With the explosive growth of AI computing demand and the continuous evolution of data center infrastructure, AI liquid-cooled server clusters have become the core of high-performance computing. Their power delivery and thermal management systems, serving as the energy conversion and control center, directly determine the cluster's computational efficiency, power usage effectiveness (PUE), operational stability, and overall cost of ownership. The power MOSFET, as a key switching component in these systems, significantly impacts power density, conversion efficiency, thermal performance, and long-term reliability through its selection. Addressing the high-power, high-ripple, and harsh operational environment (high temperature, high humidity) of AI server clusters, this article proposes a complete, actionable power MOSFET selection and design implementation plan with a scenario-oriented and systematic design approach.
I. Overall Selection Principles: System Compatibility and Balanced Design
The selection of power MOSFETs should not pursue superiority in a single parameter but achieve a balance among voltage/current rating, switching performance, thermal impedance, and package robustness to precisely match the stringent requirements of server power supplies and liquid cooling pump/fan drives.
Voltage and Current Margin Design: Based on the system bus voltage (e.g., 12V input, 48V intermediate bus, or high-voltage PFC stages), select MOSFETs with a voltage rating margin of ≥50-100% to handle severe switching spikes and transients in multi-phase VRMs and LLC converters. The current rating must withstand high ripple currents with sufficient margin, typically keeping the continuous operating current below 50-60% of the device’s rated DC current.
Ultra-Low Loss Priority: Loss directly dictates power supply efficiency and heat generation. Conduction loss is critical in high-current paths like synchronous rectification, necessitating ultra-low on-resistance (Rds(on)). Switching loss dominates in high-frequency primary-side switches, requiring low gate charge (Qg) and low output capacitance (Coss). For the highest efficiency, Silicon Carbide (SiC) MOSFETs should be considered for high-voltage stages.
Package and Thermal Management Coordination: High power density demands packages with extremely low thermal resistance and suitability for heatsink or cold plate attachment (e.g., TO-247, TO-247-4L, D2PAK). The 4-lead packages (like TO-247-4L) with a separate source sense (Kelvin connection) are preferred for critical high-frequency switches to minimize parasitic inductance and switching loss. PCB design must integrate thick copper layers and thermal vias.
Reliability and Ruggedness: Servers operate 24/7 under high load. Focus on the device's maximum junction temperature (Tj max), avalanche energy rating (EAS), body diode robustness, and long-term parameter stability under thermal cycling. Automotive-grade or equivalent high-reliability parts are recommended.
II. Scenario-Specific MOSFET Selection Strategies
The main power domains in AI liquid-cooled server clusters include high-voltage AC-DC front-end PFC/LLC, intermediate bus DC-DC conversion (48V-12V/5V), and Point-of-Load (POL) voltage regulation (VRM/VRD) for CPUs/GPUs, alongside the liquid cooling pump and fan drives. Targeted selection is required for each.
Scenario 1: High-Voltage Primary-Side Switching & PFC (650V-850V Class)
This stage handles AC-DC conversion and power factor correction, requiring high-voltage capability, fast switching, and high efficiency to reduce losses before downstream conversion.
Recommended Model: VBP165C93-4L (Single-N, 650V, 93A, TO247-4L, SiC Technology)
Parameter Advantages:
Utilizes advanced SiC technology, offering ultra-low Rds(on) of 22 mΩ (@18V), drastically reducing conduction loss compared to Si counterparts.
Extremely fast intrinsic body diode and low Qg/Coss enable high-frequency operation (>100 kHz), reducing magnetic component size and improving power density.
TO247-4L package with Kelvin source minimizes gate loop inductance, optimizing switching performance and loss.
Scenario Value:
Enables >98% efficiency in PFC and LLC stages, directly improving overall PUE.
High-frequency operation allows for compact, high-power-density power supply designs.
Superior high-temperature performance aligns well with the hot operating environment near server racks.
Scenario 2: High-Current Synchronous Rectification & DC-DC Conversion (48V-12V, VRM)
This stage demands the lowest possible conduction loss to handle currents often exceeding several hundred Amperes, especially in multi-phase VRMs for CPUs/GPUs.
Recommended Model: VBP1102N (Single-N, 100V, 72A, TO247, Trench Technology)
Parameter Advantages:
Very low Rds(on) of 18 mΩ (@10V) minimizes conduction voltage drop and I²R losses in high-current paths.
High continuous current rating (72A) suits parallel operation in multi-phase buck converters.
Standard TO247 package offers excellent thermal performance for heatsink mounting.
Scenario Value:
Ideal for synchronous rectifier in 48V-12V DC-DC converters and as low-side switches in multi-phase VRMs.
High current handling per device reduces the number of parallel components needed, simplifying design and layout.
