Ambiq yieldHUB casestudy
Ambiq Micro is leading the world in energy-efficient micro-controller (MCU) design, redefining "ultra-low-power" with its unique and proprietary SPOT platform. They began working with us a few years ago when their data started scaling (they now produce thousands of wafers a month!) and they needed an effective yield management system to help them. Senior Product Engineer, Jerry Kao shares his favourite yieldHUB features.
Jerry Kao, Senior Product Engineer, Ambiq Micro
The three best yieldHUB features, and the ones that I use most often, are:
The bluetooth low energy test is key to keeping yields under control for one of our products. Knowing the distribution of yield across sites is extremely important and we can do this for any given lot or across hundreds of lots if needed. If we see a deviation we can do some work to fix the offset and yields go back up.
The visualisation in the charts and the breakdown in the stats tables for parameters such as this gives us valuable information which we can share with our manufacturing partners as needed to help us optimise our yields. I can safely say that this type of volume monitoring of such a parameter is not possible using traditional desk-top analysis tools and spreadsheets. It’s right in front of you in seconds using yieldHUB!
When there is an issue with a wafer or a lot, it’s important to be able to find wafers that behaved similarly in the past and compare that group of wafers with more normally behaving wafers. yieldHUB has fantastic customisable search and analysis capabilities to enable all of this. Our volumes are now multiples of what they were even two years ago. The ability of the yieldHUB software to search and analyse so quickly even at such massive scale is hugely impressive.
For example, with hundreds of thousands of wafers in the database already, it can find in seconds wafers that have a particular bin or parametric behavior. Those wafers could have been tested at various points over the past four years! The common feature of those wafers might be a particular attribute of a fab parameter and that can be found with ANOVA and similar tools.
Again, this takes seconds and is faster now than ever with the new server technology yieldHUB invested in a few months ago. The image below shows how simple the interface is to search the entire database for wafers from a test program that has a specified parametric behavior. You can even do this parametric search for units that have specified bins.
Composite bin pattern chart:
We find that there are many useful tools in yieldHUB for quickly determining the true source of a yield issue. Such yield issues can sometimes be caused by the wafer foundry which is beyond our control, of course, being a fabless company. The issue can also be because of something intrinsic in the design of a product or the performance of the test fixture or tester. In all cases, we have found yieldHUB to be extremely useful in helping us find the real source of the yield issue. The tools are visual and tabular and excellent for collaboration with our foundry.
For example, we had a serious yield issue where the isolated bins 9 and 10 as having distinct patterns across the affected wafers. yieldHUB software allows you to stack wafers (even hundreds if needed) quickly and with very powerful visualisation.
The output from yieldHUB helped to convince the Fab to run commonality across their tools. Commonality is a method for checking the difference between good and bad wafers in terms of which tools were used in manufacturing the good and bad wafers. Often it identifies that one tool had a different setting or there was a different tool at a given step in fab used to manufacture the bad wafers. This analysis allowed the fab to adjust and put in finer control a key tool which affected the wafer performance. Often the foundry will not know such effects without feedback from their customers and yieldHUB's capabilities are ideal for such feedback.
The tool within the foundry was fixed and the wafer yield went back to normal. Without the speed and visualisation of yieldHUB, we have no doubt that the yield would not have been rectified in such a short time.