Reinventing a High-Performance CPU Cooler with Additive Manufacturing

A Disruptive Innovation of Thermal Management

High heat loads limit the miniaturization of portable computers, power electronic devices and high-power LED lighting. Most ambitious technological solutions from the lab are not ready for mass production and deployment in consumer products. But industrial 3D printing, or so-called additive manufacturing, can bridge the gap for thermal management components and keep lossy electronics cool even when the available space is severely limited.

The freedom of design provided by 3D printed thermal management components offers the same or superior effectiveness as conventionally manufactured components, but requires much less space. Enlarged surfaces, any-shape geometries and conformal cooling channels are among the opportunities of this manufacturing technology.

The efficiency gains that can be achieved were demonstrated with a gaming CPU cooler design for additive manufacturing. To maintain the same chip temperature, the new part requires 81% less space and 93% less weight than the best-in-class conventional cooler.

The Future of Thermal Management for Electronics

Electronic components are developing in two basic directions: Performance is increasing and size is decreasing.

Everything revolves around one central problem: heat [1]. This is true at microscopic scales, e.g. where noise on a quantum level limits how small the structures of processing units can become [2,3]. But it also manifests at macroscopic scales, where the amount of heat that can be transported away from high-performing components limits the performance per volume.

As a fundamental consequence of non-ideal systems, losses and hence heat cannot be avoided. But they could be reduced in the future. Innovative technologies that perform tasks much more efficiently than before can circumvent the excess heat limitations. For example, high hopes have been placed in the transition from electronics to photonics. Even now, photons rather than electrons are the information carriers used by the backbones of modern communication [5]. In the future, photonics might also advance into microprocessors and improve the performance per volume of computers [6]. New algorithms for quantum simulators and one day possibly even quantum computers can solve certain classes of problems more efficiently than classical, purely binary computers [7].

But these approaches are all decades away from being deployed to consumer products. The question at hand is how to radically improve thermal solutions for existing electronic devices and push the boundaries of miniaturization.

A solution would serve a huge market demand.

  • Laptops and mobile phones are becoming thinner and more powerful.
  • Communication infrastructure is a data bottleneck and needs to support ever higher transfer rates, while at the same time a landscape of bulky transmission towers must be avoided.
  • Electric cars are full of high power electronics that need to be integrated into a limited space.

In each of these markets, new thermal solutions need to be developed. Significant technical progress is required within the next few years.

The Challenges of Miniaturization: Paper-Thin Laptops

To illustrate the miniaturization trend, the historical development of portable computers is a good example. Their steady increase in performance and decrease in size has been particularly challenging. Portable computers have come a long way since the 1980s. The miniaturization of sub-systems and components was a major aspect of their hardware development.

Rapid-3D + EOS - High Performance CPU Cooler with Additive Manufacturing - the challenges of miniaturisation: paper thin laptops
The CPU clock frequency is a readily available measure of performance, and there is plenty of historical data for computers. The clock frequency has been stagnating since about 2004; instead, the number of cores has been growing exponentially [8]. The stagnation of the clock frequency has several reasons, the most prominent being excess heat creation. Higher frequencies require higher voltages, but the excess heat associated with them cannot be fully compensated by cooling technology improvements. In a nutshell, the limitations in heat transfer restrict the maximum clock frequency and hence the single- thread performance of the processor*.

We analyzed the situation for laptops as an example of powerful computers with limited available space. To measure the technological advancement of miniaturization, we normalized the CPU clock frequency with respect to weight. The resulting historical analysis shows that miniaturization of laptops is slowing down.

The data points in figure 1 are randomly picked laptop types with different display sizes from various manufacturers [10]. The chart shows how the exponential CPU clock frequency growth ended in around 2004.

Laptops are just one example, but they convincingly illustrate the importance of efficient cooling in small spaces and demonstrate how additively manufactured thermal management components can revitalize the miniaturization of electronic devices.

Small and Light WeightHeat Exchangers Come at High Cost

For the study, we focused on CPU coolers directly mounted on the chip via a heat spreader.

If we look at the solutions available today, we notice two main types of CPU coolers. One is optimized for manufacturability, the other for performance. The most economical way to mass manufacture chip coolers is by extruding aluminum, but the performance of products manufactured by this technique is limited due to its design restraints (figure 2a).

For optimum cooling performance, complex assembly parts are required. These are composed of sub-components, for example milled components (figure 2b).

In a housing with extremely limited space for the heat exchanger, neither of the two commonly used types is a good choice.Further miniaturization of the best-performing coolers would be costly, because it involves further production steps and longer machining times. This represents a restraint on the further miniaturization of computers and other electronic devices.

The goal of this study was to break the miniaturization limit and innovate a CPU cooler that matches the best-in-class products while requiring much less space and material. By designing for additive manufacturing from the start, the typical restraints of conventional production methods were overcome.

