In the era of cloud computing, with services such as video streaming, social networking, and online storage and file sharing, the demand for online data processing and storage capacity is growing rapidly. These services are hosted in huge data centers that need not only fast servers but also fast communication between servers.1 Because copper cables have high attenuation at high frequencies, the most promising solution is to use fiber optical links (called optical interconnects) to connect different parts of the data center. Today, gallium arsenide-based vertical-cavity surface-emitting lasers (VCSELs) emitting at 850nm are the standard light source in transmitters used in commercially available optical interconnects, operating at up to 14Gbit/s with a link length of up to 300m. These lasers have the advantages of low power consumption, fast direct modulation at low currents, and low-cost manufacturing.2 However, with data centers growing ever larger, often into multibuilding complexes, there is a need for high-capacity and low-cost longer-reach optical interconnects for links exceeding the traditional maximum link length of 300m.1
Figure 1. Schematic view of a high-speed vertical-cavity surface-emitting laser (VCSEL) with an integrated mode filter. The figure shows the overlap of the fundamental (green) and first and second higher-order modes (blue and red) with the mode filter. The insets show the intensity distributions of the three modes (measured by spectrally resolved near-field imaging). DBR: Distributed Bragg reflector.
The main challenge in designing longer links with 850nm VCSELs is the wide optical spectrum of the laser, which leads to severe pulse broadening because of dispersion in the multimode fiber. A standard VCSEL has a short cavity with a quantum-well gain region sandwiched between two distributed Bragg reflectors (DBRs): see Figure 1. Because of the short cavity, the VCSEL lases in only one longitudinal mode. The transverse dimensions are defined by one or several selectively oxidized layers that force the current through an oxide aperture in the center of the device. Since the oxide has a low refractive index, it also provides transverse guiding of light. Due to large transverse dimensions, the VCSEL waveguide normally supports several transverse modes. These modes have slightly different wavelengths, leading to a wide optical spectrum with an RMS width that can be as large as 1nm (see Figure 2). The large spectral width and the different transverse mode distributions lead to pulse broadening along the multimode fiber due to chromatic and modal fiber dispersion. For longer reach or higher speed, the spectral width of the VCSEL must therefore be reduced.
The most common approach is to shrink the oxide aperture diameter down to just 2–3μm.3 A waveguide with small transverse dimensions will support only the fundamental transverse mode (just like a single-mode fiber). However, driving the injected current through such a small aperture will lead to increased resistance and larger self-heating of the device, which is detrimental to its high-speed properties. Smaller devices also have lower output power and operate at higher current density, which may compromise reliability.4
We have instead integrated a mode filter in our high-speed VCSEL.5 By making the topmost DBR layer λ/2 thick (where λ is wavelength) instead of the usual λ/4, the reflectivity of the top mirror is reduced, raising the threshold for lasing. The extra layer is then etched back to a thickness of λ/4 in the center of the VCSEL, locally increasing reflectivity. Since the fundamental mode is strongest in the middle of the waveguide, it will be better reflected than the higher-order modes, lowering the threshold for the fundamental mode. Higher-order modes with intensity distributions overlapping the low-reflective region will not have enough feedback to start lasing (see Figure 1). The shallow surface etch thus acts as an integrated mode filter.6 As Figure 2 shows, the mode filter reduces the spectral width by over 70% for a VCSEL with a 5μm-diameter oxide aperture.
Figure 2. Optical spectra at a bias current of 4mA for 5μm-diameter oxide aperture VCSELs without (left) and with (right) a mode filter. ΔλRMS: RMS spectral width.
Figure 3. Bit error rates (BER) plotted versus received optical power at room temperature over a variety of fiber lengths for VCSELs without (top) and with (bottom) a mode filter. Insets: Eye diagrams of the received signal.
We tested the VCSEL, with a modulation bandwidth of 19GHz, by using it to transmit at 25Gbit/s over standard multimode OM3+ fiber. The test parameter is the bit error rate (BER), which is the number of erroneously detected bits divided by the number of transmitted bits. For the transmission to be ‘error-free,’ the BER should be less than 10−12 (less than one incorrect bit in 1000 billion bits). At a bias of only 3.5mA, the mode filter VCSEL could transmit at 25Gbit/s over 500m of fiber error-free7 (see Figure 3). The VCSEL without mode filter could not transmit error-free even at 100m due to the large spectral width. Figure 3 also shows eye diagrams of the received signal, formed by overlaying several of the received data patterns on an oscilloscope. These show severe signal distortion after 300m when using the VCSEL without a mode filter, whereas the VCSEL with a mode filter produces clear and open eyes even after 500m.
The construction of ever-larger data centers requires longer high-speed optical interconnects. Common low-cost 850nm VCSELs have optical spectra that are too wide, leading to impairments due to fiber dispersion. An integrated mode filter in the form of a shallow surface etch enables the manufacturing of low-spectral-width VCSELs with relatively large oxide apertures. We have shown that a mode-filtered VCSEL can be used to transmit data at 25Gbit/s over 500m of multimode fiber. Our next step will be to investigate in detail how the mode filter influences the dynamics and bandwidth of the VCSEL to optimize the design for both low spectral width and high speed.
This work was supported by the European Commission Seventh Framework Programme project VISIT (224211) and the Swedish Foundation for Strategic Research project LASTECH.
Erik Haglund, Åsa Haglund, Petter Westbergh, Johan Gustavsson, Benjamin Kögel, Anders Larsson
Department of Microtechnology and Nanoscience
Chalmers University of Technology
1. C. F. Lam, H. Liu, B. Koley, X. Zhao, V. Kamalov, V. Gill, Fiber optic communication technologies: what's needed for datacenter network operations, IEEE Commun. Mag.
48, p. 32-39, 2010. doi:10.1109/MCOM.2010.5496876
2. A. Larsson, Advances in VCSELs for communication and sensing, IEEE J. Sel. Top. Quantum Electron.
17(6), p. 1552-1567, 2011. doi:10.1109/JSTQE.2011.2119469
3. R. Safaisini, K. Szczerba, E. Haglund, P. Westbergh, J. S. Gustavsson, A. Larsson, P. A. Andrekson, 20Gb/s error-free operation of 850nm oxide-confined VCSELs beyond 1km of multi-mode fibre, Electron. Lett. (To be published.)
4. B. M. Hawkins, R. A. Hawthorne III, J. K. Guenter, J. A. Tatum, J. R. Biard, Reliability of various size oxide aperture VCSELs, Proc. 52nd Electron. Comp. Technol. Conf.
, p. 540-550, 2002. doi:10.1109/ECTC.2002.1008148
5. P. Westbergh, J. S. Gustavsson, B. Kögel, Å. Haglund, A. Larsson, Impact of photon lifetime on high-speed VCSEL performance, IEEE J. Sel. Top. Quantum Electron.
17(6), p. 1603-1613, 2011. doi:10.1109/JSTQE.2011.2114642
6. Å. Haglund, J. S. Gustavsson, J. A. Vukusic, P. Modh, A. Larsson, Single fundamental-mode output power exceeding 6 mW from VCSELs with a shallow surface relief, IEEE Photon. Technol. Lett.
16(2), p. 368-370, 2004. doi:10.1109/LPT.2003.821085
7. E. Haglund, Å. Haglund, P. Westbergh, J. S. Gustavsson, B. Kögel, A. Larsson, 25Gbit/s transmission over 500 m multimode fibre using 850nm VCSEL with integrated mode filter, Electron. Lett.
48(9), p. 517-519, 2012. doi:10.1049/el.2012.0529