What happens when semiconductor process cannot be "shrunk" any smaller?

[Cogman]

Intel has to, otherwise they stand to lose revenue.

So long as the revenue growth exists from people throwing away their computers from two years to replace with one today, the money will be there to keep making computer tech that compels you to throw away that otherwise perfectly working computer.

this is somewhat their problem. Because their CPUs last for 10 years, they have to convince consumers to throw away working products. Not any easy thing to do.

I will certainly be interested to see the direction they take when the nodes do stop shrinking.

 

[Idontcare]

I will certainly be interested to see the direction they take when the nodes do stop shrinking.

Reinventing DRAM with the Hybrid Memory Cube

Vertical scaling combined with continued materials scaling (using more of the periodic table) to continue to reduce joules/bit in compute so the power-density remains manageable.

 

[Mopetar]

Intel has to, otherwise they stand to lose revenue.

So long as the revenue growth exists from people throwing away their computers from two years to replace with one today, the money will be there to keep making computer tech that compels you to throw away that otherwise perfectly working computer.

How many people actually throw away two year old computers? My parents still have an old P4 machine at home that they use. I have an aunt who's still running Windows 98 on a machine she bought over a decade ago because all she uses it for is checking her email.

I usually run on a 4 year upgrade cycle and consider that relatively fast. That and when I replace my current computer, I'm not going to throw it away.

 

[Cogman]

Reinventing DRAM with the Hybrid Memory Cube

Vertical scaling combined with continued materials scaling (using more of the periodic table) to continue to reduce joules/bit in compute so the power-density remains manageable.

Assuming that they keep the same manufacturing process, wouldn't this be MUCH more expensive? I was under the impression that each layer added something like 2 weeks to the fabbing process. I could see it reducing power, just not cost.

 

[maniacalpha1-1]

Now, not being a computer engineer or having, really, any knowledge of electronics on the silicon wafer level, this may be a dumb question, but, GPUs can be SLI'ed, why can't we increase computer capabilities by having 2 or more CPUs in the same fashion? Is it not possible to write programs to take advantage of multiple CPUs?

 

[Cogman]

Now, not being a computer engineer or having, really, any knowledge of electronics on the silicon wafer level, this may be a dumb question, but, GPUs can be SLI'ed, why can't we increase computer capabilities by having 2 or more CPUs in the same fashion? Is it not possible to write programs to take advantage of multiple CPUs?

http://www.newegg.com/Product/Produc...82E16813131378

Behold, and be amazed!

CPUs have pretty big footprint on motherboards. Plus there are issues with synchronizing cache across the CPUs that kind of makes things more difficult to deal with (Honestly, I'm surprised that the dual socket motherboards are this cheap, usually they are like $300).

It really is much more of a logistical problem than a technical one.

 

[Idontcare]

How many people actually throw away two year old computers? My parents still have an old P4 machine at home that they use. I have an aunt who's still running Windows 98 on a machine she bought over a decade ago because all she uses it for is checking her email.

I usually run on a 4 year upgrade cycle and consider that relatively fast. That and when I replace my current computer, I'm not going to throw it away.

I'm the appointed "family purchaser" for my extended family and I am loathe to the job. As such I have zero incentive to urge these folks to upgrade their systems, but I'm also smart enough to know its not in my best interest to argue with them to not upgrade as often as they do.

That said, I find myself consistently buying new rigs for friends and family on a 3yr replacement cycle. This year I will be replacing 1 desktop and three laptops.

The point is to not get fixated on the absolute number, whether its every 2yrs or every 3 or every 4 the bottom line is that whatever the number is it is responsible for roughly 80% of Intel's existing $50B in annual revenue. They can't afford to allow that cycle extend by one year, their revenue would drop by $10B easy if the cadence went from 2yrs to 3yrs.

Assuming that they keep the same manufacturing process, wouldn't this be MUCH more expensive? I was under the impression that each layer added something like 2 weeks to the fabbing process. I could see it reducing power, just not cost.

You are referring to the cycle-time of adding a metal layer (which is about 1wk) whereas the photo you are looking at is showing actual dies being literally stacked on top of each other, no cycle-time adder for that.

It does raise cost, would you rather sell 4 dies each for the price of $X (netting 4x$X) or would you rather sell 4 dies stacked together for the price of maybe 1.5x$X?

The point here isn't the ASP angle but the technological capability angle. The 3rd dimension, literally making processing cubes, has yet to be exploited. As such you can expect it to eventually be exploited.

The question for us would be, after exhausting that avenue, what then?

 

[gevorg]

For any experts on this subject:

At which nm node will it be possible to integrate most of computer components on a single chip (for a typical desktop PC, not workstation/server class)? That is CPU + IGP + Northbridge + Southbridge + RAM.

 

[maniacalpha1-1]

http://www.newegg.com/Product/Produc...82E16813131378

Behold, and be amazed!

CPUs have pretty big footprint on motherboards. Plus there are issues with synchronizing cache across the CPUs that kind of makes things more difficult to deal with (Honestly, I'm surprised that the dual socket motherboards are this cheap, usually they are like $300).

It really is much more of a logistical problem than a technical one.

I knew those things existed, but having that won't do anything for you right now, will it? Do games actually code in such a way that it would help you?

 

[Cogman]

I knew those things existed, but having that won't do anything for you right now, will it? Do games actually code in such a way that it would help you?

Depends on the game. It is the same as if those cores were all in the same package.

 

[Cogman]

For any experts on this subject:

At which nm node will it be possible to integrate most of computer components on a single chip (for a typical desktop PC, not workstation/server class)? That is CPU + IGP + Northbridge + Southbridge + RAM.

