Power Tips #152: Design considerations and topology comparisons for 48V intermediate bus converters

Increasing power demands in data centers demand high-efficiency, high-density power-conversion solutions.

Figure 1 shows a block diagram of power distribution inside an IT tray. A 48V bus bar goes down the back of the rack to distribute power to the IT trays. Inside each tray is hot-swap or e-fuse circuitry to limit inrush current during tray plug-in and to protect the upstream rack during tray failures. Intermediate bus converters (IBCs) convert 48V to the second-stage voltage, usually 12V or 6V. Final-stage multiphase buck voltage regulators complete power delivery by converting the second-stage voltage to the loads, with the majority of power going to sub-1V, high-current processors. In this edition of Power Tips, I will focus on the 48V IBC, covering design considerations, comparing topologies, and discussing system trade-offs of various approaches.

Figure 1 48V IT tray power distribution. Source: Texas Instruments

The IBC power distribution network offers a wide range of power-conversion approaches inside an IT tray (Reference 1). As the system architect, you have three main design choices:

A modular or discrete solution (also known as chip-down design).
Regulated, unregulated (also known as fixed ratio) or semiregulated IBC operation.
The second-stage bus voltage to maximize system performance.

When selecting a modular or chip-down design power converter, your main trade-off will be power density vs. board design flexibility. Power modules, as shown in Figure 2a, are highly optimized solutions built on high-layer-count printed circuit boards (PCBs) (usually more than 16), offering prequalification and the highest power density. The drawbacks of power modules are a lack of flexibility, with fixed footprints and set features, as well as a higher cost per watt.

Chip-down designs, as shown in Figure 2b, are highly flexible solutions that offer footprint and feature freedom, with a lower cost per watt in high-volume production. Their drawbacks include longer upfront time and greater cost investments to qualify the design.

(a)	(b)

Figure 2 48V IBC design examples of modular (a) and chip-down design (b) approaches. Source: Texas Instruments

When considering the output regulation of the IBC, your choice depends on two main factors: the load being powered and the operating range of the IBC’s input bus voltage. When the IBC directly drives 12V loads such as cooling fans, hard drives and Peripheral Component Interconnect Express cards, only a fully regulated output voltage (Reference 1) will ensure component safety. In modern data centers, the tray voltage has a more stable, narrow range, typically 40V to 60V. This narrow input range gives you the option to use higher-efficiency and higher-power-density fixed-ratio or semiregulated IBCs. The regulated second-stage voltage regulators following the IBC stage can absorb fixed-ratio IBC output voltage fluctuations.

Your third design choice is the second-stage voltage delivered by the IBC. Equation 1 determines system efficiency (η_system):

η_system = η_IBC x η_PDN x η_VR

For a given power load, decreasing the second-stage bus voltage will lower the IBC efficiency (η_IBC), because it must deliver more current at a lower voltage to provide the same output power. Similarly, for the motherboard power distribution network (PDN), which distributes current from the first-stage IBC to the second-stage voltage regulator, the PDN efficiency (η_PDN) will also decrease because of increased I² x R ohmic losses. The benefit of a lower second-stage bus voltage is apparent when using final-stage, high-frequency, high-current voltage regulators with significantly reduced voltage-related switching losses. This results in higher second-stage efficiency (η_VR) and a potentially smaller size of the second stage.

Unlike a buck converter-dominated second-stage voltage regulator, a first-stage IBC has a wide range of power delivery approaches and thus a wider variety of power-conversion topologies available. In most modern IT applications, isolation for safety purposes is not required, so your power topology options increase further when you can consider transformerless options. Figure 3 shows four popular options for IBC module and chip-down designs.

The full-bridge converter shown in Figure 3a is a simple buck converter-derived transformer-isolated topology. The full-bridge converter’s strengths are ease of regulation and the ability to easily scale the intended output voltage by adjusting the transformer turns ratio for your chosen second-stage bus voltage. One drawback of the full-bridge converter is that transformer design is key to its performance, requiring a high-layer-count PCB that limits the topology to module-based designs. Another drawback of the full-bridge converter is that the primary devices are hard-switched, limiting power density and efficiency.

The transformer-isolated inductor-inductor-capacitor (LLC) converter shown in Figure 3b looks very similar to the full-bridge converter but uses an additional capacitor and two inductors to eliminate switching-related losses in the primary devices, enabling high efficiency and high power density (Reference 2). The LLC converter has the same transformer-related strength (an easily scalable output voltage) and weakness (it’s limited to module-based designs) as the full-bridge converter. The LLC converter operates with the highest efficiency at the resonant frequency set by the additional passive components (CR and LR), with efficiency decreasing as you move away from the resonant frequency to regulate the output voltage. For this reason, the LLC converter’s most common application in IBCs is fixed-ratio designs, always operating at the resonant frequency, ensuring the highest efficiency.

