Moore’s Law Is Alive And Well, And Intel Will Prove It Today | Hacker News Reader

Moore’s Law Is Alive And Well, And Intel Will Prove It Today

newenterprise.allthingsd.com

66 pointstilt17 comments

GODDAMMIT, Moore never 'observed' this: "Perhaps you’ve heard of Moore’s Law. This was the observation in 1965 by the Intel co-founder Gordon Moore (pictured at the Intel Museum in 2005) that the number of transistors that could be crammed onto a chip doubles and the size of those chips tended to shrink as manufacturing technology improved on a fairly regular basis: About every 18 to 24 months."

Moore's observation was that the density yielding the minimum cost per transistor doubled with each generation (which doesn't actually predict that the density itself will double, only the point at the bottom of the cost curve will move to 2x density), not that the density doubled or that everything got 2x MOAR AWESUM in some random way.

Yes, it's sad, but it's true. The number of mis-statements of Moore's Law in the computing press will double every 18 to 24 months.

Moore wrote a paper in 1965 that was the first reference to anything in semiconductors doubling every N months, but he also wrote wrote many other papers and speeches before the term "Moore's Law" came into use, and in some of them he talked about feature sizes shrinking every N months. So if you want to be pedantic you could say that the 1965 paper wan't actually referring to "Moore's Law" and that it was really formulated later, but there's no reason to say that modern usage is a misstatement.

See this paper for a good historical overview: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1677462

Moore's Law has become synonymous with any technology getting faster/smaller/cheaper logarithmically.

There are several exponential growth laws in computer science: for example, Nielsen's law for bandwidth, Kryder's law for storage. It's misleading to address them all as Moore's.

Also (George) Gilder's law.

It's also important to note:

http://news.ycombinator.com/item?id=2040593

Well, yes and no. You can read the original 1965 paper that birthed the theory here:

ftp://download.intel.com/museum/Moores_Law/Articles-Press_Releases/Gordon_Moore_1965_Article.pdf

Cost is certainly one of his concerns, but the original "law" has much to do with actually cramming more circuits onto a board.

It's less a misstatement and more a generalization. Stuff getting exponentially faster or cheaper is a very powerful idea.

Perhaps the announcement will be more concerned about the chip stacking (i.e. that is a bigger deal than going 22nm):

http://semiaccurate.com/2011/04/07/intel-goes-finfet-on-22nm...

The current prediction among computer architects I've talked to is that Moore's Law will go for another 10-20 years. I don't think anyone's surprised to hear about 22nm being achieved (not to say that these continual technology shrinkings aren't hard).

I thought the long term trend has shown for a long time the size of chip components converging to the size of an atom by the late teens. A little hard to imagine how physical transistors can get smaller than that. Possibly qubits could built from smaller particles: photons, electrons

We could still get speed ups from higher electron and/or hole mobility after we stop shrinking things by going to different materials, but yeah - that really won't be Moore's law any more.

At that point, build up.

Here's a video explaining the new transistor design: http://www.youtube.com/watch?v=YIkMaQJSyP8&feature=playe...

Pretty cool stuff.

22nm is exciting to me for power consumption.

Imagine a computer that uses nearly 50% less power than even the current 32nm state-of-the-art.

But the problem is the support chipsets (north/southbridge) aren't keeping up?

ARM is going 28nm next year too.

I wonder if this is not just the 22nm process. I wonder if they'll be unveiling something to do with memristors.

From talking to people working on on-chip non-volatile memory, my impression is that memristors are still too immature for production. If anything, phase-change memory seems more viable at this point in time.

Reply on news.ycombinator.com