//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
Within the 2023 film BlackBerry, the pinnacle 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 an organization promoting “check seconds” for semiconductor units, this obtained me pondering that there’s just one check second in a second. And whereas our prospects could settle for some extra check seconds to get a tool to market rapidly, these seconds have to settle again right 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 higher 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 accessible to fund this improve. So, the problem is, how can we offer multiple 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 assets wanted to be sooner and extra correct than the device-under-test. Whereas this requirement has definitely supplied vital challenges through the years, 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 gadget 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 business, embrace the newest machine studying and superior information analytics approaches, fashionable information sharing applied sciences, and a tradition of sharing and collaboration throughout the semiconductor worth chain.
Whereas it might be tempting to think about these merely as elective methods to optimize present check methodologies, I keep that these steps are an absolute necessity, and corporations that don’t embrace them will likely be left behind.
In a chat Nvidia founder and CEO Jensen Huang gave final yr at Columbia Enterprise College, he acknowledged, “AI shouldn’t be going to take your job. The one that 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.
If you happen to don’t reap the benefits of AI-based approaches, the businesses that do are going to remove what you are promoting.
Making it occur
How can we squeeze multiple second of worth out of 1 check second?
First, we have to use that one check second as correctly as attainable. This aim is greatest achieved utilizing edge compute functionality to make real-time choices throughout the check circulate, based mostly on a mixture of knowledge collected within the present insertion, earlier check insertions, and manufacturing or inspection steps. This brings a number of benefits:
- Worthwhile check time will 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 offers a higher stage of perception into efficiency, particularly as direct pin entry is lowered with superior packaging.
- In multi-site check, failing units will be inspected additional to collect critically essential diagnostic information fairly than sitting idle as the nice units end their check circulate.
- Machine learning-based sensible binning allows die matching for advanced heterogenous integration packages or greatest match of the units to particular end-use purposes.
The information 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 completely obtain this imaginative and prescient, the business should make progress on communications infrastructure, information codecs, traceability, and superior information analytics algorithms. If the imaginative and prescient is really embraced throughout the business, these challenges will likely be seen as alternatives for offering inventive and helpful 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 expertise to achieve extra perception from this information. Persevering with to construct on these successes can obtain a essential mass that can gasoline additional growth of the enabling applied sciences.
So, whereas there’s certainly just one check second in a second, I consider we’re simply cracking the floor when it comes to the extra worth that may be extracted from the info collected throughout these check seconds. The creation and widespread adoption of modern approaches for extracting this worth will likely be a key requirement for achievement within the chiplet age.
