3D-Printed Bare-Die Liquid Chip Coolers Smash Barriers, up to 3.5X Improvement
A series of 3D printed processor coolers was one of the most interesting presentations at the conference ITF World, hosted by chip research giant imec in Antwerp, Belgium. These prototype water blocks boost the cooling capacity of high-density processors such as CPUs and GPUs by up to 3.5x over the kind of solutions found in today’s best CPU coolers. This enables higher power densities and unlocks untapped performance in modern chips. The results of this research could lead to radically new water cooling systems for chips of all kinds.
Bare die cooling for forced liquid cooling directly With thermal processing on the processor die emerging as one of the most obvious advances to address the excess heat generated by new chips, imec has developed a new We are leading in technology. Its importance is increasing with each new generation of chips, as power consumption will skyrocket as smaller nodes mean smaller scale power reductions. In addition, smaller transistors increase power density, complicate cooling tasks, and ultimately limit chip performance.
The ultimate goal of chip designers is to be able to do more work in less space. Yet today’s chips are already power-limited, and while the chips operate within certain TDP and temperature limits, the “dark silicon” realm is turned off. This means most chips use only a fraction of that potential during normal operation. Moreover, this problem only gets worse with each generation of chips. Modern CPUs like AMD’s Epyc Genoa are already topping out at 400W, and the roadmap shows: 600W server chip in the future.
In contrast to the standard water cooling approach of cooling the processor using an internal water block with a cold plate combined with a chip heat spreader, the prototype 3D printed cooler pictured in the album below Press the liquid directly onto the die of the processor. Boosts cooling capacity by pumping coolant directly onto the surface of the processor.
3D printed water blocks enable rapid prototyping and imec uses different types of standard polymers used in 3D printing to ensure that the water blocks can cope with temperature loads. . Not sure if these designs can be printed on the best 3D printers.
The 3D printed waterblock can be customized in several different ways with a custom nozzle array (you can see it in the image). The liquid is sprayed directly onto target areas of the chip surface, such as individual cores or just above areas that generate high heat. Increases the cooling capacity of chips used for vector operations.
The water block has also been custom-fitted to take up as little space as possible and now uses o-rings to prevent liquids from entering around the water block. Naturally, imec is experimenting with several different types of sealing mechanisms and different types of his 3D printing materials for the blocks.
Almost any dielectric liquid can be used with these chillers, including process water and refrigerants. Of course, even if the liquid is not conductive, bare die liquid cooling requires sealing the area around the chip, such as the capacitors and other electronics on the PCB. However, there is no sealant at all on the top of the die to keep the coolant as close to the chip as possible. The researchers pumped the liquid directly onto the smooth die surface, but other approaches, such as adding striations to the top of the die, can further improve cooling performance.
Sealants pose long-term reliability challenges due to rapid thermal cycling and interaction with various coolants used in the system. Nevertheless, imec works methodically to find the right combination of all materials to ensure long-term reliability.
The album above contains several slides outlining the researchers’ experiments. In general, cooling more than 100W of power per square centimeter has proven to be very problematic, and distributing his 1W of power over 1 square millimeter of silicon allows for effective cooling. It follows a general rule of thumb. However, power densities skyrocket as process nodes get smaller, so increasing the ability to remove heat from the higher power concentrations is paramount to continued performance gains.
Keep in mind that more power often means more performance for the chip (which can be less efficient, so be careful). Imec researchers say they can cool 1,000W per square centimeter (100W per mm^2), or up to 500W per mm^2, but such cooling performance is not typical. Performance suffers because it doesn’t scale well across the chip.
In typical applications, these chip coolers can deliver up to 350 W per square centimeter, or about 3.5 W per mm^2. This is 3.5 times more than what is commonly seen today. As shown in the album above, this allows chip designers to push performance limits in a relatively conservative manner than single- and dual-phase cooling solutions that require scaling beyond 4W per mm^2. You can keep pushing the .
Of course, this is a simplification of how these cooling solutions work, and to properly measure the various benefits of this approach, other factors such as temperature delta and other factors must be considered. Many variables are required. But one thing is certain: this approach is one of the easiest ways to increase cooling at a modest cost increase. Other technologies, such as TSMC’s research that proposes injecting coolant through microchannels inside the chip itself, are clearly much more exotic and thus more costly and farther into the future.
Imec’s efforts are still in the research stage, with researchers working to identify the right materials, fluids and designs that will enable the creation of mass-produced cooling solutions. Early products from this study could take up to five years to filter. market.