Slashdot Mirror
Intel Makes 45nm Chip
dolphinlover writes "Intel announced today that it created its first microchip using the 45 nanometer manufacturing process that it says will go into its processors in the second half of 2007. Intel said that this development provides it with a 'considerable lead over our competitors in the 45-nanometer generation'."
I heard that AMD will be shipping their first 65nm products in late 2006 and have heard nothing about 45nm production.
Just "gittin-r-done," day after day.
AMD has traditionally been behind Intel on the bleeding edge fab stuff. Intel's dominated the fab tech race by six months or so for years and years. That is not changing here, as far as anyone I know of can see. AMD using SOI sort of blurs the line here, but in terms of process shrinks and the like Intel's ahead.
AMD's chips being better performers despite being behind some in chip fab is an important feature. But roadmaps based on imaginary pixie dust, in an industry where fabs cost $4 billion or so, are a waste of time even on slashdot.
Unless I misplaced a decimal point or misunderstand physics, isn't 45 nm only a very few generations from needing connections only one molecule thick?
Who added the "and therefore its processing power" to the quote? Was it the reporter or someone from Intel? Moores law has nothing to do with processing power.
In popular science, Moore's law is used to describe anything that resembles exponential growth. Not only that, it is applied without regards to whether the underlying technology scales in an exponential way, as long as it appears to have done so for a certain period of time, meaning whichever period gives the desired results. "Computers" and any part thereof seem to qualify on historical merit. Transistor count? Clock speed? HDD size? Take your pick. In this case I presume the journalist felt people didn't know what transistor count meant - but sorta drew the conclusion that it is related to performance or something - at least the Intel people talked about it alot, so it must be important...
Live today, because you never know what tomorrow brings
It seams to make sense that because Intel has the most money, that they can spend money on developing better manufacturing and engineering techniques than their competition. But with all of this extra money, and seamingly having better technological capabilities, AMD is still beating out Intel as far as performance.
Looks like Intel basically does all of the hard work figuring out how to do things for the first time, and AMD just has to wait until Intel is finished and then just learn from them. I of course know nothing about how to make processors, but it seams that this is the most plausible reason why Intel has trouble making chips that are as good as AMD.
This news about the 45nm manufacturing looks very bad for AMD, but I doubt it will matter very much. If Intel is doing it by the end of 2007, AMD will probably be doing it by first or second quarter 2008. And if history is any indicator, they will probably be doing it better. But I guess time will tell, maybe this 45nm technique really is too hard for a company without endless money to figure out.
--
-- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
Yes, when AMD came to my school 45 nm was set for 2008, 65nm for later this year I believe.
Just to make sure, this is not a CPU chip using 45nm technology. This is a test vehicle which contained SRAM (static RAM) and some control logic. SRAM arrays are regular and don't have the same complexity as ALU (arithmetic logic unit) and other control circuits found in CPU. So yes this is a big step because it is gives some indication about how complicated will it be to get a good yield in this process. Also note that SRAM arrays can be easily made defect tolerant by using spare rows/columns. Same is not true for CPU cores. So there is still al long way to go before an efficient working CPU with production acceptable yield is available.
But with smaller transistors you also have smaller distances in between, thus reduced emission (power). The question is which effect has the larger gradient.
Intel may have 65nm in Fab D1D right now, with plans to convert 2 more fabs to 65nm while converting D1D to 45nm, they have many more fabs that are much farther away from 65nm. Many Intel chips will have to be made on processes older than 65nm.
AMD's new Fab65 just opened last October, is already generating fantastic yields at 90nm, and it is ready for 65nm and below (this year, sooner rather than later), and (even though AMD hasn't spoken about doing it), it would not be impossible to retool their Fab30 to 65nm.
And AMD's 65nm will have SOI, just like their 90nm does. Intel does not have SOI.
--- Hindsight is 20/20, but walking backwards is not the answer.
"Intel has learned from it's mistake with NetBurst (P4) design descisions and are finally heading back in the right direction."
Exactly, they are going a step backward by going back to the Pentium-M with some modifications.
