Do cores sit on one blank layer of metal and insulation?

[MadRat]

Layers of a core seem to number between six to nine layers this past couple of years. AMD recently added a metal layer that significantly cooled down the core of the Tbred. I wonder if the extra layer is for spreading the heat along the layer closest to the silicon, or if the layer is on the top. I would imagine that having one blank layer to start the process would help cool things off significantly since the silicon is what makes contact with the heatsink. Does anyone know what is the case, if a first blank layer would help or not?

 

[pm]

The new metal layer doesn't cool down the part. Where did you see that?

 

[MadRat]

I didn't. I deducted it from the thermal properties of the metal. Even a single layer of aluminum or copper atoms would conduct heat quicker than going straight to the silicon. Its been around a year since I saw infrared pictures on the interent of microfilaments laid upon an insulator and the heat still tried to follow the metal pathway, even when the size of the pathway was incredibly restricted. I'd think it lends proof that Nature does try to move heat from hot to cold using the path of least resistance.

Its probable that eventually they will form an inside-to-outside pattern that will be etched using disimilar dielectrics applied to the base layer in order to create an electrothermal cooling effect. Moving the heat from the interior of the core to the perimeter would improve thermal disipation. Active cooling at the microscopic level would be a logical "eventual" step in cooling cores. Dielectrics are so much simpler to design than engineering phyical channels for an active coolant.

I realize that active cooling is used already in some processors for the military, but these are relatively crude. The next step in consumer processors is BBUL technology that further extends the life of passive cooling. But passive cooling will only take the processor so far. Eventually it will need some type of active cooling.

I'm just wondering if the metal layers used now are not already specifically used to cool the core by acting as a passive heatsink.

 

[Eskimo]

No, a metal layer like you describe would short all of the gates together. I suppose you could most easily categorize the added metal layers as being added on top. As in farthest from the silicon substrate. More metal layers simply allows for easier and more efficient routing. Think of it as adding additional overpasses at a busy intersection.

 

[boran]

well, I aint hightechnical eh, but I came up with a tought of integrating the CPU cooler into the core, maybe with heatpipies even into the centre of the core, I doubt this is managable because it would require a rebuild of the factories and chipmakes would immediatly become heatsink makers, and offcourse the overclockers will not like it because a manufacturer has control over the cooler.
the advantage is that the manufacturer chooses the cooler and is sure that the thing is stronge enough for the processor at the current speed.

another idea is something like the P4 cap, but where the cap is actually a mini heatpipe (maybe with some extrusions to go into the heatsink itself, this would also be a whole job, but is probably easier to implement.

 

[MadRat]

Why would a solid layer short the gates? I'd think it would be shielding if anything.

 

[Carrot44]

A Very crude example of why it would not work.
Take your Computer MB lay a sheet of foil over the back side bolt it down into the case and turn it on.

Ken

 

[rimshaker]

Originally posted by: MadRat
Why would a solid layer short the gates? I'd think it would be shielding if anything.

Sounds like you're thinking of a layer as a solid sheet of metal, but that's not the case. It's simply an extra layer of signal interconnects, like a huge spider web design except at right angles, not a solid sheet. And, please correct me if i'm wrong, but they can add the extra metal layer at any point, not just at the very top or bottom.

 

[MadRat]

So what you are saying is that the layered approach to core design does not include complete layers of insulation.

 

[Eskimo]

The ILD (inter level dielectric) layers are fairly complete across the die except where metal vias are defined which provide the paths from one metal layer to the next.

Picture an office building where the hallways allow people to walk quickly from one loaction to another without bumping into eachother. This is like the metal layers in a integrated circuit. But there are still walls, floors and ceilings that surround the hallways. These would be your insulation. And the elevators that carry people from one floor to another are like the metal vias. Ideally we'd like to design as wide of hallways as possible to fit more people. But we are trying to shrink the overall dimesions of the entire building which forces us to pack the hallways closer and closer together. There are limits to how well we can do this and sometimes we have little choice but to add more floors to our building to accomodate all the traffic.

 

[zephyrprime]

The metal layers aren't for spreading heat. I would think that they actually impede heat transfer since they add a layer of extra material. The metal layers are layers of wiring that connects the transitors together on the chip to form circuits. The reason the extra layer resulted in cooler running is because it allowed for shorter wire lengths. Since the wires on a chip have such small cross sections, they provide significant electrical resistance so shortening the wires reduces heat production and reduces power consumption and increases signal strength.

 

[pm]

Besides what everyone has already said, modern package types are flip-chip. The heatsink is touching the backside/bottom of the chip. The metal layers are on the far side where the heat transfer is less efficiently transferred through the package, or through the FC underfill.

The reason that AMD added a new metal layer to their Thoroughbred core wasn't to lower the die temperature. In fact, I'd be surprised if the core temperature dropped at all with the addition.

 

[MadRat]

I find it hard to believe that a single layer of metal atoms at the least wouldn't significantly help spread the flow of heat across the surface of the chip. The properties of most metals have several times the ability of silicon to transfer the energy of excited orbital electrons. Spreading them out evenly across the very layer that has to absorb the heat makes alot of sense to me. When it gets down to it we are cooling off the core via an insulator (silicon) contact point even when it comes to flipchip designs.

 

[Peacekeeper100]

The only problem is in current chips (non-SOI) the substrate connects the bottoms of all the gates together. If you place this layer of metal to close to the gate substrates it would change the number of electrons and holes causing the device to fail. If I am correct they to place a metal layer far from the gates now, at least Intel does with their heat spreaders.

The think you have to remember about what AMD did is the metal layer is being used. This will cause more heat because of the resistance in the metal to current flow. In theory if you only use the metal layer to spread heat, and not actually use it for circuit purposes it might keep it cooler, but the cost of adding these steps to the process outway the need to cool the chip by far.

 

[MadRat]

<<In theory if you only use the metal layer to spread heat, and not actually use it for circuit purposes it might keep it cooler, but the cost of adding these steps to the process outway the need to cool the chip by far.>>

I'm not sure that is a valid argument. If that was the case then many improvements would never have been used.

Seriously, though, I think there has to better ways to cool the core without major redesigns of the current technology. Perhaps it would be as simple as doping the backside of the core with some type of liquid penetrant. I'd of thought that a simple insulated layer of metal underneath the core would be fairly effective. I still think the idea would be worth pursuing, but don't have $12 million to blow on the project.