It is possible now (see microcontrollers) the problem is that whenever you put on one of those components, you essentially rob space for another (such as more cache, registers, or alus). Therein is the tradeoff.

 

[maniacalpha1-1]

Depends on the game. It is the same as if those cores were all in the same package.

ah, so any game that's programmed to use as many cores as exist will benefit?

Here's another thought, since things have shrunk so much anyway, why not simply use more space? What gamer would mind getting top level performance for mid range price, the only trade off being your case has to be a foot taller and 6 inches wider? lol.

 

[BTRY B 529th FA BN]

At some point the semiconductor manufacturing process is going to be so small that it's not going to be able to be shrunk anymore. Probably in the next 10 years. What will happen to the state of cpu's and gpu's? Will we just add more cores? Will the ever increasing power of computers start to slow dramatically?

There are different ways apparently like materials other than silicon. I think it's just the cost of production that is the factor holding things back.

 

[GammaLaser]

For any experts on this subject:

At which nm node will it be possible to integrate most of computer components on a single chip (for a typical desktop PC, not workstation/server class)? That is CPU + IGP + Northbridge + Southbridge + RAM.

Traditional DRAM process is incompatible with CMOS process used in CPUs. There is a thing called eDRAM which is easier to integrate but I suspect on x86 we will see on-package integration rather than on-die integration of RAM.

 

[Smartazz]

ah, so any game that's programmed to use as many cores as exist will benefit?

Here's another thought, since things have shrunk so much anyway, why not simply use more space? What gamer would mind getting top level performance for mid range price, the only trade off being your case has to be a foot taller and 6 inches wider? lol.

I think that if we do start stacking processors, then power consumption and heat output will probably go up. However, Idontcare is probably right in that there will be improvements to help negate some of the added power consumption.

 

[poohbear]

i recall reading how in the future they will use living cells, bio technology type stuff, because OP is right, we will reach a limit with semiconductors. With living cells they can store soooo much more data than semiconductors & they can work much faster... but the biotechnology is still in its infancy. It was a really interesting show on the discovery channel i believe.

 

[Cogman]

I think that if we do start stacking processors, then power consumption and heat output will probably go up. However, Idontcare is probably right in that there will be improvements to help negate some of the added power consumption.

Power consumption will likely go down. One of of the big issues with current processor design is the wiring, it looked (to me) that this author was proposing putting the transistors on one chip and the wires on another. This could result in better power performance just because you could more easily control the capacitive and inductive effects of the wires. (though, I don't know how much they account for power draw).

 

[Tuna-Fish]

Here's another thought, since things have shrunk so much anyway, why not simply use more space? What gamer would mind getting top level performance for mid range price, the only trade off being your case has to be a foot taller and 6 inches wider? lol.

The big deal about shrinking isn't space, it's cost. On a 32nm process, you get roughly twice the transistors you'd get at 45nm at roughly the same cost, allowing you to sell more power for the same price. You *can* add more cores by just adding more chips, but your costs don't come down.

 

[Smartazz]

The big deal about shrinking isn't space, it's cost. On a 32nm process, you get roughly twice the transistors you'd get at 45nm at roughly the same cost, allowing you to sell more power for the same price. You *can* add more cores by just adding more chips, but your costs don't come down.

Costs would likely come down a little as the process technology really matured and yields increased.

 

[PCGamer3000]

Quantium Computing..

We go on a new technology all together and get off X86. and X64

There have been quantum computers that can factor 3x5 into 15 right now, which is amazing enough that humans can pull this off, but current predictions are that they will not be able to do super computing tasks like rendering a 3D game in real time of the universe the whole earth a human eye resolution. They will be able to solve certain problems faster like factoring large numbers for security codes or getting a greater understanding of quantum mechanics. Possibly helping to solve questions like what is the theory of everything.

i recall reading how in the future they will use living cells, bio technology type stuff, because OP is right, we will reach a limit with semiconductors. With living cells they can store soooo much more data than semiconductors & they can work much faster... but the biotechnology is still in its infancy. It was a really interesting show on the discovery channel i believe.

I have not seen this one. This sounds like it would present some unique maintenance challenges for the consumer like a lot of water cooling solutions involving complete water loops where you have to change the water every 6 months.

Sounds like it would be like going to the pet store and buying a gold fish only it would be a lot more expensive. They would have to solve the problems of how to feed the organic processor, where does the waste go(filtration like in the gold fish tank?), and what if it dies on you(do you open up the case and pull it out and replace it or take it to some kind of PC vet?).

Some micro biologists have figured out how to isolate and stimulate certain neurons from lab mice to produce a predicted reaction. Maybe in the future these reactions could be translated into 1s and 0s and you would essentially have a small brain on your motherboard that would act as the processor.

 

[aphelion02]

There have been quantum computers that can factor 3x5 into 15 right now, which is amazing enough that humans can pull this off

Why is that amazing? The way we do 3x5 in our heads is either by means of a LUT or by breaking it down into 5+5+5. Its not a one step process. This isn't that different how current computers do it.

And then again, there are some people who can't do this .

 

[Matt1970]

ah, so any game that's programmed to use as many cores as exist will benefit?

Here's another thought, since things have shrunk so much anyway, why not simply use more space? What gamer would mind getting top level performance for mid range price, the only trade off being your case has to be a foot taller and 6 inches wider? lol.

That's fine for most PC and Laptop users. It's when you start jamming them into smart phones and ipads that space becomes an issue.