Abstract

AI servers are equipped with multiple high-performance chips including memory, GPU and CPU. Thus, they require much more power than conventional servers. The efficiency of power modules that supply stable power directly to GPUs impacts the server performance. Power modules take a 12V input and provide the low voltages required by GPUs and CPUs (e.g., 0.8–1.5V) through DC/DC converting.
Samsung Electro-Mechanics unveils the high-capacitance MLCC for the 12V product line. It effectively removes power instability (noise/ripple) caused by the current load of power consuming GPUs and CPUs, thereby improving both server performance and reliability.

The need for high-density designs for AI server power, ultra-small / high capacitance MLCCS.

Power supply for conventional servers, including AI servers, is organized in gradual stages of power.
This leads to the construction of high-efficiency, high-performance, high-reliability systems.

High-performance semiconductors mounted on server mainboards (GPU, CPU, memory and more) operate at low voltages (0.8–1.5 V) to improve power efficiency. During fast paced operations, there are peak currents of large fluctuations. Therefore, a great amount of MLCCs must be placed in order to stabilize the power supply. The increase in component integration is reducing PCB areas, which also contributes to the growing need for ultra-small, high-capacitance MLCCs that meet both requirements at the same time.

To support the hundreds to thousands of MLCCs required at the server VRM 12V input stage, Samsung Electro-Mechanics has released an ultra-small, high-capacitance product: 0402 inch, 1㎌, 25V, X7S.

This product is suitable for AI server power-module designs. It also improves power density.
More than just a component, it is a foundational technology aiming to surpass the limits of high-performance computing. It will drive the stable establishment of next-generation AI infrastructures.

Product Specifications

Characteristics data