Apple chip partner TSMC plans to launch 7nm process in 2018, 5nm in 2020
Taiwanese semiconductor foundry TSMC is reportedly preparing to roll out its 7-nanometer process node as soon as 2018, with a jump to 5 nanometers currently penciled in for 2020.
In the interim, TSMC expects its 10-nanometer fab to begin production ramp-up by the end of this year. TSMC co-CEO Mark Liu made the process announcements at an investor meeting, according to DigiTimes.
If TSMC is able to meet the aggressive launch schedule, it would put the company at the leading edge of wafer fabrication. Intel -- widely regarded as the gold standard -- is already behind schedule in its own transition to 10-nanometer production, which was initially expected to begin last year.
Many believe that either the 5- or 7-nanometer nodes will be the last commercially viable process shrink, given the limitations of physics. Such small processes magnify quantum effects, requiring substantial changes to transistor architecture and materials.
The industry has only now begun to shift to 14-nanometer production. Apple was among the first companies to take advantage of Samsung's 14-nanometer process with the A9, though some A9 chips are fabricated on TSMC's 16-nanometer line.
In the interim, TSMC expects its 10-nanometer fab to begin production ramp-up by the end of this year. TSMC co-CEO Mark Liu made the process announcements at an investor meeting, according to DigiTimes.
If TSMC is able to meet the aggressive launch schedule, it would put the company at the leading edge of wafer fabrication. Intel -- widely regarded as the gold standard -- is already behind schedule in its own transition to 10-nanometer production, which was initially expected to begin last year.
Many believe that either the 5- or 7-nanometer nodes will be the last commercially viable process shrink, given the limitations of physics. Such small processes magnify quantum effects, requiring substantial changes to transistor architecture and materials.
The industry has only now begun to shift to 14-nanometer production. Apple was among the first companies to take advantage of Samsung's 14-nanometer process with the A9, though some A9 chips are fabricated on TSMC's 16-nanometer line.
Comments
Really though, the fact that all A9X production is handled by TSMC tells you who Apple prefers, and I suspect the majority of A9 production is handled by TSMC's fabs.
That means that even if it is physically possible to find a way to continue the transistor shrink game, that doesn't tell us if Moore's Law will continue. It can only continue if playing that game continues to be profitable. And there are reasons to believe it might not be. The experience of the industry up until now has been that it's always cheaper to build a processor with N transistors on the next node than on the current node. For example, a 1 billion transistor SOC fabbed at 28 nm is cheaper than a the same 1 billion transistor SOC fabbed at 32 nm.
There has been a lot of speculation that those days might be over -- that when we go beyond 14 nm, we might find that the processors become more expensive ,not less expensive. If that's true, then the only reasons to continue moving forward are improvements in performance/watt. Of course, that's a good reason. But it might not be a good enough reasons the further we go....
I'm still amazed at the current speeds and battery capacity.
Quote from a public lecture at the Royal Society a couple of years ago on particle physics..."Science is what you do when you don't know what to do or how to do it"
For Si, the lattice constant—the physical dimension of a crystalline cell—is 5.431 Å, which means 0,5431 nm!
There is no way of getting a smaller node size because a single layer of crystalline silicon would be, at least, this big!
Of course there are other practical limitations that may (or may not, my microelectronics is a bit rusty and, possibly outdated) limit the shrinking well above this number. Of the top of my head, tunneling comes to mind.
It is well known (to some audiences) that the position of an electron, for example, can't be exactly defined. There is an intrinsic uncertainty about its position in space. If a physical barrier is thinner than this uncertainty, there is a chance that the electron will simple appear on the other side. This phenomenon is called tunneling, and there is a vast array of semiconductor devices that works based on this very principle.
So, if you make a transistor small enough, there may be the chance that the electronic current will simple tunnel through the transistor, completely ignoring if it is open or closed! And there goes its functionality!
We are indeed approaching a hard wall with silicon technology, but other alternatives may be found! The Moore's Law is a human fabrication, which has been true for a long time, but it's not a Natural Law. The important factor in it is its economic impact: Moore's Law has been allowing, for the better part of half a century, for more processing power at the same (or even lower) price, as the the years gone by. That may change...
