Academia.eduAcademia.edu

Figure 9 – uploaded by David Fattal

See full PDFdownloadDownload figure
Fig. 9. Comparison of measured AWG spectrum and comb spectrum at 1310 nm, for the setup in Figure 8. For the C-band AWG, which is periodic in free spectral range (FSR) [26], it is simply a matter of designing the comb laser to fit the appropriate period, as illustrated in Figure 9. Channel spacing increases at shorter wavelengths, e.g., from 50 GHz at 1550 nm to 60 GHz at 1310 nm. Spacing is aligned by adjusting laser cavity length, and overlap improved by tuning with current or temperature.

Figure 9 Comparison of measured AWG spectrum and comb spectrum at 1310 nm, for the setup in Figure 8. For the C-band AWG, which is periodic in free spectral range (FSR) [26], it is simply a matter of designing the comb laser to fit the appropriate period, as illustrated in Figure 9. Channel spacing increases at shorter wavelengths, e.g., from 50 GHz at 1550 nm to 60 GHz at 1310 nm. Spacing is aligned by adjusting laser cavity length, and overlap improved by tuning with current or temperature.

Related Figures (14)

Therefore, the question is how to make WDM practical for short reach interconnects? Solutions are needed at the cable, board, and eventually the CMP levels, as well as for PON. Today the bottleneck is the banks of lasers and modulators plus drivers, etc., e.g., Fig. 1, as well as temperature control. This complexity can be mitigated to some extent by integration, e.g., Infinera’s 10-laser WDM telecom transmitter on an InP photonic integrated circuit (PIC) [9]. However, at high volumes the number of lasers is limited by yield and system performance, e.g., Cyoptics has stated that their most cost-effective number of DFB InP lasers integrated on a Tx PIC is around five [10]. Alternatively, Intel proposes bonding and coupling an array of lasers to a silicon photonics chip; shown in the right drawing in Fig. 1, from [11]. To this end, Bowers’ group at UCSB recently demonstrated laser bonding/coupling of electrically pumped DFB lasers [12]. Nonetheless, the degree of difficulty in scaling this solution appears daunting.
Therefore, the question is how to make WDM practical for short reach interconnects? Solutions are needed at the cable, board, and eventually the CMP levels, as well as for PON. Today the bottleneck is the banks of lasers and modulators plus drivers, etc., e.g., Fig. 1, as well as temperature control. This complexity can be mitigated to some extent by integration, e.g., Infinera’s 10-laser WDM telecom transmitter on an InP photonic integrated circuit (PIC) [9]. However, at high volumes the number of lasers is limited by yield and system performance, e.g., Cyoptics has stated that their most cost-effective number of DFB InP lasers integrated on a Tx PIC is around five [10]. Alternatively, Intel proposes bonding and coupling an array of lasers to a silicon photonics chip; shown in the right drawing in Fig. 1, from [11]. To this end, Bowers’ group at UCSB recently demonstrated laser bonding/coupling of electrically pumped DFB lasers [12]. Nonetheless, the degree of difficulty in scaling this solution appears daunting.
In this paper we show that with a suitably engineered quantum dot gain stripe, a relatively simple semiconductor laser diode can emit a stable comb of wavelengths that are suitable for WDM. Furthermore, comb output is obtained with continuous wave (CW) emission, i.e., without obvious mode locking. For example, a measured spectrum is plotted in Fig. 2 from an early prototype. The spectrum does not change significantly over the laser’s operational envelope and lifetime, and time-variation of peak intensity is small. The resulting low relative intensity noise (RIN), also illustrated in Fig. 2 from a similar laser, makes each channel suitable for high speed modulation. The following sections describe the multi-channel comb laser, RIN and BER measurements, eye diagrams under direct modulation, operating wavelength range, photonic circuitry and components, and commodity optical coupling.
In this paper we show that with a suitably engineered quantum dot gain stripe, a relatively simple semiconductor laser diode can emit a stable comb of wavelengths that are suitable for WDM. Furthermore, comb output is obtained with continuous wave (CW) emission, i.e., without obvious mode locking. For example, a measured spectrum is plotted in Fig. 2 from an early prototype. The spectrum does not change significantly over the laser’s operational envelope and lifetime, and time-variation of peak intensity is small. The resulting low relative intensity noise (RIN), also illustrated in Fig. 2 from a similar laser, makes each channel suitable for high speed modulation. The following sections describe the multi-channel comb laser, RIN and BER measurements, eye diagrams under direct modulation, operating wavelength range, photonic circuitry and components, and commodity optical coupling.
