With the rapid evolution of AI computing and the urgent need for data center sustainability, AI server energy-saving control systems have become critical for managing power integrity and thermal performance. The power delivery and management systems, serving as the "vascular and neural network" of the server, provide precise power conversion, sequencing, and switching for key loads such as CPU/GPU VRMs, high-speed fans, and auxiliary DC-DC rails. The selection of power MOSFETs directly determines system conversion efficiency, thermal footprint, power density, and reliability. Addressing the stringent demands of AI servers for extreme efficiency, high heat flux management, and precise control, this article develops a practical and optimized MOSFET selection strategy through scenario-based adaptation.
I. Core Selection Principles and Scenario Adaptation Logic
(A) Core Selection Principles: Three-Dimensional Optimization for High Density
MOSFET selection requires coordinated optimization across three dimensions—loss, thermal, and voltage—ensuring robust operation under high ambient temperatures and dynamic loads:
Prioritize Ultra-Low Loss: For CPU/GPU VRMs and bulk DC-DC converters, prioritize devices with extremely low Rds(on) (minimizing conduction loss) and low Qg/Qoss (minimizing switching loss at high frequency), crucial for achieving >96% peak efficiency and reducing coolant burden.
Thermal and Package Co-design: Choose packages (e.g., TO220F, TO263) with low thermal resistance for high-power stages, enabling effective attachment to heatsinks or cold plates. For point-of-load (POL) switches, select compact packages (e.g., SOP8, SC70-6) to save board area in congested spaces.
Voltage and Reliability Margins: For 12V/48V intermediate bus architectures, ensure sufficient voltage derating (≥60% margin) to handle transient spikes. Devices must offer high junction temperature capability (Tj max ≥ 150°C) and robust gate oxide for 24/7 mission-critical operation.
(B) Scenario Adaptation Logic: Categorization by Power Path and Function
Divide server power paths into three core control scenarios: First, High-Current DC-DC Conversion & VRM (Power Core), requiring highest efficiency and current handling. Second, Power Sequencing & High-Side Switching (Management & Safety), requiring compact solutions for board-level power control. Third, Low-Power Auxiliary Rail & Fan Control (Support & Cooling), requiring space-efficient switches for intelligent power-gating and speed control.
II. Detailed MOSFET Selection Scheme by Scenario
(A) Scenario 1: High-Current DC-DC Conversion & VRM – Power Core Device
Multi-phase VRMs and 48V-to-12V/5V intermediate bus converters (IBCs) require handling very high continuous and transient currents with minimal loss.
Recommended Model: VBGMB1103 (Single-N, 100V, 80A, TO220F)
Parameter Advantages: Advanced SGT technology achieves an ultra-low Rds(on) of 2.9mΩ at 10V. High continuous current of 80A suits multi-phase buck converters for GPU power or high-power IBCs. TO220F package offers excellent thermal performance for heatsink mounting.
Adaptation Value: Drastically reduces conduction loss in critical power paths. In a 48V-12V IBC phase handling 30A, device conduction loss is only ~2.6W, enabling system efficiency >97%. Supports high-frequency multiphase operation, improving transient response and reducing input/output capacitance needs.
Selection Notes: Verify phase current and thermal design. Ensure gate driver capability (≥3A peak) to switch low Qg device efficiently. Implement precise current sharing and temperature monitoring.
(B) Scenario 2: Power Sequencing & High-Side Switching – Management & Safety Device
Power sequencing, hot-swap, and high-side load switching for various rails (e.g., 12V, 5V) require compact, efficient P-Channel or high-voltage N-Channel solutions for safe power-up/down sequences.
Recommended Model: VBA2311A (Single-P, -30V, -12.5A, SOP8)
Parameter Advantages: P-Channel configuration simplifies high-side drive. Very low Rds(on) of 11mΩ (10V) minimizes voltage drop. SOP8 package saves significant PCB area compared to discrete solutions. -30V rating is ideal for 12V bus control with ample margin.
Adaptation Value: Enables intelligent, sequenced power-up for SSD backplanes, memory modules, or expansion cards, preventing inrush currents. Low on-resistance ensures minimal power loss on the critical power path. Saves layout space for dense server motherboard designs.
Selection Notes: Use with an NPN transistor or dedicated high-side driver for gate control. Add RC snubber if switching inductive loads. Ensure current is within safe operating area (SOA) during hot-swap events.
(C) Scenario 3: Low-Power Auxiliary Rail & Fan Control – Support & Cooling Device
Controlling low-power rails (3.3V, 5V) for sensors, management controllers, and PWM control of high-speed cooling fans requires small, logic-level devices.
