Getting the Most Out of ATE Test Seconds in the Chiplet Age
This article is a condensed version of an article that appeared in EE Times on March 27, 2024. Adapted with permission. Read the original article here.
By Ira Leventhal, VP, U.S. Applied Research & Technology, Advantest America
In the 2023 movie BlackBerry, the head of AT&T Cingular explains to the BlackBerry CEO about Cingular’s strategy to sell data plans with the newly introduced Apple iPhone: “Do you know what the problem with selling minutes is? There’s only one minute in a minute.”
As someone working for a company selling “test seconds” for semiconductor devices, this got me thinking that there’s only one test second in a second. And while our customers may accept some additional test seconds to get a device to market quickly, those seconds need to settle back down to typical levels once devices ramp to high volume—or else those additional test seconds are coming right out of our customer’s bottom line.
In the pre-chiplet/pre-heterogenous integration world, tester resources needed to be faster and more accurate than the device-under-test. While this requirement has certainly provided significant challenges over the years, we now face the added challenge of having to be smarter than the complex, multi-chip system-under-test.
And we must meet this added challenge as 2.5D and 3D packaging are reducing direct access to device pins, and general-purpose processors are giving way to artificial intelligence (AI) processors that serve multiple specialized applications. If you don’t take advantage of AI-based approaches, the companies that do are going to take away your business.
Making it happen
How do we squeeze more than one second of value out of one test second? The data collected during those test seconds can be combined with data from across the semiconductor value chain, enabling feed-forward and feed-backward applications for optimizing design, manufacturing, and test processes.
Multiple successes have already been achieved by connecting data from two or more manufacturing or test steps and taking advantage of improvements in machine learning and edge compute technology to gain more insight from this data. Continuing to build on these successes can achieve a critical mass that will fuel further development of the enabling technologies.
I believe we are just cracking the surface in terms of the additional value that can be extracted from the data collected during those test seconds. The creation and widespread adoption of innovative approaches for extracting this value will be a key requirement for success in the chiplet age.