually, you’re in luck. yieldHUB itself is great for analysis but now lets you download to external analysis packages such as JMP and Excel with a preparation step that will take only a couple of minutes of your time. So the time saving sounds good, but, you’re asking, what type of data are we talking about?

yieldHUB Data Management & Data Integration

yieldHUB includes a relational database that contains the production datalog data from all your subcons, updated several times a day. Hence all the data management and integration is done within yieldHUB.

The following are the data that are stored and linked up relationally:

  • Fab PCM/WAT data, Wafer sort and Final Test data (e.g. STDF)
  • Characterisation data for new products
  • MES data  (e.g. actual quantities tested, package type etc)
  • Genealogy data (eg which wafers from fab are in which final test split)
  • Inline Fab data when available
  • Comments and other knowledge sources

Having all this data in a scalable relational database allows powerful analysis within yieldHUB itself. 

But I need to download data to use JMP!

Currently, you’re not that interested in the analysis that yieldHUB provides. You just want to use JMP! So you need to download the data into JMP and take it from there.

To answer this need, MFG Vision has updated yieldHUB to do a preparation step for any such download to Excel or JMP. This includes all the genealogy information also and MES information. So you have the connected data downloadable, up to 20,000 wafers at a time!

“The JMP download takes me seconds to set-up even for thousands of wafers. The MFG Vision team have done a great job in making this so easy for me and my team to run very deep analyses.” – Engineer, Korea.

A typical download would be for a few hundreds of wafers of data for several key tests in wafer sort. All the raw data can be included, with summary stats, including Cpk, bin information for each die and the fab lotid for each wafer. The download format is compatible with JMP or Excel, depending on which option you choose.

With all the time you have saved, you should have time freed up to learn about why it may not even be needed to download to JMP in the first place! yieldHUB’s powerful native analysis may save you from paying for that JMP license for next year!

Contact:  +353 61 309745 or any of our sales agents worldwide if you are interested in finding out more. We can set up a web conference call with you to demo these capabilities so you can JMP in!

(A∪C) ⊆ (A∪B∪C∪D)

By Jason Bent, Principal R&D Engineer, MFG Vision

In the manufacturing of physical, non organic, items there are, roughly speaking, three areas:

  1. Subtractive: start with a block of stuff and remove bits until you are left with the object you want
  2. Additive: build up increasing amounts of stuff until you have the thing you want
  3. Assembly: combine several items together (made by one of the first two techniques) until you have the thing you want, usually a ‘complex’ machine.

We can think of data in similar terms. Subtractive: start with a big table and filter out bits until you have just the bits you’re interested in. Additive: stick together small sets of data of a common type until you have the set you’re interested in. Last, assembly: put together data from different sources and types to create a big picture, complex machine.

In physical object manufacturing, additive techniques have enjoyed something of an resurgence in recent years. Additive 3D printing now offers rapid production of mechanically viable parts, not simply pre-production models that give stakeholders warm fuzzy feelings before signing off that six or seven figure tooling cheque.

Manufacturing Data Analysis

When we look at how manufacturing data analysis is typically done we notice something a bit odd. Generally we do additive and then when we want to investigate our data we do subtractive. More specifically we add together many sets of data into a few huge tables and then when we want to look at that data we apply some query filter to the superset to then produce a smaller subset that is of interest to us.

So what? Well the thing is this is both inefficient and also inflexible. It’s inefficient because it has redundant steps in the process but also, significantly, because it requires a super set that is huge. Big enough to contain all the data we are possibly interested in.

This brings me on to inflexibility. Look at the title of this blog; In non symbolic form ‘the union of sets A and C is a subset of the union of sets A,B,C and D’. What if we were interested in AE? We can’t do it because we never included E in our original set of everything we could possibly be interested in. With an additive approach it wouldn’t matter we hadn’t included E to start with.