Additive Manufacturing Is Ideal for Miniaturized Heat Exchangers

The economics of complex parts made with additive manufacturing by laser sintering are exactly the opposite of the economics of subtractive manufacturing (i.e. most conventional manufacturing technologies). Roughly speaking, the less material the laser has to melt, the less machining time is required. Delicate and complex parts are therefore faster and cheaper to produce than massive and simple parts (figure 3).

Additive manufacturing thus opens a door to a new world of thermal management solutions. Oldlimitationsareobsoleteandnewideascan flourish. For thermal solutions, the rewards include increased surface areas, complex shaped internal channels for optimized coolant flow and functional integration. Miniaturization and increased heat transfer performance are the goals.

Delicate and complex components are where additive manufacturing shines, whereas massive parts with simple geometries are better suited for traditional subtractive manufacturing methods.
Innovating a High- Performance Gaming CPU Cooler

The process of reinventing a CPU cooler was headed by AM Metals, a company specializing in application development for additive manufacturing. The starting point was the best-in-class CPU cooler currently on the market for gaming computers. It meets extreme heat transfer requirements to enable the common practice of overclocking.

As part of the project, new ideas for innovative cooling flow geometries were generated and the heat transfer path was optimized. As a final design result, the area directly above the bottom plate was equipped with an arrangement of specially designed pins. The pin arrangement was adapted to the specific thermal footprint of the CPU.

The goal was to reduce its size and weight while maintaining best-in-class cooling performance.

Comparison with Best- In-Class CPU Cooler

To predict and compare the performance of the 3D printed CPU cooler to the benchmark CPU cooler, a highly accurate thermal simulation was performed using CFD by TheSys, a company specializing in the development of new heat exchangers, thermal simulation of performance and validation testing of heat exchangers and thermal systems. For several years, they have focused in particular on the development of additively manufactured heat exchangers.

The thermal performance simulation considered constant boundary conditions of 1 l/min flow rate and 60°C coolant inlet temperature.* The heat generation of 120 W by the CPU was modelled as a constant heat flow density in the contact area.

The benchmark cooler was simulated with the assumption of a pure copper body. In reality, only some parts of the cooler were made of pure copper; other materials were used as well. Considering that copper has one of the highest thermal conductivities the performance prediction of the benchmark cooler is optimistic.

The first design of the 3D printed cooler was realized within 8 working days. After evaluating the simulation results, TheSys implemented two key improvements into the design. The pin geometries above the base plate were optimized and the wall geometry was adapted. Both measures adjusted the heat distribution and ultimately achieved the targeted low surface temperature at the contact area of the CPU, while also reducing the pressure drop. After just one design iteration, the surface temperature distribution, cooling performance and pressure drop matched those of the benchmark cooler closely.

For the benchmark cooler, there is an impact flow with a stagnation point at the CPU center line area. Outside of the CPU footprint, the flow velocities are constant, inducing a pressure drop in areas without heat input.

For the 3D printed cooler, the local coolant flow velocities are adapted to the local heat flow density. As a result, there is an optimized impact flow with a stagnation point at the CPU center line area and higher velocities beneath, balancing the high heat flow density in this particular area. The velocities are reduced to less than 50 % outside of the CPU contact area. A small vortex still forms in the outlet, meaning that there remains potential for some further design optimizations.

Coolant flow at CPU cooler central plane
Figure 5a: Simulation of velocities for benchmark CPU cooler in central plane
Figure 5b: Simulation of velocities for 3D printed CPU cooler in central plane

Coolant flow at CPU cooler horizontal plane

Figure 6a: Simulation of velocities for benchmark CPU cooler in horizontal plane directly above the CPU
Figure 6b: Simulation of velocities for Temperatures at CPU surface across the base plate

Figure 7a: Simulation of surface temperatures for benchmark CPU cooler in horizontal plane directly above the CPU; CPU footprint marked as a rectangle.
Figure 7b: Simulation of surface temperatures for 3D printed CPU cooler in horizontal plane directly above the CPU; CPU footprint is approximately the size of the whole base plate

3D printed CPU cooler in horizontal plane directly above the CPU

Figure 8: Innovated gaming CPU cooler optimized for 3D printing, manufactured on a standard EOS M 290 machine in pure copper from Elementum 3DCPU Cooler Size Decreased by 81% with 3D Printing

Comparing the surface temperatures in the contact area of the CPU shows that the minimum temperatures of both designs are similar. The internal pins in the 3D printed cooler were arranged to optimize heat rejection at the position of the main heat source.

The pressure drop increased slightly by 5.5 % from 4.5 mbar (benchmark cooler) to 4.75 mbar (3D printed cooler).

