Because servers are crucial for processing data communications, the server industry has grown exponentially in tandem with the internet. Although server units were initially based on PC architecture, server systems must be able to handle the ever-increasing number and complexity of network hosts.
This paper illustrates a typical rack server system in a data center, along with a block diagram of the server system. The power supply unit (PSU) is the core of the server system and requires a complex system architecture. This paper will examine five server PSU design trends: power budget, redundancy, efficiency, operating temperature, and communication and control.
1. Power Budget
In the early 21st century, the power budget for rack or blade server PSUs ranged from 200 W to 300 W. At that time, the power consumption of each central processing unit (CPU) ranged from 30 W to 50 W.
Today, server CPUs consume approximately 200 W and have a thermal design power (TDP) approaching 300 W, significantly increasing the power budget for server power units (PSUs) to between 800 W and 2,000 W. To support the increasing computing demands of servers, such as cloud computing and artificial intelligence (AI) computing on the internet, servers may incorporate graphics processing units (GPUs) to work alongside CPUs. This inclusion could increase server power requirements to over 3,000 W within five years. However, because most rack or blade server PSUs still use AC inputs rated at up to 16 A, their power budget is limited: approximately 3,600 W at 240 V AC input, considering converter efficiency. Therefore, 3,600 W will remain the power constraint for server rack PSUs in the short term.
For data center power racks, server PSU designers widely use AC power outlets rated at 20A according to the International Electrotechnical Commission (IEC) 60320 C20 standard. The PSU power budget is limited by its AC inlet current rating, allowing approximately 3,000 W in today's data center PSUs; however, in the near future, data center PSU power levels may increase to over 5,000 W. To allow for higher power budgets and achieve higher power density per PSU, you can also increase the input current rating by using a busbar at the AC inlet.
1. Redundancy
The importance of reliability and availability in server systems necessitates redundant power supply units (PSUs). If one or more PSUs fail, other PSUs in the system can take over power supply.
A simple server system can have 1+1 redundancy, meaning there is one active power supply unit and one redundant power supply unit. Complex server systems may have N+1 or N+N (N>2) redundancy, depending on system reliability and cost considerations. To maintain system operation when a PSU needs to be replaced, hot-swappable (ORing) technology is required. Since multiple PSUs are powered simultaneously in an N+1 or N+N system, current sharing technology is also necessary for the server PSUs.
Even when a PSU is in standby mode (not supplying power to the output from its main power rail), it still needs to provide full power immediately after a hot-swap event, thus requiring continuous activation of the power stage. To reduce power consumption in standby mode, "cold redundancy" is becoming a trend. The purpose of cold redundancy is to shut down the main power supply or operate in burst mode, allowing the redundant PSU to minimize standby power consumption.
1. Efficiency
In the early 2000s, efficiency specifications were slightly above 65%; at that time, server PSU designers did not prioritize efficiency. Traditional converter topologies could easily meet the 65% efficiency target. However, because servers need to run continuously, higher efficiency can significantly reduce the total cost of ownership.
Since 2004, the 80 Plus standard has provided certification for PC and server PSU systems to achieve efficiencies of over 80%. Currently, most mass-produced server power supplies meet the 80 Plus Gold (>92% efficiency) requirement, and some can even reach 80 Plus Platinum (>94% efficiency).
The server PSUs currently under development are primarily targeting the higher 80 Plus Titanium specification, which requires peak efficiency exceeding 96% at half load. Furthermore, according to the Open Compute Project (OCP) open rack specification followed by data center PSUs, PSUs need to achieve peak efficiency exceeding 97.5%. Therefore, new topologies such as bridgeless power factor correction (PFC) and soft-switching converters, as well as wide-bandgap technologies such as silicon carbide (SiC) and gallium nitride (GaN), can help PSUs achieve the 80 Plus Titanium and Open Compute efficiency targets.
1. Operating temperature
In terms of server PSU thermal management, designers define the ambient temperature at the PSU AC inlet (where the fan is located) as the server PSU operating temperature. In the early 2000s, the highest operating temperature was 45°C, and now it reaches a maximum of 55°C, depending on the server room's cooling system.
