Overclock.net › Forums › Industry News › Technology and Science News › [Arstechinica] Transistors will stop shrinking in 2021, but Moore’s law will live on
New Posts  All Forums:Forum Nav:

[Arstechinica] Transistors will stop shrinking in 2021, but Moore’s law will live on - Page 4

post #31 of 46
Quote:
Originally Posted by Asmodian View Post

My understanding is that the current wavelength isn't short enough to pattern even 16nm, that is why you have to do multipatterning. I believe 7nm is possible with the current wavelength but with even more multis in the multipattern.

I must say directed self-assembly (DSA) sounds like much cooler tech. How far away are seeds for an i7? What do you need to water them with? biggrin.gif

Current wavelength is 193nm on an Immersion tool and no its nowhere close to being able to print 14/16nm lines straight up. Double patterning started several generations ago but is mainly useful for straight lines in DRAM/NAND. Its harder to print random transistors this way.
post #32 of 46
Extreme ultraviolet lithography (also known as EUV or EUVL) is a next-generation lithography technology using an extreme ultraviolet (EUV) wavelength, currently expected to be 13.5 nm. EUV is currently being developed for possible future high volume use in 2020 for Intel's,[1] Globalfoundries'[2] and Samsung's [3] 7 nm node, TSMC's 5 nm node,[4] and SMIC's 14 nm node.

They are trying to make Ultra Violet a commercial viable option.
post #33 of 46
Nice Wiki copy...
post #34 of 46
I saw a few comments that suggested that nanometers on the box corresponds to copper traces in the chip. This is not the case. (I'm on mobile so I wont bother quoting now.

When they talk about 28 or 14nm they dont mean that the copper traces are 14nm but that the gap that electrons have to cross in a transistor is 14nm. This is the pitfall of a transistor. If the material inbetween is so small that "quantum tunneling" happens then that is the absolute limiting factor of a *transistor size.

This measurement is called "from gate to gate" search for "transistor parts" if you want to know more, its quite interesting stuff.

*transistors can still get smaller with higher resistance materials.
Edited by Yttrium - 7/26/16 at 7:48am
post #35 of 46
I may be way off base here, but I believe Neurons have multiple energy states they use to communicate, wouldn't ditching the binary on/off approach in and of itself massively improve computational power? '

Could basically reboot the die shrink race and start all over with big 60 nm multi state transistors that speak in a 0,1,2,3,4, or 5 rather than 0 and 1 to one another.
post #36 of 46
Quote:
Originally Posted by DNMock View Post

I may be way off base here, but I believe Neurons have multiple energy states they use to communicate, wouldn't ditching the binary on/off approach in and of itself massively improve computational power? '

Could basically reboot the die shrink race and start all over with big 60 nm multi state transistors that speak in a 0,1,2,3,4, or 5 rather than 0 and 1 to one another.

From what I have read, quantum computers could have up to a 13 distinct states, which could allow them to operate across many more prime numbers. The initial research suggests that they could be very powerful for purposes of cryptography, because they have the potential to factor very large prime numbers very quickly. There really is not guarantee that they would be faster or smaller for other types of computing though. I think among the biggest problems is they are having a hard time saving quantum states. It is difficult to have sophisticated software without reliable memory (at least in a Von Neumann architecture).
post #37 of 46
Quote:
Originally Posted by Yttrium View Post

*transistors can still get smaller with higher resistance materials.

Not much smaller. One of the first basic quantum mechanics problems given to new students (sophomore level) is to calculate the tunneling through an infinite potential barrier. It's basic because it just follows an exponential decay through the width of the barrier.

Even with a perfect insulator of infinite resistance, tunneling is still a problem. It's often described as leakage current, and it's charge that's not going where it's supposed to.

There really are hard limits we're about to get to. Keep in mind, the spacing between atoms in most crystalline materials, including just about everything that makes up a processor, is 0.4 nanometers. It varies a bit, 0.39, 0.41, that sort of thing, but the range really is pretty narrow.

