Data centers and the chips that power them form the backbone of our society. Semiconductors play an essential role in tackling global challenges from climate modeling to drug discovery. For decades, chipmakers relied on making transistors in integrated circuits smaller and packing more of them together to achieve more powerful and efficient chips. Approximately every two years, the number of transistors per chip doubles, which is the driving force behind the increase in computing power we have taken for granted over the last decades. This trend was first anticipated by Gordon Moore in 1965, co-founder of Intel, and hence is known as Moore’s Law. In a way, this trend has been a self-fulfilling prophecy, as industry used this aggressive exponential growth model as a roadmap for product development. 

 

About 10 years after Gordon Moore’s observation, Robert H. Dennard at IBM described a physical foundation behind this rapid scaling of semiconductors, which relates to heat: Whenever transistors become smaller, they also become more energy-efficient, so when we place more of these smaller transistors on a chip, we end up with identical power consumption. And this is essential because all power that goes into these billions of transistors is turned into heat. This heat needs to be extracted, as overheating causes chips to fail and lose performance. Dennard observed that the amount of heat generated in a chip stays constant, regardless of how small we make our transistors and how many we place together. This observation, called Dennard scaling, reassured industry that heat would not become a limiting factor. Spoiler: this didn’t hold. 

 

Enter the mid-2000’s: Transistors approached several tens of nanometers in size, which opened up a whole new spectrum of physical effects that were not accounted for in the time of Dennard and Moore’s observations. Because of these new effects, Dennard scaling was no longer valid, and every new chip started to generate more concentrated heat than its predecessors. Right now, high-performance chips turn 100 Watts of power into heat, within an area as small as one square centimeter. This number is referred to as heat flux: 100 W/cm2. This value is extremely high: An electric kettle used to boil has a heat flux of about 10 W/cm2, ten times lower. Incremental improvements to cooling design are now lagging behind. Extracting this highly concentrated heat is rapidly becoming a bottleneck for the next generation of computing, holding back industries worth over 150 billion USD. Chips are now designed in such a way that only a small percentage can be simultaneously active to prevent overheating. In other words: We’ve spent billions to reach 5nm or 3nm node sizes but are not using it to its full potential.

​

​

​

​

 

 

 

 

 

 

 

The last decades have seen a proliferation in heat sink design. Bigger, more fans, vapor chambers, and heat pipes. However, none if these methods solve the problem at its core. Conventional cooling, from air cooling to liquid cooling with cold plates, are separated by multiple layers of inefficient thermal interface materials the chips they are cooling, which greatly limits heat transfer. This has proved sufficient for past generations of chips; however, with the power of GPU/ML chips increasing from 75W to 600W+, this is no longer a viable solution as the next generation of chips will continue to generate even greater levels of heat. 

 

In addition to the impact on performance, the cooling of chips accounts for about 30% of electricity consumption in data centers, causing an enormous environmental footprint. Therefore, sustainable and high-performance heat extraction is key to satisfying our ever-increasing demand for computational power. There’s a clear need for a disruptive cooling solution to enable the future of computing in a sustainable manner. 

 

Corintis addressed the problem of cooling right at the core. Instead of big heat sinks and multiple layers of packaging, we believe cooling should be an integral part of chip design. Corintis resolves this problem by enabling leading semiconductor companies to embed optimized microfluidic cooling channels directly into the silicon of their chips, removing the need for any thermal interface materials. This enables ten times more heat extraction than the best conventional cooling solutions available today.