Silicon is the second most common element in the Earth’s crust after oxygen. Its abundance was one of the reasons that Robert Noyce, co-inventor of the electronic integrated circuit in 1959, advocated the use of silicon as a substrate material rather than germanium, which was more popular at the time. The rest, as they say, is history.
In contrast, optical components today are manufactured from a wide range of materials, depending on the functionality required. These specialist materials include indium phosphide for active components such as lasers and photodetectors, silica-on-silicon for passive functions such as the arrayed waveguide gratings (AWGs) that combine and separate wavelengths, and lithium niobate for optical modulators, which are used to encode the data on a beam of light.
The need for a diverse range of high performing but expensive materials is a major weakness for optical components, and has hampered efforts to develop photonic integrated circuits, contends Roel Baets, head of the Photonics Research Group at Ghent University, Belgium. The bottom line is that optics vendors need high unit sales volumes to fund the expensive research into these specialist materials, and telecoms lacks such volumes, he says.
The lure of silicon
Silicon is emerging as the material that could transform the economic equation for optics vendors. “With silicon photonics the aim is to pull all of the applications onto one technology platform,” said Doug Gill, member of the technical staff at Alcatel-Lucent Bell Labs. That technology platform is complementary metal-oxide semiconductor (CMOS) – the bedrock manufacturing process of the semiconductor industry.
CMOS has been used for more than 30 years to make increasingly complex digital logic chips, from the first integrated circuits, which comprised just a handful of components, to the latest four-core microprocessors from the likes of Intel and AMD that integrate over 730 million transistors. “It is a huge advantage if photonics can fall back on such a very strong, mature technology,” said Baets.
“The appeal of CMOS photonics is threefold,” explained Karen Liu, vice-president of components at market research firm Ovum. “First, silicon is much easier to work with than materials such as indium phosphide. Second, CMOS silicon in particular has very advanced process capabilities and available commercial fab capacity behind it. That translates not only to economies of scale but the sort of yields that enable complex or parallel devices. Finally, digital CMOS is great for simplified control circuits that deliver high-speed and low power consumption.”
The current upsurge of interest in silicon photonics is coming from two different directions. On the one hand there is the huge development effort by computer giants Hewlett-Packard, IBM, Intel and Sun Microsystems, which, after more than a decade of research, has resulted in a near complete set of silicon-based optical building blocks. On the other hand there are start-ups such as Kotura, Lightwire and Luxtera that are focusing on revenue-generating products.
Potential applications include optical transceivers for connecting servers and switches in the data centre, followed by consumer electronics interfaces, and optical interconnects between and within electronic chips. “Silicon photonics is different for different companies – it is not homogeneous,” said Yurii Vlasov, a researcher at the physical sciences department at IBM’s T J Watson Research Center.
The focus of IBM and the other computer companies is to determine whether optics could replace copper connections between chips or between different cores in future multicore processors. This would open up a new market opportunity for optics.
The focus of the silicon photonics start-ups is on near-term products. Although silicon photonics is a new technology, it can be adopted immediately by customers if it is brought to market as a standards-based product, such as a pluggable optical module.
Lightwire, for example, is selling a 10 Gbit/s SFP+ optical transceiver to address the 10GBase-LRM standard, which is used for connecting data-centre equipment over distances of up to 220 m on multimode fibre. Similarly Luxtera is shipping a 40 Gbit/s active optical cable with QSFP pluggable modules at the ends, targeting data centre links up to 300 m.
Luxtera has also developed a prototype cable using silicon photonics for an optical HDMI (high-definition multimedia interface) included with high-end TVs. Given the growing resolution used for high-definition TV and the increasing screen refresh rates, the data-rate requirements are increasing to 10 and 20 Gbit/s. “The volumes of Ethernet and Fibre Channel transceivers are in the millions of units per year, whereas some 200 million TVs are sold a year, with each TV having two to four HDMI interfaces,” said Marek Tlalka, vice-president of marketing at Luxtera.