"So, by the time the 45nm comes out, there will also be a new architecture to place on the new chips. We'll see how things are then."
That is an awful lot of 'forward looking'. AMD will not sit still between now and then either, on either front (process and architecture).
Intel just canceled their Whitefield processor, the only one that ever was on their roadmap to sport an integrated memory controller.
AMD is improving their architecture continuously: Introducing chips with DDR2 support very soon (motherboard manufacturers have samples), just licensed ZRAM technology to add low silicon area very large Level 3 caches, will be introducing improved versions of the HTT interconnect that makes the multi-chip Opteron systems scale so well, is going to show us a quad-core version this year, etc etc.
I don't have to wait and see, it's too clear where this is going.
--- Hindsight is 20/20, but walking backwards is not the answer.
I once had a conversation with someone who was doing developmental research for an even smaller process for some very large semiconductor manufacturer. According to him, they were one day running some measurements on the first prototype wafers. From experience with every previous process (65, 90, 130, etc.), they were expecting this particular measurement to yield a nice bell curve. Instead, they got a strongly quantized bell curve: it looked more like a histogram. The reason, they realized, was because the gate oxide was becoming only a handful of atoms thick. The quantization between two steps in that bell curve was the difference between, say, a gate oxide 5 atoms thick and 6 atoms thick.
So, yes, they are indeed coming up against some real physical limitations for CMOS technology. Of course, people have been saying that for years.
Actually it's a 45 nautical mile ship!
Dang, if you were in the middle of that ship you could look either way and it would disappear into the horizon, unless it was also absurdly tall, of course!
When they start building ships to take into account the curvature of the Earth, I'll officially be scared.
Constitutional rights may be respected, repealed, or modified; but they must never be ignored.
IIRC, the first decent microprocessors were about 30K transistors and the next generation of chips should fit 300M transistors easily. So with enough crack and enough money, you could theoretically make a 10,000 core chip now. A practical device would need a network and memory on chip, but still - thousands of cores. If the inter-core communications were asynchronous, the cores could run at amazing clock rates, too, since the longest signal paths would be only a picosecond or two.
"Is life so dear, or peace so sweet, as to be purchased at the price of chains and slavery?" - Patrick Henry
I'd say you're mostly right. I think Intel will be positioned as the desktop / lower-end / lower-priced processor, and AMD will dominate the market for servers. Sort of a role reversal from the case before. Unless Intel comes up with something like AMD's direct connect architecture and hypertransport, AMD will continue to dominate I/O performance, even if Intel does eclipse them in the standard user benchmarks. I don't personally care who I buy from at work, but at least as it is now, the Opterons are the only thing allowed in the rack for new systems, and Intel chips for laptops. Desktops don't matter as much, so it's mostly Semprons at the moment (best bang for little bucks). As soon as Intel really abandons the MHz/GHz = performance mantra and realizes that the northbridge is their biggest drawback, they'll turn it around. The mobile chips and the Core Duo are a good start, but we'll see if they stick with them.
I saw a quote somewhere from an exec at AMD, who was asked how the Pentium 4 could be improved on. He said: "Use the Pentium III". The sooner Intel realizes he was right (mobile/Core chips are more closely related to the PIII than P4), the better.
--That's the point of being root, you can do anything you want, even if it's stupid.
AMD licensed magic technology recently where SOI processes would be used to reduce the density of on-die SRAM and DRAM. The company spokesperson said that tech was usually integrated in about two years. Hey, that's 2008! So again, Intel has higher fab tech but AMD may win on architecture yet again.
A NYC lawyer blogs. http://www.chuangblog.com/
Even worse is that people actually think it's some sort of technology limitation. It's based on economics more than anything else. The Fab companies can afford to upgrade every 18 months (it's very expensive) because that's the rate at which most consumers (be they businesses or individuals) are willing to upgrade their existing equipment.
Moore's "Law" is just where the lines "how fast are people willing to upgrade" vs. "how much does a new fab cost and we remain profitable" cross. It's more of "Moore's Economic Observations" than anything technological.