Recommended Model: VBK7695 (Single-N, 60V, 2.5A, SC70-6)
Parameter Advantages: Ultra-compact SC70-6 package minimizes footprint. Low Vth of 1.7V allows direct drive by 3.3V MCU GPIO. Good Rds(on) (75mΩ @10V) for its size. 60V rating provides strong margin for 12V/24V fan PWM circuits.
Adaptation Value: Enables fine-grained power-gating of peripheral components, reducing standby power. Used as a PWM switch for 4-wire fans, allowing dynamic speed control for acoustic and thermal optimization. Its tiny size allows placement near the load.
Selection Notes: Ensure gate drive strength is adequate for target switching frequency. Add a small gate resistor to dampen ringing. For fan control, include a freewheeling diode path.
III. System-Level Design Implementation Points
(A) Drive Circuit Design: Matching Device Characteristics
VBGMB1103: Pair with high-current, high-speed multi-phase PWM controllers and drivers (e.g., Infineon XDPE, TI PMBus controllers). Minimize power loop inductance with a tight layout.
VBA2311A: Implement a simple NPN level translator or use a dedicated load switch IC with integrated control for sequencing. A pull-up resistor on the gate ensures default-off state.
VBK7695: Can be driven directly from MCU but include a series resistor (e.g., 10Ω). For higher frequency PWM, a dedicated gate driver buffer may be beneficial.
(B) Thermal Management Design: Tiered and Direct Cooling
VBGMB1103: Requires dedicated heatsink or thermal interface to server's cooling solution (cold plate/heat sink). Use thermal pads and ensure mounting pressure.
VBA2311A: A modest copper pour under the SOP8 package is typically sufficient for its power level. Ensure general airflow over the board.
VBK7695: Local copper heat spreading is adequate. Its low power dissipation minimizes thermal impact.
Overall: Place high-power MOSFETs in the primary airflow path or directly on cold plates. Monitor junction temperature via associated controller ICs.
(C) EMC and Reliability Assurance
EMC Suppression:
VBGMB1103: Use low-ESR ceramic capacitors very close to drain and source pins. Optimize snubber networks for high-frequency switching nodes in VRMs.
VBA2311A/VBK7695: Add ferrite beads in series with the switched load for noise filtering. Ensure clean, isolated ground paths for sensitive analog controls.
Reliability Protection:
Derating Design: Derate voltage by 50% and current by at least 30% from absolute maximum ratings at maximum operating temperature.
Overcurrent/Overtemperature Protection: Utilize the integrated protection features of server PWM controllers for VRM stages. For load switches, implement current limiting or use ICs with built-in fault protection.
ESD/Surge Protection: Apply TVS diodes on all external connections (fan headers, power inputs). Use gate-source resistors and TVS for GPIO-connected MOSFETs (VBK7695).
IV. Scheme Core Value and Optimization Suggestions
(A) Core Value
Maximized Power Density and Efficiency: The selected devices optimize the efficiency of every power stage, reducing total power loss and enabling higher compute density within the same thermal design power (TDP) envelope.
Enhanced System Reliability and Manageability: Enables precise power sequencing and intelligent control of auxiliary loads, improving system stability and facilitating advanced power capping/management features.
Optimized Cost-Structure for Scale: Leverages mature, high-volume MOSFET technologies, providing the optimal balance of performance and cost for large-scale server deployment.
(B) Optimization Suggestions
Power Scaling: For even higher current VRMs (>100A per phase), consider parallel operation of VBGMB1103 or evaluate higher current variants. For 48V direct conversion, consider VBN1154N (150V).
Integration Upgrade: For advanced power sequencing, use integrated load switches with I2C/PMBus interface instead of discrete VBA2311A. For fan fail detection, consider devices with integrated sense FET.
Specialized Scenarios: For redundant power supply (OR-ing) applications, use VBA2311A for its low reverse recovery charge. For harsh environments, select automotive-grade equivalents.
Advanced Cooling Control: Pair VBK7695 with temperature sensors and MCU to create closed-loop, adaptive fan speed profiles, further optimizing acoustic noise and energy use.
Conclusion
Strategic MOSFET selection is central to achieving the high efficiency, power density, and intelligent control required in next-generation AI servers. This scenario-based scheme provides targeted technical guidance for power system architects through precise load matching and robust system-level design. Future exploration should focus on the integration of Wide Bandgap (SiC/GaN) devices for the highest efficiency conversion stages and the adoption of fully integrated smart power stages (IPMs) to further simplify design and enhance monitoring capabilities, paving the way for more sustainable and powerful AI computing infrastructure.

Detailed MOSFET Application Topology Diagrams