Originally posted by humyThe thread title confused me a bit till I realised it should be AMD.
http://phys.org/news/2014-06-amd-reveals-25x-efficiency-gains.html
But, such improvements will be dwarfed by those that will be gained when spintronics finally becomes of age for spintronics promises to improve energy efficiency of microchips by more like very roughly ~10,000 fold!
How would spintronics eliminate resistance in the processor circuitry? Or is it because nothing that is not actively doing logic does not draw power?
Originally posted by twhiteheadNo, CMOS already does that, it's just that spintronics fundamentally uses a thousand times less power to do a switching or logic function.
The thread title confused me a bit till I realised it should be AMD.
How would spintronics eliminate resistance in the processor circuitry? Or is it because nothing that is not actively doing logic does not draw power?
It may also be a lot faster than the 3 odd Ghz CPU's run at today.
If we could get them up to 1 Thz and use less power in the bargain, we would have supercomputers in a cell phone....
Originally posted by sonhouseCMOS is slow.
No, CMOS already does that, it's just that spintronics fundamentally uses a thousand times less power to do a switching or logic function.
The main power usage in a CPU, is from keeping state alive when not being used and from resistance in the wires connecting it all up. That is one reason why shrinking the die size results in significant power savings.
The main way that AMD and Intel save power is by turning off sections of the CPU while not in use, and by die shrink.
The problem for AMD is the same as the problem for Intel, which is they both use the x86 instruction set and all the instructions are different lengths, which means that they are forced to have a very complicated instruction decode stage. ARM processors already use of the order of 1 Watt (these are 32bit processors, I don't know what their 64 bit architectures is like). Their main difficulty is that although they can produce a low power architecture they are not in control of the actual process technology. That part is up to the chip manufacturers. At least these should come on line in some sane time-scale.
Originally posted by twhitehead
The thread title confused me a bit till I realised it should be AMD.
How would spintronics eliminate resistance in the processor circuitry? Or is it because nothing that is not actively doing logic does not draw power?
The thread title confused me a bit till I realized it should be AMD.
Oh yes, didn't notice that. I can see how that could be confusing.
How would spintronics eliminate resistance in the processor circuitry?
It doesn't do anything to reduce electrical resistance nor increase electrical conductivity of the materials that make up each wire but, because the signals are not represented by current in a spin circuit but by the spin bias (of electrons ) , there is only need for a minute amount of electric current to send down each wire to send each signal. Some tiny amount of current is still needed to send down the electrons with spin bias but only a minute proportion of that current that would have to be sent down that wire if it was the current, not the spin bias, that represents the signal. Because the required current is many times less in a spintronic wire, the energy losses through the electrical resistance would be many times less even if the material that makes up the wire, as like some of the materials being considered to be used for this in spintronic research, is not a particularly good electric conductor! This is why spintronics promises to make microchips many times more energy efficient.
As a couple of bonus, I have also heard that spintronics also promises to make the speed of the signals and of the circuits much greater and it also promises to massively improve computer memory with greater access speed and energy efficiency although.
But I must confess, although I understand why spintronics would help make circuits use less energy, I don't understand why spintronics would help with speed in particular -I have merely heard it should do.
Anyone: does any one here know and would explain here?
Originally posted by humyThanks for the explanation.
Because the required current is many times less in a spintronic wire, the energy losses through the electrical resistance would be many times less even if the material that makes up the wire, as like some of the materials being considered to be used for this in spintronic research, is not a particularly good electric conductor!
But 10,000 fold less energy? Surely you would then start experiencing problems with magnetic interference between wires, or even from outside the chip?
Originally posted by twhiteheadNot as far as I know because I have never heard this as a possible problem but don't know why it shouldn't be a problem although there are still some other technical problems still to solve before spintronics becomes practical including making good spin pumps and good spin transistors and then use that to achieve sustainable cascade logic. Fabrication may also have to be more complex.
Thanks for the explanation.
But 10,000 fold less energy? Surely you would then start experiencing problems with magnetic interference between wires, or even from outside the chip?
Originally posted by humyI know virtually nothing about spintronics, but I'm going to guess anyway. In conventional semi-conductors it takes a while for the state of each transistor to change, of the order of a pico-second. Flipping spins presumably takes less time, so the circuit is faster.The thread title confused me a bit till I realized it should be AMD.
Oh yes, didn't notice that. I can see how that could be confusing.
How would spintronics eliminate resistance in the processor circuitry?
It doesn't do anything to reduce electrical resistance nor increase electrical conductivity of the material ...[text shortened]... ave merely heard it should do.
[b]Anyone: does any one here know and would explain here?[/b]
Originally posted by DeepThoughtAnd uses a thousand times less energy in the process.
I know virtually nothing about spintronics, but I'm going to guess anyway. In conventional semi-conductors it takes a while for the state of each transistor to change, of the order of a pico-second. Flipping spins presumably takes less time, so the circuit is faster.