Intel debuts '3D transistors' with 22nm chip recipe Intel has unveiled its 22nm manufacturing process. The process marks the debut of Intel's "Tri-Gate" transistors, first revealed as a research project over eight years ago, and the company has demonstrated its first microprocessor built with the new process, a chip codenamed "Ivy Bridge". Equipped with a three-sided gate – …
There are two types of transistors, the type that was described in the article and another which works in reverse, ie when applying voltage to the gate, it will restrict the flow of electrons from the source to drain.
Also it is very unlikely that crApple would be using Intel's new chips, as the iPhail products use ARM chips, which Intel does not make.
what with Apple's legal actions against Samsung, who currently provide their ARM chippery. So far this has revolved around the idea of Intel fabbing ARM chips, but what happens if Intel can provide lower consumption and iOS can be ported over to whatever x86 variant these new chips are running on?
This post has been deleted by its author
I do hate crApple too (El Reg = FAIL for mentioning them), but ALL mobile phones these days contain a good few ARM processors. Some as many as 9 or 10!!!!
... and Intel DO make ARM SoCs through the acquisition of Infineon's mobile business which was completed earlier in the year.
No one said that intel could not act as Apple's fab in the future, in fact there are rumors going around that indicate that this could be the case.
Also, let's not start confusing the whole verilog/netlist-level ISA argument of ARM vs x86 with what was announced here, which is a transistor-level process improvement and is completely unrelated to what processor architecture you choose to use.
A transistor consists of three basic components: the emitter, the base, and the collector.
A field-effect transistor consists of a source, a gate, and a drain. But the term "a transistor", unless otherwise qualified, refers to a bipolar transistor. Even if they haven't been using bipolar transistors to build computers for quite some time now.
"A transistor consists of three basic components: a source, a gate, and a drain. When a voltage is applied to the gate, electrons flow from the source to the drain, turning the transistor "on". When a different voltage applied, current between source and drain stopped, turning the transistor "off"."
Not the best description of transistor operation....
Yes, for a p-channel MOSFET if you apply a voltage between gate and drain, then current will flow from source to drain, but only if there is voltage applied between source and drain.
For your third sentence, I think you probably wanted to say: if you set the gate voltage to zero, then the current stops flowing from the source to drain.
Saying that by changing the voltage you get the effect is somewhat ambiguous as you do not specify whether you mean increase, decrease, set to zero etc etc
Also as CrazyOpsGuy says, there are far more types of transistors, some are switched on or off by voltage, others are switched on and off by current. The control current and voltage can positive or negative depending on the type of transistor.
To set the record straight, the Register's explanation of an FET is entirely adequate with regard to illustrating the point of deeper coupling of the gate's field into the channel, and the additional channel "width" available by folding a planar device into a "U" shape.
The nit-pickers, as is so often the case, offer a truth that is in need of even more qualification, i.e. is less generally true, than the original.
Firstly, crazy guy, depletion mode transistors are not normally included in any CMOS process - the standard "inverter" component would be a rail-to-rail short-circuit at power-up. Valid point but irrelevant.
Secondly, Mr. Savard, the usage of the term "transistor" is not as rigorously defined as you suggest, certainly within the CMOS discipline - all of the processor/memory/sub-micron world - it refers to a FET; whether by definition or mere usage I care not.
Thirdly, TWB, getting all tricky with the details of what voltages are referred to which terminals doesn't add clarity. Never mind that you yourself fall into "when a voltage is applied to the gate" - without mentioning polarity (needs to be negative for a P-FET) - not to mention that the choice of source and drain is often quite arbitrary, defined by which terminal we choose to connect the substrate to - the "reverse diode arrangement".
Oh, and if that wasn't enough detail, remember that these devices work in the "deep sub-threshold" region, they're hardly switched on at all, so conventional electronics understanding is quite a way off the mark.
This sub-threshold operation is why the on/off ratio (of current) is so poor, hence why they need to be run with high leakage currents, to make the on-current they need to drive further devices (the fan-out). The Intel device is essentially a triple-breasted "fin-FET" from long ago, worthy because it has a much improved on/off ratio and hence lower leakage//static power consumption.
Whatever it cost them to develop, they can look at spending that again on lawyers and vexatious patent suits, if they choose to supply Apple.
This is only the beginning of Dimension Z. Eventually we will layer the silicon in stacks hundreds high.
The breathlessly anticipated "Intel announces deal to fab Apple CPU" news is silly. You cannot expect Intel to announce that they're fabbing Apple chips. Apple deals don't work that way. If you can't keep your mouth shut, Apple is not going to work with with you. Apple makes their product announcements not years in advance on the component level, but on a stage with the CEO showing off products that are already manufactured and available for the press to fondle immediately after - with millions manufactured and available for order on a certain date in the near future at a certain price. In most cases somebody's got to buy the thing at retail and tear it apart with some sophisticated gear to find out what's in it.
Apple doesn't sell computers. They don't sell phones. They don't sell music players. They don't sell an OS. They sell "The Apple Experience," and part of that experience is that you don't get to peek in the oven and know what wonderfulness their gnomes are baking until it's done.
I am in the business and have respect for Intel - they may be big, but you NEED to be big to bring innovations to fruition in the semiconductor industry - Intel are consistently ahead of all other rivals (including my employer).
