New Materials For Computing

The U.S. Department of Energy rolled out a new program to develop materials for “extreme conditions” for high-performance computing, setting the stage for much more mobile versions of AI and machine learning.

This effort, if successful, has interesting implications on a number of levels. For one, the DOE’s mandate includes everything from energy security to weaponry, and high-performance computing will play an increasing role in both. In some cases, this will require recurrent neural networks and huge blocks of logic that are mobile, and until now that kind of computing has been confined to data centers.

And this points back to a big problem in high-performance computing that has never been adequately addressed. To maintain reliability, high-performance computers need to run within a narrow temperature range. Inside of corporate data centers, there is only so much cold air that can be blasted into densely-packed racks of blade servers. For mission-critical operations, the American Society of Heating, Refrigeration and Air-Conditioning Engineers—the group that sets the operating range for computer equipment—has set the operating temperature for data centers between 59° and 89.6° F (15° to 32° C), with relative humidity between 20% and 80%.

Fig. 1: Reliability of servers at different temperatures. Source: ASHRAE, 2011

Data center managers have been asking for equipment that can operate at higher maximum temperatures for at least the past 15 years, and server makers have been slowly increasing the safe operating temperature for their equipment. Current server guidelines from IBM and Dell show an operating temperature of between 5° and 40° Celsius (41° to 104° F) under continuous use. And there are restrictions for other equipment inside of data centers, as well, including power distribution units, data storage, uninterruptible power supplies, and a host of other equipment.

For data centers, this is an economic issue. The cost of cooling servers to stay within that upper limit is expensive, which is why data centers are being built in cooler windy areas, such as the Columbia River Gorge, and in places like Iceland, where outside air can be used to chill these systems.

Increasing operating temperatures would have significant effects on cooling costs and energy usage. The most recent estimates of data center power consumption are somewhere around 3% of the total energy being generated. In fact, some data center operations have been limited in size and scope by the amount of power available on the grid.

Changing the operating temperatures with the development of new materials would go a long way in making computers more robust. But it also could save huge amounts of energy everywhere. This is an effort that is long overdue, and it has implications for the design and manufacturing of chips used in the AI/machine learning systems for autonomous driving as well as inside of industrial operations such as power plants and steel mills.

The focus has been on getting more performance into chips, and more recently on reducing the amount of power necessary to achieve that performance. The next step is to make it ubiquitous, and that will require new materials, new methods and potentially new architectures.

Related Stories
New BEOL/MOL Breakthroughs?
Different materials, approaches for contacts and interconnects begin to surface for 7/5nm.
Next Challenge: Contact Resistance
New materials and tools are needed to solve an issue no one worried about in the past.
Interconnect Challenges Rising
Resistance and capacitance drive need for new materials and approaches.