Two other popular topologies, the hybrid switched-capacitor (HSC) converter (Reference 3) shown in Figure 3c and the basic buck converter shown in Figure 3d, both offer benefits for chip-down designs because of their lack of AC-dependent power transformers. The HSC converter has a natural step-down ratio of 4-to-1, making it a strong candidate for high-efficiency 48V to 12V IBCs. The addition of flying capacitors limits the power density and hinders this converter’s operation in boost mode, making it a good fit for semiregulation, as regulating only occurs in step-down buck converter mode.

Because the HSC converter has a natural step-down ratio of 4-to-1, scaling the output voltage down further to an 8-to-1 6V_OUT design (for example) is more challenging than it would be for the full-bridge and LLC converter options because the HSC converter must rely instead on a longer freewheeling period, requiring a larger output filter inductor, decreasing power density and efficiency.

The buck converter is the most common topology in power electronics, used exclusively in the second-stage voltage regulator, so it is natural to want to apply this simple and well-known approach to the IBC stage as well. The challenge with using a buck converter in the higher-voltage IBC application is that the power devices experience the highest voltage and current stresses when compared to the other topologies, limiting efficiency and power density.

(a)	(b)

(c)	(d)

Figure 3 Popular IBC topologies: full-bridge converter (a); LLC converter (b); HSC converter (c); and buck converter (d). Source: Texas Instruments

Table 1 compares the different topologies and trade-offs.

	Full-bridge converter	LLC converter	HSC converter	Buck converter
Module or chip-down design	Module	Module	Both	Both
Regulation type	Regulated	Fixed ratio	Semiregulated	Regulated
Efficiency	Medium	High	High	Low
Power density	Medium	High	Medium	Low
Output-voltage scalability	High	High	Medium	Medium
Complexity	Medium	High	High	Low

Table 1 Comparing IBC topology characteristics.

With the maturation of gallium nitride (GaN) power devices (Reference 4), which have much lower switching-related charges compared to traditional silicon metal-oxide semiconductor field-effect transistors (MOSFETs), simpler topologies like the buck converter topology become more attractive and viable options for higher-voltage applications like IBCs. See Table 2.

	100V Texas Instruments GaN semiconductor	100V silicon MOSFET	Difference
V_DS (V)	100	100
R_DS(on) (mΩ)	1.1	1.7	35% lower
Q_G (nC)	27	106	75% lower
Q_OSS (nC)	98	205	52% lower
Q_GD (nC)	2.5	26	90% lower
FOM₁ = Q_G x R_DS(on)	29.7	180.2	83% lower
FOM₂ = Q_OSS x R_DS(on)	107.8	348.5	69% lower
FOM₃ = Q_GD x R_DS(on)	2.75	44.2	93% lower
Package (mm x mm = mm²)	4 x 6.5 = 26 FET with gate driver	5 x 6 = 30 Discrete FET	13% smaller

Table 2 Comparison of 100V GaN and silicon-based IBC semiconductor options.

The IBC power distribution network offers the widest range of power-conversion approaches of the systems inside an IT tray for good reason. As power requirements and architectures rapidly evolve, the best way to optimize performance for 48V IBCs changes. And as additional variables such as highly improved GaN semiconductors get thrown into the equation, it becomes even more important to understand design considerations, topology comparisons and trade-offs.

References

Hsu, C., L. Olariu, S. Zou, et al. “48V Onboard Power Solution Requirements.” Open Compute Project, Version 1.0.0, Nov. 15, 2024.
McDonald, Brent. “Overview of a planar transformer used in a 1kW high-density LLC power module.” Texas Instruments technical article, 2025.
Li, C., and J.A. Cobos. “A Switched Capacitor and Autotransformer Hybrid Converter With DC Current in the Windings,” in IEEE Transactions on Power Electronics 37 (2), February 2022, pp. 1870-1884.
Gallium nitride (GaN) power stages, Texas Instruments.

David Reusch is a systems engineer on the data center team at Texas Instruments, specializing in power electronics. David has more than 20 years of experience in power electronics, ranging from cutting-edge gallium nitride (GaN) technology to high-reliability space-grade DC-DC converters. He received his B.S., M.S. and Ph.D. in electrical engineering from Virginia Tech.

Source link