Slashdot Mirror
Samsung Reaches Milestone For 14nm Technology
An anonymous reader writes "Samsung announced a milestone on its development of 14nm manufacturing semiconductors, claiming that it offers major advantages to system-on-chip devices using in consumer electronic products (especially lower power). They recently taped out a Cortex-A7 processor with this technology, calling it a significant milestone for the fabless ecosystem."
What is fabless ecosystem?
#
#\ @ ? Colonize Mars
#
Apple should sue them for "method and apparatus" to make something smaller.
THL phish sticks
These are ARM based chips and they are completely dominating, if Intel had 30 years of advanced technology they wouldn't be so far behind on the low powered chip side of things.
Umm, no. They had a hell of a time moving to 22nm and getting volume production up, pretty much your entire post is just paranoid delusion.
Intel just has more money to throw at the problems, and they've managed to get a lead of a few years on the other companies. They only maintain that lead because they keep pushing forward.
The problem is that makes it harder for everybody else to compete, but that's not really Intel's fault.
ARM is cheaper and that's why intel is screwed
Few computing tasks need the power of an I core CPU
To be honest, I tried to sum up some outrage over your statement... but I just couldn't.
If Intel really is THAT good that they can coast through life and still beat their competition in technological advancements... then fine.
Eventually, if your theory is right, someone will come along (maybe even ex-Intel engineers) and beat them. It is already happening in other sectors.
THL phish sticks
Mod post ignorant. If that were true, AMD wouldnt have been so far superior to Intel back in the P4 / Pentium D days.
Fact is it takes about 3-5 years for this tech to be fully realized, and Intel is currently (AFAIK) the only one with solid 22nm production simply because their R&D budget is huge. If you find that scary or whatever you can send your dollars to AMD to help them get up to speed.
Tape-out means Samsung has got a design that they think might work but hasn't actually been fabbed. Intel has had WORKI(NG 14nm microprocessors (in the lab, not in production) since mid 2012. It will be mid 2013 before Samsung has that. Intel will be in production by then. (Production starts some months before retail shipments since they have to build up inventory, get parts to integrators, etc.) Intel's lead seems to be about where it was before, not getting larger but isn't clearly getting smaller.
"Tape out" is a term of art of the processor industry. See http://en.wikipedia.org/wiki/Tape-out where the first sentence will tell you "In electronics design, tape-out or tapeout is the final result of the design cycle for integrated circuits or printed circuit boards, the point at which the artwork for the photomask of a circuit is sent for manufacture."
"Fabless ecosystem" is another term of art of the processor industry. Wikipedia is similarly helpful here at http://en.wikipedia.org/wiki/Fabless_manufacturing -- where the first sentence will read "Fabless manufacturing is the design and sale of hardware devices and semiconductor chips while outsourcing the fabrication or "fab" of the devices to a specialized manufacturer called a semiconductor foundry."
STFW FTW.
Depends on your cost metric. They're only cheaper if you don't consider cost per computation/second.
"Tape out" is in my dad's generation they used the same tech for photolithography for both PCBs and ICs. In other words a "Draftsman" (which is kind of like a CAD operator, but manual, done by hand) using what looks like black electrical tape stripes on clear mylar sheets. Then a projector blasts UV light thru the marked up sheet onto a photosensitive copper circuit board, or silicon slice, and where the UV hits the plastic polymerizes and is "permanent" and where it doesn't, it washes away. Sorta like a photo negative enlarger but more of a shrinker than an enlarger... which is another mostly dead technology. You'll meet people who rewrite history for laughs who claim the "tape" is magnetic tape of cad drafting or maybe Verilog/VHDL. In the "biz" it means the dev team has ended work and the responsibility is now entirely on the production team (assuming it achieves production level success on the first try, without any design issues ruining yield, LOL)
"Fabless ecosystem" is fru fru talk for you outsource your manufacturing to a company (usually a competitor) you trust to give you reliable access to their best processes, while trusting them not to "pirate" your IP which is your companies only resource. Its a great idea for weird stuff where you can corner the market or R+D or teaching. Strikes me as an idiotic business model for competitive "mainstream" processors or generic commodity chips.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
One could think that this announcement of 14nm development is Samsung one-upping their competition.