I have no respect for the crAppl€ marketing company.
I am in the business and have respect for Intel - they may be big, but you NEED to be big to bring innovations to fruition in the semiconductor industry - Intel are consistently ahead of all other rivals (including my employer).
I have no respect for the crAppl€ marketing company.
Well Paul, I do respect your opinion, and you may be " in the business " with an "Employer". However, looking at your post would I be correct in guessing that you are there on work experience ?
Actually, I disagree. While intel does move the industry forward with great leaps in semiconductor fab processes and core designs, Apple has also moved the software world forward in many ways, for example WebKit and LLVM just to name a couple of the most well-known ones. While it's true that both those examples were based on existing open source projects (KHTML and the original university LLVM project), Apple has enormously helped them move forward and we all enjoy both today in many other products, including Android, Chromium and Gallium 3D for LLVM.
I'd say they do help in making computers marketable, if some people like to pay premium and enjoy doing so then good for them, in turn the industry as a whole gets bigger and we get more toys to play with and more jobs to make money from.
We still have choice and freedom methinks, which is the important bit.
While this technology is a worthy contribution towards lowering power use, it is not very drastic. The change from NMOS to CMOS was a lot more important for this.
Basically, an NMOS inverter gate connects the output to Vdd (high) through a resistor and to Vss (ground) through an n-type FET controlled by the input. If the input is low, current is led from Vdd to the output through the resistor (as the FET is closed), but if the input is high, the output is grounded through the FET. There would, however, also be a current from Vdd to Vss through the resistor, which uses power and generates heat.
A CMOS inverter gates connects Vdd to the output through a p-type FET and to Vss through an n-type FET, both controlled by the input. Exactly one of these will be open depending on the input, so there is no current (except leakage) from Vdd to Vss. Also, the gate will switch faster.
When you get more complex gates, the cost (in number of transistors) of CMOS increases: An NMOS NAND or NOR gate costs two FETs and a resistor, but in CMOS it costs four FETs (two p and two n). Even so, the power usage is smaller than NMOS.
Intel's 3D transistors use basically unchanged CMOS logic, just with more efficient FETs, so it is not very radical compared to the change from NMOS to CMOS.
It would be more radical and save more power if signals were double-railed: One line being high on 1 and low on 0 and another line being high on 0 and low on 1. An inverter would simply swap the two rails, and so use no transistors. By using only reversible logic gates implemented using pass transistors (each of which uses one p and one n FET), you avoid the power loss of destroying information and can get very power-lean circuits. In theory, you can build entire CPU's from reversible logic, but in practice you would probably irreversibly store values at each clock tick to avoid losing signal strength (and to implement irreversible operations). However, this is still very much experimental with the state of the art being ALUs and not complete processors.
What seems to me to be most impressive about these gates is not the idea itself (makes sense that if you wrap your poly and dielectric around the channel you can turn it on a lot more completely at a lower voltage), but the magic that has gone into the manufacturing process to allow them to get poly and dielectric into the base layer. I'm not a circuits or process guru by any means, so please let me know if I'm wrong, but this seems like it could get really ugly quickly.
How much of the improvement is due to the new geometry, and how much to the process shrink?
How do the new 22nm tri-gate transistors compare with 22nm flat transistors?
When I read about this on the BBC, they seemed to be under the impression that the whole transistor was only 22nm across!
For heaven's sake, this is a news article about Intel. Why the obligitary Apple product placement?
Yes, you can use processors in products made by Apple. The same applies to Dell, Asus, Nokia, and plenty of other companies that sell as much or more than Apple - do they not get a special mention everytime there's an article about processors?
ARM processors ship something like a billion processors a years on phones. Apple are number 5, companies like Nokia sell ten times as many. If and when those companies start switching to Intel, will be the more news.
And Intel already powers tablet/netbook sized devices, so that you could power an Ipad with an Intel processor isn't news.
"For heaven's sake, this is a news article about Intel. Why the obligitary Apple product placement?"
Could it be because Apple are probably Intel's single biggest client? If Apple were to switch to AMD, Intel would be reporting rather smaller profits next time around.
And while Dell, Asus, Nokia, HTC, Acer, and many others also make PCs, it's Apple who are making by far the most PROFIT from it. And PROFIT is all that matters when trying to run a successful business. Everything else is marketing.
(Note the subtle hints.)
"....Could it be because Apple are probably Intel's single biggest client?...." I think you'll find companies with breadth in IT, like Dell or hp, are Intel's biggest companies, not Apple. Apple's share of the server market is non-existant, their desktop and laptop share is still tiny, and even in phones it's just the "smartphone" segment that Apple can claim to lead.
Reread: "...the 22nm Tri-Gate provides up to a 37 per cent performance increase at a low voltage compared to its 32nm planar transistor...The transistor also consumes less than half the power as a 32nm planar transistor at the same performance level... ...The cost of building the chips with the new transistors...is two to three per cent higher per wafer."
The 22nm wafers contain about 2x as many transistors, so the price per transistor is about halved vs. 34nm wafers. The the power use per transistor is halved, so you get nearly a doubling of transistors per dollar (cost to intel) and a corresponding doubling of performance per watt for the consumer.
At best this could give a ~4x performance boost per dollar (purchase price) and a ~2x improvement in performance per watt (recurring cost for users), i.e. pretty impressive.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
