White Paper Laser Sources For Optical Transceivers Giacomo Losio ProLabs Head of Technology September 2014
Laser Sources For Optical Transceivers Optical transceivers use different semiconductor laser sources depending mainly on the reach and bit rate that the device has to guarantee. In this paper we will describe the most commonly used types and their application. We will skip the physics of the laser and semiconductors; we will focus only on the technological aspects of each device. Giacomo Losio ProLabs Head of Technology September 2014 Edge-emitting lasers A laser diode is electrically a p-i-n diode. The active region of the laser diode is in the intrinsic (I) region, and the carriers, electrons and holes, are pumped into it from the N and P regions respectively. The goal for a laser diode is that all carriers recombine in the intrinsic region, and produce light. Thus, laser diodes are fabricated using direct bandgap semiconductors. The active layer often consists of quantum wells, which provide lower threshold current and higher efficiency [1]. A quantum well laser is a laser diode in which the active region of the device is so narrow that quantum confinement occurs. Forward electrical bias across the laser diode causes the two species of charge carrier holes and electrons to be injected from opposite sides of the p-n junction into the depletion region. When an electron and a hole are present in the same region, the result may be spontaneous. The difference between the photon-emitting semiconductor laser and conventional phonon-emitting (non-light-emitting) semiconductor junction diodes lies in the use of a different type of semiconductor, one whose physical and atomic structure confers the possibility for photon emission. These photonemitting semiconductors are the so-called direct bandgap semiconductors. Suitable materials include. Gallium arsenide and indium phosphide, In the absence of stimulated emission (e.g., lasing) conditions, electrons and holes may coexist in proximity to one another, without recombining, for a certain time before they recombine, then a photon with energy equal to the recombination energy can cause recombination by stimulated emission. This generates another photon of the same frequency, travelling in the same direction. This means that stimulated emission causes gain in an optical wave (of the correct wavelength) in the injection region, and the gain increases as the number of electrons and holes injected across the junction increases. 2
The gain region is surrounded with an optical cavity to form a laser. In the simplest form of laser diode, an optical waveguide is made on that crystal surface, such that the light is confined to a relatively narrow line. The two ends of the crystal are cleared to form perfectly smooth, parallel edges, forming a Fabry Pérot resonator. Photons emitted into a mode of the waveguide will travel along the waveguide and be reflected several times from each end face before they are emitted. As a light wave passes through the cavity, it is amplified by stimulated emission, but light is also lost due to absorption and by incomplete reflection from the end facets. Finally, if there is more amplification than loss, the diode begins to lase. Some important properties of laser diodes are determined by the geometry of the optical cavity. Generally, in the vertical direction, the light is contained in a very thin layer, and the structure supports only a single optical mode in the direction perpendicular to the layers. The wavelength emitted is a function of the band-gap of the semiconductor and the modes of the optical cavity. The width of the gain curve will determine the number of additional side modes that may also lase, depending on the operating conditions. Single spatial mode lasers that can support multiple longitudinal modes are called Fabry Perot (FP) lasers. An FP laser will lase at multiple cavity modes within the gain bandwidth of the gain medium. The number of lasing modes in an FP laser is usually unstable, and can fluctuate due to changes in current or temperature. Single frequency diode lasers are either distributed feedback (DFB) lasers or distributed Bragg reflector (DBR) lasers. Fig 1. Laser diode example 3
DBR and DFB lasers The standard Fabry Perot Lasers are not wavelength selective. This leads to lasing of many modes and allows for mode jumps. A possible method is to insert an optical feedback in the device to eliminate other frequencies. Periodic gratings incorporated within the lasers waveguide can be utilized as a means of optical feedback. Devices incorporating the grating in the pumped region are termed Distributed Feedback (DFB) lasers, while those incorporating the grating in the passive region are termed Distributed Bragg Reflector (DBR) Laser. DFB and DBR lasers oscillate in a single-longitudinal mode even under high-speed modulation, in contrast to Fabry-Perot lasers, which exhibit multiple-longitudinal mode oscillation when pulsed rapidly Fig 2. DBR vs. DFB Lasers The gratings or distributed