Microsoft is progressing with a new way to cool microchips that states that it can give rise to more energy-efficient data centers in the future. It is a method called microfluidics in which the liquid coolant flows directly into the silicone.
After lab tests, Microsoft found that this strategy today can remove three times better heat than cold plates currently used in data centers. The company announced this week that it is capable of developing a microfluidic cooling system for core services running a server for a fake Microsoft teams meeting.
If they can find the same success outside a laboratory, the microfluidics can cut the amount of energy required to cool the data center. This can also cause more powerful chips that the current cooling system will struggle to avoid overheating. But there are still a lot of factors that can affect how impressive this new technology is eventually in the real world.
This can give rise to more powerful chips that will struggle to keep the current cooling system overheating
Compared to your data centers, the next generation is being designed to train and run the new AI model, which is more powerful chips. Not only those GPUs use too much energy, they also become very hot. Keeping them cool is a challenge that not only affects performance, but also consumes more energy.
Typically, a data center can use fans to pass cold air on a chip. A more advanced technique that employs microsoft for high-power chips includes cold plates made of copper, which flows with fluids. Put that plate over a chip, and it removes heat.
With microfluidic cooling, liquid flows through the etched channels on the back side of a chip. Trick is ensuring that the channel is deep enough to prevent clogging, about the width of a human hair, but is not so deep that the chip becomes more likely to break. Microsoft says that it uses AI to find out that it is most efficiently cooled to direct coolant on a chip. Etched designs are also inspired by nature – mimicking the patterns of veins on the leaves, for example – which already show how practical they are in distributing water and resources. Using microfluidics, Microsoft documentation of a 65 percent decrease in the maximum temperature increase of the silicone of the GPU.
The advantage with microfluidics is that it brings fluid directly to the chip, eliminating the need for protective layers of the material between the chip and coolant when cold plates are used. Each layer, like a blanket, holds some heat, and so the coolant needs to be cool to work well inside the cold plates. The cool fluid flows into the plate; Warm fluid flows and needs to be cooled again. With microfluidics, the coolant does not require cooling as a low temperature, conserving energy.
Microfluidics may also allow the data center to handle the peaks more efficiently in demand. Calls of teams usually begin every hour or half an hour, Microsoft gives as an example. To handle those spikes in demand, they may have to install more servers to have sufficient ability on the hand, even if they will not be used all the time. The option will be to allow the existing server to work extra hard, which is called overclocking – but this can lead to overheating and damage to the chip. Microfluidic cooling, as it is more efficient, a chip may allow for more overclocking without melting.
In theory, if the server can work hard by now without melting the chips, a data center may not require many of them. And by reducing the risk of overheating, microfluidics can also allow for a more tightly packed server within the single data center. This can cut cost, literal and environmental, for the need for construction of additional features.
All these benefits can be important for the next generation microchips, which is expected to be so powerful that cold plates can be reduced. Microsoft says that microfluidics can also enable 3D chip architecture. 3D chips will be more powerful than today’s half-flat designs, but the heat is a stumble to do so. With microfluidics, however, coolant is likely to flow through the chip.
Efficiency can also be a two -edged sword
Microsoft does not have a timeline when all this may happen. After more laboratory testing, the challenge to find out how to allow the hardware and supply chain changes required to allow for microfluidics – for example, at which point the grooves will be dug into the chips in the manufacturing process? Fortunately, they can use a mixture of the same type of coolant, water and propose glycol, which is used in cold plates today.
Other researchers have also been studying microfluidics for years. For example, HP was awarded $ 3.25 million Grant To develop its own microfluidic cooling technology from the Department of Energy last year. Husm Alisa, director of Cloud Operations and System Technology in Microsoft, says, “It is good to see all these things, we are happy to see them, and where we can participate to move those things fast, which we are happy.”
Microsoft says that it “recently helps to pave the way for more efficient and durable next generation chips across the industry” blog post Avoid its progression on microfluidics. Energy efficiency is important if the company wants to operate more continuously. Like other technical companies, Microsoft’s planetary heating has increased in carbon emissions as it has bent into a generous AI. But efficiency can also be a two -edged sword. As something else becomes more efficient and inexpensive to use, people use it too much and it can eventually give rise to another large environmental footprint. It is an event called the Javon Paradox, which is also the CEO of Microsoft Satya Nadella has commented AI as a force to adopt more and more.
