Silicon Photonics: The Data Connection
Intel has developed the first silicon-based optical data connection with integrated lasers, confirming that light beams can replace electronic signals for future computers. Its prototype device could move data much faster than today's copper technology, up to 50 gigabits of data per second.
SPIE Fellow Mario Paniccia, director of the Photonics Technology Lab at Intel, answers questions about recent developments in "siliconizing photonics," a term he has coined. He also discusses Intel's support of the new foundry service, OpSIS.
What we demonstrated was the ability to communicate between two silicon chips via an optical fiber, using light instead of copper traces or wires. To the eye, the 50Gbps link consists of a thin optical fiber with each end plugged into a small circuit board about the size of a cracker. Mounted on each circuit board are several electronic components and a silicon photonics chip roughly the size of a fingernail.
On the transmitter side, electrical data signals are directed into the transmitter board, amplified by a CMOS (complementary metal oxide semiconductor) driver IC and then fed into the silicon photonics transmitter chip. The photonic transmitter chip includes four hybrid silicon lasers, formed by fusing a piece of indium phosphide onto a row of silicon waveguides.
The laser is formed by etching gratings into the silicon waveguides and used to create a laser cavity. From this, four different colors of continuous-wave laser light or wavelengths are generated.
These four continuous streams of light then pass into four integrated silicon modulators that encode the data at a data rate of 12.5 Gbps, based on the incoming electrical data stream. The four optical signals then enter a silicon multiplexer where they are combined and coupled out into an optical fiber.
All of this happens in a single piece of silicon!
The 50Gbps data stream (4 laser wavelengths times 12.5 Gbps) then travels down the fiber into a silicon photonics receiver chip. The optical signals are then separated by a demultiplexer where they are directed into four integrated silicon germanium photodetectors that convert the optical signals into electrical signals. The optical data are then regenerated electrically. So we have now transmitted 50Gbps of data from one silicon chip to another "optically."
The ability to communicate optically will dramatically change the way data and information is moved and communicated in the future. Today, as more and more video and information are being stored in the "cloud," it is getting more and more difficult to move this information around.
Having very high-speed, low-cost optical communications will enable new usage models for getting and moving this information in and around the PC, server, tablet, and handheld.
For personal devices, imagine backing up your hard drive or transferring all your music libraries in less than a second. Imagine going to a kiosk and downloading a week's worth of TV shows or a few movies to your handheld in a few seconds via a simple fiber.
Imagine having a wall-size, high-resolution TV display in your home that can be fed via a thin optical fiber. All of this could happen if you had the ability to connect your devices via silicon photonics.
In the cloud or for the data center, the ability to have low-cost fiber connections between servers would eliminate the traditional design constraints we have with CPUs and servers. This would allow completely new architectures and systems to be built which connect CPUs to memory or CPUs to CPUs, all of which would enable better performance, more efficiency, and less power consumption.
We are running into fundamental limits with copper as we continue to scale Moore's Law and drive bandwidth up in frequency. Moving data faster or further over traditional copper wires requires more power, bulkier cabling, and more complex circuits. All of this can be done more efficiently with optical technology, but today optical communication is still very expensive.
However, just as with the integrated circuit, the benefit of silicon photonics is not about a single device. It's about integration of these devices. Moore's Law is not about putting two transistors down onto silicon; it's about what happens when you integrate millions or billions of transistors.
When you do this, you enable new form factors, higher performance, and lower power all at lower cost.
The benefits can be seen today in handheld devices that have the same computing power that you probably had on a desktop a few years ago.
What has limited the adoption of photonics in and around the PC and server has been the cost.
By having the ability to manufacture photonics in silicon we can draft off what has happened in the IC industry. We will be able to have highly integrated photonic devices with very low cost that can be manufactured in high volume. This will allow us to put photonics and optical communications in and around the PC and server and everywhere you need devices connected.
The 50Gbps link was just the first step in demonstrating not only integration of photonic devices in silicon but also low-cost packaging and assembly. We continue to optimize the silicon-photonics process work to improve performance and yields and are aggressively pushing toward commercialization of this technology.
We are trying to tackle the challenges that are needed to enable this technology in high-volume production. In terms of speed, 50 Gbps was also just the beginning, and we are also aggressively pushing toward higher data rates i.e. 100Gbps, 200Gbps and eventually 1Tbps.
We are just at the beginning of this journey to siliconize photonics. Hang on; the next few years will be exciting.
While I cannot comment on specific dates, I can say we are working with our internal business units to find appropriate landing zones and product introductions that span all areas from the consumer space to the data center and to high-performance computing. While we are going as fast as we can, the specific time frame will be determined by our product groups based on the different market opportunities they see.
In addition to financial support for OpSIS, we are providing ongoing technical guidance and support as the center develops. I've worked closely over the years with Professor Hochberg of the University of Washington who has led the establishment of this center, and we will hopefully continue to collaborate on research projects in the future.
By providing a foundry service capability to the research community, we believe OpSIS will also help train and enable the next generation of silicon photonics researchers and engineers that will be needed as we move to commercialize silicon photonics in the future. Mobile and optical technologies are the future. OpSIS gives us a playing field where people can now start playing.
SPIE Fellow Mario Paniccia is director of the Photonics Technology Lab at Intel Corp. and leads a group in the area of silicon photonics. His team is focused on developing silicon-based photonic building blocks for future use in enterprise and data center communications. They have pioneered activities in silicon photonics that have led to the development of the first silicon modulator with bandwidth greater than 1GHz, the first continuous wave silicon laser breakthrough, and the first electrically pumped hybrid silicon laser in collaboration with the University of California, Santa Barbara.
Paniccia, who received the 2011 ACE Award for "Innovator of the Year," was named R&D Magazine's 2008 Scientist of the Year, and Scientific American named him one of the top 50 researchers in 2004.
He answered questions in SPIE Professional magazine about recent developments in "siliconizing photonics," including Intel's support of the new foundry service in Seattle, OpSIS. (SPIE members can view a separate article on OpSIS.)
A video interview with Paniccia from 2010 can be found on the SPIE Newsroom.
The SPIE journal Optical Engineering will have a special section in July on integrated optics, including silicon photonics.
Guest editor is Giancarlo Righini of the Istituto di Fisica Applicata Nello Carrara (Italy).
"Besides being an excellent material for photonics, silicon has the advantage of leveraging the manufacturing infrastructure developed by the silicon microelectronics industry," Righini says.
Integrated optics is based on optical waveguides fabricated on planar substrates. The concept of integrated optics began developing about 40 years ago, with a few papers published at the end of 1960s and the seminal work of researchers such as S. E. Miller and P.K. Tien.
The first efforts were aimed at producing discrete passive optical elements and at integrating a number of them through a monolithic or hybrid approach. The goal was to achieve enhanced functionality and performance, low footprint, and cost-effective volume production capability.
Over the years, the goal has remained valid, but the concept has evolved into integrated photonics, with the inclusion of active elements and possibly of some electronics too.
Indeed, a topic of intense research activity in both industry and academia in the recent years has been silicon photonics. Applications and markets include telecommunications, optical interconnects, optical processing, optical storage, biological and chemical sensing, and biomedical devices.
The aim of this special section of Optical Engineering is to provide readers with the latest achievements in the broad area of integrated optical circuits (IOC) / integrated photonic circuits (PIC).
OPEN ACCESS: This article is open-access to the general community. To read the full text of other feature articles inside SPIE Professional, including a separate article about OpSIS, please use your SPIE member login.
Have a question or comment about this article? Write to us at spieprofessional@spie.org.
