Pages: pp. 16-19
Vendors of computer-, networking-, and telecommunications-related equipment have long faced a dilemma. Much of their equipment is based on silicon and copper wiring, which are inexpensive to use but offer limited data rates. Thus, as microprocessor and networking speeds have increased, the speed of communications within chips, between chips or circuit boards, within LANs, or even along the "last mile" from ISPs to customers has not kept pace.
An option is to use optical connections for these communications. "Any signal processing you can do with electronics you can do faster with light," said Alfred University professor Alexis Clare.
However, optics has been expensive and, therefore, not suitable for any but the largest networking operations with the most traffic and the biggest potential for return on investment. Thus, optics has been used largely in settings such as telecommunications vendors' backbone networks.
Optical connections have been impractical for communications on or between chips because the shorter distances involved yield fewer bandwidth improvements, which don't justify the expense, said University of Rochester distinguished professor Philippe Fauchet.
Now, though, vendors are attempting to combine the best of both worlds and offer silicon optics, which uses complementary metal-oxide semiconductor (CMOS) technology to fabricate optical components on silicon.
This approach would speed up traditionally silicon-based systems. It could also reduce the cost of optical equipment and bring optical systems within reach of more users, including companies and service providers with networking operations smaller than those of large telecommunications providers.
Combining silicon and optics is a complex process. However, some vendors are already selling some less complex silicon-optical components.
Meanwhile, scientists are making progress on more complicated components. For example, Intel and UCLA researchers have each developed prototype silicon lasers.
Nonetheless, obstacles remain, so while some components are already available, it may be years before silicon optics can be widely and reliably used commercially.
Transistors, diodes, and most logic-control components in today's computers are silicon-based. Silicon is found in sand, which is plentiful and relatively inexpensive to process, explained Clare.
Numerous fabrication plants have worked with silicon for many years. This enables high-volume production and makes it easy for manufacturers to use the material and integrate components. Moreover, as a semiconductor, silicon is power-efficient and generates relatively low amounts of heat.
However, silicon, working with copper wiring, creates the binary ones and zeros of data via electronic pulses, so factors such as electrical resistance and capacitance limit the technology's bandwidth.
Optics generates binary ones and zeros via a modulated laser beam. This makes the technology particularly fast and able to handle large quantities of data.
Typical electronic components offer maximum data rates of 1 Gbit per second, according to University of Oklahoma professor Patrick J. McCann. Proponents say silicon optical components, on the other hand, could offer rates as high as 40 Gbps during the next few years.
In addition, light beams used in optical transmissions can be split into multiple communications channels that can be multiplexed onto a single link, thereby offering very high data capacities, according to Fauchet.
Manufacturers frequently use exotic materials like gallium arsenide and indium phosphide with optical components to maximize performance and add special properties, such as minimal light loss during transmissions. However, these alloys are expensive. "Indium, for example, is a limited resource that must be mined," noted University of Maryland professor Thomas E. Murphy.
In addition, using exotic materials in manufacturing is difficult, time-consuming, and costly. For example, Clare explained, manufacturers often must use specialized equipment that deposits the materials as vapors or a liquid, layer by layer.
US Air Force researchers pioneered silicon optoelectronics in the mid 1980s for sophisticated communications and signal processing, said Richard Soref, research scientist with the Air Force Research Laboratory's Sensors Directorate.
Since then, a variety of universities and companies, such as Bookham and Intel, have worked on silicon optics.
In most current silicon-optical devices, such as waveguides and filters, the silicon is used only as a passive medium through which light can be transmitted. These devices do not tackle the more fundamental problem of converting signals from optical to electronic and vice versa, said the University of Maryland's Murphy. This critical capability maximizes the use of optical technology and increases performance.
Many of today's systems use photodetectors that convert signals only from optical to electronic.
Another key obstacle to achieving true silicon optics has been the lack of a silicon-based laser, noted McCann. Semiconductor physics has limited the ability to build such a laser. The issue is bandgap: the energy difference between a material's conductive and nonconductive state. Direct bandgap semiconductors such as gallium arsenide and other material used in optics are efficient at emitting light, while indirect bandgap semiconductors such as silicon are not.
Some researchers are looking into techniques such as introducing other materials that will make the silicon transmit light more efficiently or adding dopants that themselves transmit light effectively, according to University of Surrey professor Graham Reed.
Meanwhile, scientists continue looking for other ways to provide silicon optics.
In silicon optics, manufacturers build both the optical and electronic components on a single silicon chip, explained UCLA professor Bahram Jalali. This reduces mass-manufacturing costs and simplifies packaging.
Vendors generally design silicon-optical components to fit on a silicon substrate as part of a standard CMOS manufacturing process, according to Arlon Martin, vice president for sales and marketing for silicon-optics vendor Kotura.
The principal optical component is the laser, shown in Figure 1, which acts as the system's light source.
Figure 1 Intel is working with six building blocks for producing a silicon-optical transceiver. (1) A laser produces the light that will eventually carry the data. (2) Waveguides guide the light across the chip along the proper route. Splitters divide the light into separate beams to carry multiple signal sets. (3) A modulator adds the signal to the light beam at high rates. (4) A recipient's photodetector decodes the encoded light and turns the optical signal back into an electrical signal for processing. (5) Passive alignment precisely but inexpensively aligns optical elements. (6) Chip-based intelligence processes encoding and decoding.
