Planar Regrowth Simplifies Photonic Integration Of InP Chips
Rapid growth of multimedia communications, video streaming and file transfer activities are driving capacity increases in broadband connectivity to the internet. And internet traffic increases are set to continue, thanks to the introduction of new communication and entertainment services that include real-time interactive high-definition video with stereographic and even holographic images. Several companies are pioneering these systems and in 2011 the Japanese broadcasting corporation NHK plans to start experimental transmissions of its ultra-high-definition broadcast, which features a picture resolution that is 16 times greater than that of HDTV.
In terms of numbers, today s internet traffic in the US already exceeds 5 Tbit of data every second, according to AT&T, and by 2015 this figure will be 1000 times greater. This will place even more demand on network infrastructure, including the routers and switches in central offices.
Routers are already substantial pieces of equipment. For example, Cisco Systems highest-capacity version for core networks, the CRS-1, weighs more than 50,000 kg and uses 1152 slots of line cards in 80 shelves to operate at 46 Tbit/s. Its energy consumption hits 0.85 MW, partly due to the very large cooling system required, and results in a staggering annual electricity bill of $1.5 million. Clearly building larger routers is a highly undesirable approach to addressing the growth in internet traffic, both from an environmental and commercial standpoint.
Fortunately, photonic integration – the construction of optical circuits that unite various discrete components, including mode-locked lasers, optical switches (wavelength converters), arrayed waveguide gratings and modulators, on a single chip – can deliver tremendous miniaturization and lead to lower cost, more-efficient systems for the internet.
The inefficiencies in today s communication systems stem from excessive electronic processing, with each bit stored, processed and then sent forward by high-speed electronics. By switching to the photonics domain, data and signal processing can serve similar purposes while consuming very little energy. These benefits have already been realized in reconfigurable optical add-drop multiplexers with smaller IP routers, and carriers around the globe are now using these components in preference to very large equivalents.
Several other benefits result from the replacement of hundreds or thousands of discrete modules with a single chip. Almost no optical power is lost at interfaces between components, and a single large chip is more affordable and easier to package than many small chips, which collectively consume more space and power. Replacing many chips with just one improves reliability by cutting the number of optical joints and a single thermoelectric cooler (TEC) can take the place of a handful of TEC units. A single photonic chip also has the potential to introduce new functionality and operate at faster speeds.
Although many benefits are associated with photonic integration, it is notoriously difficult to create these large, complex chips. The integrated fabrication process flow has to simultaneously produce all of the different types of device on a single chip and deliver a high manufacturing yield to realize the potential cost benefits.
Process optimization is crucial and the key challenge is to devise a series of fabrication steps that can maximize the performance of many different components. Any changes must deliver a universal benefit across the various active elements, including lasers, optical amplifiers, switches and passives such as waveguides, splitters and couplers.
The challenge of producing high-performance photonic integrated circuits has attracted several research teams around the globe. These include the vertically integrated communication-system company Infinera, the UK s Center for Integrated Photonics (CIP) and our team – a partnership between the University of California, Davis, the Royal Institute of Technology, Sweden, and Multiplex, Inc.
Infinera is the leading maker of photonic integrated circuits and its success at the systems level has catapulted the company to the number-one spot for deployment of long-haul fiber-optic backbone technology. The Californian company doesn t disclose the details of its InP monolithic chip technology, but its integration process probably involves MOCVD-based material growth and regrowth, shallow etching of curved waveguides, iron-doped InP regrowths and p-type InP regrowth.
Ipswich-based CIP employs a form of hybrid technology that combines silica waveguides and InP semiconductor optical amplifiers. Mixing different materials has its downsides, however, such as greater manual handling, which leads to higher labor and production costs and less-robust chips.
Like Infinera, our approach involves large-scale photonic integration, but we have the benefit of a more straightforward integration process that exploits judicious use of MOCVD and HVPE technologies. MOCVD growth produces the complex heterostructures, but it is not ideal for regrowth steps that fill in lower regions to produce the planar structures that are needed to complete the waveguides. This is because vertical growth takes place and forms "ears" that protrude in and outwards. HVPE does not suffer from this weakness thanks to lateral growth that is independent of crystal orientation.