What the number of the side of the box refers to is getting down to single digits in terms of how many atoms across it is. Contemplate on what something that's comprised of less than ten atoms looks like.
post #38 of 46
Quote:
Originally Posted by mothergoose729 View Post

From what I have read, quantum computers could have up to a 13 distinct states, which could allow them to operate across many more prime numbers. The initial research suggests that they could be very powerful for purposes of cryptography, because they have the potential to factor very large prime numbers very quickly. There really is not guarantee that they would be faster or smaller for other types of computing though. I think among the biggest problems is they are having a hard time saving quantum states. It is difficult to have sophisticated software without reliable memory (at least in a Von Neumann architecture).

The transistor method shouldn't require the quantum state issue though, just adjustable voltages though. like 0 volts = 0 .00001 volts = 1, .000015 volts = 2 etc. etc.

Fine tuned enough you should be able to send a byte instead of a bit per every gate opening which would in theory make it 8 times faster than regular transistors, so if you can shrink them down to say, 48nm or whatever, at the same clock speeds, would be able to do the same amount of work as a 7nm current CPU.

I'm probably way over simplifying it and it wouldn't work the way I'm imagining it, but the multiple states of current method neurons use could in theory do the job I would think.

Pound per pound, brain matter still smashes the the most advanced of silicon by multiple orders of magnitude in processing power at a tiny fraction of power consumption using that method, so I don't see why emulating nature wouldn't be the next logical step.
Quote:
Originally Posted by Mand12 View Post

Not much smaller. One of the first basic quantum mechanics problems given to new students (sophomore level) is to calculate the tunneling through an infinite potential barrier. It's basic because it just follows an exponential decay through the width of the barrier.

Even with a perfect insulator of infinite resistance, tunneling is still a problem. It's often described as leakage current, and it's charge that's not going where it's supposed to.

There really are hard limits we're about to get to. Keep in mind, the spacing between atoms in most crystalline materials, including just about everything that makes up a processor, is 0.4 nanometers. It varies a bit, 0.39, 0.41, that sort of thing, but the range really is pretty narrow.

What the number of the side of the box refers to is getting down to single digits in terms of how many atoms across it is. Contemplate on what something that's comprised of less than ten atoms looks like.

Don't know what college you went to but in the early 2000's it was Classical Mechanics/Newtonian Gravity ---> Electro-Magnetic Force ----> Weak and Strong Nuclear Force all before touching on quantum mechanics and relativistic physics. If I recall you gotta at least have at least Calc 3 under your belt before you even have a chance at Quantum Mechanics
Edited by DNMock - 7/26/16 at 12:02pm
post #39 of 46
Second semester of sophomore year when I took it, at a state university. Standard track for an engineering degree. Yes, we had Calc 3 before taking it.

You don't need nuclear forces, though. This is basic QM, an introduction.
post #40 of 46
Quote:
Originally Posted by Mand12 View Post

Quote:
Originally Posted by Yttrium View Post

*transistors can still get smaller with higher resistance materials.

Not much smaller. One of the first basic quantum mechanics problems given to new students (sophomore level) is to calculate the tunneling through an infinite potential barrier. It's basic because it just follows an exponential decay through the width of the barrier.

Even with a perfect insulator of infinite resistance, tunneling is still a problem. It's often described as leakage current, and it's charge that's not going where it's supposed to.

There really are hard limits we're about to get to. Keep in mind, the spacing between atoms in most crystalline materials, including just about everything that makes up a processor, is 0.4 nanometers. It varies a bit, 0.39, 0.41, that sort of thing, but the range really is pretty narrow.

What the number of the side of the box refers to is getting down to single digits in terms of how many atoms across it is. Contemplate on what something that's comprised of less than ten atoms looks like.

I know, I just wanted to prevent nitpicking from the one guys that starts talking about better resistance. ( Or starts talking about infinite resistance ahem )

Good times ahead, once we hit the barrier things like clockspeed needs to go for better power consumption, maybe we'll get asyncronus CPU's? Maybe I should become a microarchitecture engineer? Maybe I'll have some cake.. Who knows really?
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Technology and Science News
Overclock.net › Forums › Industry News › Technology and Science News › [Arstechinica] Transistors will stop shrinking in 2021, but Moore’s law will live on