Path: EDN Asia >> Design Centre >> Medical >> Assessing production outlier removal techniques
Medical Share print

Assessing production outlier removal techniques

25 Jun 2014  | Wesley Smith

Share this page with your friends

Figure 2 shows an example of a parametric wafer map that is colour coded by the measured values, with red and orange being on the high side and green and blue on the low side. The orange/red die shown in the magnified boxes are surrounded by green die, which makes them outliers when compared to their nearest neighbours, but not when compared to the average die on the wafer. NNR rules would correctly identify these die as outliers and bin them out.

Figure 2: Nearest Neighbour Residual exposes hidden outliers by comparing parametric results to other die in the same area of the wafer.

Another sophisticated approach that identifies a different class of hidden outliers is called multi-variate PAT. This technique looks for variations across groups of similar/correlated tests instead of just looking at each test individually. For example, if a part has four leakage tests that are usually closely correlated, and three of them track normally but one test is on the high side, this might be considered an outlier- again because it uses contextual data to make a more accurate decision. In the example shown in figure 3, the four leakage tests tend to track closely from part to part, except for part ID #8 in which parameter 4 is unusually high and could be considered on outlier.

Figure 3: Multivariate PAT compares two or more correlated tests to identify outliers such as part # 8 in the example above.

Multivariate outliers
The example shown above is conceptually quite simple but the underlying math used to compute multi-variate outliers is a bit complex. The test results are first normalized to a principal component value, and then a Mahalanobis Distance (MD) is computed as the distance from the mean for each test value. Parts that fall outside of the MD are considered outliers. Figure 4 illustrates an example of the multi-variate analysis for two correlated leakage tests, which you can visualise in a two-dimensional ellipse. The red dots have principal components that exceed the MD and are therefore considered outliers.

Figure 4: Example of Multivariate PAT analysis of 2 correlated leakage tests. The red parts, which fall outside of the Mahalanobis Distance, are considered outliers.

Multivariate analysis can actually be performed on any number of correlated tests, but exceeding three tests would require more than three dimensions to visualise the Mahalanobis Distance, which would make it difficult to represent graphically. At some point you just have to believe the math.

Turn loss into improvement
In highly competitive markets like automotive and mobile, OEMs are reluctant to pay more for devices with enhanced quality based on testing processes like PAT. So if PAT leads to an incremental yield loss of, typically between 0.1% and 1%, the IC supplier will usually have to absorb that cost. So how does one justify the incremental cost? One way is to accept it as the cost of doing business with high volume and/or high quality manufacturers. A ≤1% cost increase is a lot better than losing a contract. Another argument is that by reducing the number of field returns there is an associated cost savings related to reducing the amount of failure analysis required, which can be quite expensive and time consuming.

Most enlightened users of PAT and other similar techniques have, however, discovered that by analysing the outlier data and feeding back their findings to the manufacturing process they can actually improve overall yield by better centring the fab process and/or eliminating assembly related defects. So for example, instead of seeing yield drop from 90% to 89% after adding PAT binning, they may see yields improve to 92% or more after incorporating PAT-driven process improvements.

The automotive and medical industries have done a lot to show the benefits of outlier removal in improving IC quality and reliability. Other sectors of the semiconductor industry including smartphones and high volume consumer devices are now following suit as quality becomes a key competitive criterion. But given the huge volumes associated with these highly competitive markets device manufacturers are moving to the next generation of outlier removal techniques like NNR and multi-variate PAT to strike the optimum balance between quality and yield. Industry leaders are going a step further to leverage the outlier data to drive process improvements and ultimately higher yields. Welcome to PAT 2.0.

About the author
Wesley Smith is VP of Advanced Technology at Galaxy Semiconductor, and is the architect for Galaxy's Semiconductor Intelligence solutions. Prior to joining Galaxy, he worked for QC Solutions, LG Electronics, and MEMC. Mr. Smith holds degrees in Materials Sci and Econ from Clemson, and an MBA from USC.

To download the PDF version of this article, click here.

 First Page Previous Page 1 • 2

Want to more of this to be delivered to you for FREE?

Subscribe to EDN Asia alerts and receive the latest design ideas and product news in your inbox.

Got to make sure you're not a robot. Please enter the code displayed on the right.

Time to activate your subscription - it's easy!

We have sent an activate request to your registerd e-email. Simply click on the link to activate your subscription.

We're doing this to protect your privacy and ensure you successfully receive your e-mail alerts.

Add New Comment
Visitor (To avoid code verification, simply login or register with us. It is fast and free!)
*Verify code:
Tech Impact

Regional Roundup
Control this smart glass with the blink of an eye
K-Glass 2 detects users' eye movements to point the cursor to recognise computer icons or objects in the Internet, and uses winks for commands. The researchers call this interface the "i-Mouse."

GlobalFoundries extends grants to Singapore students
ARM, Tencent Games team up to improve mobile gaming

News | Products | Design Features | Regional Roundup | Tech Impact