The new horizon of optical computing
Modern business and warfare technologies demand vast flows of data, which pushes classic electrical circuits to their physical limits. Computer designers increasingly look to optics as the answer.
Hybrid networks that blend optical and electronic data move ever closer to the promise of optical computing as scientists and systems designers continue to make incremental improvements.
By Ben Ames
Modern business and warfare technologies demand vast flows of data, which pushes classic electrical circuits to their physical limits. Computer designers increasingly look to optics as the answer. Yet optical computing — processing data with photons instead of electrons — is not ready to jump from lab demonstrations to real-world applications.
Fortunately, there is a middle ground — engineers can mix optical interconnects and networking with electronic circuits and memory. These hybrid systems are making great strides toward handling the torrents of data necessary for new applications.
The trend began at the biggest scales. Fiber optics have replaced copper wiring at long distances, such as communications trunks between cities. More recently, engineers have also used optical networking to link nearby buildings. And with the introduction of a new parallel optics technology called VCSEL — short for vertical cavity surfacing emitting laser — they have even used optics to connect computer racks inside the same room. VCSEL now connects routers, switches, and multiplexers.
But the trend has stalled there. As systems designers use optics on ever-smaller applications, the next step should be to use them on PC boards and backplanes. And theoretically, the step after that would be to build computer chips that run on photons instead of electrons. Such a chip would be free of electrical interference, so it could process jobs in parallel and be blindingly fast. But experts agree it is still decades away from reality.
"At the backplane level, it's still electric," says Dave Dolfi, optical interconnects technologies department manager at Agilent Labs in Palo Alto, Calif. "Within four or five years, optics will replace that. And within another five years, optics will replace electrical connections between boards, and maybe between chips. But as far as optical computing — replacing processing or memory with optics — I'm not convinced that will ever happen."
That is primarily because of cost rather than technology, he says. Existing electric DRAM (dynamic random access memory) technology is so good that it represents "a very high bar to get over" before people would abandon the approach for something new, Dolfi says.
Most current research in this area is in optical networking. Dolfi is working on optics within the box, especially for projects in military and aerospace applications sponsored by the U.S. Defense Advanced Research Projects Agency (DARPA), he says.
Still the problem remains: faced with massive data throughput, classic electrical circuits and interconnects have weaknesses; they are power-intensive, leak electrons, and are vulnerable to radiation interference. At the highest levels of data flow, the only advantage of electronic design is its low cost.
So military designers say they are excited about optical networking because optics consume less power than electric, he says. Yet they have not been able to take advantage of that benefit until recently, because the optic/electric and electric/optic conversion was too inefficient.
They can finally do it today, because of two trends. First, electrical interconnects are demanding increasing amounts of signal processing to preserve the huge amount of data they carry, making optical options look better by comparison. Second, fiber optic technology has reduced power consumption, so optics now use less power than electric connections.
Military planners also like optical interconnects because they are nearly immune to electromagnetic radiation (EMI), says John Stratton, aerospace and defense program manager in Agilent's Santa Rosa, Calif. offices.
Modern warfighting depends on increasing volumes of data flow, as every vehicle — or even every soldier — is networked to the others for greater situational awareness.
"But if you're on a battlefield or an aircraft carrier or near a radar, the radiation can degrade the signal so much you have to retransmit it," he says. "Another strength of optical interconnects is that they're particularly good in a noisy environment."
Military designers also like optical networking because it offers great security, making data difficult to intercept.
That is particularly true for wireless optics — free-space systems that exchange information with lasers rather than with fiber-optic cables, Stratton says. Unlike radio broadcasts, which can be overheard by anyone in the area, free-space optical links go point to point. So a spy would have to stand between the sender and receiver to hear the signal. And by doing so he would reveal his presence.
Satellites use such systems today to communicate with each other. For extra security, they use a frequency range that cannot penetrate the Earth's atmosphere, he says. They use a separate, high-frequency signal to talk to their terrestrial controllers. A spy would have to be floating in space to overhear the signals.
The difficulty with free-space optics is that they must be very precise. "To make it work, you need a sophisticated tracking system," Stratton says. "The question in RF (radio frequency) is how big is the aperture or dish? But a laser has to hit its target exactly, or it's just zero signal."
Another potential military application for free-space optical networks would be on-demand local area networks (LANs) on the battlefield. Such a system would channel data through a backbone of aircraft and ships, but would still rely on satellites, since it is very difficult to track a moving aircraft with enough precision to uphold a laser link.