Maybe quantum computing, or "spintronics", will come to our rescue in less than 10 years! Or maybe not...
Apple will be forced to go to the A series CPU for the MacBook line. The discrepancy between the performance of the core i7 and the A10X, not to mention the A11X, will simply be too great.
As far as GPU performance goes, Intel chips are even further behind.
Intel should have granted Apple an x86 license or built the mobile CPU as originally requested by Jobs. It's now too late. The momentum behind the A series and the newly released S series is simply too great to stop it now.
The Intel apologists are hoping TSMC's 10 nm process will not able to be introduced until 2017. From the looks of things, TSMC might be able to introduce 10 nm for the iPhone 7 at the end of 2016. And if 10 nm InFO goes into the next version of the iPad pro running the A10X, from a pure performance perspective, not to mention battery life, Intel won't be able to field a competitive CPU of any type for a portable machine.
Unlike the heady days of the PowerPC when IBM and Motorola, allowed complacency to set in, Apple won't let it happen this time. The A series CPU will hold the performance crown and the discrepancy with x86 will grow worse with the development of each new generation.
- many of these molecules are in the single nm-range (the entire electronic components, not just the leads!)
- intrinsic transistor/switch functions
- large quantum confinement
- atomic-scale barriers
- bi-stable electronic states (memory w/o need for power)
- no need for long-range crystalline perfection (molecules self-assemble in smaller, isolated complexes)
- self-repairing
- a whole range of "active" molecules already pre-existing, eg light harvesting, chemo-sensors …
Single molecular-electronic components have been demonstrated since more than 10 years back. However, there are quite a few hurdles left to overcome, and at least another decade of work before entire circuits can be produced. But it definately has the potential of replacing the Si logics we use today.It isn't like Intel who is the sole supplier of x86 chips. Apple has options when it comes to TSMC. And losing Apple's business means a serious loss of profit.
Hence, I take TSMC's promises as being more truthful than not. And Apple will deploy the 10 nm node at the earliest opportunity. And it may mean 10 nm goes into Apple Watch initially if TSMC ramps up 10 nm at the end of this year with the iPhone 7 and the next version of the iPad pro remaining on 16 nm but with InFO. As you point out, 10 mm likely is deployed in the iPhone 7S.
I too am salivating over 10 nm Apple devices. I would really like to see the capabilities of iOS extended or OS X adapted to the A series processor.
I know it's coming. Just hard to be patient.
TSMC: 10nm is on-track for volume production start in Q4 2016
Lithography is likely to become the next limiting factor in being able to shrink things down and after that the 'current method' has pretty much reached the max of its physics potential.
Other methods are in development but who knows which one is next. Carbon nanotubes or organics could allow Moore's law to continue on, just not using current manufacturing techniques.
"Carbon nanotubes" and "organics" are in fact molecules. Their intrinsic electronic properties are intended to be used to create logic functions. That's why it's called molecular electronics. On the contrary, the "current methods" you are referring to are solid-state based.
This Exactly, and to be precise Moore's Law is ALL about economics, the performance that comes with it was more of a bonus from that point of view.
And Moore's Law is dead in that sense, at least not functioning at its original prediction. 28nm was the last node that brings cheaper wafers. Everything below is going up, luckily TSMC is working on 16nm FFC, which will some day be cheaper then 28nm few years down the road.
On to 10nm, TSMC 10nm is more like a half node. So no, on a technical level it doesn't complete with Intel 10nm, not to mention Intel's 10nm is much more feature rich and coming in and SHIP in early 2017. Although it does make very good headlines for tech site alike.
It is the 7nm that is interesting because it will be, truly for the first time ever TSMC will be ahead of Intel in terms of Fab Tech. ( Not in terms of Node numbers, but in actual features set and technical performance. )
That's all people really care; pissing contests are left for engineers.
Intel has refused to fab for someone else because it's a low profit proposition; they may have to eventually with TSMC pressing them.
I wouldn't be surprised if within 2-3 years Apple switches out the their low end laptop for their own A chips. That will hurt Intel.
Having the best X86 chip doesn't matter if 95% of people are not using it directly (still being used in servers and niche applications).
That's where Intel will be if they don't wise up; maybe it's already to late.