Our efforts began by developing and refining a fabrication process for producing the building blocks for photonic chips, including mode-locked lasers, Mach-Zehnder interferometer (MZI) optical switches/wavelength converters, arrayed waveguide gratings (AWGs), amplitude modulators and phase modulators. Producing all of these devices involves a well defined process flow of selective etching of active quantum wells, MOCVD regrowth of p-type InP and a contact layer, waveguide definition and dry etching, HVPE iron-doped InP lateral regrowth, and metallization (figure 1). Using this process flow, many building blocks can be put together into a larger system-on-a-chip.
One of HVPE s strengths is its ability to produce complete planarization, regardless of the mesa s orientation and height. Because of this, our complex devices can accommodate AWGs and MZIs that feature waveguides curving to all directions.
Other photonic integrated circuit developers, such as Infinera, don t enjoy the benefits of a highly effective lateral growth process. This hampers further integration and additional MOCVD regrowth steps are needed after shallow etching to produce planarization. The absence of HVPE also dictates the use of high-mesa waveguides (as pursued by NTT researchers), which are more difficult to define precisely and produce higher scattering losses.
Some researchers avoid issues associated with MOCVD regrowth by turning to quantum-well intermixing technology. Passive and active elements can be integrated by locally converting wells into passive material through intermixing with the barrier. However, this tends to lead to high background losses typically beyond 10 dB/cm.
During the last few years we have built a variety of photonic integrated circuits with our relatively simple fabrication process, including 10 Gbit/s optical code-division-multiple-access (O-CDMA) transmitters and receivers, optical arbitrary waveform generators, and optical router systems.
O-CDMA transmitters and receivers have the potential to support optical local-area networks with a bandwidth of 10 Tbit/s, which is substantially broader than today s Ethernet or DSL networks. This technology provides secure links, just like its wireless equivalent, because every user has a specific code set and communication only takes place between users with matching codes. These are defined by applying phase shifts and amplitude changes on each spectral element of a broadband, coherent optical pulse.
Our transmitter features a mode-locked laser that produces ultrafast optical pulses, which are shaped with an integrated spectral encoder – an array of phase modulators sandwiched by AWGs. The receiver contains a spectral decoder with the same structure as the encoder and detection occurs when the mode-locked laser triggers the MZI switch.
Based on this technology we have demonstrated a 320 Gbit/s all-optical passive network testbed that uses spectral phase encoded time spreading O-CDMA and features up to 32 users operating at 10 Gbit/s (or equivalently 32,000 users at 10 Mbit/s). This set-up delivers a bit error rate below 10–12 for a 320 Gbit/s O-CDMA network using Reed-Solomon forward-error correction.
The optical arbitrary waveform generator that we have developed operates at 1 THz bandwidths. It is suitable for rapid, secure communication of vast amounts of data and ultrahigh bandwidth LIDAR (light detection and ranging) – an optical technique used to determine the distance to a target. AWGs and high-speed (10 GHz) amplitude and phase modulators are combined in this monolithic chip.
We have also built an optical router that features photonic integrated chips and surrounding electronics. It has a routing capacity that can be scaled to 42 petabit/s, with switching speeds of 1 ns and routing latency of 50 ns. The chip s tiny footprint and power requirements make it an ideal replacement for today s power hungry, bulky electronic routers in telecom and enterprise networks and it could be suitable for deployment in the home and on military planes.
Our three types of photonic integrated circuits illustrate the potential of this technology, which has the capability to create supercomputers on a chip. The technology is still in its infancy, but we are fascinated to see how it develops. We will continue to play our part in this evolution by pushing photonic integrated circuit performance forward and investigating its commercial opportunities.
R G Broeke et al. 2007 IEEE J. Sel. Topics Quantum Electron. 13 1497.
N K Fontaine et al. 2007 Optics Letters 32 865.
S J B Yoo et al. 2006 J. Lightw. Technol. 24 4468.
View pdf of article