Densely Packed Electronic Systems

P. Salmon
Electronic Innovations, allied with Nif/t,LLC,
United States

Keywords: water-cooled, power density, energy efficiency, thermal design, server, supercomputer, microelectronic module

Summary:

Electronic Innovations (https://electronicinnovations.tech) was formed in 2022 to bring to market new technology for densely packed electronic systems, especially water-cooled servers and supercomputers. This endeavor builds on over 20 years of exploration of product ideas for interconnecting, testing, and cooling semiconductor systems. Dissipation of heat is the biggest problem facing designers of high-performance computing systems. Processors such as the Nvidia Hopper chip at 700W are consuming more and more power. The proposed solution for dissipating power compares very favorably with that of the Frontier supercomputer, introduced by Hewlett Packard Enterprises in 2021 and currently the world's fastest computer. For example, power usage can approach 500 teraflops/W versus 50 teraflops/W with Frontier, and water usage can approach 7.5 petaflops/gpm versus 0.25 petaflops/gpm for Frontier. The proposed solution discards conventional packages and heat sinks and employs a copper metallization at the interface between chips and cooling water. This tightly couples chips to cooling water, more efficiently than any known alternative solution. In one configuration there are only two films between the chip and the water: (i) a thermal interface material comprising a die attach film only 20µm thick; (ii) a copper metallization having a thickness of approximately 100µm. These combine with the thermal resistance of the chip itself to comprise a total thermal resistance of only 0.0093°C/W; this means that a 700W chip can operate with a maximum junction temperature only 7°C above the water temperature, an astonishing result. Certain behaviors relating to the impingement of water against a copper surface are not well understood. In particular, a convective heat transfer coefficient is currently unknown for the chosen materials and operating conditions. Additionally, the water reduction predictions are so dramatic they require experimental validation. However, proving the thermal solution and harnessing its benefits will impact electronic systems across a wide spectrum of applications, including servers, automotive, military, and space. Water-cooled systems are the primary objective, but air-cooled systems can also benefit from the approach. The proposed technology has the potential to: (i) reduce design costs by reducing and simplifying the thermal structure; (ii) reduce manufacturing costs by employing regularized structures amenable to automation, and by eliminating materiel such as packages, heat sinks, daughter boards, and custom enclosures; (iii) reduce the required infrastructure by, for example, reducing the footprint of a data center by at least 10X.