Bragg reflectors (DBRs) are used for one or both cavity mirrors. The grating consists of corrugations with a periodic structure. They are used because of their frequency selectivity of single axial mode operation. The period of grating is chosen as half of the average optical wavelength, which leads to a constructive interference between the reflected beams. A DBR Laser can be formed by replacing one or both of the discrete laser mirrors with a passive grating reflector. Besides the single frequency property provided by the frequencyselective grating mirrors, this laser can include wide tunability. Since the refractive index depends on the carrier density, this can be exploited to vary the refractive index electro optically on the sections by separate electrodes. A distributed feedback laser (DFB) also uses grating mirrors, but the grating is included in the gain region. Reflections from the ends are suppressed by antireflection coatings. Thus, it is possible to make a laser from a single grating, although it is desirable to have at least a fraction of a wavelength shift near the center to facilitate lasing at the Bragg frequency. The pure DFB structure in fact will lead to the oscillation of two symmetrical modes, but not at the Bragg frequency. Adding a perturbation like a quarter wavelength shift leads to single mode operation at Bragg frequency. DFB Facts DFB Lasers are easier to fabricate and show fewer losses and therefore have a lower threshold current. The DBR is widely tunable, but relatively complex since a lot of structure must be created along the surface of the wafer. For this reason DBR Lasers are only formed when their properties are required. Both lasers however work in single mode. 4
VCSEL VCSEL has several advantages over the production process of edge-emitting lasers. Edge-emitters cannot be tested until the end of the production. If the edge-emitter does not work as per specification, the production time and the processing materials have been wasted. VCSELs however, can be tested at intermediate steps to check for material quality and processing issues. Additionally, because VCSELs emit the beam perpendicular to the active region of the laser as opposed to parallel as with an edge emitter, tens of thousands of VCSELs can be processed simultaneously. A 3 wafer can yield approximately 15.000 VCSELs but only about 4.000 edge emitting lasers of comparable power. Furthermore, the yield can be controlled to a more predictable outcome. Other VCSEL advantages include higher reliability, simple fiber coupling and packaging, all this results in lower cost. The drawback of VCSELs is that the longer the wavelength becomes, the more complicated is the fabrication. As of now they are not used in the 1550nm region. There are many designs of VCSEL structure; however they all have certain common aspects in common. The cavity length of VCSELs is very short typically 1-3 wavelengths of the emitted light. As a result, in a single pass of the cavity, a photon has a small chance of a triggering a stimulated emission event at low carrier densities. Therefore, VCSELs require highly reflective mirrors to be efficient. In edge-emitting lasers, the reflectivity of the facets is about 30%. For VCSELs, the reflectivity required for low threshold currents is greater than 99.9%. Such a high reflectivity can t be achieved by the use of metallic mirrors. VCSELs make, use Distributed Bragg Reflectors (DBRs). These are formed by laying down alternating layers of semiconductor or dielectric materials with a difference in refractive index. VCSELs for wavelengths from 650 nm to 1300 nm are typically based on gallium arsenide (GaAs) wafers with DBRs formed from GaAs and aluminium gallium arsenide (AlxGa(1-x)As). The refractive index of AlGaAs does vary relatively strongly as the Al fraction is increased, minimizing the number of layers required to form an efficient Bragg mirror compared to other candidate material systems. Furthermore, at high aluminium concentrations, an oxide can be formed from AlGaAs, and this oxide can be used to restrict the current in a VCSEL, enabling very low threshold currents. Since the DBR layers also carry the VCSEL advantages include higher reliability, simple fiber coupling and packaging, all this results in lower cost. current in the device, more layers increase the resistance of the device therefore dissipation of heat and growth may become a problem if the device is poorly designed. The figure below describes a realistic VCSEL implementation. 5
Fig 3. Description of VCSEL implementiation. Fig 3. VCSEL device structure, bottom emission (from Wikipedia) Fig 4. In the tables we report a comparison of the laser types used in optical communications. Fig 4. Description of laser types 6