Smaller Chunks

It looks to me that the reason we do this is historically driven by the tools we use. SQL type databases, spreadsheets and the like are strongly biased toward subtractive techniques, or filters to use the more common term. Maybe we need to turn our thinking on its head. Instead of trying to work out everything that we may possibly be interested in and lumping it all together, to be carved out later, why not store our data in smaller junks. The trick is to organise those junks in a way you can get at them efficiently.

Turns out computers are actually very good at doing this. In fact when you want to combine your small junks together into a bigger junk to work with you don’t even need to move those small junks around. They can stay where they are.

Secret Sauce

There is one last trick. I mentioned in the first paragraph assembly as our third technique. Well here’s where the secret sauce is. Combining like data is easy enough but combining disparate sources in an on-demand additive manner is harder. But that’s all I’m saying on that, if you want to know more you’ll have to speak to us.

Effective Semiconductor Yield Management: 8 key questions

Effective Yield Management ensures that you minimise scrap and lower the cost of manufacturing. Whether you have one product or a thousand, the effort is worth it.

Take a product with a volume of 100 wafers per month where the yield is 93% and each wafer costs $5000 from the foundry. A yield improvement to 96% means you will save $15,000 per month or $180,000 a year just on that product, while delivering the same number of good dice to your customers.

Large companies will see millions of dollars or savings per annum when an effective yield management strategy is employed. For this you need to have a suitable information system for yield.

Does your company have such a system? If put on the spot, would you be able to answer the following questions for your top volume product in the next few minutes using your current yield management system?

  1. What is the yield up to today and what is the trend over the quarter?
  2. What are the highest failing bins and their trends?
  3. What are the highest failing tests and their trends?
  4. What impact is Fab having on the wafer sort and final test yields?
  5. Are the parameters being tested sensitive to the test software or hardware?
  6. Is the yield always uniform across the wafer? If not, why not?
  7. What percentage of good units are failed incorrectly because sites aren’t yielding consistently?
  8. Who is alerted if there are yield, site, bin or parametric issues in production and when and for what were the last alerts sent out?

If you and your engineers can answer these questions in minutes, then you are most of the way to effective Yield Management. If it would take you hours or days to get the information, then that is time that can be saved with a proper information system. 

In a highly competitive landscape, why have your engineers waste hours and hours searching for data and generating reports before they can get cracking on finding root cause for yield loss?

Waste that time no more. We at yieldHUB would be delighted to help and such a system can be all up and running sooner than you think.

Contact:  +353 61 309745 or any of our sales agents worldwide if you are interested in finding out more. We can set up a web conference call with you to demo these capabilities.

Characterisation of New Products

By John O’Donnell, April 8th 2018

I recall doing Characterisation on many occasions on analog and mixed signal devices twenty years or so years ago. The goal of Characterisation then and now is to make sure that the variation in behaviour of semiconductors due to changes in Material or Test Conditions is fully understood. Examples of Material Conditions are Process Corners which represent the extremes of fabrication parameters at which the devices are still expected to function. Examples of Test Conditions include Voltage, Temperature and Operating Frequency.

Twenty years ago Characterisation used to take two or three weeks and it was and is a major item in the critical path. Designers are waiting eagerly for what you find. With tools up to now it’s from a few days to a week to complete characterisation once testing is finished. Twenty years ago a fast database was out of the question and Excel was your friend. Unfortunately for many companies it still is their friend and means they spend too much time still generating the reports.

With modern tools the characterisation report should take only a few hours once testing is complete.


Figure: Normal Probability Plot from yieldHUB with some groups highlighted

The following are Huge Time Savers in any such Characterisation tool:

  • Simple for the test engineer to generate compatible datalogs (e.g. STDF or text based) so that the software automatically detects both the Material and Test Conditions associated with each test result
  • Fast, Interactive Analysis across all conditions with Flexible Filtering and Regrouping
  • Multiple chart types for maximum visualisation
  • Automated report generation across hundreds and hundreds of tests with varied visualisations throughout the report
  • Reports may be saved on-line and be combined into a Master Report
  • Ability to add comments on-line to the report & to each test individually
  • Ability to easily replace a subset of the data if retesting is required for a certain set of conditions
  • Ability to deal intelligently with outliers
  • Easy navigation throughout the report
  • Easy sharing of the report with colleagues

Using the above principles in the design of such a characterisation tool, many days of work in report preparation and rework can be reduced to only hours. This greatly accelerates discussions around the key issues with designers and foundry engineers. In the semiconductor industry, where time to market is so important in new product introduction, this makes a tangible difference.

