Integrated photonics won’t replace or supplant chips but will soon become an essential technology to maintain the steady performance increases in computing we’ve come to take for granted, says Martijn Heck, scientific director of the recently launched Eindhoven Hendrik Casimir Institute.
If Moore’s Law isn’t dead already, it will die soon. Or so we’re led to believe. Well, if the electron is running on its last legs, shouldn’t we find another particle to exploit our computational needs? The photon seems an obvious choice, partly because there’s not much to choose from, and partly because it seems perfect for the job. Photons zoom relatively unimpeded through many materials, and less resistance means less heat dissipation. The existence of photonic integrated circuits prove that we already know how to manipulate light particles.
Integrated photonics as the successor to CMOS-based electronics seems perfectly reasonable at first glance. Even someone with a fair command of physics could believe it. But it’s simply not going to happen. The notion keeps popping up nonetheless. Sometimes politicians, lobbyists or journalists are under the impression that it could or will happen, perhaps because the researchers describing the potential of integrated photonics to them dumbed it down a little too much.
Presenting electronics in this manner, as something on its way out, understandably hits a nerve in the electronics community. Just ask Bram Nauta, professor of IC design at the University of Twente.
As a result, integrated electronics and integrated photonics sometimes end up being pitted against one another. But there’s absolutely no reason why they should be, says Martijn Heck, professor of photonic integration at Eindhoven University of Technology (TUE) and scientific director of the newly minted Eindhoven Hendrik Casimir Institute (EHCI).
Here’s a quote from the press release announcing the launch of the EHCI: “The end of traditional scaling in communications and computing comes in sight. To continue the trend in computational power and energy-efficient communication, emerging photonics and quantum technologies are the future.” Isn’t that the kind of statement that irks the electronics community?
Heck: “It shouldn’t. Transistors are great and electronics is the most advanced technology ever conceived by man, but let’s not be naive: there’s a limit to scaling. In fact, we’ve bumped into a number of limits already. It just so happens that, so far, the chip industry has managed to come up with innovative solutions to keep increasing computational power. For example, when thermal issues prevented further increases in clock speed, multicore computing was introduced. There’s nothing wrong with that, but I think it’s fair to say that the end of traditional scaling is approaching, and alternative approaches are needed to pick up the slack. Deployment of integrated photonics is one such approach.”
How will integrated photonics support this ‘modern scaling’?
“As the amount of data that chips need to process grows, so does the need to move it. It’s getting increasingly difficult to transfer all that data over copper wires. A bandwidth bottleneck is looming, and there’s a power problem. Already half of a chip’s power is spent on data transfer, and that won’t be solved by scaling transistors. So, there’s a wall in sight, and that’s making photonics-based I/O an appealing, perhaps unavoidable alternative.”
“This is simply a continuation of a long-standing trend. Initially, fiber-optic cables were used for long distances only, to connect countries and cities. Now, it’s quite common to use fiber to connect individual homes and buildings. Within data centers, fiber optics has taken over server-to-server connections. The next logical step is chip-to-chip. Since the demand for bandwidth keeps growing, it’s inevitable that photonic links will be used over increasingly short distances. Maybe one day even intra-chip.”
How about computation? Can photons be used to compute?
“Certainly. The question is: can photonic chips be competitive with CMOS? This is an important research topic at the EHCI. We want to find out whether we can build competitive photonics-based neuromorphic AI chips. By competitive, I don’t mean that they would replace electronic processors. They would serve a complementary role for specific tasks, similar to GPUs for graphics. In the future, CPUs may divert more tasks to specialized processing units.”
The EHCI also focuses on quantum technology. Is there any synergy between integrated photonics and quantum technology?
“Photonics will definitely play an important role in advancing quantum technology. Photonic quantum processors – which are based on photonic integrated circuits – have a lot of potential. For other quantum computing implementations, photons are used to read out and manipulate qubits. To build a full-fledged quantum computer with thousands or even millions of qubits, the control electronics will have to be scaled. Integrated photonics is therefore a key enabler for quantum technology.”
“It’s a two-way street, though. Integrated photonics is constantly looking to make smaller and more energy-efficient components. If we extrapolate, we’ll eventually end up with signals consisting of a mere handful of photons. At that point, quantum effects will need to be taken into account for reliable operation.”