"When an electrical current passes through a semiconductor laser, it emits a coherent beam of light (photons)," explained the University of Maryland's Murphy. "Information can be encoded onto the optical signal in two ways: by simply turning on and off the electrical current that drives the laser or by modulating the continuous laser beam's [intensity] in an external device (the modulator). No one has succeeded yet in building a silicon laser that can be directly driven by an electrical current."
Manufacturers—such as IBM, Intel, Kotura, and Luxtera—are currently developing new types of modulators that can convert electrical signals to optical.
Amplifiers compensate for the loss of light that occurs during the transmission process by producing more photons and thereby strengthening signals.
Individual optical filters let only a single set of signals pass onto specific wavelengths within a laser beam. This separates signals into discrete groups of transmissions.
Waveguides direct the light as it passes through open spaces within a system, leading it around corners or along paths other than a straight line, noted Harold Hosack, director of interconnect and packaging science for Semiconductor Research, a research-management consortium.
Optical switches move data between light paths, sending data streams to their proper recipients.
Proponents see four primary ways in which silicon optics could be used in the near future.
Silicon-optical components could replace existing transceivers in 1- and 10-Gbit Ethernet LANs. Because LANs typically include many nodes with many connections, the use of silicon optics could improve overall communications enough to justify the expense, UCLA's Jalali explained.
Equipment vendors are interested in building silicon optics for LANs in part because many corporations use these networks and thus represent a lucrative market.
While microprocessors have gotten faster, overall computer-performance increases have been limited because of slower communications via data buses and along copper wiring between chips and circuit boards.
Using optical communications instead of electrical wiring would be faster, and silicon optics could keep the implementation affordable.
Silicon-optical interconnects could replace copper wiring on chips, according to Semiconductor Research's Hosack. A single chip would contain most of the basic optical components except for the light source.
Short, high-density interconnects should be left in copper wiring because the necessary optical components would be too large and require too much power, noted Hosack.
Research into on-chip silicon optics is ongoing. "Prototype waveguides and detectors for optical interconnects are available," Hosack said. "However, the necessary extensive studies of reliability and development of cost-effective manufacturing methods have not been done."
In addition, researchers have only just begun developing prototype silicon lasers. Manufacturers have had trouble beaming lasers onto a chip without causing heat, alignment, and other problems.
Intel's approach introduces photons onto a chip via a laser. It then uses transistor-like devices to remove the buildup of electrons that the light source creates, explained Mario Paniccia, the company's research director. If not removed, the electron cloud would absorb light and interfere with the generation of the continuous laser beam that is necessary for proper modulation and data transmission, he said.
Intel expects to have silicon lasers ready for commercial use by the end of this decade," noted Victor Krutul, the company's director for enterprise-initiatives technology management.
In addition to replacing components in silicon-based systems to make them faster, silicon optics can also replace traditional optical components in optical systems to make them more affordable.
For example, Kotura is using silicon optics to make arrays of variable optical attenuators, which incrementally adjust the power of the signal passing through an optical system, according to the company's Martin.
The VOAs are made of a silicon chip with parallel waveguides. Applying an electronic current across the waveguide attenuates the light, Martin explained.
Using silicon makes the manufacturing process less expensive and improves the VOAs' performance.
Manufacturers such as Intel are also using silicon optics to make modulators for optical systems.
Lasers use significant electricity, which generates heat. This is a problem for chip-based silicon optics because manufacturers try to reduce processors' power usage and heat generation.
Manufacturers must precisely align optical elements such as waveguides and lasers so that light transfers precisely from one component to another. "If they are misaligned, then light misses the receiving device and is lost," explained Reed. Proper alignment can be a challenge during the mass production of silicon-optical components.
Some silicon-optical elements, such as waveguides, are relatively large and thus occupy considerable real estate on chips, particularly as processors shrink in size. This causes design challenges
In addition, replacing selected electrical wires with optical interconnects can cause the loss of some of the desired performance advantage because of the time it takes to transform electrical signals to optical signals and vice versa, said Semiconductor Research Corp.'s Hosack.
There can also be loss of light that the material in waveguides absorbs. Moreover, Reed said, waveguides that interact with one another can generate noise that interferes with signal detection.
Proponents say that during the next few years, silicon-optical components will shrink in size and become more efficient, which will encourage more widespread use. And, they add, manufacturers and researchers will demonstrate increasingly complex integration of optical technology with electronics.
Meanwhile, usage could increase as more researchers and companies work on silicon optics.
According to UCLA's Jalali, the greatest impact will be in enabling low-cost optical transceivers for communications over short distances such as those in interchip and intrachip interconnects.
Because so many chips are made and sold worldwide each year, the biggest potential market for silicon optics is on-chip interconnects, predicted Richard Wawrzyniak, an analyst at Semico Research, a semiconductor-market research firm.
Complex commercial silicon-optical devices and applications will hit the market within two years, Reed said.
In the future, manufacturers could develop ways to use silicon optics on backplanes and to connect computers and peripherals. Also, noted Reed, ISPs could use silicon optics to cover the last-mile connection between their facilities and their customers. This would be much faster than DSL- or cable-based broadband, he said.
Wawrzyniak predicted that revenue from silicon optics will increase from $10 million this year to $1.8 billion by 2010.
"There will be an explosion of interest in silicon photonics in the next few years," said Reed. "Maybe in 10 years, your laptop will rely on silicon photonics as much as it relies on silicon electronics today. Silicon is simply an excellent material for mass production."