Figure 3 shows an overall emission spectrum of a quantum dot laser taken at CW current of 85 mA, which is 5x threshold. Optical output power is 50 mW from the two facets. The spectrum comprises a series of longitudinal FP modes separated by 0.22 nm or 40 GHz. The Oth order mode (maximum intensity) is centered at 1265.5 nm. FWHM of the 10 highest power modes, -7th to +2nd, is about 2 nm. An external FP etalon was adjusted to transmit only one longitudinal mode and suppress the others [22]. For example, Fig. 3 includes a filtered intensity of the -10th longitudinal mode. At least 19 dB of side mode suppression was achieved for all filtered modes. Fig. 3. Emission spectra of a nominal 1260 nm quantum dot comb laser (left) and one of the spectrally filtered modes (right) using a fiber FP tuneable filter [22].
Figure 3 shows an overall emission spectrum of a quantum dot laser taken at CW current of 85 mA, which is 5x threshold. Optical output power is 50 mW from the two facets. The spectrum comprises a series of longitudinal FP modes separated by 0.22 nm or 40 GHz. The Oth order mode (maximum intensity) is centered at 1265.5 nm. FWHM of the 10 highest power modes, -7th to +2nd, is about 2 nm. An external FP etalon was adjusted to transmit only one longitudinal mode and suppress the others [22]. For example, Fig. 3 includes a filtered intensity of the -10th longitudinal mode. At least 19 dB of side mode suppression was achieved for all filtered modes. Fig. 3. Emission spectra of a nominal 1260 nm quantum dot comb laser (left) and one of the spectrally filtered modes (right) using a fiber FP tuneable filter [22].
Fig. 4. Modal RIN typical of the filtered -7" to +2" modes in Fig. 3 (left) and eye diagram for the -6" mode. and a % decision level, minimum BER as a function of total RIN can be estimated as BER = erfc[(2V2RIN) 'J/2, where erfc is the complementary error function. A calculation shows that a total RIN of less than 0.004 would yield error-free transmission (BER < 10°’) [22].From the data in Fig. 4, total RIN is 0.0021 over the full frequency range of analysis (1 MHz to 10 GHz). These results indicate that this multi-longitudinal mode, quantum dot comb laser is a practical source for high-speed WDM data transmission.
Fig. 4. Modal RIN typical of the filtered -7" to +2" modes in Fig. 3 (left) and eye diagram for the -6" mode. and a % decision level, minimum BER as a function of total RIN can be estimated as BER = erfc[(2V2RIN) 'J/2, where erfc is the complementary error function. A calculation shows that a total RIN of less than 0.004 would yield error-free transmission (BER < 10°’) [22].From the data in Fig. 4, total RIN is 0.0021 over the full frequency range of analysis (1 MHz to 10 GHz). These results indicate that this multi-longitudinal mode, quantum dot comb laser is a practical source for high-speed WDM data transmission.
Fig. 5. Spectra and total RIN measurements of nominal 1320 nm comb lasers with channel spacing of 27 GHz (left) and 40 GHz (right). RIN measurements are used to screen and optimize QD comb laser fabrication process and design modifications. For example, recent measurement of two nominal 1320 nm lasers with different lengths are illustrated in Fig. 5, yielding combs with 27 GHz and 40 GHz channel spacing. The detector is a 10 Gb/s surface mount coplanar APD with a preamp receiver from Bookham, and the RF spectrum analyzer is an HP8563E. Total RIN is 0.001 + 0.0004 for the central channels. RIN of the 1328.05 nm peak for the left spectrum in Fig. 5 is shown in Figure 1. These results illustrate that within 3 dB of the peak, as many as 50 comb channels with low RIN are available for error-free transmission. Power is approximately 1 mW/channel in this case.
Fig. 5. Spectra and total RIN measurements of nominal 1320 nm comb lasers with channel spacing of 27 GHz (left) and 40 GHz (right). RIN measurements are used to screen and optimize QD comb laser fabrication process and design modifications. For example, recent measurement of two nominal 1320 nm lasers with different lengths are illustrated in Fig. 5, yielding combs with 27 GHz and 40 GHz channel spacing. The detector is a 10 Gb/s surface mount coplanar APD with a preamp receiver from Bookham, and the RF spectrum analyzer is an HP8563E. Total RIN is 0.001 + 0.0004 for the central channels. RIN of the 1328.05 nm peak for the left spectrum in Fig. 5 is shown in Figure 1. These results illustrate that within 3 dB of the peak, as many as 50 comb channels with low RIN are available for error-free transmission. Power is approximately 1 mW/channel in this case.