Contact:  +353 61 309745 or any of our sales agents worldwide if you are interested in finding out more. We can set up a web conference call with you to demo these capabilities.

Partnership in Taiwan between MFG Vision and Best-IIC

There has been excellent progress made in Taiwan this week between MFG Vision and our partners Best-IIC. 

MFG Vision Directors Kevin and Jerome visited to meet with staff, partners and customers. 

Kevin Robinson, Director of Customer Success says “It is always great to be in Taiwan, to visit our customer success manager and to meet our excellent partners from Best-IIC. Our joint visits went very well and we look forward to a continued busy time together”. 

Pictured left to right are: Jerome Auza – Director of Engineering and Enid Chen – Customer Success Manager, both of MFG Vision and Sam Chen & Gary Lin from Best-IIC with Kevin Robinson – Director of Customer Success MFG Vision.

Taiwan is a hot bed of factories (“test houses” or “test subcons”) dedicated to assembling and testing semiconductors for companies around the world. Taiwan also has many fabless semiconductor companies who specialise in the design of semiconductors and outsource their manufacturing to local companies in Taiwan and mainland China. Both types of company need to have a complete understanding of yields to succeed in the very competitive  semiconductor industry.

Real Time

MFG Vision is in a unique position to grow in Taiwan as we have the technology required for engineers in the test houses to keep on top of yields and collaborate with their fabless customers using our on-line yieldHUB platform. It includes real time technology. For example, if a batch of 30,000 semiconductor chips is about to go on a tester and will take four hours to complete testing, the test house will want to detect any issues when they actually happen, not afterwards, to maximise efficiency and minimise cost. This is extremely hard to do and to do so seamlessly, without additional infrastructure and at scale. In a test house in Taiwan we have proven this is actually possible.

Our Best-IIC collaboration means that this message (and more messages of our capabilities built up over 13 years) are being transmitted across the test industry in Taiwan since 2017 from a trusted local Taiwanese sales partner, which is leading to exciting times for MFG Vision and Best-IIC in the years ahead.

Figuring out binary datalog formats without a specification

by Jerome Auza, Director of Engineering , MFG Vision

The Challenge

Being in the realm of semiconductor data with a wide range of customers, companies often throw interesting technical challenges at us. The most complex one so far this year is probably a request (OK, a requirement!) to interpret binary test datalog files so that they can then be analysed from our yieldHUB database system.  The company provided us with little information on the actual binary format.  We had to interpret the binary files based on their equivalent ASCII files and our knowledge on how datalog data is usually stored.

We needed to parse the  binary files directly because generating the ASCII version is a manual operation an desktop software.  It could take thousands of man hours to convert their existing archive of the binary files into a format compatible with entry into our database.

After several days of combing through the binary files searching for patterns and the values that we know from the ASCII version, we were able to write our own converter for the binary files.  Our converter was written as a command line program so that we can integrate it into the automated datalog processing and allow our customer to analyze their historical data using the tools in yieldHUB.  The journey was extremely challenging but when we finally discovered all the needed information, it felt all the more satisfying.

Lessons Learned

In this project, the key lessons we learned are the following:

 — the “endianness” of the binary files need to be understood first.  We got stuck into a confusing trial and error method because some files were little endian and the others were big endian.  It was unexpected that a single system can have both but it did happen in this case.  

 — the script or software used to read and interpret the binary files need a way to specify the endianness of the binary data.  We were initially using an older version of the scripting language to detect the numbers and we never found the correct values until we realized we needed to use the newer version.

