“Building semiconductors is like playing Russian roulette. You place a gun to your head, pull the trigger, and uncover out four many years later on if you blew your brains out.”
— Former Electronic CEO Robert Palmer
During its Q2 convention contact very last night time, Intel dropped several bombshells in swift succession. Initially, the company’s 7nm process node — the same node the firm was projecting these self-confidence about in March — is apparently a total 12 months guiding wherever Intel intended it to be at this point in its advancement cycle. As a consequence, the company’s 7nm launches will be delayed at least 6 months.
The firm did not explain the defect in depth, only noting that it experienced performed a root lead to assessment and believes there are no “fundamental roadblocks” to correcting the trouble and commercializing the node. In speaking about the trouble, Swan mentioned that Intel experienced designed “contingency plans” to deal with this issue.
If 7nm is delayed, what is Intel bringing up to exchange it? The firm experienced a couple text to share on that topic as effectively. Tiger Lake (cell) and Ice Lake (Xeon) will both of those debut this 12 months, as beforehand introduced. Upcoming 12 months, we’ll see Alder Lake, Intel’s initial 10nm desktop platform, and a new 10nm server CPU, Sapphire Rapids. Rocket Lake — the intended 10nm backport crafted on 14nm and anticipated late this 12 months — wasn’t described. There is loads of proof that Rocket Lake exists, but it might not have been wherever Intel wanted to aim attention.
Intel Commits to Employing Third-Party Foundries
Ever since Broadwell and 14nm, traders have periodically requested if Intel would faucet third-celebration foundries or look at heading fabless. Intel’s response to these issues has usually been that it does faucet third-celebration foundries for some goods (which it does), and to point out that its virtually special status as a semiconductor IDM (Integrated System Producer) gave it an advantage other foundries can not match. If you want to go through the argument, I basically wrote about it again in 2014. Hunting again at people articles, I wince. It is not that I believe they were being inaccurate when penned — it is that the technique they explain went from ticking like a effectively-tuned clock to knocking like an motor about to toss a rod.
This time, Intel’s entire method to the topic was distinctive. Alternatively of reassuring traders that the goods crafted at third-celebration foundries would be reduced-cost components, Intel openly acknowledged that it would use what ever know-how stack was essential to provide effectiveness management, be that thoroughly internal producing, thoroughly external, or a mixture of the two. CEO Bob Swan emphasized that this program is part of Intel’s motivation to overall flexibility and argued that its willingness to create what he named contingency plans is a sign that the firm is established to provide greatest value to both of those traders and consumers.
I don’t disagree. At this point, offered the issues Intel has confronted with its possess producing, the firm would be silly and likely negligent if it unsuccessful to discover every single possibility. That does not change the truth that 6 many years in the past, Intel declared its process node management would continue on on 14nm and into the future, when in 2020, the firm CEO spoke of guarding the company’s roadmaps and goods from its possess process node problems. “We have figured out from the issues in our 10-nanometer transition,” Bob Swan stated, “And have a milestone-driven method to make sure our merchandise competitiveness is not impacted by our process know-how roadmap.”
That is a stark, dramatic shift. Intel talked up a very good sport very last night time, with a discussion of “disaggregation” (chiplets) and its plans to acquire advantage of superior packaging know-how like Foveros in future goods. The firm is not lying when it claims that superior packaging methods are critical to continuing to advance silicon scaling and improve effectiveness. Concepts like wafer-scale processing are staying evaluated all over again exactly because there’s a will need for new packaging methods to continue to keep the field moving ahead.
CEO Bob Swan declared Intel would be “pretty pragmatic” about deciding wherever to build parts based mostly entirely on what form of components it necessary to bring to current market, and did not examine the issues of ramping models at multiple foundries at the same time. Intel’s possess fabs are regarded for deploying Intel-distinct know-how and producing processes intended to improve the velocity of their possess parts, not decrease cost for mass buyer producing.
Swan is correct that this is the most pragmatic point Intel could do, offered its possess producing problems, but a large amount of traders will be looking at the company’s “contingency plans” meticulously. It is high-quality to faucet a third-celebration foundry for chipsets or Atom. It is no trouble to farm out reduced-finish producing to make far more worthwhile use of existing producing property. The prompt Intel has to activate one particular of people contingency plans to deal with top-edge “big core” producing because its possess fabs can not deal with the work, people substantial factories go from a toughness to a weak spot on the harmony sheet.
It is believable that Intel may well bring an improved 10nm process to desktop, because its authentic 2017 slides always implied that 10nm++ would, in truth, be a little bit outstanding to 14nm for effectiveness and electric power use in desktop electric power envelops. But all over again — the firm admitted just 6 months in the past that 10nm wouldn’t be the accomplishment that its traders have arrive to be expecting. Today, Intel sang the praises of 10nm. A few months in the past, the firm was fifty percent-completely ready to bury it. It is unquestionably attainable that COVID-19 brought on some of this hold off, but Intel did not discover coronavirus as a principal explanation guiding its roadmap slip.
I believe what Intel did nowadays was place a pretty clean confront on a radical company realignment. If I may well be permitted a bit of poetic license and a Deus Ex: Human Revolution estimate: “It’s not the finish of the globe, but you can see it from listed here.”
Dependent on Swan’s possess remarks and timelines, Chipzilla has 24-36 months to show why it should really even now possess its possess fabs. By late 2022 / early 2023, TSMC should really be delivery 3nm. Even if we presume that Intel’s 7nm is very good sufficient to examine straight to TSMC’s 5nm, that even now puts the Taiwanese firm a total node in advance.
Is this the finish of Intel? Not by fifty percent. Intel’s financials are excellent, the firm is tremendously worthwhile, its knowledge center enterprise carries on to expand, and its dollars movement is wholesome. AMD was in considerably far more issues following Bulldozer bombed in 2011 than Intel is now, even facing even more delays and the dilemma of no matter if it will continue being an IDM around the extended expression. The company’s process engineers might be having difficulties, but its CPU layout groups are even now fantastic.
But having a excellent CPU layout crew is a needed but inadequate component of holding the management placement in CPUs that Intel has extended commanded. The firm is able of mounting aggressive comebacks, but if Intel wishes traders to see its foundry amenities as a needed part of the enterprise instead than a drag on its gains, it is time to pull out all the stops and resolve its factories. No, Bob Swan did not say that explicitly, nowadays.
He did not have to.