Customer hack: using outlier detection for virtual reprobe

Customer story June 2022

Vicor Corporation was founded in 1981 as a local company in Massachusetts when technology was advancing slowly but surely. Fast forward to today and the giant corporation is a world leader in technological innovation. The company is known for designing, building, and supplying high-performance power modules that are used by customers across computing, robotics, automotive, and more.

Dieter Hirsbrunner, Vicor’s product and test engineering manager, works at the company’s Rhode Island site and leads the product engineering team. He’s been using yieldHUB for Vicor’s yield management needs since July 2020.

Dieter began his career at Cherry Semiconductor (now On Semiconductor) and moved to Texas Instruments before going to Vicor. 

“I came [to Vicor] as we were starting to select yield enhancement software. What we had was limited to a single license and a single computer. So basically, we evaluated three or four software packages,” Dieter said.

“For a product engineer, I have specific things I want it to do like going back to the wafer fab data and that’s why we ended up with yieldHUB. yieldHUB was the easiest one to use and lent itself well to experimentation. 

“It was online and we had access to it everywhere and it was able to do a lot of the stuff we wanted it to do. We had a long demo and we found without the training we were able to figure out how to do a lot of the stuff. A couple of other companies wouldn’t even provide a demo without cash so that was a big red flag.”

Typically, yieldHUB helps customers to manage the entire outlier removal process from initial wafer lot characterization to final-test yield monitoring. With it, customers can see an increase in the quality and reliability of their semiconductors and ICs.

However, Dieter said for a special project he experimented with yieldHUB by using outlier detection for virtual reprobe so his team didn’t have to physically probe wafers again.

“When you’re testing to limits for automotive when it’s different, it’s bad. If you have ten thousand parts that are reading four-and-a-half volts and you have a couple of hundred parts that are reading 4.7V, then that’s considered bad, they don’t want those. Even if it’s within spec and that range is between 3V and 5V.

“They want everything to be nice and uniform so you can plug it in and go. So what PAT does is it takes a wafer that was already tested and gets rid of those that are different. We have a specific need on parts so we probed our wafers, and a whole bunch of them were at 4V, and some were at 4.5V. 

“The engineer wanted the weird parts assembled. We’d already probed these wafers multiple times, at multiple temperatures. The more you probe them, the more problems you may have with damaging the pads [which are used for wire attach later].”

In yieldHUB outlier detection was purposefully designed not to let failed die pass to avoid the risk of escapes going into production but in this case, Vicor was undertaking an engineering study and that’s why they wanted to trick the system and manipulate data.

While the safeguard is there for good reason, the die can be assembled and studied in a controlled environment. What Dieter did was access raw data in the form of the wafer map and reset it, then download the datalogs and reset the data so it would be recognized as being passed data. It means you retain the parametric values and that’s easily loaded back into yieldHUB. 

By doing this, Dieter moved around the safeguard so previously failing die can pass and the data is available to be used for the engineering study. Not having to probe the wafers once more was a big advantage.

“It saved us time and not having to probe the wafers again,” he said.

Dieter took our marketing manager through the process online, which can be viewed in this video below. However, we’ve summarized this in the above paragraphs.

Dieter shared some of his other tips for using yieldHUB everyday.

“It’s important to ensure the program file names are set up. A lot of people, when they probe a wafer at room, they have a test program for that and then they may probe the same wafer at hot and they may have a second test program for that so a separate test program on every separate wafer… but we generally have multiple test steps in a program. 

“We have some parts that we test at room, hot, and cold, and they all use the same test program and that gets confusing so in our file names we ended up having the device, and then right after it the test step and then the lot number and the sub lot number. 

“So we ensure in our engineering product name column that we put the bump hot so these are considered separate products so that way the data doesn't get blended together. So the only way to differentiate the data is by putting the test step into the engineering product. That helped solve a lot of our issues with data integrity.”

When it comes to choosing a yield management platform, Dieter suggests looking at systems that are both accessible and intuitive. 

“If you’re able to play with it in a demo, see if you can figure it out on your own.”

“I like the fact that yieldHUB is there whenever and wherever I want it. The alerts go to my email. I got about 20 while we were on this call.”

Interested in learning more about yieldHUB? We offer demos so you can try yieldHUB risk-free. Make a smart decision today!