— The 8 byte floating point value 0.12 may not be exactly 0.12 in the binary representation but could be 0.11999999999.  Because we were aware of this from the start, we did not look for floats or doubles at the beginning but rather for ASCII text and integers.

— A good visualization tool is key to finding the patterns in the binary data, especially for the part of the data that represented multiple records.  The tool we used allowed us to see the obvious patterns very quickly. It also aided us in validating byte positions and block sizes for the multiple record data.

Does your company have such an archive of data that you can’t interpret or analyse any more?

Ramp your Multi-site Test Yield Faster with “Test Environment” Gage R&R

Ramp your Multi-site Test Yield Faster with “Test Environment” Gage R&R

by Kevin Robinson, Director of Customer Success, MFG Vision

Gage R&R has been used for many years in the semiconductor industry to measure variation during test for repeatability and reproducibility. These measurements are then used to assess the suitability of the test for production. But often the focus is on qualifying tests on one site only leaving issues for production when multi-site testing is used. See an example below of a parameter where the site 2 measurements are different to the other sites and give many false fails. This would fail Gage R&R.


Many Factors at Once

Gage R&R typically requires that tests are qualified with only one test environment factor at a time, so it is often the case that not all factors are assessed. This is simply due to time and resource constraints. It is rare that test engineers will have a few weeks to do many studies. So why not consider more factors at once? The main reason is the complexity of interpreting the results for multi-factor analysis. However with modern analysis systems this is no longer an issue. The new version 3 of Gage R&R in yieldHUB for example will support the entire process from debug of tests, to qualification of tests, to qualification of the test environment.


So back to the multi-site problem shown above. The efficient test development route is to assess repeatability on one or two units and fine-tune tests as needed. Then carry out a study with more devices and a key factor of variation. And finally vary everything in a single study and analyse the data. By being able to select data and run reports in a few mouse clicks generating tens of fields of statistics makes it much more efficient to analyse the data. Data collection can be ambitious to prove that the entire environment is stable and can be qualified. But if a problem is seen the cause can be traced and only if necessary are more detailed gage studies carried out.

If your yield ramp-up of multi-site testing is not looking smooth in initial production then yieldHUB’s Gage R&R V3 is there to help you. The unique combination of MSA Gage R&R analysis along with additional max range analysis brings more insight into the data and the causes of issues in test. It is possible to launch multi-site testing into production with high yield from day one and this can be achieved with less effort than ever before.

Contact:  +353 61 309745 or any of our sales agents worldwide if you are interested in finding out more.

What types of companies do business with MFG Vision?

Here’s an overview of our customer types who come from many countries around the world:

  • High profile international conglomerates who have semiconductor divisions often supplying chips to their own divisions, who like to have a system that helps them analyse millions of chips a month
  • Fast growing fabless start-up companies who want to concentrate on design wins and like to use our drillable dashboards to manage subcontracted manufacturing. They don’t want to be concerned about all the idiosyncrasies of STDF and all the other data formats live today
  • Companies who sell into the military market in the USA and are paranoid about their data and the ITAR regulations associated with such data, so need a trusted data partner
  • Fast growing system-on-a-chip and module companies who have complex supply chains and need to correlate across a multitude of steps to extract the highest final yield possible to keep their operating margins high
  • Companies who haven’t enough engineers to watch everything and so like to have automated analysis of what is going through the factory. We make sure their engineers are automatically notified to analyse what is actually important
  • Companies who have no IT department and like that we can host everything for them safely and securely on AWS. Of course we also offer in-house hosting where we don’t even need to see any real data ourselves
  • Companies who like to collaborate closely with their suppliers. Our third party on-line licenses allow such suppliers to analyse data from our customer to help our customer achieve high yield and quality delivery faster than ever. Examples are wafer suppliers to our customers who are provided third party licenses to see the effect variations in their fab are having on our customer’s products
  • Companies who see value for money in SAAS technology, or hybrid SAAS when they host our software in their own facilities. These companies typically are moving to the cloud or hybrid cloud for many of their systems, not just Yield Management software.

