Quantum-dot comb laser with low relative-intensity noise for each mode
The field of silicon-based photonics is currently enjoying an explosion of research interest.1 Future progress in information processing and increased computer speed depend both on integrating micro- and optoelectronic devices and developing optical integrated circuits. Electrical interconnects have inherent bandwidth limitations that are spurring the quest for higher-bandwidth optical alternatives. The dense wavelength division multiplexing (DWDM) approach appears to be especially attractive for cost-efficient high-bandwidth chip-to-chip and board-to-board communications. In contrast to parallel optical solutions, DWDM relies on a single, simple, and cheap laser capable of emitting many spectrally separated channels. The use of 100 channels at 10Gb/s modulation would give 1Tb/s total bandwidth in one link. Because efficient lasing in silicon-based devices remains a challenge,2 future optical integrated circuits will likely include a III-V semiconductor laser integrated into a silicon chip.
An attractive candidate for such a multiwavelength emitter is an edge-emitting (or Fabry–Perot, abbreviated FP) semiconductor laser. Wavelengths that correspond to different longitudinal modes of the FP resonator can be filtered out, separately modulated, and used as independent optical channels. In this case, channel separation is naturally determined by resonator length. All channels can be stabilized and tracked simultaneously.
A prerequisite for application of a multiwavelength light source in DWDM systems is low relative-intensity noise of each longitudinal mode (modal RIN). In single-frequency lasers with a large side mode suppression ratio (>40dB), the values of RIN are rather low. These lasers are suitable for telecommunications applications. In case of the conventional quantum well (QW) multimode (multifrequency) lasers, the value of RIN rises significantly due to random redistribution of mode intensity between longitudinal laser modes. In contrast, in quantum dot (QD) lasers, the nonlinear gain saturation effect is pronounced, and a strong decrease in RIN for individual FP modes can be achieved.
We grew a QD laser structure by molecular beam epitaxy.3 The active area consisted of 10 planes of InAs/In0.15Ga0.85As QDs. Waveguide FP lasers with a 3μm-wide ridge were fabricated, with a cavity length of roughly 1mm. For eye-diagram and bit-error-rate (BER) evaluations, an individual longitudinal mode of the optically isolated laser was sifted out using a fiber FP tunable filter. The spectrally filtered mode was amplified by passing it through optical amplifiers and then externally modulated.
Figure 1(a) shows an overall emission spectrum of the QD laser taken at a continuous wave current of 85mA. The output power is 50mW per two laser facets. The spectrum comprises a series of longitudinal FP modes separated by 0.22nm. The zeroth-order mode, i.e., the mode of the maximum intensity, is centered at 1265.5nm. Its full width at half maximum is about 2nm. An external FP etalon can be adjusted to transmit only one longitudinal mode. We experimented with longitudinal modes that fall into a wavelength interval from 1263.3 to 1266.4nm. These modes are indicated in Figure 1(a) by an index ranging from −10 to +4. Figure 1(b) shows a filtered intensity of the −10th longitudinal mode.
Modal RIN for the spectrally filtered FP mode is shown in Figure 2. At low frequencies, the RIN spectrum is nearly flat at a level of −105dB/Hz. With higher frequencies, the modal RIN drops from −120 to −145dB/Hz in the 0.1–10GHz range. If a received power is sufficiently high, the BER is governed by the total RIN, i.e., an integral over the relative intensity noise spectrum. Our calculation has shown that the total RIN of 0.4% would be acceptable for error-free transmission (BER<10−15). Using the data in Figure 2, the total RIN was calculated to be 0.21% over the full frequency range of analysis (0.001–10GHz).
An individual longitudinal mode after spectral filtering was modulated at 10Gb/s by a 231−1 pseudorandom binary non-return-to-zero sequence (see Figure 3) and shows an eye pattern. We obtained error-free operation with a BER of less than 10−13. A similar BER was achieved for the modes with the highest intensity (mode order ranges from −7 to +2): see Figure 1(a).
The results demonstrate that we can reduce the requirements for precise lasing wavelengths because channel separation is naturally pre-determined by only one parameter, which is cavity length. All channels can be stabilized and tracked simultaneously. Sufficiently long wavelengths make such lasers compatible with optical and optoelectronic components based on silica fiber or silicon-based planar waveguides. The multimode QD laser shows promise for the development of simple and cost-effective high-bandwidth optical interconnects for silicon-based photonic integrated circuits.
Future optimization of the comb laser is aimed at increasing the intensity of individual FP modes and further improving their RIN level, as well as increasing the number of channels suitable for optical transmission. In cooperation with academic and industrial partners, we have also started to develop transmitters on photonic integrated circuits in III-V materials and on silicon photonic chips. In each case, the transmitter comprises a laser fiber-coupled to a demultiplexer that separates the comb input channels, modulators for each of the channels, a multiplexer that combines the modulated channels, and a coupler into the output fiber. We expect to demonstrate the feasibility of our comb laser's for short- to medium-reach WDM transmission, such as 100Gb/s Ethernet and passive optical networks.
Alexey Kovsh is CTO at Innolume Inc. He is known in the compound semiconductor community for his contributions to semiconductor QD technology, which have been recognized by a half-dozen scientific awards from various institutions.
