By: Adrian (a.delete@this.acm.org), June 14, 2022 12:06 pm
Room: Moderated Discussions
--- (---.delete@this.redheron.com) on June 14, 2022 9:58 am wrote:
>
> So your argument would be not that cobalt traces were difficult to fab, but that they
> behaved too poorly? That's logically consistent, but seems like something that should
> have been caught a *lot* earlier! So much about this whole episode we just don't know.
>
There are no reasons for the cobalt traces to be significantly more difficult to fabricate than those of copper. Both metals require much more sophisticated methods than were necessary for aluminum, which was used until a quarter of century ago, but such techniques have been well known for many years.
The Intel slide shows poor resistance for cobalt, which justifies its replacement with "enhanced" copper, but that should have also been known 5 years ago.
The Intel slide also shows poor reliability for simple copper, at those small pitches. Maybe this is the reason why they have not been able to go back to copper immediately after seeing that they cannot obtain high clock frequencies.
Maybe replacing cobalt with copper in Intel 7 would have resulted in CPUs with a too short lifetime, so they have been forced to keep the cobalt until they have discovered how to make the improved linings used in the "enhanced copper" metal layers.
The best companies publish post-mortems after their great failures. Those are in fact very good PR for a company. It is useless to pretend that the failures did not happen, as the customers are already well aware of them. On the contrary a detailed post-mortem analysis, explaining all the details of the causes of the failures is the best method to restore the confidence of the customers that such failures will not happen again.
Unfortunately it seems very unlikely that Intel will ever release an explanation of the causes of their inability to predict the performance characteristics of their 10 nm process.
>
> So your argument would be not that cobalt traces were difficult to fab, but that they
> behaved too poorly? That's logically consistent, but seems like something that should
> have been caught a *lot* earlier! So much about this whole episode we just don't know.
>
There are no reasons for the cobalt traces to be significantly more difficult to fabricate than those of copper. Both metals require much more sophisticated methods than were necessary for aluminum, which was used until a quarter of century ago, but such techniques have been well known for many years.
The Intel slide shows poor resistance for cobalt, which justifies its replacement with "enhanced" copper, but that should have also been known 5 years ago.
The Intel slide also shows poor reliability for simple copper, at those small pitches. Maybe this is the reason why they have not been able to go back to copper immediately after seeing that they cannot obtain high clock frequencies.
Maybe replacing cobalt with copper in Intel 7 would have resulted in CPUs with a too short lifetime, so they have been forced to keep the cobalt until they have discovered how to make the improved linings used in the "enhanced copper" metal layers.
The best companies publish post-mortems after their great failures. Those are in fact very good PR for a company. It is useless to pretend that the failures did not happen, as the customers are already well aware of them. On the contrary a detailed post-mortem analysis, explaining all the details of the causes of the failures is the best method to restore the confidence of the customers that such failures will not happen again.
Unfortunately it seems very unlikely that Intel will ever release an explanation of the causes of their inability to predict the performance characteristics of their 10 nm process.
