Microsoft Exploring Fuel Cell-Powered Datacenters via the Server Rack
Sean James, Senior Research Program Manager
Global Foundation Services
Recently, together with colleagues in Microsoft Research, we published a research paper titled No More Electrical Infrastructure: Towards Fuel Cell Powered Data Centers that describes how we are taking an unconventional approach to power a datacenter entirely by fuel cells integrated directly into the server racks. This brings the power plant inside the datacenter, effectively eliminating energy loss that otherwise occurs in the energy supply chain and doubling the efficiency of traditional datacenters.
As a Senior Research Program Manager, I am responsible for datacenter design advancements through research and development. These R&D programs include unconventional, yet efficient approaches such as the Data Plant project in Cheyenne, Wyoming that demonstrates the first zero carbon datacenter by integrating the datacenter with a wastewater treatment plant. The Data Plant is a design that can utilize other sources of natural biogas or flare gas to reduce energy sourcing complexity.
The Data Plant research and this current study share three business requirements: the designs must improve service availability, reduce infrastructure costs, and meet our commitments to sustainability. This balancing act is challenging, because improving datacenter availability typically means adding more infrastructure, the antithesis of reducing costs.
This study takes our Data Plant concept a step further and explores how to collapse the entire energy supply chain, from the power plant to the server motherboard into the confines of a server single cabinet. Based on our models detailed in the paper, we show how integrating a small generator with the IT hardware significantly cuts complexity by eliminating all the electrical distribution in the grid and datacenter. This is where the fuel cell comes in. Fuel cells are not limited by the Carnot Cycle Efficiency limit that conventional generators are. Fuel cells are very clean, reliable and perfect for small form factor applications. By integrating fuel cells with IT hardware, we can cut much of the power electronics out of the conventional fuel cell system. What we are left with is a very simple and low cost datacenter and fuel cell system. As the fuel cell industry becomes more mature, especially small form factor fuel cells for automotive and IT applications, the cost of fuel cells will drop. You may end up with one someday delivering clean electricity and heat to your home.
What did we discover? Our research determined that an integrated server rack with IT equipment and smaller form fuel cells offered several improvements over traditional datacenter designs, including:
- Higher power availability through reducing points of failure. With the failure potential limited to a single rack and leveraging Microsoft's virtualization technology, our customers would experience zero impact in the event of an electrical disruption. We could then collapse the entire energy supply chain from the power plant into a single module.
- Lower infrastructure costs would be achieved through the elimination of electrical distribution, back-up, and transformation in the datacenter and power conditioning equipment in the fuel cell.
- Dramatic improvement in efficiency in not only the power utilization effectiveness (PUE), but also from the power plant to the motherboard resulting from placing a high efficiency fuel cell on the hardware. This would enable us to double the total system efficiency from the power plant to the chip.
- The first universal datacenter design can be achieved since methane is a fungible energy source, contrasted with the variety of grid frequencies and voltages around the nation and globe. This exact design can be mass produced and deployed anywhere in the world with access to methane. Yes, it's biogas ready, and it can also utilize solar and wind electrolysis. One of the huge challenges of mass-producing datacenters is the long lead time needed to purchase the electrical equipment. This equipment, usually custom, cannot be ordered until the design is adapted for the intricacies of the local grid, including voltage and frequency, extending build schedules, and thus hampering quick time to market.
The main distinction between this data plant concept and previous architecture ideas is the notion of bringing the power plant inside the datacenter, instead of putting the data center in the power plant. A lot of energy is lost in today's datacenter energy supply chain. To illustrate, let's consider the life of one watt of energy potential destined for a server in the conventional grid system.
First, chemical energy in the fuel is converted to thermal energy in the combustor, then converted to mechanical energy in the engine, then magnetic energy in the alternator, then converted to electrical energy alternating from positive voltage to negative voltage 60 times a second, converted back to magnetic energy, then back to electrical energy in the transformer to convert to high voltage at the plant substation, transmitted hundreds of miles, transformed four more times to various voltages from the grid to the back of the server, converted from an alternating current (VAC) to a direct current (VDC) in the server power supply unit (PSU), then finally to perform the real work at the chip. At the chip level, we are left with less than .2 watts of the original unit of energy or over 80 percent of the energy is lost from the power plant to the server!
In the new datacenter design approach outlined in our paper, chemical energy is first converted to direct current electrochemically and sent a few feet to the server power supply. With our one watt of energy we are now getting almost .4 watts or double the efficiency of traditional datacenters.
Additionally, natural gas has the highest energy-to-carbon ratio of any hydro carbon fuel. It is simply a cleaner alternative to coal and diesel, and by leveraging fuel cell technology we can achieve even lower emissions per watt. The electrolytic process in fuel cells produce no diesel particulate matter, nitrous oxide, or sulfur oxide. While there is still a CO2 waste stream, the output is significantly lower and cleaner than other power generation technologies.
With the potential to double the efficiency of traditional datacenters, we see tremendous potential in this approach, but this concept is not without challenges. Deep technical issues remain, such as thermal cycling, fuel distribution systems, cell conductivity, power management, and safety training that needs to be further researched and solutions developed. But we are excited about working to resolve these challenges. This is only the early stages in our exploration of this concept and I look forward to sharing updates in the future. For more information on our best practices for greener datacenters, please visit our team's web site at www.microsoft.com/datacenters.