SAMSUNG ELECTRO-MECHANICS

▷ The power consumption of AI servers is five times higher than that of general servers, and AI servers employ 12.5 times more MLCCs than general servers. [Figure1, 2] MLCCs used in AI servers require high capacitance, high voltage, and miniaturization to handle higher power, and these needs are continuously increasing.  

▷ AI servers use high-performance ICs such as CPUs, GPUs, and TPUs, and during high-computation operations, significant instantaneous current changes occur. High-capacitance MLCCs minimize voltage drops and quickly compensate for current fluctuations, improving power stability. MLCCs of general capacitance have an excellent high-frequency response. However, they lack the ability to store energy in low-frequency bands if they have lower capacitances. In contrast, high-capacitance MLCCs allow sufficient energy storage even in the low-frequency range, providing stable performance across a wider bandwidth.

▷ Additionally, by using high-capacitance MLCCs in AI servers, the number of required components can be reduced, simplifying PCB design and increasing spatial efficiency for mounted areas. For example, as shown in Figure3 below, applying four MLCCs (180uF x 4 @ 1Vdc) in place of one SP-CAP (330uF) can reduce the required area by approximately 35% while improving capacitance by 218%. Furthermore, impedance characteristics are enhanced by 90% compared to SP-CAP. [Figure3] 

Product Line-up

Related Keywords

#Product Detail

#Product Inquiry

#Previous Article