Other silicon photonics products to reach the market in the last year include a 10 Gbit/s receiver containing a silicon-based variable optical attenuator (VOA), developed by Kotura in collaboration with Japanese firm Oki Optical Components. The silicon VOA acts as a dimmer switch to protect the detector from transients and has a faster response time than a MEMS-based VOA. “What we have done [with our VOA] is to address areas where we can bring value but it is not a huge market,” said Arlon Martin, vice-president of marketing and sales at Kotura.
Kotura’s real ambitions are longer term. The start-up wants to be a “photonic ASIC” design company whose roadmap will target 100 Gigabit Ethernet (GbE) and then optical interconnects. “Our research goals include gaining a hands-on understanding of the pros and cons of optical components in silicon, as well as silicon’s performance as an integration platform,” Martin said.
The silicon photonics toolbox
If silicon is to become an optical workhorse, then it must be able to support light generation, transmission and detection in an efficient, miniature package. Some of silicon’s optical characteristics are favourable, including low optical loss at infrared wavelengths and a high refractive index. The latter makes it possible to design optical waveguides with a small radius of curvature, which enables high packing density and miniaturization of designs. But other optical properties are more problematic. For example, silicon’s electro-optic coefficient is zero, so companies must find new ways of making optical modulators.
Significant progress has been made in recent years, however. In July 2007 Intel showed a silicon-based Mach–Zehnder modulator that could operate at speeds up to 40 Gbit/s, and then went on in May 2008 to integrate eight of these devices onto a single chip. Meanwhile Lightwire has developed a unique structure for a low-power silicon modulator, which consumes just 30 mW at 10 Gbit/s compared with more than 100 mW to drive an indium-phosphide-based electro-absorption modulator at the same data rate.
Photodetection is another key function but unlike the waveguide and modulator, it cannot be implemented using silicon alone. Luxtera and Intel have developed photodetectors by adding germanium to silicon. Furthermore, in December 2008 Intel presented an avalanche photodiode design that it claimed is an industry first in that the device’s optical performance – a 340 GHz bandwidth-gain product – is better than the performance obtained from equivalent devices made in specialist materials.
The major optical function still missing from silicon photonics’ toolbox is the laser. “Silicon can do everything except generate light,” said Martin. “Some people think that since you can’t build a laser [in silicon], the technology is not that useful, but nothing is further from the truth.”
IBM’s Vlasov agrees, pointing out that in the chip world electrical power is not generated on the processor. IBM would prefer the light to be generated off the chip, so that the laser can be cooled separately from the microprocessor. Heat dissipation is one of the key design considerations for optical transmitters, which need stable temperatures to deliver reliable performance.
An external laser can be coupled to silicon using flip-chip bonding. However, Intel believes that external lasers will be too expensive for chip-level optical interconnects. “With 25 [optical] lines each at 40 Gbit/s, using 25 individual lasers: you can’t do that cost effectively,” said Mario Paniccia, Intel fellow and director of its Photonics Technology Lab.
With this in mind, Intel, in collaboration with the University of California, Santa Barbara, has developed a technique in the lab, first described in September 2006, that uses a plasma to fuse an indium phosphide wafer on top of silicon. The indium phosphide layer is then etched into islands where required, exposing the silicon underneath. Intel is now working on adding a mirror to the laser to route the light source to other optical functions such as a modulator. IMEC, a nanoelectronics research centre in Belgium, also detailed an alternative approach in 2006 that adhesively bonds the indium phosphide laser to silicon. Both approaches need further work before being commercially deployed.
Lightwire and Luxtera have found a way to minimize the cost of the external laser in their transmitter chips by sharing the laser’s output among multiple channels. Luxtera’s 40 Gbit/s active optical cable splits the light from a single laser among four waveguides, each with its own modulator. Lightwire’s product plans call for something similar, targeting active optical cables and QSFP modules, where light from a single laser is divided between four or even 12 channels.