  • With a single design iteration achieving performance equivalent to that of the benchmark 3D printed CPU cooler, the benefits are clearly visible:
  • The size of the cooler was reduced by 81% from 58 cm3 (benchmark cooler) to 10.8 cm3 (3D printed cooler).
  • The weight of the cooler was reduced by 75 % from 450 g (benchmark cooler) to 107 g (3D printed cooler made of copper).
  • The weight was even reduced by 93% to 32 g when using aluminum for the 3D printed cooler and accepting a slight in- crease of the surface temperature by 0.4°C.

Key advantages

Innovative cooler for gaming CPUs made by 3D printing.

Results for both designs

Table 1: Comparison of the technical performance between the benchmark CPU cooler and the 3D printed CPU cooler built in copper and aluminium, respectively. *The volume without legs to mount the cooler on the chip has been compared. **Weight including legsThe innovations presented above can be imme- diately translated into commercial products. Given a known, inhomogeneous heat flow density in the CPU contact area, 3D printed designs can be specifically tailored to the heat distribution. This is impossible in practice for conventionally designed coolers because of the ramifications for the production chain, but tailoring to specific CPUs or other chips is easy to accomplish with additive manufacturing. The changes in the production chain are purely digital. Customized parts can be released into production without human interaction or change-over costs.

 

New Design Freedom Paves the Way for Radical Innovations

In just a few days development time, the cooling performance of the best-in-class conventional CPU cooler was equaled with a 3D printed cooler that only requires a fraction of the space. The key was to combine the expertise of TheSys in thermal solutions with the additive manufacturing experience of AM Metals and EOS.

Eliminating the inhibiting effects of conventional manufacturing opens a door to further miniaturization and radical new concepts
like biomimetics.

Once the creative power of thermal engineers has been unleashed by overcoming the dogma of conventional manufacturing, radical innovations will be just moments away.

For example, we are surrounded by designs optimized over millions of years of evolution that can be translated into ingenious technical designs for thermal management. Thanks to 3D printing, we are now capable of manu- facturing these geometries. Figure 9 shows a sea anemone that evolved to maximize its surface area to filter out nutrients from passing water. Perhaps this structure would be ideal for a low-flow heat exchanger?

In fact, there are countless applications where heat transfer space comes at a premium.

Just consider gaming laptops, HP LEDs, lasers, autonomous driving, power electronics [4], chemical micro reactors. We believe that additive manufacturing can bridge the gap between current market demands for miniaturized thermal management solutions and future technologies that might circumvent the problem in a fundamentally new way.