Higher operating temperatures reduce the energy costs of server cooling systems. Over time, energy costs, as operating expenses, are expected to exceed capital expenditures for data centers (such as hardware). This is based on Power Usage Effectiveness (PUE) standards.
PUE = Total Data Center Power / Actual IT Power
A lower PUE value indicates a more efficient data center. PUE values are estimated at different operating temperatures. For example, a data center with a PUE of 1.25 can only allow its cooling system to consume 10% of its total power. This means that server PSUs require higher operating temperatures.
1. Communication and Control
For many years, communication and control have played a crucial role in server power supplies. In the early 2000s, internal information from the PSU was transmitted to the system side via the system management bus interface. In 2007, the Power Management Bus (PMBus) interface added functionality, including configuration, control, monitoring and fault management, input/output current and power, board temperature, fan speed control, real-time code updates, overvoltage (current, temperature) and protection. Subsequently, to meet the growing demands of data center power racks, the Controller Area Network Bus (CANBus) became part of server power supply communication.
Power management controllers have also evolved alongside communication buses. In the early 2000s, analog controllers primarily controlled server PSUs. As increasing control demands led to greater communication requirements, it became easier to implement these needs using digital controllers. Using digital control also reduces the debugging workload for hardware engineers, potentially lowering labor costs during the PSU design and verification phases.
Future Development Trends of Server Power Supplies
As server power budgets grow while capacity remains constant, power density requirements will become increasingly stringent. The power density of newly developed server PSUs has increased from single digits in the early 2000s to nearly 100 W/in³. Improving converter efficiency through advancements in topology and component technologies is a solution for achieving high power density.
Similar to current, power, and efficiency trends, an ideal diode/ORing controller needs to deliver high current in a small package. An ideal diode/ORing controller must also integrate monitoring, fault handling, and transient handling functions to reduce the overall component count and PCB area required to implement these functions.
For example, PFC circuitry in server PSUs has evolved from passive PFC to active bridge PFC, and then to active bridgeless PFC. Isolated DC/DC converters have evolved from hard-switching flyback and forward converters to soft-switching inductor-inductor-capacitor resonant and phase-shifted full-bridge converters. Non-isolated DC/DC converters have evolved from linear regulators and magnetic amplifiers to buck converters with synchronous rectifiers. Subsequent improvements in overall efficiency have reduced internal power consumption and the work required to address thermal issues.
The technology for components used in server PSUs has also evolved, from IGBTs and silicon MOSFETs to wide-bandgap devices such as silicon carbide MOSFETs and gallium nitride FETs. The non-ideal switching characteristics of IGBTs and silicon MOSFETs limit the switching frequency to below 200 kHz. While wide-bandgap devices offer switching characteristics closer to ideal switching, their use allows for higher switching frequencies, which helps reduce the number of magnetic components used in the PSU.
As operating temperatures rise, components in server PSUs need to withstand greater thermal stress, driving advancements in circuitry. For example, a traditional approach involved connecting mechanical relays in parallel with resistors to suppress input inrush current during startup. However, due to their bulk, reliability issues, and lower rated operating temperatures, solid-state relays are now replacing mechanical relays in server PSUs.
The 3.6kW single-phase totem-pole bridgeless PFC design features a power density of >180W/in³, while the 3kW phase-shifted full-bridge PFC employs an active clamping design with a power density of >270W/in³, designed to meet the redundant power supply specifications commonly found in servers.
In a 3.6 kW PFC design, solid-state relays can withstand high operating temperatures. Here, the LMG3522R030 GaN FET supports the use of a bridgeless totem-pole PFC topology. This "baby boost" reduces the size of large capacitors to achieve higher power density.
In the 3kW phase-shifted full-bridge design, the LMG3522R030 GaN FET helps reduce circulating current and enables soft switching. An active clamping circuit acts as a lossless buffer, enabling higher converter efficiency and lower synchronous rectifier voltage stress. All of the above control requirements are achieved using a C2000 microcontroller as the digital control processor.
