//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

Within the 2023 film BlackBerry, the top of AT&T Cingular explains to the BlackBerry CEO about Cingular’s technique to promote information plans with the newly launched Apple iPhone: “Have you learnt what the issue with promoting minutes is? There’s just one minute in a minute.” As somebody working for a corporation promoting “check seconds” for semiconductor units, this acquired me considering that there’s just one check second in a second. And whereas our prospects might settle for some extra check seconds to get a tool to market shortly, these seconds must settle again all the way down to typical ranges as soon as units ramp to excessive quantity—or else these extra check seconds are coming proper out of our buyer’s backside line. 

Quickly rising design complexity, heterogenous integration packaging, and better adoption of system-level check insertions are all drivers for rising check seconds, but there is no such thing as a secret stash of cash obtainable to fund this enhance. So, the problem is, how can we offer a couple of check second’s value of worth in a single second? 

As within the film, the reply has to do with information. Within the case of semiconductor check, it’s about maximizing the quantity of worth that may be extracted from the info collected throughout check.

Within the pre-chiplet/pre-heterogenous integration world, tester sources wanted to be quicker and extra correct than the device-under-test. Whereas this requirement has actually supplied important challenges over time, we now face the added problem of getting to be smarter than the advanced, multi-chip system-under-test. And we should meet this added problem as 2.5D and 3D packaging are lowering direct entry to system pins, and general-purpose processors are giving technique to synthetic intelligence (AI) processors that serve a number of specialised purposes.

By Shingo Kojima, Sr Principal Engineer of Embedded Processing, Renesas Electronics  03.26.2024

By Dylan Liu, Geehy Semiconductor   03.21.2024

By Lancelot Hu  03.18.2024

To fulfill these challenges we should, as an trade, embrace the most recent machine studying and superior information analytics approaches, trendy information sharing applied sciences, and a tradition of sharing and collaboration throughout the semiconductor worth chain. 

Whereas it could be tempting to think about these merely as non-obligatory methods to optimize present check methodologies, I keep that these steps are an absolute necessity, and firms that don’t embrace them might be left behind. 

In a chat Nvidia founder and CEO Jensen Huang gave final yr at Columbia Enterprise Faculty, he said, “AI isn’t going to take your job. The one who makes use of AI goes to take your job.” 

I see the identical for corporations vying for enterprise because the semiconductor worth chain transforms and explodes in complexity in a number of methods. 

Should you don’t reap the benefits of AI-based approaches, the businesses that do are going to remove your small business.

Making it occur

How will we squeeze a couple of second of worth out of 1 check second? 

First, we have to use that one check second as correctly as doable. This objective is greatest achieved utilizing edge compute functionality to make real-time selections in the course of the check move, based mostly on a mixture of information collected within the present insertion, earlier check insertions, and manufacturing or inspection steps. This brings a number of benefits:

• Priceless check time could be saved by changing search algorithms for check ranges and trim values with machine learning-based predictions. 
• Environment friendly entry to on-chip sensor information supplies a better degree of perception into efficiency, particularly as direct pin entry is diminished with superior packaging. 
• In multi-site check, failing units could be inspected additional to collect critically vital diagnostic information slightly than sitting idle as the great units end their check move.
• Machine learning-based sensible binning permits die matching for advanced heterogenous integration packages or greatest match of the units to particular end-use purposes. 

The info collected throughout these check seconds can then be mixed with information from throughout the semiconductor worth chain, enabling feed-forward and feed-backward purposes for optimizing design, manufacturing, and check processes.

To totally obtain this imaginative and prescient, the trade should make progress on communications infrastructure, information codecs, traceability, and superior information analytics algorithms. If the imaginative and prescient is really embraced throughout the trade, these challenges might be considered as alternatives for offering inventive and precious options. A number of successes have already been achieved by connecting information from two or extra manufacturing or check steps and benefiting from enhancements in machine studying and edge compute know-how to realize extra perception from this information. Persevering with to construct on these successes can obtain a crucial mass that may gasoline additional growth of the enabling applied sciences.

So, whereas there may be certainly just one check second in a second, I consider we’re simply cracking the floor by way of the extra worth that may be extracted from the info collected throughout these check seconds. The creation and widespread adoption of progressive approaches for extracting this worth might be a key requirement for fulfillment within the chiplet age.