Who typically uses our software then in the above companies?

  • Product and Test Engineers for detailed and fast root cause analysis of any yield or parametric issues that arise in the course of a normal day
  • Managers and higher ups who use our Bird’s Eye View dashboards for an accurate and up to date glimpse of what is going on across their company in terms of yield and throughput
  • Foundry engineers who work between the Fabless company and their Fabs, to correlate any fab issues with the quality and yield of the semiconductors that end up as finished goods for sale.

What’s the typical result?

  • A complete understanding by our customers of their yields and efficiency of manufacturing. And if not, we close the gap being a listening and agile company dedicated to customer success.

Can you see the wood for the trees?

At MFG Vision we receive many requests from customers. Some requests are for help with a problem that can might be able to be solved with the right data analysis. Some are requests for “make me a chart of this!” Both are interesting opportunities. Let me share an example of the latter type. We were asked to make a chart to include the data shown below:
MFG_Curve_Chart_AWhat does this show us? It looks like the data is unstable, then increases and then starts to disappear. Not much else to be read into it.
Now try removing extreme outliers:

Now we can see that the majority of the early data is very stable and that there is a more gradual move to increase in the data.
But is this the appropriate chart for this amount of data? Our customer thought so until we showed them another option:
Now we can see the behavior of the majority of the data far more clearly. The earlier plots made it very difficult to see how the values are increasing steadily before dropping off again.

This is a great example of what can happen when people want to see all of their big data. That’s not to say that it is never appropriate to show all the data. But finding the right type of chart and the right way of representing the data to give the right level of information is key. If you’re frustrated at looking at the wrong level of detail, either too much or too little, in your data then contact us at MFG Vision and we’ll see how we can help you.

Make Vs Buy of your Yield Management System


Is it ever worth re-inventing the wheel? Every industry outsources key functions over time. Airlines outsource baggage handling, construction companies outsource pipe-laying. Many fast growing technology companies outsource enterprise software and many are tempted to build their own. Yield Management Software (“YMS”) is no different. 

Why have a YMS system at all?

Whether built internally or bought from a vendor, YMS is essential for saving costs and delivering product on-time and with high quality. Manufacturing more wafers just to get enough product to a customer is not good enough. Your margin will disappear and your competition will eat your lunch. On the other hand, providing your engineers with real time in-depth engineering information from manufacturing has an indisputable ROI. Informed decisions will be made quickly by your skilled engineers to improve your margins and optimise delivery and quality.

Why not build your own?

If I am involved in running an up-and-coming semiconductor company and have a few engineers who are skilled in IT, then I could have them work for a few months on bringing all our data together from our subcons and providing a database accessible by everyone working on yield. These internal IT specialists would be tasked to provide a web-site which is searchable by manufacturing stage and lotid. Any engineer throughout the company could download any data they want when they want it. Indeed I could just hire maybe 5 people to do this for six months. That would cost maybe $300,000 and then I would have a great system for my ten or so end users, right?

Well, let’s see!

Since you’re an up-and-coming company, it may well fulfil your needs initially, but your data volume will grow and your requirements are going to change. So you think: “OK, I’ll hold on to one or two engineers and then 50% of their time can be on the ongoing debug and support of the system.”

So now it’s costing $100K+ a year. Vendors are calling talking about how they can link up the data across manufacturing but one of your engineers in Taiwan says he also can do that within the next few months. You buy new testers and new formats other than industry standard formats like STDF appear. The IT guys working on the system are not sure about the actual DB design as they were more familiar with the UI and training the users. Then the engineer who wrote the scripts for characterising devices has just retired. And his scripts are not fully documented. The Taiwan engineer got dragged into another project. It’s difficult to find a replacement as it’s specialised work. The cost of all this is only going one way.

