5 main developments in energy provide design for servers


As a result of servers are important for dealing with information communications, the server trade has grown exponentially in parallel with the Web. Though server items have been initially based mostly on a PC structure, a server system should be capable to deal with the growing quantity and complexity of community hosts.

Determine 1 exhibits a typical rack server system in an information middle and a block diagram of a server system. Energy provide items (PSUs) are on the coronary heart of a server system and require a posh system structure. This text will look at 5 server PSU design developments: energy finances, redundancy, effectivity, working temperature, and communication and management.

Determine 1 A server system block diagram is proven together with how a server is positioned in an information middle. Supply: Texas Devices

  1. Energy finances

Within the early 21st century, the ability finances of a rack or blade server PSU was within the 200- W to 300-W vary. At the moment, energy consumption per central processing unit (CPU) was within the 30- W to 50-W vary. Determine 2 exhibits CPU energy consumption developments.

Determine 2 The CPU energy consumption developments are proven for the early 21st century. Supply: nap.nationalacademies.org/learn/12980/chapter/6#87

Right this moment, a server CPU’s energy consumption is round 200 W—with thermal design energy closed to 300 W—significantly growing the server PSU’s energy finances to a spread of 800 W to 2,000 W. In an effort to assist increasingly server computation necessities equivalent to cloud computing and synthetic intelligence (AI) calculations on the Web, servers can embrace graphics processing items (GPUs) to work alongside CPUs. This inclusion might improve a server’s energy demand past 3,000 W inside 5 years. Nonetheless, since most rack or blade server PSUs are nonetheless utilizing an AC inlet with as much as a 16-A present score, they are going to have restricted energy budgets: round 3,600 W at a 240 VAC enter, accounting for converter effectivity. So, 3,600 W will nonetheless be a server rack PSU’s energy restrict within the close to time period.

For the information middle energy shelf, server PSU designers extensively apply the Worldwide Electrotechnical Fee (IEC) 60320 C20 AC inlet with 20-A present score. PSU energy budgets are restricted by their AC inlet present score, which permits about 3,000 W in at the moment’s information middle PSUs; however within the close to future, an information middle PSU’s energy stage might improve to over 5,000 W. To permit a better energy finances per PSU and obtain larger energy density, you may also use a busbar for the AC inlet to extend the enter present score.

  1. Redundancy

The significance of reliability and availability in a server system necessitates redundant PSUs. If a number of PSUs fail, different PSUs within the system can take over to ship vitality.

A easy server system can have 1+1 redundancy, which means that there’s one energetic PSU and one redundant PSU within the system. A posh server system may need an N+1 or N+N (N>2) redundancy, relying on system reliability and value issues. In an effort to maintain the system working usually when a PSU must be changed, the system wants a hot-swap (ORing management) approach. And since a number of PSUs ship energy concurrently in an N+1 or N+N system, server PSUs additionally require a current-sharing approach.

Even a PSU in standby mode—not delivering energy to the output from its major energy rail—nonetheless requires instantaneous supply of full energy after a hot-swap occasion, thus requiring fixed activation of the ability stage. In an effort to scale back the ability consumption of the redundant energy provide in standby mode, “chilly redundancy” performance is changing into a development. The aim of chilly redundancy is to close down the primary energy operation or function in burst mode, enabling the redundant PSU to attenuate standby energy consumption.

  1. Effectivity

Effectivity specs within the early 2000s have been simply above 65%; on the time, server PSU designers didn’t prioritize effectivity. Conventional converter topologies might simply fulfill the 65% effectivity goal. However as a result of a server must function repeatedly, larger effectivity can significantly scale back complete price of possession.

Since 2004, the 80 Plus normal has offered certifications for PC and server PSU programs that may obtain over 80% effectivity. Server PSUs in mass manufacturing at the moment principally obtain the 80 Plus Gold (>92% effectivity) requirement, and a few may even obtain 80 Plus Platinum (>94% effectivity).

Server PSUs beneath improvement at the moment primarily goal the even larger 80 Plus Titanium specs, which require over 96% peak effectivity at half masses. Desk 1 exhibits the assorted 80 Plus specs.

Desk 1 The 80 Plus specification ensures above 80% effectivity. Supply: www.clearesult.com/80plus/

Additionally, based on the Open Compute Challenge (OCP) open-rack specification that information middle PSUs are following, a PSU wants to attain over 97.5% peak effectivity. Subsequently, new topologies equivalent to bridgeless energy issue correction (PFC) and soft-switching converters, together with broad bandgap applied sciences equivalent to silicon carbide (SiC) and gallium nitride (GaN) may help PSUs obtain 80 Plus Titanium and open-compute effectivity targets.

  1. Working temperature

Within the context of server PSU thermal administration, designers outline the ambient temperature on the PSU AC inlet—the place the fan is positioned—because the server PSU working temperature. The working temperature began at 45°C most within the early 2000s and at the moment reaches 55°C most, relying on the cooling system within the server room.