Another interpretation is that companies need to exercise "continuous disclosure" in order to be taken seriously in the share markets and not fall foul of the market regulators which insist that companies reveal important information as soon as is practicable so that investors and possible investors get a true picture of the company's market worth. In most cases, a good-news story is a great way to have the market clamouring to invest, and so assists the company to raise the capital needed to get its developments to market.
It also does not hurt to rub the nose of the opposition.
Looking at space, radio, science and computing from a 'down-under' amateur enthusiast perspective.
They recently taped out a Cortex-A7 processor with this technology, calling it a significant milestone for the fabless ecosystem."
I'm very good at the English language but I have no idea what this means. How do you 'tape out' a processor? What's a 'fabless ecosystem'? (The rainforests are rather wonderful, I hear.)
"Taping out" is the process of laying out the actual lines that will become the paths of the circuit. This used to be done with actual tape and photographed and reduced in size. Somehow I doubt they actually used that method with this and more likely the work was all done on a computer.
I can only assume a "fabless ecosystem" is a fancy way of saying "the industry of making something when you don't make anything" or chip design/IP creation.
I'm very good at the English language but I have no idea what this means. How do you 'tape out' a processor? What's a 'fabless ecosystem'? (The rainforests are rather wonderful, I hear.)
Tape out is an expression that ASIC (Application Specific Integrated Circuit) designers use. It means to finalize a version and send it out to be manufactured. There is some disagreement where it came from, but it most likely referred to the fact that magnetic tapes were used to transfer the (largish) electronic design descriptions to the manufacturers site. These tapes were used because they were the only medium at the time that could store the entire files on one volume instead of having to split the data among volumes.
"Fabless ecosystem" refers to the lack of a "fab" or fabrication facility. There are two types of setups for designing and manufacturing integrated circuits. The first is the "classical" model where a design company designs and manufactures the chips themselves. The other type of model is one where two separate companies design and manufacture the chips. This is called fabless, since the designer does not have a fab themselves (and in theory are free to go with any of a number of fabs and or manufacturing methods). Being fabless has advantages and disadvantages. The main advantage is that your company does not have to eat all of the overhead of owning and operating a fab (Each one costs around the same as a modern aircraft carrier, and employs the same number of people). If you have limited production runs you don't have to pay the full overhead, but instead that cost is split among all of the customers of a given fab. It also allows you to switch between fabs if your current fab is not meeting your particular requirements (cost, failure rate, QA, whatever). The biggest disadvantage is that your company doesn't have any control over the fabrication process, so you have very limited say as to how they manufacture your parts. The fab is also free to manufacture parts for your competitors, which makes it very difficult to have significant performance differences between competitors. Intel maintains a large part of its performance edge by using proprietary fab techniques that other companies do not have access to.
-=Geoskd
I wish I had a good sig, but all the good ones are copyrighted
apple has never fabbed a chip in the entire history of the company, nor would they even begin to know how to, fucktard
you know AMD is only 10 months younger than intel right? or that Acron computers, where the ARM guys came from has been developing their cpu since the 1980's?
cause you make it sound like they just popped up out of nowhere yesterday, or maybe that's just your uninformed tinfoil hat conspiracy that intel, 2pac and sea lab are really ruling the world.
on Apple's patents on "making something", "thinking about making something", and "dropping acid to free your mind to think about making something"?
Samsung fanbois here are also Steve Jobs fans, since he led the revolution they're all trying to claim now for themselves
A processor is designed using a programming language like Verilog or VHDL. These languages provide standard logic cell libraries that support floating-point, integer arithmetic and multiplication.
