Power System Design for Multiple Voltage Levels

🔌 The state-of-the-art power system in the data centre utilises 400 V AC connected to the MV grid via a low-frequency transformer (LFT) and distributed power factor correction (PFC) rectifiers at the rack level, achieving an overall efficiency of approximately 97.1% from MVAC input to the rack-level 400 V/48 V DC-DC conversion. Increasing the AC distribution voltage to 690 V may enhance the overall efficiency to about 97.8 % due to reduced distribution losses, as losses in identical busbars decrease with the square of the voltage. PFC rectifiers suitable for 690 V AC can be employed with three-level topologies, maintaining high conversion efficiency. Alternatively, an 800 V DC (±400 V DC) distribution system can result in slightly lower distribution losses than the 690 V AC system. Additionally, there are other advantages to DC, such as the straightforward and efficient integration of battery energy systems.   💡 In principle, three conceptual approaches to MVAC-LVDC conversion can be considered. The first involves retaining the LFT and centralising the PFC rectifier functionality with a high-power SiC unit. This approach achieves an MVAC-LVDC conversion efficiency of approximately 98.2 % and an overall efficiency of around 97.9 %, with an estimated power density of about 0.25 kW/dm³. The second option employs robust 12-pulse rectifier systems complemented by active filters (AFs) to achieve power factor correction, forming a hybrid transformer. This partial-power-processing technique enables a high MVAC-LVDC conversion efficiency of approximately 98.5 % and an overall efficiency of about 98.2 %, with a power density estimated at 0.22 kW/dm³. Finally, solid-state transformers (SSTs) with medium-frequency transformers (MFTs) represent a fully controllable option. Current MVAC-LVDC SST prototypes have demonstrated full-load efficiencies of around 98 %, possibly reaching 98.5%, resulting in an overall efficiency of approximately 97.7 % or 98.2%. However, the power density of the overall SST system based on modular topologies tends to be comparatively lower than that of the hybrid transformer solution, despite the very high power density of the modules. #solidstate #powerelectronics #datacenters #lowvoltage #directcurrent #efficiency #powerdensity

Data centers are rapidly becoming a major driver for DC power distribution and DC microgrids. Hyperscale facilities consume 20 to 100MW each, wit most of that power ultimately delivered at ~1V at the point-of-load for xPUs and memory. To improve efficiency and to reduce distribution cost, designers push voltage levels upward, with the conversion to 1V as close to the silicon as possible. Hereby, the power conversion system architecture is of crucial importance! Across the industry, facility-scale DC distribution is converging on either +/-400VDC (Google, Meta, Microsoft) or 800VDC (NVidia). In a future architectures, these DC buses will likely be fed from the medium voltage AC grid via solid-state transformers (SSTs). Today, the power delivery from 800V to 1V is envisioned to move from 800 V → 48 V → 12 V → 1 V. A rack-level conversion from 800V to 48V, is followed by a tray- or GPU card-level conversion from 48V to 12V. The final conversion from 12V to 1V happens on the GPU card, as close to the silicon as possible. Exact voltages may vary a bit. This structure evolved from traditional AC-fed architectures. However, it has two big drawbacks: it still requires substantial copper at rack- and tray-level and it has multiple conversion stages. Both add loss and cost. Skipping a stage—for example, jumping from 800 V directly to ~12 V—sounds attractive, but creates challenges for converter semiconductors and magnetics. A multi-module series architecture may be more promising! On the high-voltage side, modules connect in series, naturally dividing the input bus (e.g., 800 V into ~100 V segments). Each module converts directly to 12V, a much more favorable design point for both semiconductors and magnetics. These modules can be integrated directly on the GPU board, minimizing the amount of copper needed to transport power within a rack. A series architecture taps into low-voltage power devices which are more more efficient and more reliable than high voltage ones. Moreover, power converter transformer ratios are less extreme which simplifies magnetics. On the flip side, a series architecture requires a more complex communication and control. But embedded digital control, and high-speed communication are becoming inexpensive, making the control challenge solvable. Power system design is ultimately about managing the “conservation of misery”. Design challenges remain, but you can choose where the burden sits. The arrival of smart, all-digital power modules unlocks new possibilities to redistribute that burden more intelligently. #DC, #800V, #microgrids, #datacenters, #nvidia

