The development of photonic integrated circuits will transform supercomputing and the semiconductor industry in ways that are important strategically and commercially.
Excerpted from The Future Postponed, Massachusetts Institute of Technology, 2015
Michael R. Watts: Associate Professor of Electrical Engineering
Optical fibers have dominated long-distance telecommunications for decades, because light—photons—can transfer much more data at lower cost than electrons in copper wires. A typical optical fiber, for example, can transmit as many as 100 wave- length channels each carrying 100 Gigabits of information—and each channel more than 10 times as much as a copper wire. Moreover, the optical fibers can transmit signals with little distortion or loss over many kilometers. Now optical technology, in the form of photonics or photonic integrated circuits, is poised to move into computing in a big way—an opportunity with major strategic, commercial, and environ- mental advantages for the country that leads in this technology.
This transition is being driven by advances in the development of photonic chips—photonic integrated circuits (PICs) made from silicon and indium phosphide that can now be reliably
and inexpensively produced—but also by the need for ever faster supercomputers and the ever-expanding Internet data centers. In fact, internal data transfer speeds are becoming the limiting factor for both supercomputers and data centers. A data center is, in effect, a massive network that interconnects millions of storage nodes and shuttles data back and forth, an activity that, with copper wiring, also consumes huge amounts of electric power. Already such centers can consume as much as 100 megawatts of power each, enough power for a small city, and all U.S. data centers together account for about 2 percent of national electric power consumption, a number that is growing rapidly. In supercomputers, the internal data communication networks will account for nearly all of the power consumption—and the cost—of these machines within another five years. So switching from copper wires to photonic circuits, which can operate at much higher bandwidth with lower power consumption and lower costs, makes a lot of sense.
But the opportunity does not stop there. Even within a given storage node, or server “blade”, of a data center, the wires that connect processors to memory will face similar constraints and will need photonic solutions. Ultimately photonic circuits may be needed even at the intra-chip level, within microprocessors. Super-computers face similar constraints to those of data centers, since they are massive networks of processing nodes that need to share data across the network rapidly and accurately. Here, too, photonic circuits are already needed to enable faster data transfer, especially for still larger computers that can solve complex problems more effectively. And in supercomputers too, data transfer capacity between processors and memory will become a growing constraint. The inevitable transition to photonic circuits is likely to completely reshape much of the $300 billion semiconductor industry, now dominated by U.S. companies.
￼It is not clear that the U.S. can maintain a leadership or even a competitive role in supercomputers, despite their strategic importance, or ultimately even in the semiconductor industry.
Photonics has implications beyond super-computing and datacenters in all kinds of measurement and sensing applications. One example is microwave photonics, where applications to radar signal processing on aircraft could be important. Radar units are typically located at numerous locations on an aircraft, and the signals from these units collected for central processing and interpretation. Transmitting these signals with photonic circuits and optical fibers instead of wires offers greater signal fidelity, lower power consumption, and greater freedom from interference or jam- ming. Photonic-based clocks and oscillators also have greater stability and accuracy than their microwave counterparts. Optical sensing and distance-ranging techniques imbedded
in photonic circuits are finding their way into applications of autonomous and semi-autonomous vehicles and robots, enabling capabilities that will dramatically improve both safety and
productivity. By 2025, nearly every new vehicle on the road could have silicon photonic-based chips providing three-dimensional information about a vehicle’s surroundings.
Historically, the U.S. has led in the field of pho- tonics. Now both Europe and Japan have much larger R&D programs in this space. The recent photonics initiative announced by President Obama proposes to increase funding by about $20M a year for photonics manufacturing, but does not adequately address the need for more fundamental work—such as intra-chip communications, quantum, and ultrafast optics— needed for continued progress. Such research has historically been carried out by universities, which also train the next generation of talent for the field, while industry has typically built on that research and focused more on manufacturing. So as things stand, it is not clear that the U.S. can maintain a leadership or even a competitive role in data centers or supercomputers, despite the strategic importance of the latter, or ultimately even in the semiconductor industry, as it shifts to incorporate photonics into electronic chips.