Next-generation data centers are within reach thanks to new energy-efficient switches

Data centers—specialized spaces for storing, processing, and distributing data—enable everything from cloud computing to video streaming. In the process, they consume large amounts of energy as they transmit data back and forth within the hub. As the demand for data grows exponentially, so does the demand for data centers to become more energy efficient.

Data centers house servers, powerful computers that talk to each other through interconnects, which are physical connections that allow data to be shared. One way to reduce energy consumption in data centers is the use of light to transmit information with electrical control optical switches control of the flow of light, and therefore of information, between servers. These optical switches must be multi-functional and energy-efficient to support the continuous expansion of data centers.

In a newspaper published online on July 4 in Nanotechnology of naturea team led by scientists at the University of Washington reported the development of an energy-efficient silicon-based non-volatile switch that manipulates light using a phase-change material and a graphene heater.

“This platform really pushes the boundaries of energy efficiency,” said co-author Arka Majumdar, a UW associate professor of physics and electrical engineering. computing technology, and faculty at the UW Institute for Nanoengineered Systems and the Institute for Molecular and Engineering Sciences. “Compared to what is currently used in data centers to drive photonic circuits, this technology would significantly reduce the energy needs of data centers, making them more sustainable and environmentally friendly.”

Silicon photonic switches are widely used in part because they can be manufactured using well-established semiconductor manufacturing techniques. Traditionally, these switches have been tuned using the thermal effect, a process in which heat is applied (often by passing a current through a metal or semiconductor) to change the optical properties of the material in the switch and thus alter the path of light. However, not only is this process not energy efficient, but the changes it produces are not permanent. As soon as the current is turned off, the material reverts to its previous state, and communication and the flow of information are broken.

To solve this problem, a team that includes researchers from Stanford University, Draper Charles Stark’s lab, the University of Maryland and the Massachusetts Institute of Technology has created a set-it-and-forget-it switch capable of maintaining a connection without additional power. They used a non-volatile material with a phase change, meaning that the material is transformed by brief heating, and it remains in that state until it receives another heat pulse, after which it returns to its original state. This eliminates the need to constantly add energy to maintain the desired state.

Previously, researchers used doped silicon to heat the phase change material. Silicon itself does not conduct electricity, but when selectively doped with various elements such as phosphorus or boron, silicon can both conduct electricity and scatter light without excessive absorption. When a current is pumped through the doped silicon, it can act as a heater to switch the state of the phase change material on top of it. The catch is that it’s also not a very energy efficient process. The amount of energy required to switch the phase transition material is similar to the amount of energy used by traditional thermo-optical switches. This is because the entire 220 nanometer (nm) thick layer of doped silicon must be heated to convert just 10 nm of the phase change material. It takes a lot of energy to heat such a large volume of silicon to switch a much smaller volume of phase change material.

“We realized that we needed to figure out how to reduce the volume that needed to be heated to improve the efficiency of the switches,” said lead author and co-author Zhuoran (Roger) Fang, a UW doctoral student in electrical and computer engineering. mechanical engineering.

One approach is to make a thinner silicon film, but silicon does not scatter light well if it is thinner than 200 nm. So instead, they used a 220nm undoped silicon layer to propagate light and introduced a graphene layer between the silicon and the phase material to conduct electricity. Like metal, graphene is an excellent conductor of electricity, but unlike metal, it is atomically thin—it consists of just a single layer of carbon atoms arranged in a two-dimensional honeycomb lattice. This design avoids wasting energy by directing all the heat generated by the graphene to change the material with a phase change. In fact, the switching energy density of this setup, calculated by dividing the switching energy by the volume of the switching material, is only 8.7 atajoules (aJ)/nm³, a 70-fold reduction compared to widely used doped silicon heaters, current state-of-the-art equipment. This is also within one order of magnitude of the fundamental switching energy density limit (1.2 aJ/nm³).

Although using graphene to conduct electricity causes some optical loss, which means absorbing some light, graphene is so thin that not only is the loss minimal, but the phase material can interact with the light propagating through the silicon layer . The team found that a graphene-based heater can reliably switch state of the phase change material for more than 1000 cycles. This is a marked improvement over doped silicon heaters, which have been shown to last around 500 cycles.

“Even 1,000 is not enough,” Majumdar said. “Practically speaking, we need about a billion endurance cycles, which we’re currently working on.”

Now that they’ve demonstrated that light can be controlled using a phase-change material and a graphene heater, the team plans to show that these switches can be used to optically route information through a network of devices, a key step toward their use in data centers. They are also interested in applying this technology to silicon nitride for routing single photons for quantum computing.

“Being able to tune the optical properties of a material with just an atomically thin heater is a game changer,” said Majumdar. “The exceptional performance of our system in terms of energy efficiency and reliability is truly unheard of and could help advance both information technology and quantum computing.”

Additional co-authors include UW electrical and computer engineering students Rui Chen, Jiajiu Zheng and Abhi Saxena; Asir Intisar Khan, Kathryn Neilson, Michelle Chen, and Eric Popp of Stanford University; Sarah Geiger, Dennis Callahan, and Michael Moebius of the Charles Stark Draper Laboratory; Carlos Rios of the University of Maryland; and Juejun Hu of MIT.