Global positioning satellite (GPS) receivers communicate with satellites today, but they are passively listening to broadcast signals from a range of sources. An optical network would have to track specific satellites with great precision. Engineers most likely would tackle that problem with similar technology to what laser-guided weapons use today, he said.
Cost slows new adoptions
The downside to wire-based optical networking is its cost. Optical interconnects are more expensive than electronic interconnects, Dolfi says. For long-distance high-bandwidth use, the investment is worthwhile, yet for short distances of only tens of meters, the costs can be three to five times as much.
"That's an improvement, since it used to be an order of magnitude more expensive," Dolfi says. "But it's still expensive if you don't need the performance. For instance, the computer market is extremely cost driven, so optical has its work cut out for it to get the price down."
The best way to reduce cost is through the lasers that generate the signals.
"Until recently, it's been done with single-channel, serial links. But with parallel optics, you need a widespread adoption of laser arrays," Dolfi says. "To some extent wavelength division multiplexing does this, but that's all on one board. So people have to learn to wield a large numbers of lasers, and that's a relatively new challenge; they haven't had a commercial incentive to do it before."
Once the commercial sector learns to generate low-cost laser arrays, military designers will choose optics for its obvious benefits: security, bandwidth, lightweight and EMI immunity.
Bandwidth drives applications
In the meantime, bandwidth is driving existing applications of fiber-optic networking, agrees Earle Olson, product manager for ruggedized optics in the Global Industrial & Commercial Business Unit at Tyco Electronics in Portsmouth, N.H.
As naval, ground-based, airborne, and commercial avionics designers seek faster and lighter designs, they are turning to gigabit Ethernet, a fiber optic short-range (500 meters), high-bandwidth (1,000 megabits per second) LAN backbone.
One of the first affordable backplane optical interconnects was Agilent Labs' PONI platform (above). This parallel optics system achieves high-capacity, short-reach data exchange by offering 12 channels at 2.5 GBd each.
The telecommunications industry primarily drives applications of such relatively low-cost interconnects and transceivers, specifically for data exchange, says Dennis Renaud, director of global engineering and product management in Tyco's Fiber Optics Business Unit.
The latest applications are in commercial avionics, where designers use optical networks as a common backbone to carry data throughout the airplane, Olson says. The sensors and wiring are still electronic, but can trade data as long as they have the right connectors.
Such applications will happen first in the commercial world, since technical committees can agree on common standards, such as ARINC. But military products are typically unique, so they cannot communicate with each other, he says.
Creating a hybrid computer
In fact, DARPA researchers may have a solution to that problem. They are continuing the trend of replacing copper conduits with fiber optics at ever-smaller scales.
One research program on chip-scale WDM (wavelength division multiplexing) has the goal of developing photonic chips, says Dr. Jagdeep Shah of DARPA's Microsystems Technology Office.
Today's optical interconnects rely on components placed on different boards, so optical fiber connects the laser, modulator, multiplexer, filter, and detector, he says. That takes up lots of space and power. Instead, he is trying to fit several components on a single board. Such a chip would be very attractive for airplane designers, since it would save size, weight, and power. It could make a particularly big difference on a plane like the U.S. Navy EA-6B Prowler electronic warfare jet, which is packed with electronics for radar jamming and communications.
One major challenge in that application is "format transparency," Shah says. "Usually fiber optics transport digital data in ones and zeros, but many military sensors generate analog data. So our goal is to develop components to handle either one."
The next challenge will be integrating those components at a density of 10 devices per chip, which is an order of magnitude improvement over current technology. That will be hard to do because energy loss and reflection can easily degrade laser quality.
DARPA engineers have also founded a research program on optical data routers. Any optical interconnect includes an intersection where many fibers come together at a node, which must act a like a traffic cop to steer various signals to their goals. Electronic routers from companies like Cisco and Juniper currently do that job. These routers are very precise, but have limited data capacities, Shah says.
The group's goal is to create an all-optical dataplane, so the device no longer has to convert data from electrical to optical and back again. Such a device would combine the granularity of electronics and scalability of optics, he says.
That type of optical logic gate would let engineers do nonlinear processing of signals without converting them, says Dr. Ravindra Athale, also in DARPA's Microsystems Technology Office.
That would be a critical achievement because it would solve the current bottleneck between line rates and switch rates. "Current switch fabrics are electronic, and they are just going at one gigabit per second. But the input from an optical fiber is 10 gigabits per second. So an optical router could eliminate that mismatch," he says.
