High-speed lasers for future optical interconnects
Directly modulated semiconductor lasers that operate at ≫10Gb/s are needed to meet the requirements of developing standards for high-capacity short-reach data-communication links. For these multimode fiber-based connections, vertical-cavity surface-emitting lasers (VCSELs) offer significant advantages, including high modulation speeds at low currents, low power consumption, and high beam quality, as well as low manufacturing cost. Commercially available VCSELs cannot operate beyond 10–20Gb/s. However, this will be required in future universal-serial-bus devices and high-definition video links, for example.
The speed of a semiconductor laser can, in principle, be increased by scaling down its volume so that a higher internal photon density (and thus a higher resonance frequency) is achieved. However, this comes at the expense of reduced reliability because of the laser's increased current density.1 Recent developments include high-speed VCSELs that can operate at up to 40Gb/s but with unacceptably high current densities and at nonstandard wavelengths (980nm and 1.1μm).2,3 As an alternative, we have explored boosting the speed of standard 850nm VCSELs (which are compatible with high-bandwidth multimode fibers) without increasing their current density. We showed that sophisticated tailoring of the design for high-speed operation implies that 10Gb/s should be far from the upper limit for reliable long-term operation.
The structure of traditional gallium arsenide (GaAs)-based 850nm VCSELs consists of an active region—containing one to three GaAs quantum wells (QWs)—that is sandwiched between two highly reflecting distributed-Bragg reflectors (DBRs) composed of hundreds of aluminum-GaAs (AlGaAs) layers of different compositions. The resulting effective resonator length is very small (typically only 1–3 wavelengths long), which offers advantages compared to other semiconductor lasers in terms of low operating currents and power consumption, and high modulation bandwidth. However, to electrically pump the active region, the DBRs must also conduct current. The large number of layers in the DBRs leads to high electrical resistance and problems with, for example, self-heating. These issues become more pronounced as the VCSEL size is reduced.
In addition to the photon density, the intrinsic speed limitation of a semiconductor laser is determined by the differential gain of the active region and the photon lifetime of the resonator. To improve the intrinsic properties, we doubled the differential gain by employing In0.1Ga0.9As (In: Indium) instead of GaAs QWs. To maintain emission at 850nm, we marginally reduced the QW thickness and slightly modified the QW barriers.4 A given differential gain corresponds to an optimal photon lifetime. We finetuned the resonator's photon lifetime by shallow etching on the VCSEL's top surface (~40nm).
The laser's resonance frequency increases with the current's square root. However, it will eventually saturate because of self-heating effects. Good thermal management is, therefore, crucial to reach higher speeds. To improve the thermal properties, we used a complex modulation-doping scheme in the DBRs to reduce the electrical resistance. We also replaced several AlGaAs layers in the bottom DBR with aluminum arsenide to increase the thermal conductivity. In addition to thermal and intrinsic speed limitations, it is important to isolate electrical parasitics from the distributed resistances and capacitances associated with the device structure and geometry. (These parasitics effectively form a low-pass filter and limit device performance at high modulation frequencies.) One of the major contributions to parasitic speed limitations originates in the oxide aperture, a thin layer of selectively oxidized AlGaAs confining both the current and the optical field. We used a double- rather than single-layered oxide aperture to reduce the oxide capacitance by almost half. For further capacitance reduction, we included four additional oxide layers but with an aperture that was twice as large. Figure 1 shows a schematic of our VCSEL's structure.
With these speed-improvement measures, we achieved a record 3dB modulation bandwidth of 22.5GHz at 850nm. The aperture diameter was 7μm, and if the aperture of the device were increased to 13μm, the bandwidth would still be as large as 20GHz. The resulting VCSEL operated at a current density below that used in today's commercial 10Gb/s devices (10kA/cm2 industry benchmark for reliability). For a device with a 9μm aperture (20.5GHz bandwidth), we demonstrated successful data transmission at very high data rates. In Figure 2 we show eye diagrams and bit-error rates (BER) for on-off keying transmission over 100 and 50m multimode fibers at 25 and 32Gb/s, respectively.5 We achieved error-free transmission (BER<10−12) at both bit rates for 11kA/cm2. We accomplished 40Gb/s transmission over 200m multimode fiber by instead implementing a more spectrally efficient modulation format (16 quadrature amplitude modulation subcarrier multiplexing) compared to on-off keying (see Figure 3).
Development of low-cost 850nm VCSELs capable of transmitting data at very high bit rates without sacrificing reliability is needed to meet near-future demands for short-distance data communication. By implementing minor modifications into the standard 850nm design, we have demonstrated that transmission rates of up to 40Gb/s can be attained for acceptable current densities. Our next step will be to further reduce thermal effects and electrical parasitics to improve performance.
This work was supported in part by the European Seventh Framework Programme's project VISIT (FP7-224211) and the Swedish Foundation for Strategic Research.