The number of people who need to use the system is now above twenty but the support for them is erratic. You will have to invest another couple of hundred K very quickly in specialist hires. Vendors are quoting a fraction of the per annum cost for the current number of users you have for your own YMS system and are comfortable with your data formats. But you say, no, I’ll plough ahead with my internal YMS solution.


Six months later you have decided to move your test engineering to Taiwan to be closer to the foundries. So half the users will be offshore and you don’t have training and support available for them in the YMS system, not even taking into account their time-zone. In addition the website is starting to crash because your volumes have more than doubled and the design was only capable of operating with half a terabyte of data.

New Opportunity Missed

Also there is a cheaper multisite test solution for testing products from a low margin but fast growing product line and the YMS system is not capable of analysing the data at all. The compatibility of the new data format with the YMS system was an afterthought. Now an opportunity has come up for that product line to move to automotive but since they can’t see the production performance easily the decision is being slowed, potentially missing a higher margin opportunity.

Advantages of using a Vendor

If however a good and established vendor were used from day one, the setup cost would have been a lot less and every one of the issues encountered would not have been a big deal. YMS vendors, if they are worth their salt, can set up a system for a small but fast growing company in a few weeks maximum. They would not just provide a web-site where you can download the data. They would provide an interactive website where you can directly analyse it without downloading the data. You would not have to pay them fully until they delivered to your requirements.

Customers like You

The vendor would link up the data across manufacturing, assuming you have a good MES/ERP system they can use. New formats to you would not be new formats to the YMS vendor. They probably have 30 or 40 customers like you with all types of formats and access to the best ideas for your next YMS challenge. They might have to tweak some parsers but this should not cost the earth.

You would ideally have a multi-year contract with the YMS vendor. Include minimum support requirements in the contract and you should have no worries.

The YMS vendor will also have highly skilled support in different continents. The database design will be able to scale to Terabytes, so don’t worry about fast ramps when you get your contract from Cupertino!

If you move into automotive the vendor will have all your requirements available in terms of data and monitoring. If you have particular needs they will work with you quickly and efficiently on them.

Needs Evolving Over Time

So as your needs evolve, your YMS vendor probably already has a solution or can develop one quickly. For example you may need a special type of SPC alert. You can include in the contract the requirement that the system is extended and upgraded at a reasonable frequency. The vendor should have the skill set to do almost anything you dream up that would help with your yield, quality and even reliability. They will probably charge NRE for custom requirements.

IP considerations

If you are concerned about the IP of the data being generated, ask the YMS vendor if you can host everything yourself. The vendor should be able to support this and not even have to ever see your real data.

Financial Considerations

If your vendor already supplies to much larger companies than your own, their financial stability should not be a concern, but sometimes a vendor may be under pressure by its investors to charge a lot of money. Make sure you know the ongoing costs of upgrades and maintenance. Know who is backing the vendor financially. Some vendors have sold out recently so make sure you cover such a situation in the contract.

Other considerations

Does the vendor have customers who will attest to their after sales service? Have they demonstrated that they are responsive and agile? How many years have they been in business?

Summary of Advantages of Buy instead of Make

  • Setup speed   Can setup very quickly, probably days or weeks not months (unless you’re a very big company)
  • YMS Design   Scalable and Robust, built on tens of man-years’ of experience
  • Data Variations Has probably seen them all before & dealt with the extremes
  • Software Development Probably uses latest S/W development methods (Agile etc)
  • Support  Probably has Worldwide support. Not dependent on certain people.  Teams of support engineers
  • Maintainability Probably very stable, scalable system which requires very few resources to support
  • Integration Probably has worked with your MES/ERP before and can integrate with it
  • Depth and Breadth of Analysis Tools  If the vendor is in business for years they will probably have 99% of what  you will ever want & if you want something custom, they will be able to do that for you for a minor cost


If after all the above you still are tempted to develop the YMS solution in-house, just make sure that the database design is scalable and that the engineers developing this internally are not going to be distracted from their YMS job by other IT needs later on. 

We would obviously recommend contacting a vendor like ourselves and let you concentrate on making the best semiconductors possible.