Every second, terabytes of data — the equivalent of downloading thousands upon thousands of movies at once — travel around the world as light in fiber-optic cables, like so many cars packed onto a super-fast highway. When that information reaches data centers, it needs a switching system, just as cars need traffic lights, to exit the highway in an orderly fashion.
Until now, the photonic switches used for routing optical signals have been hindered by a fundamental tradeoff between size and speed: Larger switches can handle higher speeds and more data but also consume more energy, occupy more physical space and drive up costs.
Speeding Up the Information Superhighway
In a new paper in Nature Photonics , researchers at the University of Pennsylvania School of Engineering and Applied Science (Penn Engineering) describe the creation of a novel photonic switch that overcomes this size-speed tradeoff. And at just 85 by 85 micrometers, the new switch's units are smaller than a grain of salt.
By manipulating light at the nanoscale with unprecedented efficiency, the new switch speeds up the process of getting data on and off the literal information superhighway of fiber-optic cables that encircles the globe. "This has the potential to accelerate everything from streaming movies to training AI," says Liang Feng , Professor in Materials Science and Engineering (MSE) and in Electrical and Systems Engineering (ESE) and the paper's senior author.
Quantum Mechanics Meets Optics
The new switch relies on non-Hermitian physics, a branch of quantum mechanics that explores how certain systems behave in unusual ways, giving researchers more control over light's behavior. "We can tune the gain and loss of the material to guide the optical signal towards the right information highway exit," says Xilin Feng, a doctoral student in ESE and the paper's first author. In other words, the unique physics at play allows the researchers to tame the flow of light on the tiny chip, enabling precise control over any light-based network's connectivity.
The upshot is that the new switch can redirect signals in trillionths of a second with minimal power consumption. "This is about a billion times faster than the blink of an eye," says Shuang Wu, a doctoral student in MSE and co-author of the paper. "Previous switches were either small or fast, but it's very, very difficult to achieve these two properties simultaneously."
Using Silicon for Scalability
The new switch is also notable for being made partly of silicon, the inexpensive and widely available industry-standard material. "Non-Hermitian switching has never been demonstrated in a silicon photonics platform before," says Wu. In theory, the incorporation of silicon into the switch will facilitate scaling the device for mass production and wide adoption in industry. Silicon is a key component in most technologies, from computers to smartphones; building the device using silicon makes it fully compatible with existing silicon photonic foundries, which make advanced chips for devices like graphics processing units (GPUs).
From Concept to Prototype
On top of the silicon layer, the switch consists of a particular type of semiconductor, made of Indium Gallium Arsenide Phosphide (InGaAsP), a material that is particularly effective at manipulating infrared wavelengths of light, such as those typically transmitted in undersea optical cables.
Joining the two layers proved challenging, and required numerous attempts to build a working prototype. "It's similar to making a sandwich," says Xilin Feng, referring to adding the layers to one another. Only, in this case, if any of those layers were misaligned by even a tiny degree, the sandwich would be entirely inedible. "The alignment requires nanometer accuracy," Wu notes.
Transforming Data Centers
Ultimately, the researchers say, the new switch will benefit not just academic physicists, who can now further explore the non-Hermitian physics upon which the switch depends, but companies that maintain and build data centers, and the billions of users who rely on them. "Data can only go as fast as we can control it," says Liang Feng. "And in our experiments we showed that the speed limit of our system is just 100 picoseconds."
This study was conducted at the University of Pennsylvania School of Engineering and Applied Science and supported by the Army Research Office (ARO) (W911NF-21-1-0148 and W911NF-22-1-0140), the Office of Naval Research (ONR) (N00014-23-1-2882) and the National Science Foundation (NSF) (ECCS-2023780, DMR-2326698, DMR-2326699 and DMR-2117775).
Additional co-authors include Tianwei Wu, Zihe Gao, Haoqi Zhao and Yichi Zhang of Penn Engineering and Li Ge of the City University of New York.