Whatever you can implement in C/C++ software, you can implement in hardware, with various optimizations like parallel processing, pipelining.
At the same time as the processor is being designed, verification tests are written to test every logic block. Tape-out is that special time at the end of the project when the complete system has been designed and all the verification tests pass. Then the designers don't have anything else to do. They just have to wait until the first chip dies are baked, packaged into resin and heatsinks, and mounted on test boards.
In the past, the whole directory system would be tarballed onto a tape drive and the package sent over by courier to the chip fab where the files would be compiled into silicon via a set of lithographic masks used in order to etch each layer of the chip.
hmm TSMC has a rather large profit margin for 1-5b to even open it.
9.53B in complete profit in 2010 so within 1 year not only could said foundry pay its self off, it could also pay off all employees and leave 9.5 BILLION in liquid assets per year.
Revenue $13.982 billion (2010)
Operating income $4.444 billion (2009)
cost to open a foundry is also found on wikipedia between 1-4 billion dollars, which given yearly profit margins...is one insane investment...someone is living a very nice life now.
-Noc
TVs, Blu-ray & DVD Players, Home Theater Systems, Laptops, Chrome Devices, Media Players, Cameras, Camcorders, SD Cards, Laptops, All-in-One PCs, Tablet PCs, Monitors, Printers, Memory & Storage, Washers & Dryers, Refrigerators, Microwaves, Dishwashers, Ranges, Vacuums, LED Lighting, TV Accessories, Cell Phone Accessories, Tablet Accessories, Printer Toner & Supplies, Laptop Accessories, Digital Camera Accessories -- http://www.samsung.com/us
It's not like Intel had any extra gears to put in when AMD was spanking their ass some years back, they had a process lead and sustained that lead even as AMD was putting out much better CPU designs but no more than that. But the CPU business has been very much so that the one who invests more, earns more and then has more to invest more again and Intel has simply beat AMD by spreading the costs of R&D across more chips. For a while AMD beat it by developing a better design on a lower budget while Intel floundered but in the end economics of scale won out.
This is not just Intel, the number of semiconductor players has been shrinking drastically with processor size and that trend is only going to continue, five out of the top six biggest semiconductor companies increased their market share last year. More and more go with foundries and the foundries are getting fewer and bigger too. Now ARM chips might not be the most powerful chips in the world, but they make billions of them so their processing technology is pretty good. They're going to give Intel a good run for their money, it's certainly no walkover.
Live today, because you never know what tomorrow brings
My kids have iPods and an iPhone and a Macbook Air lying around the house. Not my cup of tea personally. And yes - Apple sues too much over stupid (and increasingly invalidated) patents. And yes, it reminds me of SCO's business strategy, and its a really really bad business strategy.
Trademarks can be lost or weakened by not taking action, this does not apply to patents.
I will agree that there are a lot of clueless people around tho.
you're right that few computing tasks need the power of an I core until you start looking into the performance and do what I'm planning. My next build revolves around an E3-1245v2 Xeon that will be underclocked to around 10 percent. Should meet my current performance (x2-240) for less then 15 watts. Right now, I'm projecting a total of 60w for the CPU/GPU/SSD/3x 2TB Drives. This system will be run off a solar power system (off-grid home) and used as the Gaming/Media Center
Mod me up/Mod me down: I wont frown as I've no crown
Actually it does apply to patents. I have no idea where you get your info from.
Are agnostics skeptical of unicorns too?
Then our SSDs will survive a whole SEVEN program/erase cycles.
I did some "taping out" in the 70s, the black tape was for things on both sides of the PCB (e.g. holes and edge connectors), while blue and red tape were used for top and bottom traces.
In the 80's someone once said.
No one will ever need more than 1GB of memory....
I think you will find that was 1 megabyte. - the original IBM PC could only address 640KB.
1GB would have been unthinkable until a few years ago. I can remember being massively impressed by a 64MB machine in the mid '90s.
