You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
If you are correct, that level of hubris and lack of industry knowledge is unbelievable. I have attributed it to the IFS group not doing their homework, and not hiring the right people to drive the strategy (regardless of how they looked on paper). Come to think of it, I'm not sure which explanation is more damning.
The previous time Intel went into the foundry business they did not hire correctly. Most of the key people were career Intel and it did not go well. This time around IFS hired foundry experienced people, I know quite a few of them, most are no longer there. Two of the people built the TSMC EDA/IP ecosystem. One is back at TSMC and the other retired early.
The problem was that Intel manufacturing did not listen to them and most importantly Pat Gelsinger did not listen.
Bottom line: Intel had a serious cultural problem that has now been fixed by Lip-Bu Tan. Hubris pretty much covers it and should have been Pat's Secret Service code name. Hubris is in the building.........
You are also making the mistake of comparing Intel's behavior under Gelsinger with their behavior under Lip-Bu Tan. The way things are handled now is completely different.
Intel has tried to enter the foundry business several times. Are you referring to the 2012 Intel Custom Foundry initiative?
Intel certainly has (or had) company culture issues, but I believe the deeper root cause is its IDM business model. No matter who Intel hired, everyone was ultimately forced to follow the same flawed path, leading to a long series of poor decisions.
There are still IDMs that perform well in mature, analog, mixed‑signal, and industrial/specialty semiconductor markets. But aside from Samsung (which uses its large memory business to subsidize advanced logic) and Intel (which has been struggling for many years), there are essentially no IDMs left competing successfully in advanced logic.
So why should we expect that, after 30–40 years of industry transition toward the fabless model, the IDM model in advanced logic could suddenly become prosperous again?
Intel announced the cancellation of 20A in a single blog post. That’s not a serious way to build a reputation as a “trusted” foundry business.
Why can’t Intel keep 20A as an internal development project instead of putting it on the official 5N4Y product roadmap? Four nodes in four years is still very impressive, I guess.
I suspect Arrow Lake's poor sales may have been a partial factor in the 20A cancellation/timing. I can't be sure, but Arrow Lake launched poorly, was received poorly as it underperformed even Intel's internal performance estimates, and received discounts fairly quickly in the market.
They probably did intend "high volume" on 20A at some point, with ARL -- hence, it went on the roadmap.
(We also saw Meteor Lake-S on Intel 4 canned during the 5N4Y effort, though Intel 4 ramped somewhat as mobile is big for Intel).