Robust construction ensures reliability under high ripple current stress.
Scenario 3: Liquid Cooling Pump & High-Performance Fan Drive (12V/48V BLDC/PMSM)
Cooling pumps and fans are critical for thermal management. Their drivers require reliable high-side/low-side switching, fault tolerance, and efficient operation.
Recommended Model: VBM82152M (Single-P, -150V, -15A, TO220F, Trench Technology)
Parameter Advantages:
High voltage rating (-150V) provides ample margin for 48V or higher pump motor drives, handling back-EMF safely.
Low Rds(on) of 160 mΩ (@10V) for a P-channel device reduces losses in high-side configuration.
TO220F (fully insulated) package simplifies heatsink installation without isolation pads, improving thermal management and safety.
Scenario Value:
Excellent for high-side switching in H-bridge motor drive circuits, simplifying gate drive design compared to using an N-MOS with a bootstrap circuit.
The insulated package is advantageous in liquid cooling systems where condensation or accidental coolant contact is a concern, enhancing system safety and reliability.
Enables efficient and reliable speed control of pumps and fans, crucial for maintaining optimal coolant temperature.
III. Key Implementation Points for System Design
Drive Circuit Optimization:
SiC MOSFET (VBP165C93-4L): Must use dedicated, high-speed gate driver ICs with negative turn-off voltage capability to maximize switching speed, minimize loss, and prevent false triggering. Careful attention to PCB layout to minimize power loop and gate loop parasitics is paramount.
High-Current N-MOS (VBP1102N): Use drivers with strong sink/source capability (≥4A) to ensure fast switching in parallel configurations. Implement active balancing techniques when paralleling multiple devices.
High-Voltage P-MOS (VBM82152M): Can be driven by a level-shifted signal from the controller. Ensure the driver can handle the required voltage swing and provide adequate pull-up strength for fast turn-off.
Thermal Management Design:
Tiered Strategy: SiC and high-current MOSFETs must be mounted on dedicated heatsinks or cold plates. Use thermal interface materials with high conductivity.
PCB Thermal Design: For all packages, employ maximum copper pour area, multiple thermal vias under the thermal pad (for surface-mount types), and connect to internal power planes for heat spreading.
Monitoring: Implement junction temperature sensing or model-based thermal monitoring to dynamically adjust fan/pump speed or workload, preventing overtemperature.
EMC and Reliability Enhancement:
Snubber Networks: Use RC snubbers across drain-source of primary-side switches (VBP165C93-4L) to dampen high-frequency ringing and reduce EMI.
Protection Circuits: Incorporate comprehensive protection: TVS diodes for voltage clamping, accurate current sensing for overcurrent protection, and overtemperature shutdown.
Decoupling: Place high-quality, low-ESL capacitors very close to the drain-source terminals of all power MOSFETs to provide local high-frequency energy and reduce voltage spikes.
IV. Solution Value and Expansion Recommendations
Core Value:
Maximized Power Efficiency: The combination of SiC for high-voltage switching and low-Rds(on) trench MOSFETs for rectification/conversion pushes system-level efficiency beyond 96%, significantly reducing operational energy costs and cooling load.
Enhanced Power Density: High-frequency operation enabled by SiC and compact, high-performance packages allows for smaller, more powerful power supplies, freeing up valuable space within the server chassis.
Uncompromising Reliability: The selected high-voltage, high-current, and insulated package devices, combined with robust thermal and protection design, ensure continuous 24/7 operation under demanding AI workloads.
Optimization and Adjustment Recommendations:
Higher Power Scaling: For CPU/GPU VRMs exceeding 1000A, consider even lower Rds(on) devices or advanced packaging like DirectFET or PowerStage modules for ultimate current density.
Integration Path: For design simplification, consider integrated DrMOS or Smart Power Stages for POL applications, which combine MOSFETs, drivers, and protection.
Future Technology Adoption: Monitor the development of Gallium Nitride (GaN) HEMTs for even higher frequency (MHz range) operation in intermediate bus converters, enabling further size reduction.
Liquid Cooling Specifics: For pumps immersed in dielectric coolant, ensure selected MOSFET packages and associated components are compatible with the specific coolant chemistry to prevent corrosion or degradation.
The selection of power MOSFETs is a cornerstone in designing the power and thermal management systems for AI liquid-cooled server clusters. The scenario-based selection and systematic design methodology proposed herein aim to achieve the optimal balance among power density, efficiency, thermal performance, and rugged reliability. As AI computing demands escalate, the adoption of wide-bandgap semiconductors like SiC and GaN will become increasingly critical, providing the foundation for next-generation, ultra-high-efficiency data center infrastructure. In the era of AI, superior hardware design remains the bedrock of computational performance and operational sustainability.

Detailed Topology Diagrams