[Speculation] Moore's Law - Is it meant to 'encourage' or 'bottleneck' overall CPU technology improvement?

[kakandung]

Or to 'regulate' CPU tehnology advancement, so that it improves incrementally, not by a big leap per big leap,

So that Intel (or any other CPU brands) doesn't make too good of a CPU (eg: was Windows XP too good?)
So that they can keep people buying brand-new techs every year or so, keep the business and industry alive, just like about every other industries.

Instead of maxing out their genius engineers doing as smart as they able to, businessman control the company (yeah) and ensure the money-making things go on.
Of course this is only doable if a company has no balanced competitor (eg. Intel-AMD today). BUT even if a competition does happen, they still have to make sure not to saturate tech market with a single BIG amazing thing.

My curiosity is eg: Why didn't Intel jump to design 14nm process back then, when they already know it would happen someday?

Your thoughts?

 

[bootcher]

As manufacturing process gets smaller, the challenges that come with it grow bigger. Producing smaller and smaller CPU's is not something you decide to do, you have to have the technology and ability to do so. It is a very costly process and if yields are too low you are looking at major losses. So in y opinion Moore's Law remains relevant, even though it is harder nowadays to keep up with it because of our own technological limitations. From a business point of view smaller production process yields higher profit, since there are more CPU's being produced per wafer. However pushing for lower production process when you might not get good results can simply cost too much and might have been considered as a too high risk.

 

[kakandung]

If so, is the 'small increments' of CPU technology per year is not something they intentionally decide to do?
For example: Is it impossible to 'die shrink' directly jump, 65nm down to 14nm without inbetween increments?
Is this limitation a 'hard' limitation (physical, technical, knowledge) or a business consideration one (money, risk, market, etc)?

 

[bootcher]

It is primarily a 'hard' limitation, there was no technology readily available to jump from 65 to 14. It would be too big of a jump. If decided to do so there would be a big risk of high failure rates on the wafers, causing the market to have shortage of CPU's, causing money problems because there are no CPU's being sold, etc. AMD has had poor yields in the past, if I remember correctly at one time they had only about 50% of one their new APU's would work correctly, causing heavy losses on the production process, and also having to cancel orders because there are not enough CPU's being manufactured.

 

[gamerk316]

Past experiences jumping nodes by both Intel and NVIDIA didn't end well for either company. So they take a mature architecture, port it to a new node, and once they have experience refine that architecture on that same node. Rinse and repeat, and you get constant progress and significantly less chance of failure.

Remember also that Intel and other fabs research nodes years in advance. It takes time, and much past 10nm we start to run into fundamental physical barriers that will cause production yields to be very low. That's why almost no one has fabs that small, due to cost. The age of the die shrink is coming to an end over the next decade.

 

[Pinhedd]

Or to 'regulate' CPU tehnology advancement, so that it improves incrementally, not by a big leap per big leap,

So that Intel (or any other CPU brands) doesn't make too good of a CPU (eg: was Windows XP too good?)
So that they can keep people buying brand-new techs every year or so, keep the business and industry alive, just like about every other industries.

Instead of maxing out their genius engineers doing as smart as they able to, businessman control the company (yeah) and ensure the money-making things go on.
Of course this is only doable if a company has no balanced competitor (eg. Intel-AMD today). BUT even if a competition does happen, they still have to make sure not to saturate tech market with a single BIG amazing thing.

My curiosity is eg: Why didn't Intel jump to design 14nm process back then, when they already know it would happen someday?

Your thoughts?

Moore's Law is not a law. It started out as an observation and over time morphed into an industry wide goal that has at various times times been exceeded and at other times missed. Industry trends have missed Moore's Law repeatedly for the past couple of years, but not for lack of trying. It turns out that reliably fabricating complex circuits at such a small scale is an incredible challenge from a material science and chemistry perspective.

The various fabrication giants (Intel, Global Foundries, TSMC, IBM, and Samsung) each use their own process that is comprised primarily of closely guarded trade secrets. Intel has maintained about a 2 year gap over its competition in terms of fabrication; this has led to some cooperation between GloFo, TSMC, and Samsung purely to catch up.

 

[CountMike]

If it was a real law you would see tons of lawyers all over it, they would not miss such an opportunity to make money. !!