06 Dec 2015
Power Density Dominates Server Design: Outlook for 2016
Martin Hägerdal, President, Ericsson Power Modules looks at the growing requirements for improved efficiency for server power supplies.
The coming year, 2016, will see server processors arrive carrying more than 20 individual cores on chip and support for the latest DDR4 low-voltage memories. Designers will be working on servers that employ 30-core processors. Many systems will couple this volume of core devices with other processor architectures to maximise throughput in both data-centre and communications applications.
The two sectors are converging as communications vendors use data-centre architectures to drive the migration to software-defined networking. The incorporation of field-programmable gate arrays (FGPAs), graphics processing units (GPUs) and custom accelerators alongside many core central processing units (CPUs) is a symptom of the growing focus on energy efficiency in data-centre and communications applications. The designs not only call for high energy density and low voltages, but precise control over each voltage rail and the ability to control many rails independently at the point of load (POL).
Because of the lowering of supply voltages to less than 1V, high current levels are now inevitable, which places intense pressure on the performance of the POL DC-DC converters that supply the Systems on Chips (SoCs), both in terms of steady state and transient adaption.
Optimisation in one area or of a single DC-DC converter is no longer enough to guarantee performance. Because of the high current levels, multiple converters need to work together and easily adapt to the specific requirements of the target system.
Some rails will need very high current levels, demanding that multiple DC-DC converters be coupled using phase-spreading techniques to maximise power at high efficiency. By phase-shifting the outputs of multiple converters, the load can draw high peak currents without producing high peaks of electromagnetic interference that could occur with single-phase switching architectures. Phase spreading allows higher efficiency at lower loads as phases can be switched off as the current requirements falls during periods of low system activity. This is vital in many core systems as the operating system will switch off unused cores to save energy, reactivating them as the software demand changes.
What makes it possible to achieve high efficiency in a multivendor environment is the trend towards digital control. In much the same way a common digital protocol – the Internet Protocol – has enabled a wide variety of systems to communicate, AMP provides a common digital language between DC-DC converters. Communication through the PMBus, in turn, provides a tight link between the system management unit and each of the power devices in the chassis.
Digital power control
The capability of digital control extends much further into the core of the DC-DC converter. Until recently, converter designs employed an analogue core. Analogue control provided a reasonably efficient and low-cost method of deploying switched-mode conversion, which replaced the more loss-intensive linear architecture.
Figure 1 Digital power control
Most switched-mode converters employ pulse width modulation (PWM) in which a controller samples the output voltage and compares this to a reference voltage to generate an error signal. That in turn is compared to the output from an oscillator that generates a ramp waveform. When the ramp signal passes that of the error voltage, the MOSFET responsible for charging the inductor is switched on and is, in turn, switched off when the ramp single falls in the other direction. Changes in the error voltage as the voltage rises and falls control the width in time of the current pulse delivered to the load through the inductor.
Although it has proved effective for decades, analogue PWM control has reached its limits. Stability is a key concern in the design of the control loop for any analogue PWM-based converter. But this stability comes at the cost of flexibility and responsiveness. Switching power supplies based on analogue control techniques often use a compensation network to adjust the frequency response of the loop so that they can to achieve a good transient response without compromising stability. Designing this compensation network can be a time-consuming exercise in trial and error.
Figure 2: Digital power is particularly effective for improving efficiency under low load conditions (Source Ericsson (EGG October 2014)
Over the long term, the performance of the compensation network can drift due to variations in temperature or device ageing. As the compensation components are soldered in place, there is little that can be done to readjust the behaviour of the network when ageing effects take hold. There is little scope for tuning the power converter architecture during manufacture, which is problematic for today’s production processes where flexibility is key.
Digital processing can provide much more responsive algorithms for DC-DC regulation, allowing exploration of a wide range of control algorithm strategies. These are more easily tuned to the specific needs of the target power application because much of the tuning can be carried out using software. The more advanced control techniques also make it possible to reduce the number of output filtering components and so optimise space on the PCB, a key factor when multiple DC-DC converters need to be deployed in parallel to support current levels of 500A or more on a single blade.
With digital control, the software can use models of the converter electronics and load to more closely analyse the needs of the system. The result of the close match between model and behaviour ensures more precise control and, in the case of DC-DC converters such as the BMR46x series, enables a reduction in external passive components. System-level attributes such as allowable ripple current can be estimated and tuned using the Ericsson Power Designer (EPD) software tool, enabling the selection of the most appropriate filtering capacitors.
The co-operation needed between power converters in the system will be facilitated by standards such as those being developed by the Architects of Modern Power (AMP) Consortium, of which Ericsson is a founder member. The consortium aims to create a fertile ecosystem for simple, intuitive and high-efficiency multi-source power solutions based on digital technology.
Support for more dynamic behaviour on the server blade will increasingly be mirrored by greater flexibility in the intermediate bus architecture. A shift to dynamic bus voltage (DBV) control will provide the possibility to adjust the power envelope to meet changing load conditions in the most efficient way possible. DBV achieves this by altering the intermediate bus voltage in response to load changes. Advanced digital power control and optimised hardware combined with a series of software algorithms make it possible to implement DBV cost-effectively.
With each of these innovations, power architecture is ready to meet the challenges of 2016 and further into the future as high-density many core systems become the core of our digital world.
Page 1 of 1
About the author
Martin Hägerdal is President of Ericsson Power Modules, an independent company within the Ericsson Group. He has more than 20 years of experience working within the telecom and power industries and spent nearly a decade spearheading successful sales growth at Ericsson Power Modules. Before joining the company, he was responsible for international sales with Ericsson AB, selling telecom infrastructure equipment in Africa, the Middle East and Europe, and also worked in the R&D division at Ericsson AB. He holds an M.Sc. degree in Electro Technical Engineering from Lund Institute of Technology and a B.Sc. in Business Administration from Stockholm University.
Formed in the late seventies, Ericsson Power Modules is a division of Ericsson AB that primarily designs and manufactures isolated DC/DC converters and non-isolated voltage regulators such as point-of-load units ranging in output power from 1W to 860W for use in information and communication technology (ICT) applications in distributed power architectures. The products are aimed at (but not limited to) the new generation of ICT equipment where systems’ architects are optimising board designs for optimised control and reduced power consumption. The most important application areas are wireless and fixed networks and also in a broad range of industrial applications from medical to process control, and even in automotive applications.
Most popular articles in Power management
Share this page
Want more like this? Register for our newsletter