Everyone in the industry knows this number. Very few have ever asked why. Why are DC BESS systems almost always limited to 1500 VDC? 1500 VDC is not a standard. It’s a boundary. And it defines why PCS systems land around ~690 VAC. —---- If you’ve worked on utility-scale solar or BESS, you’ve seen this everywhere: → 1500 VDC battery / PV strings → ~690 VAC PCS output It looks like convention. It’s not. It’s the result of two independently established voltage ceilings — shaped by physics, standards, and economics. —-- ⚡ 1. The 1500 VDC boundary (DC side) At first glance, higher voltage is always better: → Lower current → Lower I²R losses → Smaller cables So why stop at 1500 V? 👉 Because 1500 VDC is effectively the upper limit of “low-voltage DC” in practical system design. And that matters. At this level, you still have: ✔ Off-the-shelf components (fuses, breakers, contactors, inverters) ✔ Standardized certification paths ✔ Manageable insulation and clearance requirements ✔ Established supply chains This boundary is reflected across standards: IEC 61730 / UL 61730 — PV module safety (extended to 1500 VDC) IEC 62109 / UL 62109 — converter safety envelope NFPA 70 (NEC) Article 690 — ≤1500 VDC avoids MV treatment —-- 👉 Go beyond 1500 VDC, and you leave that world: Go beyond 1500 VDC — and you’re no longer optimizing… you’re redesigning the entire system. → Fewer standardized components → Custom or limited equipment availability → Larger creepage/clearance distances → More complex insulation coordination → Harder and more expensive certification 💰 That’s why the industry moved from 1000 V → 1500 V: real BOS savings (~$0.05/W), with fewer cables, combiners, and devices — without stepping into a completely different design regime. —-- 🔌 2. The ~690 VAC boundary (AC side) On the AC side, PCS outputs typically land around 400–690 VAC (3-phase). Again — not arbitrary. 👉 690 VAC sits near the upper bound of low-voltage AC systems. Defined by: IEC 60038 — standard nominal voltages (400/690 V) EU Low Voltage Directive — applies up to 1000 VAC IEC 62109 / UL 1741 — certification envelope This keeps the AC side within: ✔ Mature switchgear ecosystem ✔ Widely available protection devices ✔ Lower certification complexity ✔ Limits of DC voltage that make DC/AC conversion efficient —-- 🔄 3. How this defines the PCS envelope This is the key connection: DC (≤1500 VDC) → PCS → AC (≤690 VAC) → MV transformer → grid Why this pairing works: ✔ Efficient conversion ratio ✔ Compatible with semiconductor voltage classes ✔ Keeps both sides within low-voltage design space ✔ Enables standard MV step-up integration —-- 🧠 4. The real takeaway These values are not arbitrary. 👉 They define the boundary where systems can still be built with standard components, known clearances, code compliance and scalable economics Two independently derived limits - One tightly integrated system.

Data center power demand is continuing to rise rapidly. Facilities are getting larger as more and more compute is needed to support our modern digital world, and as AI training becomes more power intensive. However innovations in server power distribution architectures could greatly increase server power density while improving data center efficiencies. Both Google and Nvidia are proposing new server power architectures centered around 400V DC and 800V DC power distribution systems respectively. These servers could consume a 1 MW of power per rack. This design would minimize conversion losses, reduce the number of power supply units (PSU)s needed, and allow for optimization of server rack layouts and future scalability. The 400V DC design was presented at the 2024 Fall Open Compute Project (OCP) Summit. For the near term, the design would utilize a Sidecar system with an off-rack PSU and Battery Backup Units (BBUs) for AC/DC conversion. The Sidecar system would bring the PSU and BBU into its own dedicated cabinet improving efficiency. Eventually the aim is to have a single building level AC/DC converter which could utilize a solid state transformer. The 400V DC design uses equipment designed for the EV industry, has lower isolation requirements, and 2x wire for power delivery. Google, Microsoft, and Meta are all collaborating on the effort to produce the Sidecar system with a project known as Mount Diablo. Meanwhile Nvidia is proposing an 800V DC architecture. This design would require lighter cables and components and have less heat loss along with better power density. In both design the power is distributed on the DC network which is then stepped down by DC/DC converters at the rack level. One challenge at such high power densities is cooling. New cooling systems need to be designed to manage the heat from a 1 MW server rack and the use of liquid cooling will be imperative. The Project Deschutes coolant distribution unit (CDU) is a new cooling system design with redundant components and uninterruptible power supplies (UPS) for high availability. Some challenges exist with the new 400V DC designs however. The higher voltages bring safety concerns, equipment compatibility challenges, and additional cost and complexity. New product designs will be needed to interface with these 400V DC systems. One interesting possibility is powering data centers with a microgrid that directly taps into the 800V DC network simplifying backup power architectures. For those of us in the power industry these new designs could require updated models for dynamic and EMT studies. Also we should think about what does it mean for the grid to have very dense high power loads in a single building compared to a spread out campus? Does it change any aspect grid management or facility design? You can learn more about these new designs from the OCP presentation here: https://lnkd.in/g-eZYaME