The semiconductor industry works on razor-thin margins, so any improvement in chip quality and yield is critical to making a profit. The same chips are inserted into a myriad of boards, sensors and other electronic devices, and are integral to supply chain performance and customer order fulfillment in electronics.
“During 2016, we used our analytics software to check over 50 billion devices that came in the form of chips, boards and systems,” said David Park, vice president of worldwide marketing at Optimal+, a big data enterprise software firm that works with major semiconductor and electronics manufacturing companies.
Park said that in the twelve-month period that Optimal+ was performing its device testing, clients collectively were able to save more than $250 million in addition to improving overall chip quality and reducing the number of tests that had to be performed on devices during manufacture.
“What the analytics did is catch critical errors in products before the products entered the supply chain where they could cause more damage,” said Park.
Here’s how the process works:
“In real time, we perform analytics on the product as the product gets manufactured,” said Park. “It doesn’t matter whether we are testing chips, circuits, boards or systems. What the software does is monitor and test function data, log files, parametric data, and test result data. It also collects measurement data for quality checking purposes. Most often, our analytics software comes in the form of an on premises system, but we can provide it in the cloud as well. The end goal is to collect data from many different and diverse sources and to then parse and analyze the data in an outage transfer distribution factor (OTDF) database. We can perform monitoring and testing throughout the entire life cycle of the product.”
Once a product is in the field, constant checkups on performance can generate alert messages for preventive maintenance before an outage happens.
“The Samsung Galaxy 7 problem, where devices were bursting into flames, is a good example of this,” said Park. “The failure in the Samsung case was not due to a manufacturing issue because the device had been manufactured to spec. What went wrong was related to device quality. This is why continuous quality monitoring of products and their performance throughout their entire life cycles is so important.”
Today, however, the most common area for analytics and quality monitoring of electronic devices is on the manufacturing floor, where timely quality checks during chip, board, and other device processing can detect problems early and save time and money.
“A major semiconductor manufacturer that we work with has an abundance of manufacturing expertise on board, and it historically functioned with a belief that if testbeds and device checks were “not built there,” that the process wouldn't work,” said Park. “But the company also had a problem in production that it couldn't solve. The tester machines on the floor were yielding 95 percent of devices at 10 a.m. in the morning, but yields were slipping down to 85% by 3 p.m. in the afternoon.”
This anomaly was a consistent pattern, so it wasn't just a case of getting a bad lot of chips. What the Optimal+ analytics revealed was that something was wrong with the functioning of the actual chip tester unit.
“The specific issue was a “freeze” that was occurring with the chip tester unit because the tester had insufficient memory to do the processing that was required,” said Park. “In some cases, we have found that not only do chip, board, system and device yields grow when these problems are detected and fixed—but the company might even be able to get rid of several testers without investing in new ones because of improved tester performance.”
Whether it is in production machinery investments or in improved chip, board and device quality and yields, the annual savings for electronics and semiconductor companies can be significant.
“There will be a tipping point as companies see these savings and also the positive contribution to supply chain performance,” said Park. “One customer told me, “When we first started talking, we didn’t think we could afford the analytics. Now, we can’t afford not to use them.’”