Having a complete collection of optical building blocks is not the only challenge for silicon photonics. One of the barriers to commercialization, according to Luxtera, is finding a CMOS fab that is able to manufacture such products in commercial quantities. Another challenge is testing the resulting devices at the wafer scale, before the wafer is broken up into individual devices. Because no such photonic tester is available commercially, Luxtera has developed its own by modifying a standard electric prober. Optical grating couplers couple light from a probe fibre into a waveguide on the surface of the die.
The commercialization issues facing the computer players will be more complex still, due to the high volumes required in computing. “Packaging at the PC level will require a pick-and-place machine,” said Intel’s Paniccia. “With 100 million units a year, you can’t afford to spend too long; 5–10 seconds [per device] for assembly and 5–10 seconds for testing.”
There is also device qualification, which means running reliability tests for 5000 hours under various humidity and thermal conditions. Any change to the CMOS process, such as adding a new material that can create stresses, requires a fresh round of reliability tests. “Any problem must be fixed and then the testing starts all over again,” said Paniccia, which is why he refers to the process as the “valley of death”.
Disruptive influence
Industry analysts view silicon photonics as a potentially disruptive technology. “It is changing the way vendors do photonic integration and has the potential to disrupt the [telecom] supply chain,” said Simon Sherrington, an independent analyst and author of a recent Light Reading Insider report on silicon photonics.
Although silicon devices may fall short of the performance of optical devices made using the best specialist materials, they promise lower-cost, higher levels of integration, and to close the performance gap as the technology matures. “As [system vendor] players in the telecom supply chain start to buy such products, established component suppliers could find themselves unseated,” he said.
Despite the primary focus today being on short-reach optical interconnects in computing and data centres, Sherrington expects silicon photonics to find use in telecoms, first in optical access applications such as 10 Gigabit PON, before spreading to metro and ultimately long haul.
Liu stresses that a key merit of silicon photonics is its potential to make highly integrated devices “for pennies”. Liu views the technology as a separate thread to existing photonic integration approaches, whose progress she describes as “painfully slow” (see FibreSystems Europe, October 2008 p16). Silicon photonics provides the industry with fresh ground to explore, she says. “With the relentless growth in traffic, telecom needs something new,” said Liu, “You can’t just take what we have and continue to cost-reduce.”
However, she warns that replacing existing optical components with silicon-based ones requires more than matching the incumbent technology. “It has to be significantly better in terms of performance or price,” she commented. Intel’s Pannicia agrees – the firm’s goal is for silicon photonics to achieve 90% of the performance of equivalent devices made from exotic materials at less than one-tenth of the cost.
Indeed, the early silicon photonics products on the market are ones where their performance is better than what’s already available. For example, Lightwire points to its SFP+ transceiver for 40 km applications, which consumes under 0.5 W compared with 1.2–1.5 W for a conventional transceiver. “As VCSELs [vertical-cavity surface-emitting lasers] move up in speed, the reliability decreases and power dissipation accelerates the problem,” said Dave D’Andrea, the company’s senior director of sales and marketing. Lightwire is also developing a 12 × 10 Gbit/s SNAP-12 parallel optics module that uses a single Fabry-Peroy laser to feed a CMOS photonic circuit with 12 modulators. This will consume 60% of the power of a VCSEL array and promises to be more reliable.
Silicon photonics could also emerge as a enabling technology for telecoms system vendors eager to differentiate their products. During the boom years most of the equipment companies, including Alcatel-Lucent, sold off their in-house components divisions, leaving them dependent on merchant suppliers for optical parts.
“Silicon photonics offers the potential opportunity for system vendors to design directly into manufacturing,” said Gill. In other words, he believes that one day silicon photonics could have a similar role for optics that application-specific integrated circuits (ASICs) play in electronics. Once the software tools become available, system houses could design their own silicon photonics devices in the same way that they currently design their own ASICs.
Kotura shares this vision and is already starting to turn it into reality. The vendor offers a “photonic ASIC” design service, which allows customers to develop their own designs using a growing library of silicon photonics building blocks, including attenuators, AWGs, low-loss fibre couplings, mode transformers, splitters, switches, taps and vertical couplers.