Figure 9: Biomimetic heat exchanger design inspired by sea anemonesSources

[1] Schelling, P.K., Shi, L., Goodson, K.E., 2005. Managing heat for electronics. Materials Today. doi:10.1016/ S1369-7021(05)70935-4
[2] Cho, J., Goodson, K.E., 2015. Thermal transport: Cool electronics. Nature Materials 14, 136–137. doi:10.1038/ nmat4194
[3] Tsutsui, M., Taniguchi, M., 2012. Single molecule electronics and devices. Sensors (Switzerland). doi:10.3390/s120607259
[4] Kang, S.S., 2011. Advanced Cooling for Power Electronics, in: 2012 7th International Conference on Integrated Power Electronics Systems (CIPS). pp. 8 pp.-8 pp. doi:10.1074/jbc.M509557200
[5] Yariv, A., 2007. Optical Electronics in Modern Communications, 6th ed, New York: Oxford University Press. Oxford University Press, Inc, New York. doi:0195106261
[6] Sun, C., Wade, M.T., … Stojanovi, V.M., 2015. Single-chip microprocessor that communicates directly using light. Nature 528, 534–538. doi:10.1038/nature16454
[7] Ekert, A., Jozsa, R., 1996. Quantum computation and Shor’s factoring algorithm. Reviews of Modern Physics 68, 733–753. doi:10.1103/RevModPhys.68.733
[8] Karl Rupp, 42 Years of Microprocessor Trend Data, 15. Feb. 2018 (https://www.karlrupp.net/2018/02/42-years- of-microprocessor-trend-data/)
[9] Retail prices taken from ALTERNATE GmbH, Germany (as of 2019 April 25th), EAN 4251312600075and EAN 0872767007727
[10]
1980: Portal R2E CCMC, portable microcomputer with Intel 8085 processor, 8-bit, clocked at 2 MHz. Technical data from musée de l’informatique (France) blog.
1981: Osborne 1. Technical data from PCWorld, dating 31.05.2010, A History of Portable Computing, by Lincoln Spector.
1982: Athena I. Technical data from Russian Vintage Laptop Museum. http://vintage-laptops.com
1983: Gavilan SC Computer. Technical data from Russian Vintage Laptop Museum.
1984: Data General One. Technical data from Russian Vintage Laptop Museum.
1985: Toshiba T1100. Technical data from Toshiba T1100 Product Information Sheet and Toshiba Science Museum (for the weight and year of release).
1986: Toshiba T3100. Technical data from Toshiba T3100 Product Information Sheet and Toshiba Science Museum (for the year of release).
1987: Amstrad PPC 640. Technical data from Russian Vintage Laptop Museum.
1988: GRiD Case 1520. Technical data from Russian Vintage Laptop Museum.
1989: Apple Macintosh Portable Model M5120. Technical data from Russian Vintage Laptop Museum.
1990: Dell 316LT. Technical data from Russian Vintage Laptop Museum.
1991: Goldstar GS520. Technical data from Russian Vintage Laptop Museum.
1992: GRiD model 1660. Technical data from Russian Vintage Laptop Museum.
1993: Apple Powerbook 145B. Technical data from Russian Vintage Laptop Museum.
1994: Apple PowerBook 520. Technical data from Glen Sanford, website apple-history.com/520.
1995: Apple PowerBook 190. Technical data from Glen Sanford, website apple-history.com/190.
1996: Toshiba INFINIA 7200. Technical data from Toshiba INFINIA 7200 Product Specifications Sheet.
1997: Compaq Armada 7730MT. Technical data from Russian Vintage Laptop Museum.
1998: Apple Macintosh Powerbook G3 (Wallstreet Series II “PDQ”) . Technical data from Russian Vintage Laptop Museum.
1999: RAYTHEON CHAMELEON. Technical data from Russian Vintage Laptop Museum.
2000: Fujitsu-Siemens Lifebook B-2131. Technical data from CHIP magazine review article on this product, dating 27.06.2000.
2001: Apple iBook. Technical data from CHIP magazine review article on this product, dating 28.05.2001.
2002: Sony Vaio PCG-SRX41P. Technical data from CHIP magazine review article on this product, dating 01.08.2002.
2003: Toshiba Satellite 2430-201 (15″). Technical data from CHIP magazine review article on this product, dating 28.04.2003.
2004: Targa Visionary 811. Technical data from CHIP magazine review article on this product, dating 04.06.2004.
2005: Fujitsu Siemens Amilo A7645. Technical data from CHIP magazine review article on this product, dating 01.08.2005.
2006: HP nx9420 (17″ display). Technical data from CHIP magazine review article on this product, dating 01.08.2006.
2007: Acer Aspire 5685WLMi. Technical data from CHIP magazine review article on this product, dating 26.06.2007.
2008: Sony Vaio VGN-FW21ER. Technical data from Russian Vintage Laptop Museum.
2009: Apple MacBook Pro (2009). Technical data from Glen Sanford, website apple-history.com/mbp_15_ mid_09.
2010: Acer Aspire 7740G-334G50Mn (17″). Technical data from CHIP magazine review article on this product, dating 29.06.2010.
2011: Apple MacBook Pro (2011). Technical data from Glen Sanford, website apple-history.com/mbp_15_ late_11.
2012: Samsung Serie 9 900X4C (15″). Technical data from CHIP magazine review article on this product, dating 29.06.2012.
2013: Toshiba Satellite U840W-10F (14″). Technical data from CHIP magazine review article on this product, dating 01.07.2013.
2014: Lenovo G505s (20255) (15+”). Technical data from CHIP magazine review article on this product, dating 30.07.2014.
2015: Dell Inspiron 15 (7548). Technical data from laptopmedia.com review article on this product, dating 27.02.2015.
2016: ASUS X555. Technical data from laptopmedia.com review article on this product, dating 27.05.2016.
2017: Dell Inspiron 17 5767. Technical data from laptopmedia.com review article on this product, dating 17.06.2017.
2018: Dell Inspiron 15 3567. Technical data from laptopmedia.com review article on this product, dating 15.03.2018.
2019: Lenovo IdeaPad 330s. Technical data from laptopmedia.com review article on this product, dating 08.01.2019.

Dr. Matthias Hoeh
Senior Business Development Manager
Matthias is scouting applications and developing business models for industrial 3D printing. One of his focus topics is thermal management innovations. He is responsible for developing holistic solutions to realize the business ideas of EOS’ customers.

In his early career Matthias was trained as a physicist and spent four years as a researcher at Heidelberg University, Germany. He then joined Konica Minolta as a product manager, responsible for in-line test systems in electronics manufacturing. Since more than 15 years he is also active as an entrepreneur in various fields. In 2016 Matthias entered the world of industrial 3D printing and joined EOS.

Contact:
matthias.hoeh@eos.info

Dr. Peter Ambros
TheSys GmbH www.thesys-engineering.de
Contact:
p.ambros@thesys-engineering.de

Dr. Florian Wendt
AM Metals GmbH www.am-metals.de

Contact:
florian.wendt@am-metals.de

Share

Leave A Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.