Such a system would not be optical computing, but it would be close. If researchers could integrate hundreds of those optical logic gates on a chip, the device would be an order of magnitude denser than the chip-scale WDM project, says Shah.
And in fact, that may be as close as we ever get to purely optical computing. Over 40 years of research, proponents of optical computing have tried to simply replace electric components in the existing architecture. But this level of innovation would use optics as interconnects, in a fundamental change of the way computing works, Athale says.
"Just as today's computers are called electronic, even though they have optical displays and memory (on CD-ROM), you could call that new creation an optical computer," he says. "It's a tall order, but hey, that's what makes it exciting."
Computing with photons
But not everyone has given up on optical computing. NASA researchers are on the verge of demonstrating a crude optical computer.
"We have already built a couple of circuits, and we need only three circuits to make our prototype," says Hossin Abduldayem, a senior scientist at NASA's Goddard Space Flight Center in Greenbelt, Md. "We're very close, but we need more time to do it."
Abduldayem's team has created an "and" and "exclusive or" circuit, and is now building a converter (one to zero and zero to one). Once it is done, they can build many combinations.
"It's impressive and feasible, and is very close to being demonstrated," he says. Abduldayem will share his research in a paper at the annual Military and Aerospace Programmable Logic Devices (MAPLD) conference in September.
Researchers at the Johns Hopkins University Applied Physics Laboratory in Baltimore are also making progress. They are demonstrating the feasibility of quantum computing, which represents data as quantum bits, or qubits, each made of a single photon of light.
In experiments over the past year, they have demonstrated quantum memory, created various types of qubits on demand, and created a "controlled not" basic logic switch, says Bryan Jacobs, a senior research physicist. And last month, they proved they could detect single-photon states, counting the number of photons from an optical fiber.
"So we're working on all aspects of the problem; we can generate, store, manipulate, and detect photons," he says. "Right now we're still in proof-of-concept demonstrations."
So how do you store light? Fortunately, an optical computer needs to store data as light only for very short times, he says. A tougher challenge is to switch the photon without changing it. Qubits exist in different states depending on their polarization, which is the orientation of their electromagnetic field. But optical fibers can change that orientation, basically erasing the data. The Johns Hopkins team stored photons in a simple free-space loop, Jacobs says.
Fortunately, photons are easy to generate. If you stand outside on a clear day and hold your arms in a loop, the sun will shine 10 sextillion photons (10 to the 21st power, or 10,000,000,000,000,000,000,000) through the circle every second. So they created them with a laser "not much more powerful than a laser pointer," put a filter in front of it, then shined it through a crystal to generate various states of light.
The team's next challenge is to do those logic operations better. Once they get error rates low, the system will be scalable enough to operate with large numbers of photons. In the meantime, quantum cryptography is the most likely commercial application of this work, he says.
In fact, some projects already exist. On June 5, researchers at Toshiba Inc.'s Quantum Information Group in Cambridge, England, demonstrated a way to send quantum messages over a distance of 62 miles.
Quantum messages usually degrade quickly over distance, yet the quantum code could let people share encryption codes while operating at this length. Until now, they have had to encode those keys with complex algorithms, and then send them over standard electrical cables. The optical method's strength lies in the ability of eavesdroppers to change the properties of stolen messages only by reading them; every trespass, therefore, would be detected.
But one challenge remains. As long as systems designers use electrical sensors, they must translate data from electric to optic.
On April 28, a team of scientists at the University of Toronto announced their creation of a hybrid plastic that converts electrons into photons. If it works outside the lab, the material could serve as the missing link between optical networks and electronic computers.
"Our study is the first to demonstrate experimentally that we can convert electrical current into light using a particularly promising class of nanocrystals," says Ted Sargent, a professor of electrical and computer engineering at the university. "With this light source combined with fast electronic transistors, light modulators, light guides, and detectors, the optical chip is in view."
The new material is a plastic embedded with nanocrystals of lead sulphide. Those "quantum dots" convert electrons into light between 1.3 and 1.6 microns in wavelength, which covers the range of optical communications. The report was published in the journal Applied Physics Letters.
Likewise, NASA researchers say they are relying on new materials to handle photons.
They are conducting experiments on the International Space Station with colloids – solid particles suspended in a fluid. The right alloy could be built as a thin film, capable of handling simultaneous optical data streams.