The cooling will be done by flowing water between the layers of tiny chips
IBM scientists are developing a new technology for cooling computer chips arranged in a three-dimensional structure - component above component - by means of water flow between the different layers.
Dissipating the heat with a liquid is many times more efficient compared to cooling with air, which is used in ordinary chips - which opens the way to break through the traditional predictions as defined in the framework of Moore's Law, regarding the future development of computing power.
This development is particularly important since the aggregate heat output of a three-dimensional chip is expected to be extremely high: almost a whole kilowatt per cubic centimeter of chip - ten times that of any other device ever built by man. In fact, the energy production density (the amount of heat per given volume unit) in such a chip is higher than that of nuclear reactors.
Researchers at IBM's research laboratories in Zurich, in collaboration with the Fraunhofer Institute in Berlin, presented a new and particularly effective technique for cooling chips in a three-dimensional structure - with the help of water. The prototype shown now draws and injects tiny amounts of water - directly between the silicon layers. The new structure allows processing units and memory components to be placed one above the other, which until now could only be placed next to each other. Thus, it is possible to reduce the dimensions of the chip and improve its performance, thanks to the shortening of the physical distance between the processor and the memory, and thanks to the possibility of operating at higher speeds, without the heat generated with the increase in the clock rate endangering the chip or disrupting its performance.
About a year ago, IBM introduced innovations in the manufacturing processes of multi-layered chips with a three-dimensional structure, which make it possible to shorten the distance that the information is required to travel within the chip to one-thousandth of the distance typical for chips with a two-dimensional structure, and to multiply 100 times the number of data channels available in each chip.
The scientists at IBM are investigating different concepts and approaches for placing memory units on top of processors - in order to eventually reach chips that integrate a large number of processing cores on top of each other. The problem is that such complex building scenarios pose a real challenge in the field of cooling. Traditional cooling methods cannot be expanded and upgraded within the existing technology - and what is required is interlayer cooling. To date, no chip manufacturer has been able to present a working model of such a cooling system.
The IBM team flows water into the cooling system, which is made of tubes whose cross-section is similar to that of an average head hair - only 50 microns. These tubes are arranged between the layers of the chip, and efficiently divert the heat generated during the activity. The improved heat conduction properties of the water, allow to cool and remove heat of 180 watts per cm4 of the chip layer, which occupies a total of XNUMX cmXNUMX. The classic approach, in which only the outer side of the chip is cooled, obviously does not allow the construction of a three-dimensional structure, in which the circuits are arranged one above the other.
In order to prove the effectiveness of IBM's new technology, the researchers flowed water through an experimental structure with dimensions of 1x1 cm, which included one cooling layer standing between two heat sources. The cooling layer, 100 microns high, includes 10,000 tiny horizontal capillaries, through which the water passes. The researchers were able to overcome the technical problems of designing and cooling a completely sealed cooling structure, which would ensure an efficient, uniform and fast flow of water - and maintain complete sealing, in order not to risk leakage and short circuits in the electrical circuits. BMW compares the challenge of the complexity of this cooling system to that of the human brain: millions of nerves and neurons transmit signals between them - without interfering or coming into unwanted contact with tens of thousands of blood vessels that provide energy and regulate the heat level.
The creation of the different layers is carried out using production technologies known in the world of microelectronics - except for the addition of a unique process required to create tiny holes, or a bore that allows signals to pass between the different layers. The isolation of these channels was carried out by leaving a "silicon wall" around each inter-layer connection, and adding a layer of silicon oxide that isolates the electrical connection from water. The production process of this complex structure was carried out with a level of precision of 10 microns - ten times the level of precision required in the connection array of chips of the generation known today.
The assembly of the different layers on top of each other was carried out with a new technology developed especially for this purpose, when at the end of the entire process the three-layer processing set is placed inside a silicon cooling structure. The water flows to one side of the cooling structure, passes between the different layers, and is pumped out to the other side.
17 תגובות
My city has a water pumping system that also allows the city to be 3 times bigger without strengthening the pumping. It's not that expensive...
Suppose that tomorrow some friend decides that he has a nuclear bomb: with special suits we can escape from the affected area quite quickly (to avoid radiation), and with the flow of water in pipes inside the underground shelters we can cool the air so that we also survive the initial heat wave and the fires outside the shelter. Not only that, think about the possibilities of replacing existing fire extinguishing means inside tanks for example.
to open
The water will freeze just because it was in the trunk of your car on a freezing night, even without you turning it on.
