Lonten super junction MOSFETs, which define AI computing power with higher energy efficiency

2026.03.20

With the explosive growth of artificial intelligence (AI) and large model training, the power consumption of AI servers is rising at an unprecedented rate. From NVIDIA H100 to B200, the power consumption of a single card has exceeded 1000W, and the power of the whole cabinet has even rushed to more than 100kW. In this race between computing power and electricity, power conversion efficiency has become a key bottleneck. With their extremely low conduction loss and excellent switching performance, super junction MOSFETs have become the core devices of high-voltage DC buses (PFC level) and primary switches (LLC grade) for AI servers.

AI server power supplies typically use a two-stage architecture or three-phase direct power supply, with pre-PFC (Power Factor Correction): The AC input (220V AC or three-phase 380V AC) is rectified and boosted to form a stable high-voltage DC bus (typically 400V DC). Post-Stage DC/DC (Isolated Conversion): Converts the 400V bus voltage to the low voltage required by the AI chip (e.g., 48V or 12V), and then stepped down to less than 1V via the onboard VRM. Super junction MOSFETs are mainly deployed on the primary side (primary side) of the PFC stage and DC/DC.

AC/DC power supply

The three core advantages of Lonten super junction MOSFETs empower AI computing power
As a leading power device supplier in China, Lonten Semiconductor's 600V/650V super junction MOSFET product line can cover different power levels and application scenarios to meet the diverse needs of AI server power supplies. With its excellent comprehensive performance, this series of products achieves three core advantages and accurately adapts to the stringent requirements of AI server power supply:

Extremely low on-resistance: Industry-leading RDS(on)· A quality factor effectively reduces conduction loss and improves power conversion efficiency.

Excellent switching performance: Low gate charge (Qg) and low output capacitance (Coss) support for high-frequency switching and increased power density.

Robust Body Diode: Integrated fast recovery body diodes (HP&HF series) meet the stringent requirements of soft switching topologies such as LLC, ensuring system reliability and stability under high-speed switching.

Lonten super junction MOSFET AI server power supply selection scheme
For titanium-level power modules in different power segments, Lonten Semiconductor has launched a variety of core adaptation models, such as: LSB60R063HP, LSB65R070HP, LSB65R041HF, LSB65R031HF, LSB60R031HP, LSB60R022HF, which fully reflects Lonten's deep technical accumulation and accurate insight into the AI server market in the field of super junction MOSFETs. According to the mainstream design of the industry, AI server power modules (PSUs) usually adopt the classic architecture of "PFC + LLC", the PFC stage may adopt a two-phase or three-phase bridgeless interleaved topology, and the LLC stage is mostly a full-bridge resonant topology.

Note: The 8-card AI server in the table adopts a 1+1 redundant power supply configuration (two power modules), so the total usage is twice that of a single power supply. In practical applications, high-power servers often use N+1 redundancy (e.g., 4 modules), and the total usage will increase exponentially.

 

If you are interested in the selection or application of Lonten 600V/650V super junction MOSFETs, please contact us for samples and technical support. 

Hotline: +86-029-8665 8666-8101 Email: sales@lonten.cc