Applications

‌Server power supplies

Working Principle of Server Power Supplies‌

AC Input & EMI Filtering:‌

• The power supply receives AC power from the grid (typically mains voltage, e.g., 110V/220V AC).
• This power first passes through an ‌EMI (Electromagnetic Interference) filter‌. This filter suppresses high-frequency noise interference from the grid and prevents high-frequency noise generated by the power supply itself from propagating back onto the grid and interfering with other equipment.

Rectification:‌

• The filtered AC power enters a ‌rectifier bridge‌ (typically composed of multiple diodes), converting it into pulsating DC (Unidirectional Pulsating DC).

Power Factor Correction (PFC):‌

• This is a critical stage, especially for modern data centers pursuing high efficiency and compliance with environmental regulations.
• Without PFC, the rectifier circuit causes severe distortion of the input current waveform, failing to align with the input voltage waveform. This results in a low ‌Power Factor‌ (significantly below the ideal 1.0). A low PF means significant electrical energy is not effectively utilized; instead, it circulates uselessly between the grid and the load, increasing line losses and causing harmonic pollution to the grid.
• The ‌PFC circuit‌ (usually Active PFC - APFC, based on a boost converter topology) regulates the pulsating DC. It shapes the input current waveform to closely resemble a sine wave and aligns its phase with the input voltage, elevating the power factor close to 1 (e.g., >0.95). This significantly improves power utilization efficiency and reduces grid pollution.

Primary DC-DC Conversion (High Voltage Side):‌

• After PFC correction, a relatively smooth, high-voltage DC is obtained (typically around 380V DC, termed the ‌High-Voltage DC Bus‌).
• This high-voltage DC feeds into a ‌high-frequency switching circuit‌ (typically using topologies like Half-Bridge, Full-Bridge, or LLC Resonant). Within this circuit, power transistors (e.g., MOSFETs) rapidly switch ON and OFF at very high frequencies (tens to hundreds of kHz), driven by a control chip.
• This high-speed switching action effectively "chops" the high-voltage DC into high-frequency AC square waves.

Transformer Isolation & Voltage Transformation:‌

• The high-frequency AC square waves are delivered to the primary winding of a ‌high-frequency transformer‌.
• The transformer transfers energy to its secondary winding via electromagnetic induction. It serves two core functions:

‌Electrical Isolation:‌ Completely isolates the input grid (hazardous high-voltage) from the server's internal circuitry (safe low-voltage), ensuring safety.

‌Voltage Transformation:‌ Steps down the high-voltage AC square wave to the required low-voltage AC square wave (e.g., 12V AC) based on the turns ratio of the secondary winding(s).

Secondary Rectification & Filtering:‌

• The low-voltage AC square wave output from the transformer's secondary winding goes through a ‌secondary rectification circuit‌ (typically using Schottky diodes or Synchronous Rectification MOSFETs), converting it back into pulsating DC.
• This pulsating DC then passes through an ‌output filter circuit‌ (an LC filter composed of inductors and capacitors). This filter removes high-frequency switching noise and ripple, yielding the ‌stable, clean low-voltage DC‌ required by the server motherboard (primarily +12V, with smaller amounts of +5VSB/+3.3V historically, though the dominant trend now leans towards a single +12V rail, with other voltages generated locally on the motherboard by Voltage Regulator Modules - VRMs).

Monitoring & Protection:‌

• Sophisticated monitoring circuitry within the power supply (typically implemented via a dedicated power management IC) continuously tracks parameters such as output voltage, current, and temperature.
• It provides comprehensive protection to prevent damage to both the power supply and server components. Common protection features include:

‌Over Voltage Protection (OVP):‌ Prevents excessive output voltage that could damage equipment.

‌Under Voltage Protection (UVP):‌ Prevents insufficient output voltage causing system instability.

‌Over Current Protection (OCP):‌ Protects against output short circuits or overloads.

‌Over Power Protection (OPP):‌ Safeguards against total output power exceeding the supply's rated capacity.

‌Over Temperature Protection (OTP):‌ Prevents damage due to internal overheating.

Architecture of Server Power Supplies

‌Power Supply Unit (PSU) Itself:

• High Power Density:‌ Delivers increasingly high rated power (from hundreds of watts to several kilowatts) within a constrained space (typically 1U or 2U height).
• Redundancy Architecture:‌ This is the core characteristic. Server chassis typically house ‌two or more‌ completely independent PSUs operating in parallel. Common configurations include:

‌N+1 Redundancy:‌ The system requires N power modules to meet the load demand but includes one additional module (+1) as a spare. If one PSU fails, the remaining units immediately assume the full load, ensuring uninterrupted system operation.

