Why not make a big jump on manufacturing size?

[ulillillia]

The manufacturing size is apparently 28 nm right now. Why don't companies make a big jump instead of small, incremental ones? That is, why not jump from 28 to 10 or even 5 nm? ... Or is 5 nm currently being worked on, but only in the purely experimental phase and it takes 5+ years to reach the market?

 

[a4mula]

Tick/Tock. Intel has a roadmap and they stick to it. This preserves Moore's Law while allowing them fleece the market along the way. Here's a alot of marketing mumbo jumbo if you're really interested though:

http://www.intel.com/content/www/us/en/silicon-innovations/intel-tick-tock-model-general.html

PS: 14nm is the next tick.

 

[popatim]

1: R&D first has to come up with a way to make these die shrinks even possible
2: new manufacturing techniques must be developed for each of these die shrinks.
3: once the bugs have been mostly worked out, it costs many many millions of dollars just to retool 1 fabrication plant.

so thats why change is slow and intel 'milks' the market. They need to make a profit to keep their stockholders happy and they need to keep researching to keep us users happy.

 

[hunter315]

Die shinks tend to be 0.7 of the last one, this helps keep the math a bit simpler and more consistent while doubling the amount of transistors you can fit in an area on the chip. The small jumps let them learn important things as they go that they will need for the next jumps. Everytime there is a die shrink there is a long period of working out the issues to improve yields, if you were to make a larger jump it would take much longer to get it all sorted out.


I also doubt that the design teams would be able to make use of a massive increase in available transistors. For CPUs the manufacturing nodes coming up after 32 nm are 22 nm, 16 nm, then 11 nm. If you were to try to jump from 32 nm to 11 nm you would have the option of putting almost 8x as many transistors on the same sized chip in a similar power envelope, but it would take several years to finally figure out and fix all the problems with the system so the initial yields would be abysmal.

 

[rds1220]

The manufacturing size is apparently 28 nm right now. Why don't companies make a big jump instead of small, incremental ones? That is, why not jump from 28 to 10 or even 5 nm? ... Or is 5 nm currently being worked on, but only in the purely experimental phase and it takes 5+ years to reach the market?

Look up Moore's Law. Your trying to jump 10 steps ahead of future. A 1 month old baby can't just get up and walk it has to crawl first then walk. Your trying to walk before you can crawl.

 

[hunter315]

Moore's law is NOT a law, it was merely an observation by Moore that the number of transistors on a chip seemed to double approximately every 24 months, not that it had to. He figured it would continue into the future for about another decade, we have already passed that point and its still approximately true, but "Moore's law" is merely an observation and does not actually have any bearing on how the IC industry functions.

 

[a4mula]

Moore's law is NOT a law, it was merely an observation by Moore that the number of transistors on a chip seemed to double approximately every 24 months, not that it had to. He figured it would continue into the future for about another decade, we have already passed that point and its still approximately true, but "Moore's law" is merely an observation and does not actually have any bearing on how the IC industry functions.

Nothing is strictly a "Law" in the sense you describe it, Intel does use it to drive their manufacturing process.

 

[jaguarskx]

1: R&D first has to come up with a way to make these die shrinks even possible
2: new manufacturing techniques must be developed for each of these die shrinks.
3: once the bugs have been mostly worked out, it costs many many millions of dollars just to retool 1 fabrication plant.

so thats why change is slow and intel 'milks' the market. They need to make a profit to keep their stockholders happy and they need to keep researching to keep us users happy.

That's more or less the reasons.

 

[hihis]

He figured it would continue into the future for about another decade, we have already passed that point and its still approximately true .

 

[ulillillia]

Best answer selected by ulillillia.