The next working temperature reduces the vitality prices of a server cooling system. In comparison with the capital expenditures of an information middle (equivalent to {hardware} gear), vitality prices as an working expense are anticipated to be larger than capital expenditures over time. In accordance with the ability utilization effectiveness (PUE) normal:

PUE = Complete Datacenter Energy/Precise IT Energy

As proven in Desk 2, a decrease PUE quantity means an environment friendly information middle. Determine 3 is an estimation of the PUE quantity beneath completely different working temperatures. For instance, an information middle with a PUE of 1.25 can solely enable 10% of general energy consumption on its cooling system. This suggests the necessity for a better working temperature in a server PSU.

Desk 2 A decrease PUE quantity interprets into an environment friendly information middle.

Determine 3 An estimation of the PUE quantity beneath completely different working temperatures exhibits diminished cooling prices with a better working temperature.

  1. Communication and management

Communication and management have performed an necessary function in server energy over time. Within the early 2000s, the PSU’s inside info was transmitted to the system facet by way of the System Administration Bus interface. In 2007, the Energy Administration Bus (PMBus) interface added capabilities, together with configuration, management, monitoring and fault administration, enter/output present and energy, board temperatures, fan velocity management, real-time replace code, overvoltage (present, temperature), and safety. Then, in response to elevated demand for information middle energy cabinets, the Controller Space Community bus (CANBus) grew to become part of server energy communication.

Energy-management controllers have additionally developed together with the communication bus. Within the early 2000s, analog controllers primarily managed server PSUs. As increasingly management calls for elevated the necessity for communication, it grew to become simpler to understand these calls for with digital controllers. Utilizing digital management additionally reduces a {hardware} engineer’s debugging efforts, doubtlessly decreasing labor prices in the course of the PSU design and verification levels.

Future improvement developments for server PSUs

As server energy budgets develop whereas the quantity stays fastened, energy density necessities will turn into stricter. Energy density has elevated from single digits in the beginning of the 2000s to almost 100 W/in3 on newly developed server PSUs. Enhancing converter effectivity by way of topology and part know-how evolutions is the answer to attaining excessive energy density.

As was the case with the present, energy and effectivity developments, the best diode/ ORing controller must ship excessive present in a small bundle. The best diode/ORing controller should additionally combine options equivalent to monitoring, fault dealing with and transient dealing with to scale back the general part rely and PCB space wanted to attain these functionalities.

For instance, a PFC circuit in a server PSU has developed from passive PFC to active-bridge PFC to energetic bridgeless PFC. Remoted DC/DC converters have developed from hard-switching flyback and ahead converters to soft-switching inductor-inductor-capacitor resonant and phase-shifted full-bridge converters. Non-isolated DC/DC converters have developed from linear regulator and magnetic amplifiers to buck converters with synchronous rectifiers. Subsequent will increase in general effectivity scale back inside energy consumption and the hassle required to resolve thermal points.

Part applied sciences utilized to server PSUs have additionally developed, from IGBTs and silicon MOSFETs to broad bandgap gadgets equivalent to silicon-carbide MOSFETs and gallium-nitride FETs. The nonideal switching traits of IGBTs and silicon MOSFETs restrict the switching frequencies beneath 200 kHz. Whereas broad bandgap gadgets have switching traits nearer to very best switches, utilizing broad bandgap gadgets can allow larger switching frequencies to assist shrink the variety of magnetic elements used within the PSU.

Because the working temperature will increase, elements in a server PSU must deal with larger thermal stress, which additionally drives circuit evolution. As an example, a standard implementation applies a mechanical relay in parallel with a resistor to suppress the enter in-rush present throughout startup. However due to their cumbersome dimension, reliability issues and decrease temperature score, solid-state relays at the moment are changing mechanical relays in server PSUs.

The three.6-kW single-phase totem-pole bridgeless PFC designs with >180-W/in3 energy density and 3-kW phase-shifted full bridge with energetic clamp designs with >270-W/in3 energy density goal to satisfy widespread redundant power-supply specs in servers (Determine 4).

Determine 4 The block diagram exhibits the three.6-kW and 3-kW reference designs. Supply: Texas Devices

Within the 3.6-kW PFC design, a solid-state relay accommodates a excessive working temperature. Right here, the LMG3522R030 GaN FET permits the usage of a bridgeless totem-pole PFC topology. A “child increase” reduces the majority capacitor quantity for larger energy density.

Within the 3-kW phase-shifted full-bridge design, the LMG3522R030 GaN FET helps decrease the circulating present and makes it potential to attain delicate switching. An energetic clamping circuit performing as a lossless snubber permits larger converter effectivity with decrease synchronous rectifier voltage stresses. All of the aforementioned management necessities are achieved by way of C2000 microcontroller performing as digital management processor.

Richard Yin is a programs engineer in Energy Design Providers at Texas Devices.

Associated Content material


Please enter your comment!
Please enter your name here