‌2N Redundancy:‌ The system employs two completely separate, identically sized power systems (Path A and Path B), each capable of independently supporting the entire load. This represents the highest level of redundancy, typically used for mission-critical systems.

‌‌Hot-Swappable Design:

• Power modules are designed for ‌hot-swapping‌, meaning they can be safely removed and inserted while the server is running, without requiring a shutdown. This is crucial for achieving high availability, allowing rapid replacement of failed units during operation, minimizing downtime.

‌‌Input Source Design:

• Dual-Cord Input:‌ Many redundant PSUs support connection to two independent AC input sources (e.g., from different UPSs or PDUs). This ensures continued operation if one mains feed fails. Combining redundant PSUs with dual-cord input creates a robust power delivery guarantee.

‌‌Power Distribution Architecture:

• Within the Server:‌ The main PSU output is typically +12V (or HVDC). This power is delivered via dedicated connectors on the server motherboard (often near the CPU and memory sockets). Voltage Regulator Modules (VRMs) on the motherboard then convert this to the lower voltages required by the CPU and memory (e.g., around 1V).
• ‌Rack/Data Center Level:‌ PSU inputs are typically sourced from ‌Power Distribution Units (PDUs)‌ mounted at the top of the server rack. The PDU distributes power from UPS systems or the mains grid to individual servers. Modern intelligent PDUs offer metering, monitoring, and remote control capabilities.

‌‌Battery Backup Unit (BBU):‌ In certain high-end storage servers or scenarios requiring

Market Development Trends for Server Power Supplies‌

‌Higher Efficiency & Stringent Energy Standards:

• 80 PLUS Titanium as the Benchmark:‌ Efficiency requirements continue to rise. Titanium certification demands extremely high conversion efficiency (>94% @ 50% load, >90% @ 10% load) across various load levels, including very light loads (10%), minimizing energy waste and reducing data center PUE and operational costs (TCO).
• ‌Lower Standby Power:‌ Requirements for power consumption during server idle or shutdown states are becoming increasingly stringent.

‌Higher Power Density:

• Soaring power demands from CPUs, GPUs, and other compute chips (especially for AI/ML/HPC applications) push single-server power requirements drastically higher (from traditional kilowatt levels to 3kW, 5kW, and beyond).
• PSUs must deliver higher power output while maintaining standard form factors (e.g., CRPS/CRPS+), posing significant challenges for thermal design, component selection, and topology optimization.

Exploration of High-Voltage Direct Current (HVDC) Power Delivery:‌

• Traditional AC distribution involves multiple conversion losses (AC->UPS DC->UPS AC->PSU DC). HVDC (e.g., 240V/336V/380V DC) distributes DC power directly at the data center level. Server PSUs connect directly to HVDC, eliminating the PFC stage and some conversion steps. This promises potential overall efficiency gains (1-5%) and infrastructure simplification. While adoption has been slow, progress continues in specific scenarios (e.g., large internet company custom designs) and research.

‌Integration of Liquid Cooling:

• The thermal challenges posed by high power density are intensifying. Traditional air cooling becomes inefficient as power exceeds 2-3kW per PSU or in ultra-high-density racks.
• ‌Liquid-Cooled Power Supplies:‌ Applying liquid cooling (cold plates or immersion) directly to PSU modules leverages liquid's superior thermal conductivity to dissipate heat. This significantly enhances cooling efficiency, enabling support for higher power densities and quieter operation. Liquid-cooled solutions are seeing rapid growth in HPC and AI clusters.

Intelligence & Manageability:‌

• PMBus (Power Management Bus) as Standard:‌ PSUs communicate via the PMBus interface with the server's management controller (BMC), reporting detailed input/output voltage, current, power, temperature, fan speed, and fault status information.
• ‌Remote Monitoring & Control:‌ Administrators can remotely monitor each PSU's health, control fan modes, and even remotely power cycle PSUs using standard interfaces like IPMI or Redfish. This is vital for large-scale data center operations.
• ‌Predictive Maintenance:‌ Leveraging intelligent data and AI analytics to predict potential PSU failures enables proactive replacement, preventing unplanned downtime.

Modularity & Standardization:‌

Sustainability & Decarbonization:‌

• Use of longer-lasting, more environmentally friendly materials.
• Pursuit of low carbon footprint throughout the entire lifecycle (manufacturing, operation, recycling).
• Ongoing research exploring the potential of ‌hydrogen fuel cells‌ as a clean primary or backup power source for data centers.

The advantages of SMC