Jörg Schwartz, email@example.com
The idea of integrating several photonic functions – such as generating, guiding, modulating, filtering and detecting light – within single small-waveguide devices has been around for some time and continues to be an active area of research. For some, integration carries the promise that optics will change our world much as electronic integrated circuits (ICs) have; for others, it simply offers the opportunity to make the devices that process photons smaller, less expensive or faster, thereby advancing communications, avionics or sensing.
However, researchers are considering a large number of materials and techniques, and one can’t help thinking that the entire field is quite diffuse. This dilemma was described by Giancarlo C. Righini of Italy’s National Research Council as he wrapped up a European meeting on integrated photonics: “The bottleneck of photonic integration is related to the (too) many degrees of freedom – many different materials and technologies, many different component types, many different wavelength ranges and applications. Even if it is the sign of scientific richness and broad technological knowledge, it would be important to try to focus on a few issues; e.g., by building up on a stable technological platform ….”
Silicon versus indium phosphide
Historically, optical ICs based on dielectrics, such as glass or silica on silicon, are easy and cheap to make. They feature low losses but do not offer an easy means of controlling the light. In contrast, integrated devices such as nested modulator structures based on crystalline lithium niobate (LiNbO3), which generate sophisticated ways to encode data on light beams for high-capacity transmission, have been on the market since the beginning of this decade. Pure LiNbO3 itself, however, cannot generate or detect light, which limits its scope as an integration platform, despite maturity and high-performance modulation characteristics.
Cascaded Mach-Zehnder modulators, monolithically integrated in InP, generate return-to-zero DQPSK signals at 40 Gb/s. Courtesy of Bookham.
This is where semiconductor materials enter the scene, with processing well understood from electronics IC production. Silicon, a material widely used in that field, pitches itself as particularly low-cost and of high quality. Silicon-on-insulator wafers offer a very high refractive index contrast between their silicon and silicon oxide materials (3.45 and 1.45, respectively), which facilitates making waveguide structures down to hundreds of nanometers. On one hand, such small structures support adaptation of electronic IC processing methods; on the other, they may pave the way for using nonlinearities to overcome one of silicon’s big weaknesses: Because it lacks a direct bandgap, silicon cannot generate light directly or provide light amplification. Furthermore, the bandgap is too large, which means that, although silicon is known as an almost classical detector material, silicon-based photodiodes for fibres and communications turn out to be difficult to make in the near-IR spectral range (1310 to 1550 nm).
Indium phosphide (InP) is the other material platform getting a lot of traction these days. A semiconductor as well, it offers everything mentioned above that silicon is missing in terms of light generation, modulation and detection. Waveguiding properties also are good, but nanostructuring is more difficult than for silicon, and mechanical handling of them is tricky. This limits InP’s ability to be as highly integrated as silicon and therefore does not support size compatibility with what is widely referred to as CMOS (complementary metal oxide semiconductor) electronics.
So why does this matter? “It doesn’t for some, but it does for others,” explains photonics integration expert Dr. Helmut Heidrich of Fraunhofer Institute HHI, the Berlin-based communications research institute. One of the key applications currently driving integrated photonics is moving increasing amounts of data from point A to B. The need for all-optical processing is not imminent, as electronic processors still have a lot of room for improvement, but a bottleneck that will be reached much sooner is the density and speed at which data can be moved into and away from the processor. Such chip-to-chip communication currently is implemented as copper wires, which are reaching their limits at data rates exceeding 20 Gb/s.
At the other extreme from these few-centimetres-long connections are long optical communications links spanning up to thousands of kilometres, which impose completely different challenges on photonic circuits. Infinera, an optical transmission systems vendor in Sunnyvale, Calif., USA, developed its own special large-scale photonic IC that includes everything required to transmit/receive ~10 wavelengths over such transport networks; i.e., parallel integration. The company, although successful in its market segment, has not had followers among its direct competitors, which all seem to take the mainstream approach by depending on component and subsystem suppliers, such as Bookham Technology or JDSU.
Those competitors also are expending much effort developing integrated subsystems, many of which are based on an InP platform. However, as Andrew Carter, Bookham’s vice president of technology, points out, they are following a more serial integration route than Infinera’s photonic IC. For Carter, one of the arguments is that integration simplifies new sophisticated modulation formats such as differential quadrature phase shift keying (DQPSK), favoured for moving optical networks to speeds of 40 or even 100 Gb/s per wavelength.
“Generating those formats can be very complex and costly using discrete components. Integration drives down the size and assembly costs as well as allowing full-band wavelength tunability from the outset,” he said.
As for materials, one could say that InP, with its stronger “active” performance to date, is the material of choice for applications where this dominates the equation; that is, where distances are long and transmission is challenging. On the other hand, silicon is a strong contender for “passive” applications or highly integrated interconnect applications at the junction of highly integrated electronics. But there is another alternative, which offers the best of both worlds.
Hybrid integration – fix or solution
This approach involves moving away from full monolithic integration using a single material toward what is termed hybrid integration – as demonstrated by Carlsbad, Calif.-based Luxtera Corp. by flip-chip bonding InP lasers onto a silicon integrated photonic circuit to make high-performance light sources. Such hybrid integration comes in a vast range of flavours that differ in the materials used or in the way the various materials are married together, both physically and optically.
Another motivation for going hybrid is economics. Solving connectivity challenges, along with delivering small and elegant solutions, is only one side of the coin in today’s world of communications, where cost pressure is dramatic. In other words, any integrated solution must offer a significant cost advantage over its alternatives. To do so, not only the material and its processing must be considered but also the yield that can be achieved in volume production.
In response, the UK’s Centre for Integrated Photonics Ltd. (CIP) developed a hybrid integration platform consisting of manufacturable and material-optimized submounts, which fit together and facilitate complex functions. Graeme Maxwell, CIP’s vice president of integration technology, explained in the August/September 2008 issue of EuroPhotonics (“Hybrid Integration Platform Facilitates Photonic Circuits,” p. 34) that yields of more than 95 per cent are required before a monolithic approach weighs in against discrete solutions. This was concluded from a study he performed in which the hybrid method makes integration work even at yields in the 65 to 95 per cent range.
Bookham’s Carter agrees that high yields are mandatory for monolithic approaches and that the thresholds for hybrid versus monolithic will depend on many factors, including complexity and market volume. He confirms to EuroPhotonics that chip yields are not an issue for his company’s monolithically integrated InP devices because these are well developed and not too complex: “We work very hard on yield – at both design and process levels. For example, the vast majority of tunable lasers on our 3-in. wafers all function correctly, with maybe only 10 or 20 out of 2400 devices exhibiting some anomaly in performance. This bodes extremely well for more complex integration.”
So it looks like, where monolithic solutions can be managed, InP has the biggest opportunity in the integrated photonics market today and most likely will further increase its footprint as long as it can help reduce costs and improve functionality. Hybrid solutions – incorporating not only InP and silicon on insulator but also LiNbO3, silica, polymers and other materials – can improve performance by using the best suitable materials and enable even more complex integrated solutions, but they usually increase production cost.
Looking forward, silicon is promising for short high-speed interconnects and may even be able to extend beyond. This would be easier if some of the laws of physics could be bent – which seems possible given that the big guys such as Intel or IBM would like to do so.