Internal Thermal Management of IBM P-Server Large Format Multi-Chip Modules Utilizing Small Gap Technology

Abstract

The large Multi-Chip Modules (MCM) used in the IBM p-Server computer systems, and their predecessors, have required rather unique cooling solutions and module hardware designs in order to meet the thermal, mechanical and reliability requirements placed on the package. The module internal thermal solution has evolved from a spring-loaded metal contact technology to a thermal compound based design using a novel gap adjustment technology employing a soldered conduction component. This current MCM makes use of a novel technology called Small Gap Technology (SGT). This technique makes it possible to control thermal compound interface thicknesses or gaps to a very tight tolerance from chip-to-chip and module-to-module. Heat flux values that have been handled vary from approximately 20 to 53 W/cm2 depending on the type of chip and the system performance level. Even higher heat fluxes have been projected for next generation products. The hardware and processing techniques employed to manufacture these modules are quite unique. These products are typically on the order of 100mm chip carrier size or 140mm overall module footprint on a side (approximately 90 cm2 of carrier area) and contain 8 chips and numerous discrete devices. The process fixturing and equipment must be able to handle the relatively large thermal mass of the components. The sequence of processing steps must take into account limitations on the material properties of the various module components. This paper will describe the SGT thermal management solution. The hardware and process employed to make the gap adjustments and the thermal interface material used in these high heat flux applications will be discussed. In addition, supporting thermal/mechanical modelling, thermal performance data and reliability data will be presented.

Challenges in Cooling Design of CPU Packages for High-Performance Servers

“…The thermal resistance of such a stacked design was verified to be lower than that of a direct chip to copper cap thermal paste design. In order to further reduce the thermal resistance between chips and the MCM cap, so-called small gap technology (SGT) is used in the IBM POWER5-based MCM packages and the z-Server MCM packages [10,11], where the MCM cap features the special cooling piston for each CPU chip, allowing a significant reduction of the gap between the chip and MCM cap. An alternative MCM cooling scheme was presented by Kobayashi et al [12] for the Hitachi M5800 high performance server, in which a dual-layer thermal interface design was applied between the chips and a micro-fin assembly, and then a common cap.…”