Comment by adrian_b | Hacker News Reader

Comment by adrian_b | Hacker News Reader

Parent Full thread

This would the correct explanation if Intel would have made no mistakes.

However, it is beyond reasonable doubt that Intel has made some extremely ugly management mistakes during their failed attempts to transition from 14 nm to 10 nm, which have absolutely nothing to do with the competition from TSMC or others.

Maybe Intel would have lost their technological advance regardless how the company had been managed, due to the difference in volume between smartphone chips and PC chips, but now we cannot know which has mattered more, the internal mismanagement or the change of the semiconductor device market favoring their competition.

I wonder if it will ever become public knowledge which were exactly the technology problems of Intel and who are those guilty for the decline at the company, because the failures could not have had any real technical causes but only bad management causes.

During their worst years Intel has been capable even of presenting to the public fake technical presentations about the alleged great characteristics of their future 10 nm CMOS process, which is something that I would never have expected from a company like Intel. The marketing messages from any company are expected to be full of lies, but the technical presentations are expected to match reality.

During many years, unlike TSMC, Intel has behaved as if it were completely unable to predict the precise characteristics of the CMOS process that they will able to manufacture in the following year.

How could this happen is very hard to understand, as the manufacturing processes and the future devices made with them can be simulated long in advance of their implementation, and the models can be tuned continuously while the processes evolve, by fabricating and measuring various test structures. Unlike for Intel, for TSMC the performance predictions for future processes have always been reasonably accurate.

There was the infamous demo when they ‘forgot’ to mention a CPU was overclocked. Bit of a red flag for the culture that permits that.

https://www.tomshardware.com/news/intel-28-core-cpu-5ghz,372...

> I wonder if it will ever become public knowledge which were exactly the technology problems of Intel and who are those guilty for the decline at the company, because the failures could not have had any real technical causes but only bad management causes.

Why do you think this? It was a yield issue, and the best guess is it was caused by multi patterning. (Intel themselves said so.)

I found an analysis from SemiWiki titled "Intel 10nm Yield Issues" pretty convincing. See https://semiwiki.com/semiconductor-manufacturers/intel/7433-...

It was not only a yield problem.

The yields are indeed hard to predict, so if that would have been the only problem, then there would not have been so much reasons to blame the Intel management.

As I have already mentioned, Intel has behaved as if being completely unable of predicting the electrical characteristics of the devices made with their future 10-nm processes.

The first generation of 10-nm products was Cannon Lake, launched in 2018. I have actually bought out of curiosity the Intel NUC with Cannon Lake U, together with an Intel NUC with Coffee Lake U, a 14-nm CPU launched simultaneously with Cannon Lake U.

The 10-nm CPU had pathetic performances in comparison with the 14-nm CPU, i.e. much lower clock frequency at identical power consumption or much higher power consumption at identical performance.

This too low performance should have been determined very early during the design of Cannon Lake and the project should have been canceled even before reaching the tape-out phase.

Even the 2nd variant of the 10-nm process, used for Ice Lake in 2019 was inferior to the by then 5-year old 14-nm process. This should also have been known enough in advance to cancel Ice Lake before tape-out.

Converting any Intel factory to mass production of 10-nm wafers should have never begun before having a working 10-nm process able to make electrically-better transistors than in the old 14-nm process.

Intel has begun the mass-production of chips using the 14-nm process in 2014. By then they must have already started the design of the 10-nm process, but only in 2020 they have succeeded to make better 10-nm transistors than their 14-nm transistors (while still having a lower maximum clock frequency) and only in 2021 they have succeeded to exceed the old transistors in all parameters.

It is impossible to understand why have they wasted huge amounts of money for failed projects like Cannon Lake and Ice Lake and for factories using uncompetitive process variants, instead of concentrating all their resources to their top priority of finding a way to manufacture 10-nm transistors that are better than their existing 14-nm transistors, which is normally a precondition for starting any design process of a CPU using a new process and for building a new factory.

It all looks like if most people working at Intel would not have been aware that they actually do not have any working 10 nm manufacturing process, and they were continuing all their usual activities, like designing CPUs, but with fake transistor models, or building new factories, while there was some gang who knew that the 10-nm process does not exist outside fake presentations, but they hoped that there will be some miracle and the 10-nm process will just work when the time comes to use it.

Intel has a decades long history of very poor top level engineering management. They got petrified about the issue of how much DRAM they'd be able to put in a system in the 1990 (remember Rambus RDRAM?), and among the consequences was a 1 million part recall of motherboards shipped to Dell just before they were going to be shipped to customers, plus another related one I can't remember the details of yet.

This speaks to the sorts of dysfunction you're speaking of; certainly someone at Intel realized these parts weren't going to work in the real world.

Reply on news.ycombinator.com