On second thought, maybe make the computer with a microwave that will heat it below a certain temperature. It is advisable to use antifreeze in the water.
In general, before starting, check if there is no leakage.
Maybe also make a water tank like in cars that we fill before use.
And maybe... Will we still give up the water?
With a smile, and happy holidays!
Sabdarmish Yehuda
Matan: Nice! Well done! We all salute! It's just a shame that when you don't want to get involved, you get involved anyway. What is this - lack of finger control?
I don't want to interfere, but I just wanted to say that I have already thought about all the problems you raised here and I have solutions for all of them.
So, have a good week.
The responses here are "water grinding"...
Yehuda. Do you think someone will develop a computer that will be damaged at a temperature below zero degrees? What is this, a middle school final project?
Larry
At a temperature below zero, the ice formed from the water spreads by at least 10% relative to the liquid state. Therefore, if by chance the computer was not in operation, as happens from time to time, and the water would freeze and the component would end its function forever.
And by the way, do IBM experts imagine how difficult it will be to replace a broken component?, and to solder the components?.
I will let others experience the "wet" ingredients for several months, before I join such a "wet dream".
may we have a nice week
Sabdarmish Yehuda
Avi:
It is worthwhile to find a solution to the bookkeeping problem that Yehuda mentions.
Yehuda:
I was already thinking about the booker's problem when I drafted the initial response but note that the writer of the comments in question didn't leave me much of a choice.
Yehuda:
It is clear that I cannot answer on their behalf and not even accept offers for them.
My guess is that they are aware of all the issues including the possibility of the freeze.
At this point, Ari (which I also thought about but did not elaborate on) it is not correct to say that the problem is solved from the beginning because you also have to take into account the time that the device is turned off. I guess there are several options and maybe even several solutions that are being considered. Not many systems are required to be at temperatures below zero and it is possible to add a substance to the water that lowers the freezing point.
Anyway - air is what it is today and the whole project is aimed at improving the situation and not leaving it as it is. As mentioned, the water doesn't just improve, it improves many times over.
Hello to Mr. Yehuda Svardamish
Annie sees a problem with stagnation because the chip generates heat and there are no moving parts so stagnation is good for the chips.
We are not as smart as IBM scientists but we are able to come up with different thoughts on the subject.
For example, are IBM scientists aware that their system may not work or even break down in sub-zero temperatures?
Wouldn't it be better to use a liquid whose freezing temperature is several tens of degrees below zero?
I still think that the flow of air through the tubes should be preferred, less risks and complications of the system, although, I admit, it is also less effective.
But if air will be enough then why get involved?
Besides that, Michael, not everyone's comments are numbered, so maybe you could define them by name?, for example in place 5 and 6 say: to comment, that way everyone will understand you and won't have to start counting.
good week
Sabdarmish Yehuda
For 5 and 6:
That's why I wrote that this is a response to some of the questions and not to all of them.
I don't have data on the heat removal ability of all liquids and I didn't want to write (even though I'm sure it's true) that IBM scientists are smart enough to understand that there are other liquids as well and that if they chose water it must have been after in-depth considerations that take into account both their heat capacity and the extent of their expansion Under heat, both the speed of their flow in the capillaries, both their price, and their chemical properties (to ensure a minimal reaction with the pipes and much more.
.
Michael, did you read the question? The question referred to the type of liquid……
And not why liquid is used and not air.
.
.
But Michael - my question referred to the type of liquid...
And not why liquid is used and not air.
.
An answer to some of the questions can already be found in the second paragraph of the article:
"Dispersing the heat with a liquid is many times more effective compared to cooling with air, which uses normal chips"
Water in such dimensions behaves completely differently than the way we know it in the accepted everyday dimensions.
Back in the XNUMXth grade, we saw how water in a long, narrow test tube does not obey the usual "laws".
It is assumed that various additives have been added to the water that will improve its flow properties in such dimensions.
.
Why did they choose water to absorb the heat generated in the chip? Aren't there better fluids for heat transfer? Maybe oil for example or other liquids?
This method can also be good for two-dimensional chips, that is, if they have already succeeded in making the capillary tubes inside the chip, it is desirable to pass air through them. The cooling will be greater than the existing one and there will be no future danger of water leaking from the complicated external and internal tubing system to the chip, or of creating dangerous pressure from steam that may be created in the system as a result of the heating.
By the way, such a system, which includes smaller and smaller tubes, begins to take the form of three-dimensional fractals. interesting.
Sabdarmish Yehuda