Fig. 6. Modulated comb spectrum driven by a 1.25 Gb/s, +2 Vpp, 2-1 PRBS (left) and frequency response from a DC-to-10 GHz sweep (right). Figure 6 shows the laser’s modulated comb spectrum (left) and frequency response (right). The laser is modulated by a 1.25 Gb/s, +2 Vpp, 2'°-1 pseudorandom binary sequence (PRBS) from an Anritsu MP1632C. The modulated spectrum is averaged over many periods since of course the optical spectrum analyzer (Agilent 86142B OSA) does not register the dynamics. Nonetheless, the plot shows negligible spectral broadening or distortion, indicating that chirp is very small compared to channel spacing. The frequency response is obtained from a frequency sweep, DC to 10 GHz. The laser’s 3 dB bandwidth is seen to be approximately 1.25 GHz so this sample was deemed suitable for further 1.25 Gb/s direct modulation experiments.
Fig. 6. Modulated comb spectrum driven by a 1.25 Gb/s, +2 Vpp, 2-1 PRBS (left) and frequency response from a DC-to-10 GHz sweep (right). Figure 6 shows the laser’s modulated comb spectrum (left) and frequency response (right). The laser is modulated by a 1.25 Gb/s, +2 Vpp, 2'°-1 pseudorandom binary sequence (PRBS) from an Anritsu MP1632C. The modulated spectrum is averaged over many periods since of course the optical spectrum analyzer (Agilent 86142B OSA) does not register the dynamics. Nonetheless, the plot shows negligible spectral broadening or distortion, indicating that chirp is very small compared to channel spacing. The frequency response is obtained from a frequency sweep, DC to 10 GHz. The laser’s 3 dB bandwidth is seen to be approximately 1.25 GHz so this sample was deemed suitable for further 1.25 Gb/s direct modulation experiments.
Fig. 7. Preliminary 1.25 Gb/s eye diagrams obtained by modulating the entire spectrum (left) and a single filtered mode (right). A Newport D-15ir detector was used for the entire spectrum eye measurement and an APD for the filtered mode case. These are interesting results since they suggest that, in addition to the use of comb lasers as CW injection seed sources, directly modulated comb lasers have potential applications to alternative WDM-PON transmission systems. Issues of power, channel count, and temperature stability are, of course, paramount, but the QD laser is well suited to address all three. These WDM-PON systems will compete with 10G PON, namely 10G EPON (IEEE) or NG- PON (FSAN/ITU-T) in the future high speed PON market, hence, motivation to optimize the comb laser for it.
Fig. 7. Preliminary 1.25 Gb/s eye diagrams obtained by modulating the entire spectrum (left) and a single filtered mode (right). A Newport D-15ir detector was used for the entire spectrum eye measurement and an APD for the filtered mode case. These are interesting results since they suggest that, in addition to the use of comb lasers as CW injection seed sources, directly modulated comb lasers have potential applications to alternative WDM-PON transmission systems. Issues of power, channel count, and temperature stability are, of course, paramount, but the QD laser is well suited to address all three. These WDM-PON systems will compete with 10G PON, namely 10G EPON (IEEE) or NG- PON (FSAN/ITU-T) in the future high speed PON market, hence, motivation to optimize the comb laser for it.
Thus, for the foreseeable future WDM optical interconnects powered by quantum dot comb lasers are available in the O-band only. This is not a fundamental problem since the slightly higher fiber loss at 1310 nm versus 1550 nm is irrelevant for short reach. The difficulty arises from the convenience of experimentation or system prototyping at 1550 nm by virtue of equipment availability and capability. For example, amplification may be useful, so EDFAs, which do not work well at 1310 nm, make 1550 nm an attractive wavelength. More to the point, demultiplexing of the comb laser output requires a grating or a filter. Arrayed waveguide gratings (AWGs) are most convenient, e.g., Fig. 8, but virtually all off-the-shelf AWGs are designed for C-band telecom operation. Similar circumstances dictate off-the-shelf modulators and drivers, e.g., our 1550 nm LiNbO; modulator for the BER measurements in Section 3. Fig. 8. 1550 nm, 16-channel AWG with 50 GHz spacing used for DEMUX of 1310 nm comb (lower left).
