Intel Downplays Reports Of 18A Chip Yield Woes Following Broadcom Chip Test
Intel's foundry has been struggling ever since. Once it became stable, Intel renamed 10nm to "Intel 7" for marketing purposes, and thus the successor was called "Intel 4," with the next two nodes being named 20A and 18A for having feature sizes better measured in Angstroms than in nanometers. 20A was canceled earlier this year, though, and the idea was that Intel would be refocusing its efforts on its 18A node.
Well, according to a report from Reuters, Broadcom sent wafers to Intel to be etched using its 18A process, and the results that the networking giant got back from Chipzilla were unsatisfactory. The specific language that the outlet used was "not yet viable to move to high-volume production." That certainly doesn't sound good, but it also doesn't really tell us much; what if Broadcom simply had high expectations, or there were actually other issues that resulted in Broadcom's dissatisfaction?
More information may have come in the form of a story in the Chosun Ilbo, an old and respected newspaper in South Korea. In a report on the retirement of CEO Pat Gelsinger, the newspaper gives a yield figure of just 10% for Intel's 18A process. If that's true, that's truly terrible, but we have to point out that Reuters did not report that Broadcom had canceled any orders from Intel.The foundry performance, into which money and manpower were poured, did not meet Gelsinger's expectations. The industry analyzed that the yield of the 18 Angstrom (equivalent to 1.8 nanometers) process that Intel was planning to mass produce next year was less than 10%. As a result, its customer Broadcom canceled its semiconductor orders for Intel.
— Chosun Ilbo Economy (via Google Translate)
It's a little hard for us to believe, though, as Intel itself has stated that "18A is powered on, healthy, and yielding well." That was in a statement reporting on the Broadcom story, and it echoes what the company said back in September when it elected to end development on its 20A process. If 18A was "yielding well," it certainly wouldn't be in the range of 10% usable dies.
Chosun Ilbo is a respected publication in South Korea, but the paper does not cite its sources. Other Korean publications in the past have reported that Samsung's fabrication capabilities were more advanced than those of TSMC or Intel; Korea could certainly use a morale boost after its legislature voted out its president when he attempted to impose martial law on the country this week. We're not accusing anyone of anything, but the paper certainly could have reason to fabricate this figure.
Indeed, according to Ben Bajarin, principal analyst for market intelligence firm Creative Strategies, the defect rate given by Intel back in September of less than 0.4 defects per square centimeter translates to approximately 60% yields. He suspects that Broadcom struggled to port its own libraries over to Intel's fab, and that that was the ultimate reason for Broadcom deciding to part ways with Intel for now.
Whatever the case, we'll know how good Intel's 18A process actually is eventually. The company's next-generation Nova Lake and Panther Lake processors are both expected to be fabricated on that technology. If Intel goes through with those plans instead of shunting production to TSMC yet again, then we'll finally have the chance to see if Intel can still compete with Asia.
