Slashdot Mirror
ARM Expects 20-Nanometer Processors By Late 2013
angry tapir writes "ARM chips made with an advanced, 20-nanometer manufacturing process could appear in smartphones and tablets by as soon as the end of next year, the head of ARM's processor division said Monday. The more advanced chips should allow device makers to improve the performance of their products without reducing battery life, or offer the same performance with longer battery life."
The question is: when Arm will become real Intel competitor? There is already a lot of android laptops, but still lame ones. Sooner or later Arm will become Intel's big problem...
Does anybody know who the 20nm fabs ARM is expecting to provide these chips are? It was my understanding that Samsung, Globalfoundries and TSMC were still working on a larger process(28mm?) and Intel has been very cagey about fabbing any 3rd-party stuff except for a handful of FPGAs and other high-margin oddballs that don't compete in Intel's area of business in any meaningful way.
They have problems delivering 28nm right now so take the 20nm predictions with a pinch of salt. "However, the transition to 28nm does not appear to be going smoothly. ARM heavyweight Qualcomm was the first to introduce a 28nm design, the stunning Snapdragon S4 based on its Krait core. But the outfit is now struggling to meet demand for S4 chips and it is basically becoming a victim of its own success. Other ARM players, such as TI, Nvidia, Samsung and Apple, have yet to introduce a single 28nm part." -- http://www.fudzilla.com/home/item/27414-arm-hopes-to-see-20nm-processors-next-year
I often wonder, with traces being made smaller all the time, how does this affect radiation resistance? Are we going to hit the point soon where just laying the chip open in the sunlight creates enough of random electron/hole generation, so that the device becomes useless? We already know that chips must be hardened to work in space, how long until this is true for Earth-tied ones? If someone has an answer, it would be interesting to know.
Some background you'll never see anywhere else, written by me:
First of all these funny numbers come from the ITRS. This is not just random numbers, process stages are not a "preferred number system" like resistor values where statistics determines the weird values. Process stage size steps are indirectly determined by physics. ITRS is an industry association of companies who actually make this stuff. Wikipedia has a page for each stage, yes there is a wiki page called "22nm" or something like that. This "20 nm" process is actually a "half step" from the 22nm process. The next "real" step is 16 nm.
(opinion alert!) Now this is a half step from 22nm to 16nm and is considered a failure. Put your efforts into cheaper higher yield more economic 22 or advance the field to 16, don't screw around halfway at 20. Another interpretation is oxide thicknesses are getting too small at the 22nm process to make anything smaller like 16nm, essentially they're giving up on 16nm because its economically impossible(end opinion alert!) The rest of my post is pretty much factual, as far as I know.
Another interesting thing about process sizes is this is a half-pitch (essentially a radius) of an array cell. Its dumb and/or marketing to spec half-pitch instead of pitch if you're talking memory. One cell of memory using a "20nm process" is actually 40 nm across. You'll read all kinds of foolishness about how the interconnects are 20 nm across, or a unit memory cell is a 20nm on a side square, or the oxide layer being 20 nm across (which would actually be Fing huge by current standards). Basically almost all size comparisons will just be random crap and no journalist or marketing PR guy ever makes a correct analogy using half pitch, they'll say absolutely anything other than the correct answer, which has made me laugh for decades now.
Everyone knows everything comes from China. Including semiconductors. Well, actually, no. There's a nice list of plants at wikipedia. You'll see a lot of US addresses. Yes you can probably buy a knock off 555 or 741 from China, but they have almost no small scale plants at all. Pretty much processors come from the USA and a scattering of small time players around the globe. That's interesting. We (USA) make really tiny processors and really giant industrial machinery and not a whole heck of a lot in between. You want a 500000 ton mining dragline? We got it. You want a 22nm processor? We got it. You want a shoe? No we don't make those in this country anymore.
http://en.wikipedia.org/wiki/List_of_Semiconductor_Fabrication_Plants
Finally processes are a moving target and different mfgrs and different products are at each stage. There are a couple plants being built for 16 nm process and there are prototypes of "real 16 nm chips" floating around. 22nm process memory was shipping two years ago, 22nm process CPUs are much harder to design. Intel is already shipping 22nm process family CPUs, so AMD gets a golf clap for promising to catch up later this year with something microscopically better. To the best of my knowledge 11nm is not out of the lab yet even for fooling around with.
And that's about all I know from making some investments in mfgrs since the 80s, some of which worked, some not so good. Not currently investing in this market, but I still keep up with the times and I do a lot of electronics in my basement.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
They are capitalized at about $10 Billion, so either they are betting everything to go into high-end processor manufacturing, or they are merely hoping that somebody builds their chips for them so that they can collect the licensing fees.
FWIW I'm a big ARM fan, but I don't believe they can make 20nm happen any time soon without the assistance of a major industry player. Perhaps they are trying to sound out a partner with this announcement.
TSMC, UMC and Samsung are some of the biggest players in the field. And almost all the DRAM is manufactured in South Korea.
Jan
I'm not sure you understand how ARM's business model works. They don't manufacture chips themselves, and they don't even hire somebody else to manufacture chips for them. They also don't design chips for a specific process node. They just produce a design and leave it up to a company like Texas Instruments to figure out how to build them at a certain process node (or hire some fab company to do it).
The 20nm statement is just a prediction. They're saying they expect their customers to get 20nm parts out in 2013.
http://slashdot.org/submission/2095781/tech-scare---quantum-depression-from-quantum-communications
This is perhaps one of those hints as mentioned.
http://www.aisnota.com/slashdot/ Welcome to Logic and the Future
Yes I know they aren't really building fabs - I did mention in my post that they were hoping to collect the licensing fees from their 20nm designs, which AFAIK is exactly how they make money.
My point was that ARM can't make this happen in a given timescale - but they CAN encourage their licensees to support them, which seems to be what they are doing.
http://reisen-nach-frankreich.com/ Reisen nach Frankreich
.. The more advanced chips should allow device makers to improve the performance of their products without reducing battery life, or offer the same performance with longer battery life.
Unless these ARM chips are going to be fab-ed at Intel, I'm not so sure that ther will be much of a perf or battery life change. Intel is leading the technology pack with a FinFET which has the ability to reduce leakage in these types of technlogy nodes to make a real advance and tradeoff (same perf w/o longer battery life). Other foundaries (e.g, TSMC's 20nm node), aren't so advanced.
For example, TSMC's 20nm HKMG node is suffering from the increasingly narrow line widths and their lack of FinFET techology. Basically, you get more gates (~2x), but not much else in this new technology node (~15% better power and perf). Instead of their normal offering of a high-perf and low-power flavors of their process, @20nm, TSMC is only offering 1 process because it's not possible for them to make an meaningful tradeoffs. Other foundaries (e.g., Samsung, and Global Foundaries) aren't in much better shape.
Otherwise, why would IBM be willing to unleash the Nazgul on the Hercules Emualtor?
Here is a rundown of the mainframe legal landscape.
x86 is the fox in the mainframe henhouse, just like it is with RISC. Just today we hear that Windows is the #1 server OS.
Aren't you glad that you aren't trying to sell a MIPS server right now?
ARM has lots of engineers working with the foundries at the low levels to co-design process design rules. Not only do they have synthesizable IP, but also physical IP for specific processes, which they sell to customers. Note that most of their customers don't even have access to the physical design view (the transistor-level details) of their processors. This is managed at the fab-level where the physical substitutions are made to the masks after customer submission. Only the biggest customers have access to the synthesizable code. Academic researchers don't even get access to it (with a few exceptions)!