"Intel probably have the processors that are 'coming out' in 2017 already laying on a shelf in a warehouse somewhere by the millions."
In the off chance you meant this literally, no. It would be extremely stupid to stock up real hardware that far ahead. I do suspect however that Intel has the technology already in the labs, which I suppose is what they base their so-called "roadmaps" on. The future is already here, it just hasn't been stress-tested yet.
In last news we had on the topic, Intel was at 22 nm.
It's Koren for fabulous ecosystem.
that was their total profit though, so even if they had 5+ plants my point is it would pay for its self so fast that it would be utterly negligible.
-Noc
Off topic and sarcastic as it may be, I always thought that boat is so ugly, finally they realized it's not worth all those dollars http://edition.cnn.com/2012/12/21/tech/innovation/steve-jobs-yacht/index.html
Hmm. Your ideas are intriguing to me and I wish to subscribe to your newsletter.
Cwm, fjord-bank glyphs vext quiz
I beg to differ.
I'd rather pay 200 bucks for a processor that actually gets stuff done, than pay 25 for one that tends to work well with very specific stuff. And let's not even go into energy efficiency.
Bollocks. Apple was not willing to negotiate licensing terms. Neither did they want to pay the market price for Samsung's own patents.
They have been catching up quite quickly. Most of their issues are not in manufacturing but in chip design. They simply did not have in-house designs for all the additional functionality you can find on your average ARM SoC. They also did not have any low power CPU designs which they could manufacture. Their solution was quite simple: you pick up a 1990s Pentium processor design and port it to a modern manufacturing process. The result was a chip with more performance and about the same power consumption as the top notch ARM CPU core designs.
Dead wrong. There were be almost no patented medicines then since patents are usually filed before Phase II or III clinical trials take place, which often take years to run. This is why there is a "patent term extension" may be granted to recover some of the lost sales during clinical trials.
No it doesn't. Stop intentionally spreading misinformation.
Go fuck yourself clueless bitch: http://news.cnet.com/8301-13579_3-57491406-37/2010-apple-license-offer-to-samsung-$30-per-smartphone-$40-per-tablet/
Are agnostics skeptical of unicorns too?
that was their total profit though, so even if they had 5+ plants my point is it would pay for its self so fast that it would be utterly negligible.
Only if everything goes to plan. In many Fabs, there are configuration and start up problems that can delay production for months or even years. In the microprocessor industry, a year might as well be an eternity. And remember that these fabs employ a small army, so you have 1000+ people making 40k / year + benefits, and you find your fab has a manpower cost of 100 mil /year whether you have good production or not. If you have a bad start up, you could very easily go bankrupt before you can turn a profit, and if the delays are bad enough, you find yourself badly behind the technology curve, and can only produce lower margin products such as NVrams, and Lower end FPGAs and the like. In short, it is usually very profitable, but it definitely not a sure bet, and don't forget that were talking about a loss that would annihilate all but the largest companies.
-=Geoskd
I wish I had a good sig, but all the good ones are copyrighted
Tapeout has an archaic meaning. It probably used to mean when the layout database was copied to tape, to send for creating masks. Since we don't use tapes anymore, tapeout just refers they passed sign-off and signals the beginning of production.
Actually, they generally still use tapes. You're starting to see more and more SSDs and USB type drives being used, but Tapes are the tried and true method. Today, Hard Drives (Spindle and SSD) are just getting to the same density as tapes, but tapes are far cheaper per MB, and are more likely to still be good 30 years down the line, which appeals to the archivists and MBAs in the crowd.
-=Geoskd
I wish I had a good sig, but all the good ones are copyrighted
I've got an Acron box within 3 feet if me (oh, sorry - my last day in the office was Friday).
Regardless - they were sweet little things.
They recently taped out a Cortex-A7 processor with this technology, calling it a significant milestone for the fabless ecosystem."