Thus, for the foreseeable future WDM optical interconnects powered by quantum dot comb lasers are available in the O-band only. This is not a fundamental problem since the slightly higher fiber loss at 1310 nm versus 1550 nm is irrelevant for short reach. The difficulty arises from the convenience of experimentation or system prototyping at 1550 nm by virtue of equipment availability and capability. For example, amplification may be useful, so EDFAs, which do not work well at 1310 nm, make 1550 nm an attractive wavelength. More to the point, demultiplexing of the comb laser output requires a grating or a filter. Arrayed waveguide gratings (AWGs) are most convenient, e.g., Fig. 8, but virtually all off-the-shelf AWGs are designed for C-band telecom operation. Similar circumstances dictate off-the-shelf modulators and drivers, e.g., our 1550 nm LiNbO; modulator for the BER measurements in Section 3. Fig. 8. 1550 nm, 16-channel AWG with 50 GHz spacing used for DEMUX of 1310 nm comb (lower left).
Fig. 10. Comparison of the 1310 nm comb laser spectrum before (left) and after (right) the 1550 nm ring resonator waveguide circuit. Channel extinction will increase for a 1310 nm ring design. In the case of ring resonators designed for 1550 nm, there is no alternative but to redesign for 1310 nm operation. For example, consider comb laser channel extinction in 3 um diameter rings made of silicon-on-insulator (SOI) strip waveguides [27] provided by Beausoleil, et al. at HP Labs. Since the micro ring is designed for 1550 nm, the extinction ratio drops from 10 dB to 3 dB at 1310 nm. This is ‘simply’ addressed by a redesign with different waveguide width and gap. The Q factor measured from experiment was 3100 at 1310 nm compared to 4700 at 1550 nm. To match the micro ring resonance wavelength, the comb emission spectrum is tuned with a TEC temperature controller. Since channel spacing of the micro ring sample (10 nm) is much greater than that of the comb laser (0.32nm), only one channel overlapped. The spectrum before and after the ring resonator is shown in Figure 10. The dropped channel is circled. We expect to have at least 10 channels matching simultaneously when we test 1310 nm samples with 80 GHz channel spacing, but this preliminary result shows the feasibility of utilizing our quantum dot comb laser as the broadband source in HP micro ring based WDM system.
Fig. 10. Comparison of the 1310 nm comb laser spectrum before (left) and after (right) the 1550 nm ring resonator waveguide circuit. Channel extinction will increase for a 1310 nm ring design. In the case of ring resonators designed for 1550 nm, there is no alternative but to redesign for 1310 nm operation. For example, consider comb laser channel extinction in 3 um diameter rings made of silicon-on-insulator (SOI) strip waveguides [27] provided by Beausoleil, et al. at HP Labs. Since the micro ring is designed for 1550 nm, the extinction ratio drops from 10 dB to 3 dB at 1310 nm. This is ‘simply’ addressed by a redesign with different waveguide width and gap. The Q factor measured from experiment was 3100 at 1310 nm compared to 4700 at 1550 nm. To match the micro ring resonance wavelength, the comb emission spectrum is tuned with a TEC temperature controller. Since channel spacing of the micro ring sample (10 nm) is much greater than that of the comb laser (0.32nm), only one channel overlapped. The spectrum before and after the ring resonator is shown in Figure 10. The dropped channel is circled. We expect to have at least 10 channels matching simultaneously when we test 1310 nm samples with 80 GHz channel spacing, but this preliminary result shows the feasibility of utilizing our quantum dot comb laser as the broadband source in HP micro ring based WDM system.
Fig. 11. WDM transmitter schematic (left) for a single comb laser source instead of an array of lasers. DEMUX, modulator array, and MUX are integrated on a single PIC (cartoon on right), which might mount a flip-chip modulator driver die. The comb laser effectively offers a new “integrated” laser paradigm. Its WDM transmitter is illustrated in Fig. 11, where the laser/driver arrays in the conventional transmitter, Fig. 1, are replaced by the single comb laser/driver and a demultiplexer. This appears to be a good tradeoff for WDM since it replaces many lasers, integrated or discrete, by a single laser and one more DEMUX device. The DEMUX and MUX are AWGs (or echelle gratings), which need to be integrated with the modulator array on the PIC. See Fig. 11 for a cartoon. The comb design doubles (DE)MUxX real estate and insertion loss, so compactness and efficiency are crucial.
Fig. 11. WDM transmitter schematic (left) for a single comb laser source instead of an array of lasers. DEMUX, modulator array, and MUX are integrated on a single PIC (cartoon on right), which might mount a flip-chip modulator driver die. The comb laser effectively offers a new “integrated” laser paradigm. Its WDM transmitter is illustrated in Fig. 11, where the laser/driver arrays in the conventional transmitter, Fig. 1, are replaced by the single comb laser/driver and a demultiplexer. This appears to be a good tradeoff for WDM since it replaces many lasers, integrated or discrete, by a single laser and one more DEMUX device. The DEMUX and MUX are AWGs (or echelle gratings), which need to be integrated with the modulator array on the PIC. See Fig. 11 for a cartoon. The comb design doubles (DE)MUxX real estate and insertion loss, so compactness and efficiency are crucial.
