Traditional power-supply system design is based on “silver boxes” or “bricks,” which are self-contained, autonomous, voltage-in/voltage-out devices with standardized mechanical and electrical connections. Changes in system requirements such as higher density, higher input voltages, lower load voltages, higher load currents, and (especially) higher efficiency require tighter mechanical and electrical integration of supplies and the circuitry they power.
This article contrasts two forms of telecom power-supply architecture: the intermediate bus architecture (IBA) and the Factorized Power Architecture (FPA). They’re analyzed from the perspective of powering a typical telecom “central-office” system.
Table Of Contents
The most common implementations of IBA and FPA use lightly regulated 42- to 50-V backplane supplies (with a minimum 36- to 55-V range) for distribution within a line-card or server mainboard. This architecture is a telecom standard, designed to reduce distribution losses, while keeping the voltage at a value unlikely to cause serious shock.
IBA and FPA both place the final isolation, stepdown, and regulation on the line card or server mainboard (Fig. 1 and Fig. 2). Also, both architectures perform electrical isolation, voltage stepdown/current multiplication, and regulation but in a different order, essentially a mirror image of each other (see the table).
IBA positions the fixed-ratio voltage stepdown and electrical isolation first, followed by a regulator that outputs the lower voltage (and higher current) required by the load. FPA begins by regulating the backplane voltage, then lowering that voltage to the load requirements with a fixed-ratio transformer (which also provides electrical isolation).
IBA uses an isolated, non-regulated bus converter. It reduces the backplane voltage by a fixed factor of four or five, creating a bus that’s called “intermediate” because it interconnects several downstream regulators with one bus converter or a few bus converters. This 9- to 13-V bus is distributed through the line card or server motherboard, powering regulated, non-isolated point-of-load (niPOL) buck converters that provide the final step-down.
Each IBA stage performs two functions: isolation/transformation, and voltage step-down/regulation. The niPOLs are off-the-shelf single-phase or multi-phase synchronous buck converters, which can handle a wide variation in input voltage, while providing tight load regulation.
The bus converter is a fixed-ratio “electronic transformer” that creates the 9.6- or 12-V intermediate bus from the 48-V backplane voltage. It electrically isolates the bus from the backplane. Because the intermediate bus voltage is relatively close to the converter’s output voltage, the IBC has to supply a relatively high current to the converter. It is therefore placed as close as possible to the niPOLs it supplies to minimize I2R losses.
Both converter types are technically mature. They offer high performance (with peak efficiencies well above 90%), while selling at commodity prices. Also, IBCs are typically modular while niPOLs can be discrete or modular.
Furthermore, IBCs are available with power ratings up to1 kW. They can effectively power several niPOL regulators. Meanwhile, niPOL regulators are standard synchronous buck converters tailored to specific loads.
FPA can be considered a mirror image of IBA. Its first stage is a non-isolated buck/boost converter, which tightly regulates the backplane voltage to its card or mainboard value. Distribution efficiency is increased, in the same way high-voltage power lines increase efficiency. The distributed voltage is relatively high compared to the final output voltage, so less current flows along the board traces. This reduces I2R losses.
The second stage, called VTM Current Multiplier, uses the same topology as the IBC (electronic dc transformer). It provides voltage reduction/current multiplication (by a much higher, fixed factor), plus isolation.
In addition to reduced distribution losses, FPA offers several other advantages over IBC. First, a factorized bus is usually higher than 40 V, reducing the copper thickness and number of layers required on the mainboard. Next, the higher bus voltage makes possible a wider range of regulated load voltages. Finally, regulation occurs at the single buck/boost converter at the head of the chain, rather than at multiple step-down regulators across the board. This can simplify thermal management.
Power systems must electrically isolate the load from the source (for both safety and grounding), step down the voltage and regulate the voltage. In power conversion using pulse-width-modulation (PWM), the laws of physics impose certain tradeoffs. In particular, large input/output voltage ratios (up or down) imply efficiency losses, due to a high form factor (the ratio of RMS current to average current).
The smaller the duty cycle (for example, 1:12 ≈ 8%), the larger the RMS value for the same average current. Because I2R losses are proportional to the square of the RMS (“heating effect”) current (not the average current), conversion efficiency decreases.
Traditionally, the most-efficient way to step voltages up or down is with ac transformers. (Efficiencies of 99% are common.) There is no variable duty cycle, so there is no variation in efficiency.
This suggests that the best way to get high efficiency is to factor (divide up) the system function to take full advantage of the electronic transformer, relieving the regulator of having to also provide a large voltage step-down, with its attendant I2R losses.
FPA assigns isolation and voltage transformation to a point-of-load dc electronic transformer with high “turns ratio” (32:1 or 40:1). The form factor can be kept low, and regulation can be provided on the primary side, thanks to advanced resonant structures. The regulator turns into a simple buck/boost, whose function is to “tweak” the supply voltage (up or down), without having to supply heavy regulation, which the transformer has already achieved.
FPA power distribution operates at a higher voltage from backplane to POL electronic transformers. The resulting lower current reduces conductors’ cross-section requirements and overall distribution losses.
Both IBA and FPA are viable solutions to high-density embedded power systems. IBA generally can be considered a better fit with many relative low-power loads, as buck regulators can accomplish the task effectively and flexibly. Power systems benefit from FPA when the load is “power-hungry” silicon such as CPUs, memory, FPGAs, and ASICs. Designers should carefully evaluate system requirements and tradeoffs to choose the architecture that best suits their needs.
- W. Tabisz, M. Jovanovic, and F. Lee, “Present and future of distributed power systems,” in Proc. Applied Power Electronics 1992 Conf., Boston, Mass., pp.11-18.K
- M. Sayani and J. Wanes, “Analyzing and determining optimum on-board power architectures for 48V-input systems,” Proc. Applied Power Electronics 2003 Conf., Miami Beach, Fla., Vol. 2, pp.781-785
- Ashraf Lotfi, “Intermediate-bus architecture solves power-conversion challenges,” EE Times Design, October 2004
- Rais Miftakhutdinov, “Improving System Efficiency with a New Intermediate-Bus Architecture,” Texas Instruments Inc. Seminar, 2009
- V∙I Chip Inc., “Configuring Vicor IBC Converters with niPOLs”, Application Note AN:001, Andover, Mass., 2011
- V∙I Chip Inc., “Factorized Power Architecture and V•I Chips”, White Paper, Andover, Mass., May 2007
- Paul Yeaman, “Datacenter Power Delivery Architectures : Efficiency and Annual Operating Costs,” Darnell Digital Power Forum, September 2007