I'm very good at the English language but I have no idea what this means. How do you 'tape out' a processor? What's a 'fabless ecosystem'? (The rainforests are rather wonderful, I hear.)
There are many steps in the design of any modern digital device. Transferring information from step to step was often accomplished with a digital tape(s) full of files. When one step finishes they send a tape out to the next step. Thus "tape out".
The first steps are often logical and classic designs were done on yellow pads of paper by hand. S. Cray was famous for this. This is where the data bus details are set down. Think of it as the primary.h file for the hardware.
Other steps are silicon to transistor and transistor to basic gate lay out. These are often the magic of the silicon fabrication houses and differ from one to another and change in subtle ways so they can be made. This magic can involve subtle interference figures that light at the sub-nanometer level has and is specific to the wavelength and purity of the light sources, thickness of mask materials and a thousand other details.
The mask is often generated from register-transfer level (RTL) files. Google searching for RTL design flow will open some doors.
The RTL is often generated from a very high level hardware description language, VHDL... This is where the yellow pad design and the tool chain first interact for many designs.
Some designs depend on purchased blobs of logic. That intellectual property (IP) can be RTL or VHDL or some mix. The ARM processor world is a sea of IP glued together on a basic bus architecture. The bus arch. of many processors is a lot like system buses once were. Look at things like the VME bus specifications most of which can be found on the web. The VME specification has dimensions, power, voltages, signal levels, timing, handshakes and more. Today system on a chip designs are a lot like the older bus based systems. A VME bus system house would buy a disk controller (think IP blob) add memory boards (think IP blob again) etc...
Back to "tape out"... I think CrankyFool in post (#42377687) below has it right. Today the files are sent via scp or ftp. In the early days a physical magnetic tape was sent. Taping out can cover a lot of steps. The files to make a printed wiring board could be taped out. The machine shop instructions for a case/ chassis could eventually be taped out and numerically controlled machines happened.
The machine shop process is a good example. Start with a casting. The fire and smoke of a smelter and molten iron/ brass/ zinc is a specialized world. The castings would then ship to a machine shop where mating surfaces would be cut to precise specifications holes tapped for bolts. Then off to the engine assembly where different parts from different materials like crank shafts, would be assembled....
Computer chips would often move from the silicon foundry to a test step, then to a package vendor/ house to be mated with a specialized package. Today a package is itself a complex device that bonds out pads on the silicon up/ down/ left right in a way that manages thermal needs and permits interconnect to the PWB below and for low power SOC device a memory chip on top or a heat sink of both... In the day IBM ECL (emitter coupled logic) systems used tricks like spring loaded thermal plungers and a module filled with helium (Google IBM TCM) to keep the astounding amounts of heat produced by ECL in check. S. Cray (his teams) was another master of heat management. Some of the Cray machines looked like the hardware store for a dairy farm. They used a lot of stainless steel from milk process equipment to pump fluids through the system for heat management. the
Truth is stranger than fiction, but it is because Fiction is obliged to stick to possibilities; Truth isn't. Mark Twain.
No, it had zilch to do with magnetic tapes. Back in the 70s/80s state-of-the-art was to design your IC circuits on a drafting table by hand using opaque tape (opaque to the light frequency that you were using during the manufacturing step). You would then coat your silicon in a special film, and shoot light through your design, through a reducer lens and onto the silicon. Follow up with an acid bath to wash away anything that wasn't shadowed and you had your circuit.
There was also "tape" used to create one-off circuit boards for the hobbyist / HAM radio crowd. You'd start with a copper covered circuit board (either one-sided, or two-sided). Tape out (or ink out) your design on the board, then dunk it in a bath to wash away the excess copper. After which you would have a circuit board with useful copper traces, ready for drilling and mounting of components. Radio Shack used to sell kits to do this stuff.
Wolde you bothe eate your cake, and have your cake?
catching up fast definitely, 30 years ahead of everyone else is utter bullshit dribbling out of the OP's mouth.