Researchers Create 3-Dimensional Chips

Spy der Mann writes "Professor James Lu and other researchers of the Rensselaer Polytechnic Institute, managed to create three-dimensional chips (coral cache) to optimize the design of future processors and prevent overheating. "Make the interconnect wire shorter, and you cut the delay time," says Lu. "A simple way to make them shorter is to stack the transistors.""

Comments

Heat

[skraps]

Hopefully there will be a parallel advance in cooling technology.

3-d

[PunkOfLinux]

I think what they mean is that instead of the processor being on a single plane (a silicon wafer) it's on 2 or more wafers (stacked on top of each other or somesuch)

Huh...

[Peale]

I thought they'd been doing this all along.

Guess I was just ahead of my time...in my head.

Re:Huh...

[InvalidError]

7-9 is the number of routing (metallization) layers. The 32-40 figure is probably the number of masks required for the whole fabbing process, including substrate doping masks, insulation masks, metallization masks, etc.

In any case, doing "cubic" chips is not really going to be practical: volume increases faster than surface (heat transfer) area. If the power density increases faster than the transfer surface, the core will be even more likely to overheat unless the extra circuitry is low-power and can serve as a sort of heat-spreader like caches do in current CPUs.

Also, there is the matter of IO density, the core needs to be large enough to place all these IO and power pads.

Large, rectangular, thin chips provide plenty of heat spreading and IO bonding area thanks to large caches and all the transistors being on the same layer.

Also, going cubic (adding semiconducting layers) would add many extra masks, at least two or three per layer. With vertical transistor spanning three layers, using these would require between eight and 12 extra masks which in turn becomes more than 50 extra processing steps. This could substantially increase failure rates by multiplying the risk of one layer contaminating another, mask misallignment and other small process variations.

If adding layers was easy and cheap, AMD, Intel and the others would not go so far out of their way to fit their designs into the fewest layers possible. The same generally applies to PCBs.

How does that prevent overheating?

[Kjella]

Essentially, they say this packs it denser. And a cube vs a flat processor = less surface/transistor. I see only factors which makes this *harder* to cool. Maybe someone can explain...

Kjella

Simple?

[Detritus]

"A simple way to make them shorter is to stack the transistors."

There must be a new meaning of the word "simple" that I'm not familiar with.

Re:shorter wires = less resistance

[jumpingfred]

Most of the heat is disapated accross the transisors. Shorter wires may reduce the capacitance which would lower the amount of charge moving which would lower power.

Re:shorter wires = less resistance

[geekee]

"There is, sort of. If the wires are shorter, they have less resistance end-to-end assuming they have the same thickness, are made from the same material, etc etc. Less resistance means less heat (and maybe core voltage could be lowered slightly too, since there would be less of a voltage drop). However, I honestly don't know how much heat comes from the actual junctions versus circuit pathways."

I don't think people are worried about the heat dissipated in the actual wire. High resistance wires require you to use additional buffers to generate signals with acceptable rise/fall times due to rc charging effects. This costs more power.

Re:Hey...

[ezberry]

Technically you are congratulating him for doing what he is paid to do - no more. I mean, it's an interesting story, but I don't know if he deserves congratulations because he didn't chose to not green-light it.
Maybe the parent was being facetious, but I can't tell.

Re:Sorry that isn't covered in High School Physics

[ColaMan]

You need to look at the whole picture.
It's not really a case of "where the current is going" - the current flows through the entire circuit, from one side of your voltage source to the other. The important thing to remember is that the current never changes through the whole circuit. The number of electrons/second (amps) is constant through the whole circuit. Only the voltage drop matters as you traverse the circuit. The part of the circuit with the biggest voltage drop across it consumes the most amount of power.

So, you get a small voltage drop across your wires, which gets turned into a small amount of heat. You normally have a large voltage drop across your load, which gets turned into useful work.... plus a bit of heat- nothing's 100% efficient.

For example, in an electric motor, the bulk of it is converted to mechanical work... which is still measured in watts, and *that* eventually gets converted to heat (by friction somewhere). The remainder gets lost due to the resistance in the motor windings.