DWDM Application The emission wavelength of a DFB laser can be tuned acting on temperature, this fact has two direct consequences: first for application that requires transmission at a precise lambda, thermal control has to be provided (for example a thermoelectric cooler (TEC)), second changing the temperature can lead to a device able to produce different wavelengths. The second method give the possibility to realize tunable lasers, setting the working point at specific temperature and eventually - keeping it at a stable wavelength over time using a wavelength locker. One widely used structure comprises and Fabry-Perot etalon, it includes a beam splitter, an etalon, a reference photodiode, and an etalon photodiode. ROADM (reconfigurable add drop multiplexers) became the core of the optical transport systems. One of the first commercial full C-band tunable lasers consisted in a selectable array of DFB lasers that are combined in a multimode interference coupler. The DFBs are powered one at a time and each is manufactured with a slightly different grating pitch to offset their output wavelengths by about 3 or 4 nm. The chip is then temperature tuned by some 30 40 C to access the wavelengths between the discrete values of the array elements. With N-DFB elements, then, a wavelength range of up to about 4N nm can be accessed, or with 8 10 elements the entire C-band can be accessed [2]. The reference photodiode measures the laser output directly (after splitter) and the other measures the transmission through the etalon. The coupling ratio of the splitter in the wavelength locker is designed such that at each exact ITU channel, the optical power levels falling on the two photodiodes are equal. As the laser frequency changes while the etalon detector photocurrent varies periodically, the ratio of the two etalon and reference photodetector currents remains constant at the lock point. Therefore, by monitoring the change in the ratio of the two photocurrents, the wavelength of the laser can be monitored and stabilized. Lasers that are tunable over multiple wavelengths, and progressively over the whole C-band appeared at the beginning of the last decade. They soon became widely used in DWDM systems since they allowed the reduction of part numbers (before a different line card or transceiver was needed for every different wavelength) and together with Fig 5. DFB Laser array (source Fujitsu Laboratories) Another example is an external-cavity laser. In this case a gain block is coupled to external modeselection filtering and tuning elements via bulk optical elements. The cavity phase adjustment, necessary to properly align the mode with the filter peak and the desired ITU grid wavelength, can be included in one of several places e.g. on the gain block or by fine tuning the mirror position. In most external-cavity approaches the mode selection filter is a diffraction grating that can also double as a mirror. In this case, a retroreflecting mirror is translated as it is rotated. 7
This combined motion changes the effective cavity length in proportion to the change in center wavelength of the mode-selection filter to track the movement of a single cavity mode. An obvious concern with these structures is their manufacturability and reliability, given the need for assembling numerous microoptical parts and holding them in precise alignment. Fig 6. Explaination of Litterman-Metcalf cavity. Fig 6. Littman-Metcalf cavity (Source New Focus) More recent approaches that are well suited for monolithic integration are variations of the DBR structure. In the SGDBR the wider tuning range filter is provided by the product of the two differently spaced and independently tuned reflection combs of the SGDBRs at each end of the cavity (front mirror and rear mirror). Good side-mode suppression has been demonstrated, and tuning of over 40 nm is easily accomplished, but due to grating losses resulting from current injection for tuning, the differential efficiency and chip output powers can be somewhat limited. In the case of the SGDBR, this is easily addressed by the incorporation of another gain section on the output side of the output mirror. A variation of this concept is the Y-branch structure, where the combination of two slightly different reflectors located in the two Y arms selects the wavelength that can be emitted. Structures like this are suitable for monolithic integration of a Mach-Zehnder modulator in a smaller footprint and low power way compared to hybrid packaged or fiber-coupled devices. In addition, the chip can be tailored for each channel across the wavelength band by adjusting the biases to the two legs of the MZM. The compact size of devices like this made it possible their integration in a transmitter whose small size was compatible with 10Gbit/s pluggable interfaces (XFP and SFP+). 8
Fig 7. Description of Monolithycally Intergrated Transmitter. Fig 7. Monolithycally integrated Transmitter (Agility) Fig 8. Recap of the different tunable laser schemes. Fig 8. Tunable laser technology comparison References [1] Larry Coldren; Scott Corzine; Milan Mashanovitch (2012). Diode Lasers and Photonic Integrated Circuits (Second ed.). John Wiley and Sons. [2] Coldren et al. Tunable Semiconductor Lasers: A Tutorial JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 1, JANUARY 2004 9