Fig. 12. Example of a high channel count comb spectrum covering 1185-1222 nm with >1 mW/channel. The scaling advantage of comb laser WDM over conventional multi-laser WDM follows from the large number of channels that can be created at negligible added cost. Figure 12 shows a 200 channel comb spectrum case. For example, instead of 10 lasers modulated at 40 Gb/s (400 Gb/s total), the same throughput is achieved by one laser with 40 channels, each modulated at 10 Gb/s. Nine lasers are saved at the expense of 30 additional, but slower, modulators/drivers, 40 total, plus the fixed DEMUX cost. This illustrates the general tradeoff, lasers against modulators, but the comb WDM modulators are relatively low speed and less expensive. A more extreme case is, of course, 100 channels modulated at 10 Gb/s (1 Tb/s total), requiring 100 modulators and drivers. But at this stage it is ‘just’ a matter of integrating more of the same components, at which semiconductor fabrication processes excel. Nonetheless, careful scaling analyses will have to be done to fully understand the tradeoffs.
Fig. 12. Example of a high channel count comb spectrum covering 1185-1222 nm with >1 mW/channel. The scaling advantage of comb laser WDM over conventional multi-laser WDM follows from the large number of channels that can be created at negligible added cost. Figure 12 shows a 200 channel comb spectrum case. For example, instead of 10 lasers modulated at 40 Gb/s (400 Gb/s total), the same throughput is achieved by one laser with 40 channels, each modulated at 10 Gb/s. Nine lasers are saved at the expense of 30 additional, but slower, modulators/drivers, 40 total, plus the fixed DEMUX cost. This illustrates the general tradeoff, lasers against modulators, but the comb WDM modulators are relatively low speed and less expensive. A more extreme case is, of course, 100 channels modulated at 10 Gb/s (1 Tb/s total), requiring 100 modulators and drivers. But at this stage it is ‘just’ a matter of integrating more of the same components, at which semiconductor fabrication processes excel. Nonetheless, careful scaling analyses will have to be done to fully understand the tradeoffs.
We fabricated the filter bank shown in Fig. 14-(a), comprising 4 rings with nominal diameters around 4.5 um. The difference in diameter of 2 adjacent rings is 2.6 nm and is calculated to give a nominal channel separation of 80 GHz. The filter bank was fabricated using a converted scanning electron microscope e-beam tool. The silicon waveguides are 450 nm wide and 250 nm tall and sit on a 3-1m silicon oxide layer. Figure 14-(b) shows the transmission spectrum of the filter bank. The filters have a rejection dip of more than 5 dB and a FWHM of approximately 0.09 nm that corresponds to 14 GHz. Quality factor is approximately 15000. The channel spacing is irregular due to fabrication errors and ranges from 40 to 260 GHz. Fig. 13: (a) Microphotograph of a filter bank comprising four 4.5 jum ring resonators. (b) Transmission curve of the filter bank.
We fabricated the filter bank shown in Fig. 14-(a), comprising 4 rings with nominal diameters around 4.5 um. The difference in diameter of 2 adjacent rings is 2.6 nm and is calculated to give a nominal channel separation of 80 GHz. The filter bank was fabricated using a converted scanning electron microscope e-beam tool. The silicon waveguides are 450 nm wide and 250 nm tall and sit on a 3-1m silicon oxide layer. Figure 14-(b) shows the transmission spectrum of the filter bank. The filters have a rejection dip of more than 5 dB and a FWHM of approximately 0.09 nm that corresponds to 14 GHz. Quality factor is approximately 15000. The channel spacing is irregular due to fabrication errors and ranges from 40 to 260 GHz. Fig. 13: (a) Microphotograph of a filter bank comprising four 4.5 jum ring resonators. (b) Transmission curve of the filter bank.
Fig. 14. Surface-mount optical coupler for low cost optical coupling to substrates using injection molded parts and automation tool assembly.
Fig. 14. Surface-mount optical coupler for low cost optical coupling to substrates using injection molded parts and automation tool assembly.
Fig. 15. Example of SMOC waveguide coupling by focusing a collimated fiber mode through the silicon back-side lens onto 45° mirror (both by wafer-scale processing) and into a front-side waveguide [34].
Fig. 15. Example of SMOC waveguide coupling by focusing a collimated fiber mode through the silicon back-side lens onto 45° mirror (both by wafer-scale processing) and into a front-side waveguide [34].
Related topics:
Quantum DotNoise Measurement
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved