Chip-based optical frequency combs for high-capacity optical communications

2.1 Optical linewidth

Linewidths of 100 kHz are easily attainable and perfectly adequate for 256 QAM data modulation [5], and even narrower linewidths allows for either higher QAM orders [23] or much simplified DSP, with the extreme case of foregoing linear phase noise correction altogether for linewidths in the sub-kHz range and for 256 QAM [23]. Compared to ECLs and DFB lasers, in the context of achievable information rate (AIR) and optical phase conjugation transmission systems, 1–2 dB of OSNR gain has been achieved for a sub-kHz laser [24]. Again, using such narrow linewidth lasers as the seed in a comb generator provides a source of multiple narrow linewidth lines with these attractive features.

2.2 Power per comb line and OSNR

We investigate power per comb line for various length of transmission, as shown in Figure 2, and compare to the case of a single laser with enough power to not need additional amplifiers before launching into the transmission fibre with 0 dBm optical power per line, inspired by [5]. A comb line with realistic powers of −10 dBm or less straight out of the comb source, and an optical carrier to noise ratio (OCNR estimated as power per comb line relative to the quantum noise limit of −58 dBm) of about 45 dB, needs a pre-amplifier as well as a compensating amplifier after the lossy data modulator before launch at 0 dBm power into the transmission span, which reduces the OSNR. Note that a polariser in front of the modulator reduces the noise sent to the modulator. One could also consider using an optical filter to reduce the ASE noise further, but this would be practically challenging and thus omitted in this analysis. Assuming EDFA noise figures of 4.5 dB, and transmission spans of 80 km with link EDFAs to compensate the loss in the fibre (0.2 dB/km), Figure 2B displays the evolution of the OSNR for the laser case and power per comb line cases of −10, −20, −30, and −40 dBm. The laser, with fewer amplifiers in the transmission line, yields the highest possible OSNR. At short reach it’s about 2 dB better than the −10 dBm OSNR per comb line. The −20 dBm OSNR per comb line is yet 2 dB lower than the −10 dBm case, but for both −10 and −20 dBm the gap to the laser OSNR quickly diminishes with more transmission links. After about 12 spans the −th dBm case has less than 1-dB penalty to the laser case, and after about 15 spans, the −ca dBm case has less than 1-dB penalty to the laser case. So after about 1000 km, the comb case is pretty much as good as the laser case. This is because the overall transmission gets more and more dominated by the many EDFAs in the link. Figure 2C shows a plot of the required power per comb line, P₀, to maintain a 1-dB OSNR penalty after transmission of the comb case to the case of using individual lasers. As seen, going beyond about 20 links, 1600 km, combs with line powers less than −25 dBm suffice to create high-quality transmission. For trans-Atlantic distances, −30 dBm (OCNR of 28 dB) or less is enough. Figure 2D shows the difference in dB of the comb OSNR to the laser OSNR as a function of power per comb line for varying numbers of links. Generally, at short reach there is a large penalty of several dB, which grows with lowering the power per comb line. For longer reach and high power, this difference decreases. A plateau is seen, on which there is a low penalty to the laser case. This plateau shows there is a large region where a comb source could offer good value for money. It is worth noting that −13 dBm power per comb line is feasible and has been experimentally demonstrated [25], but as mentioned, even for −20 dBm after 15 spans the penalty to a laser is less than 1 dB and at short reach, −10 dBm is about 2 dB worse than the laser case. So even for short reach, or a single link, it is possible to get very decent OSNRs for a range of reasonable levels of powers per comb line (more than 30 dB OSNR).

Figure 2:

Investigating the power per comb line for various transmission reaches.

(A) Schematic setup of the investigated systems, comparing a comb source, e.g. a ring resonator, to a laser with high enough power to neither need a pre-amplifier nor an amplifier to compensate for the losses in the modulator. A polariser is assumed in front of the modulator, essentially rejecting half of the noise prior to that point. The launch power is set to 0 dBm. (B) Receiver OSNR for various powers per comb line compared to the laser case, and inserted dashed lines for OSNR levels required to support 256, 64, 16 QAM. For the 256 QAM case, the laser can go 4.6 × 80 km, the −10-dBm comb case can go 3.5 × 80 km and the −20-dBm comb case can go 2.4 × 80 km with high enough received OSNR to sustain the high order modulation. (C) The required power per comb line for a 1-dB OSNR penalty compared to the laser case. (D) Comparison of the comb OSNR to the laser OSNR for different powers per comb line as a function of transmission reach.

So depending on the need, one can determine the requirements to the power per comb line this way. E.g. if 25 GBd is used, 31 dB OSNR will be needed to support 256 QAM at a bit error rate below 1E-3, yielding a generalized mutual information (GMI) of 8 bits/symbol. This can be supported by −10 dBm power per comb line. For 64 QAM, −20 dBm power per comb line is adequate. These powers per comb line are attainable by several existing comb sources. Note that, higher spectral efficiency can be achieved with a narrower frequency spacing and smaller symbol rate [26].

2.3 Multiplexing capacity

The calculated capacity shown in Figure 3B suggests a potential achievable capacity of 989 Tbit/s for this setup. The experimental conditions were not suitable to try to go closer to capacity than 661 Tbit/s, so as expected there was an implementation penalty. The curve in Fig. B shows that for unrepeated transmission, one could still obtain about 800 Tbit/s over more than 50 km.

As discussed, the achievable power per comb line sets a limit on how much data can be carried by each line, and the number of generated lines and their spacing (free spectral range) compared to the realistic data modulation rate, dictates the achievable WDM capacity. Since each line can be separated without loss of power, using wavelength splitters and filters, the scaling of the total capacity for WDM is simply the number of lines multiplied with the single line capacity.

For space division multiplexing with a single comb source, each line will be power divided into the number of desired spatial channels, and hence each channel will have equivalently less line power, and accordingly lower OSNR. However, the number of channels is increased, and so the overall capacity can still grow, until the line power is depleted.

The total capacity for SDM and WDM becomes:

where n_WDM is the total number of WDM lines, m_SDM is the total number of SDM channels, R_s is the modulation symbol rate, and the SNR is directly related to the OSNR shown in Figure 2, as the noise bandwidth B_ref divided by the modulation rate R_s. For SDM, using a single comb source, the SNR will decrease with more SDM channels, due to the power splitting, so the SDM capacity increase depends on terms both inside and outside the logarithmic function, and it is expected that C will saturate for a certain P₀ and a given number of SDM channels, m_SDM. The capacity increases most efficiently by increasing parameters outside the logarithmic function. Therefore, the number of WDM channels is very important, meaning that a good comb source should have a broad bandwidth with many separate lines. The number of SDM channels and the modulation rate scales the same way, but where the modulation rate has an upper limit given by state-of-the-art electronics (today allowing for roughly 30 GHz efficient modulation with high bit resolution), there is no obvious upper limit for SDM. We investigate the total capacity based on the SNR straight out of the comb as an instructive indication of the data carrying capacity of a comb source.

The analysis of the data carrying capacity of a single source is performed using a setup similar to that in Figure 2A, but where the comb is split in m_SDM branches just after the pre-amplifier, G_pre. To compensate for the loss of the power splitting by the SDM-splitter, each SDM branch gets an additional amplifier, G_SDM, which in part amplifies the signal to the desired level, but also introduces more ASE noise, and amplification of the noise prior to this step. Each branch then splits the wavelength channels into n_WDM channels, with no loss, and each line is data modulated, with a polariser at the input to the modulator (effectively removing half the ASE noise). The modulator is assumed to be a dual-polarisation IQ-modulator using the signal power as efficiently as possible. All WDM lines are then recombined in a lossless WDM-multiplexer, and the loss of the whole modulator section is compensated for by an EDFA, G_mod = 15 dB. After this step, all m_SDM branches are recombined in a lossless power combiner and launched into the transmission link with P_launch set to 0 dBm. So essentially the model is like Figure 2A with SDM multiplexing stages in between G_pre and the link.

For R_s = 32 GBaud, FSR = 35 GHz and 300 WDM channels (corresponding to C and L-bands, the Shannon capacities for m cores shown in Figure 3A is obtained for five different line powers for a single link of 80 km length followed by an amplifier in the receiver. As seen, the total capacity scales with the number of SDM channels, and even though each comb line splitting to a new SDM channel decreases the available OSNR for that channel, the scaling continues favourably to 1000s of SDM channels. For instance, at just above 1000 SDM channels, a capacity of 100 Pbit/s is reached. And this is for realistic per comb lines powers of about −20 to −10 dBm. So the data carrying capacity is really tremendous.

Figure 3:

(A) Capacity for SDM multiplexing. The capacity quickly scales to high numbers, Pbit/s to 10s of Pbit/s, and then starts to slowdown in growth as a result of diminishing OSNR from splitting the initial OSNR per comb line into many cores. However, a steady growth in capacity continues, and extreme rates can be obtained. For the parameter values used here, it is found that 100 Pbit/s may be reached with just over 1000 SDM channels and −20 to −10 dBm initial power per comb line, and 1 Pbit/s is reached for only 10–15 SDM channels for the same comb line powers. (B) The Shannon capacity calculated for a recent 30-core fibre transmission experiment using a comb source as in the study by Hu et al. [2]. A mode-locked laser with an initial comb of about 40 lines at 10 GHz FSR, power per comb line of −6 dBm, and optical carrier to noise ratio of 50 dB is used as the seed for a self-phase modulation unit that broadens the comb to 400 lines. As described in the study by Hu et al. [2], these lines are grouped in four lines per channel, and using optical time-division multiplexing to reach 40 Gbaud 50 GHz spaced WDM channels. With 80 such 40-Gbaud WDM channels in 30 cores, it was experimentally found that 661 Tbit/s could be successfully transmitted through a 10 km long fibre.

Figure 3B shows the Shannon capacity for a different case, where a mode-locked laser comb is used to seed a nonlinear chip to create a wider spectrum by self-phase modulation. In the study by Hu et al. [2], using an AlGaAs nano-waveguide, about 400 lines were created at 10 GHz spacing with about −15 dBm power in each line [2]. If these 400 lines could all be modulated with 8 GBd, leaving some guardband, the total capacity for 30 spatial channels would be 1.18 Pbit/s after transmission through a 10 km piece of 30-core fibre with roughly 0.3 dB/km loss. In the study by Hu et al. [2], however, as it was not possible to carve optical spectra with sharp enough filter shapes to separate 10 GHz channels, instead four lines were used to make up one channel, and the 10G symbols were time domain multiplexed to match the 4 × 10 GHz lines with a 40 GBd symbol rate. Every fifth line was suppressed as a guardband, leaving 80 WDM channels. The power per WDM channel is then 4× the power per 10-GHz line, i.e. −15 dBm + 6 dB = −9 dBm. For 30 SDM channels the total capacity after a 30-core fibre 10 km transmission becomes 989 Tbit/s. 661 Tbit/s were achieved in the experiment in the study by Hu et al. [2], due to available modulation capacity at the time, but nonetheless, this experiment and capacity calculation clearly demonstrates the high data carrying capacity of frequency combs.

2.4 Energy efficiency

It has been speculated that a comb source could save a lot of energy in transmitters relying on individual lasers for individual WDM channels. As we have seen in Figure 2 above, a comb source certainly does not produce the same amount of line power as an individual laser does, impairing the achievable OSNR. For long reach links, though, we saw that even for comb powers of less than −20 dBm the OSNR penalty was less than 1 dB for more than 1000 km transmission. For short reach, the combs start needing more power to compare to individual lasers. However, even for unrepeatered metro-range transmission, very decent OSNRs may still be obtained for combs, and 661 Tbit/s was experimentally demonstrated feasible over 10 km 30-core fibre. So combs can offer performance at a reasonable level. Especially, when multiplexing scenarios are considered as a large gain in capacity is obtained for terms outside the log₂ in the capacity formula. But can combs be more energy-efficient than single lasers. It is difficult to compare mature commercial solutions to lab-setups. However, there certainly is a potential, as all the comb lines are equidistant and stable, so no additional wavelength locking is required on a per channel basis, and a fewer current sources and laser temperature stabilizers are needed. As mentioned above, this also gives rise to certain digital signal processing benefits, such as joint DSP and nonlinear impairment mitigation. On the other hand, a comb needs a strong pump signal, and amplification of the comb lines, which costs energy. But assuming e.g. replacing 100 s of lasers with a single comb source, as potentially done for the hero experiment reaching 10 Pbit/s transmission [27] using a single source, with all their power sources etc., and knowing that the comb can perform well in some ranges, we expect there would be a considerable savings potential. One essential parameter in this consideration is the pump to line conversion, as this directly impacts the achievable OSNR. This depends on the nonlinear material used, the scheme of comb-generation, and losses in the device.

3.1 Quantum dot and quantum dash mode-locked lasers on InP substrates

Typically, quantum dots and quantum dashes are grown by chemical beam epitaxy (CBE) or molecular beam epitaxy (MBE) on InP substrates [30], [35]. Liu et al. [30] demonstrated a quantum dash passively mode-locked laser (MLL) frequency comb with a frequency spacing of ∼34 GHz, driven by an electrical direct current (DC). Figure 5A shows the cross-sectional scanning electron microscopy (SEM) image, showing the facet of a fabricated Fabry–Perot (FP) InAs/InP quantum dash laser. The schematic cross-sectional view of a shallow-etched ridge-waveguide InAs/InP QD laser is shown in Figure 5B. The InGaAsP waveguide contains five stacked layers of InAs QDs as the gain medium, surrounded by n- and p- type InP cladding layers. The MLL frequency comb has a 6-dB bandwidth of ∼13 nm and 48 comb lines with an optical carrier to noise ratio (OCNR) of more than 40 dB, as shown in Figure 4B. The average optical linewidth of each comb line is around 1.55 MHz, as shown in Figure 5B. The MLL frequency comb was used for a PAM-4 signal transmission over a 25-km standard single mode fibre (SSMF), with an aggregated capacity of 5.376 Tbit/s [30]. By modulating with a 16-QAM signal, the MLL frequency comb was also used for a 10.8 Tbit/s transmission over a 100-km SSMF [35]. In addition, a MLL frequency comb can be used not only as a laser source in a transmitter, but also as a local oscillator in a coherent receiver [36], though the linewidth would need to be sub-100 kHz.

Figure 5:

A passively mode-locked InAs/InP quantum dash laser.

(A) Schematic of the cross section. (B) Optical linewidth of filtered individual comb lines. (Left inset) frequency noise spectra from three filtered comb lines at the wavelengths of 1551.824, 1552.924 and 1553.474 nm (Right inset) Single sideband (SSB) phase noise of the RF beating. Reprinted with the permission from Ref. [30], Copyright 2020, Optical Society of America.

The current limitation of using integrated MLL-based frequency combs for large capacity optical communication systems is still the relatively broad optical linewidth, which is on the order of 1–20 MHz and at least an order of magnitude broader than most external cavity lasers (ECL), widely used in coherent transmission systems today. The optical linewidth of an MLL frequency comb limits the modulation format useable in transmission experiments to mainly intensity modulation, such as pulse amplitude modulation (PAM) signals, or simple phase modulation formats, such as QPSK or 16-QAM with degraded performance [30], [35], [37]. Nevertheless, several schemes have been proposed to solve this problem, such as introducing an external cavity to reduce the optical linewidth [38] or using a digital phase tracking algorithm to improve the transmission performance [39]. Kemal et al. [38] built a free-space external cavity for a QD-MLL reducing the optical linewidth by almost two orders of magnitude to ∼40 kHz, which was then used for 11.2 Tbit/s 32-QAM signal transmission over a 75 km SSMF. As a fully integrated alternative to free-space optical external cavity lasers, low-loss silicon waveguides have been hybridly integrated with a laser cavity to enable a III-V-on-Si MLL frequency comb with an optical linewidth of <400 kHz [40], which looks very promising.

3.2 Quantum dot mode-locked lasers directly grown on Si

Recent breakthroughs on direct growth of quantum dot materials on CMOS-compatible silicon substrates provide an appealing direction from both cost and performance perspective, as it combines the advantages of active quantum dot materials with the mass-production ability of silicon [41]. Liu et al. [42] demonstrated a 20-GHz passively mode-locked quantum dot laser which is directly grown on an on-axis (001) CMOS-compatible silicon substrate, with the schematic shown in Figure 6A. The laser operates in the O band with a 3-dB bandwidth of 6.1 nm and 80 comb lines within the 10-dB bandwidth. The average optical linewidth of each comb line is around 10.6 MHz, as shown in Figure 6B. The QD-MLL-on-Si frequency comb was used as the WDM light source for a terabit PAM-4 signal transmission over a 5-km SSMF.

Figure 6:

(A) Schematic of the 20 GHz quantum dot passively mode-locked laser on silicon (not to scale). (B) Optical spectrum and optical linewidth of individual comb lines. Reprinted with the permission from Ref. [42], Copyright 2019, Optical Society of America.

5.1 Bright dissipative Kerr soliton micro-combs

A soliton is a waveform that maintains its shape while propagating, upheld by a balance of dispersion and nonlinearity in the medium [95], [96]. The dissipative Kerr soliton (DKS) relies on a double balance of parametric gain and cavity loss, as well as dispersion and nonlinearity. Figure 8A shows the schematic setup of DKS micro-comb generation in a high-Q silicon nitride (Si₃N₄) microresonator [91]. The integrated micro-ring resonator is pumped by a tunable CW laser which is amplified by an erbium-doped fibre amplifier (EDFA). At the output of the microresonator, a notch filter is used to suppress the remaining pump light. In order to generate the bright DKS micro-comb, the pump laser is swept through the resonance from a blue-detuned wavelength to a predefined red-detuned wavelength [97], [98], [99]. Figure 8B shows the single soliton and multiple solitons. When the pump laser is tuned across the cavity resonance, high-noise modulation instability (MI) combs are first observed. When the pump laser is tuned backwards, bistability of the cavity enables the formation of soliton states from a multiple-soliton state to a single-soliton state. The power spectrum of the DKS comb state is shown in Figure 8C and exhibits a 3-dB spectral bandwidth of ∼6 THz. The DKS comb states are stable for many hours in a laboratory environment, which is key for using the sources in optical transmission experiments. The pump-to-comb conversion efficiency of the DKS micro-comb is limited to 0.1%–0.6% owing to the fundamental principle that bright DKS generation occurs only with the pump laser being far-detuned from the resonance.

Figure 8:

Bright DKS-based micro-comb generation in high-Q silicon nitride microresonators.

(A) Schematic setup of the bright DKS micro-comb generation. (B) Illustration of single soliton and multiple solitons. (C) Optical spectrum of a single-soliton frequency comb after the suppression of residual pump light. Reprinted (adapted) with the permission from Ref. [91], Copyright 2017, Springer Nature.

The DKS micro-combs are used in three optical transmission experiments [91]. In the first experiment, the DKS micro-comb with ∼100 GHz spacing is modulated with 16-QAM signals and the data with an overall net rate of 28 Tbit/s is transmitted on 94 carriers spanning telecom C and L bands. In the second experiment, the number of carriers is doubled by interleaving two DKS micro-combs. The overall net rate of 50.2 Tbit/s is achieved with a total of 179 carriers. In the third experiment, the DKS micro-combs are used both as the source in the transmitter and as the multi-wavelength local oscillator in the coherent receiver. This is a benchmark result, clearly showing the potential of these combs. In these experiments, the pump power to the ring was 32 dBm, but this can be lowered with further optimization of the losses of the ring.

Furthermore, a pump source integrated with a micro-ring has been demonstrated on an integrated hybrid III-V/ Si₃N₄ platform, relying solely on supplying the pump source with a current from a standard AAA battery, thus ensuring low-power operation [100]. In this demonstration, a gain section based on a III–V reflective semiconductor optical amplifier (RSOA) is coupled to a Si₃N₄ laser cavity, which consists of two Vernier microring filters for wavelength tunability and a high-Q nonlinear microresonator. In order to achieve long-term stable operation, active feedback circuits are needed, which may be integrated together in future work.

Very recently, a turnkey operation regime for soliton micro-combs co-integrated with a pump laser has been demonstrated [102]. Solitons are immediately generated by turning the pump laser on when the device is working at an appropriate operating point, therefore, the photonic and electronic control circuitry is not needed. In the experiment, integrated soliton micro-combs with CMOS-compatible fabrication [101] and frequency spacing at electronic-circuit rates (40 GHz down to 15 GHz) are butt-coupled to a commercial distributed-feedback (DFB) laser via inverse tapers. Due to its compact footprint and the absence of control electronics, the pump laser and micro-comb chipset is mounted into a butterfly package (Figure 9A) to facilitate measurements and enable portability. The demonstrated turnkey operation is automatic, and the entire soliton initiation and stabilization is realized by the physical dynamics of laser self-injection locking in combination with the nonlinear resonator response. Therefore, the turnkey DKS micro-comb is highly robust with respect to temperature and environmental disturbances. Figure 9B shows the optical spectrum of a single-soliton state with a repetition rate of 40 GHz, and multi-soliton states with repetition rates of 20 and 15 GHz are also demonstrated.

Figure 9:

Integrated turnkey soliton micro-comb with the repetition frequency at electronic-circuit rates.

(A) Image of the DFB pump laser and micro-comb chipset mounted in a compact butterfly package. (B) Optical spectrum of a single soliton state with the repetition frequency of 40 GHz. Reprinted (adapted) with the permission from Ref. [102], Copyright 2020, Springer Nature.

As mentioned in the previous section, the pump-to-comb conversion efficiency is an important metric. The pump-to-comb conversion efficiency of bright DKS micro-combs has a theoretical limit of 5%, as bright-soliton generation only occurs when the pump laser is detuned from the optical resonance [61], [99], [103]. Nevertheless, the overall pump consumption of the bright DKS micro-comb can already compete with massively parallel arrays of commercially available integrated tunable lasers [91], and as mentioned in Section 2.3, per comb line powers on the order of −20 dBm suffice for many cases.

5.2 Dark pulse micro-combs

In contrast to bright soliton micro-combs, which are generated in anomalous dispersion microresonators, dark-pulse Kerr combs (sometimes referred to as dark solitons) are generated in normal dispersion microresonators. Since the initial stage of micro-comb generation is triggered by modulation instability, normal dispersion has often been thought as unsuitable for micro-comb generation due to the lack of modulation instability. Nevertheless, modulation instability can occur in normal dispersion microresonators with the assistance of mode coupling [103], [104], [105], and mode-locked dark pulses can be generated. The dark-pulse micro-comb is e.g. initialized by tuning the CW pump wavelength into the resonance from the blue side and a feedback loop ensures long-term stable operation. In the feedback loop, a part of newly generated comb line is filtered and detected by a photodiode and the output of the photodiode is used as feedback to the laser wavelength setting, therefore, the pump laser will stop sweeping when the frequency comb is generated. A spectrum of a dark-pulses micro-comb is shown in Figure 10A, which operates in a low-noise state with a measured frequency spacing of ∼229 GHz. The dark pulses can be understood as time domain low-intensity dips in a high-intensity homogeneous background (Figure 10B), which exhibits unique advantages such as higher pump-to-comb conversion efficiency and flatter comb spectrum.

Figure 10:

Dark pulse micro-comb.

(A) Optical spectrum of the dark pulse micro-comb. (B) Directly measured intra-cavity time-domain waveforms. Reprinted (adapted) with the permission from Ref. [25], Copyright 2018, Springer Nature.

In a recent experimental investigation [25], a dark-pulse micro-comb reached ∼20% pump-to-comb conversion efficiency with weakest comb line power of −10 dBm, and the comb lines were subsequently modulated with 64-QAM data and used in an optical transmission experiment with an aggregated data rate of 4.4 Tbit/s. In a very recent experiment, a dark-pulses micro-comb with 105.2 GHz spacing and covering telecom C + L bands was modulated by probabilistically shaped 64 or 256 QAM and transmitted over a 37-core fibre, with an aggregated net rate of 1.84 Pbit/s [93].

5.3 Soliton crystal micro-combs

Soliton crystal micro-combs introduce a new regime of soliton physics into the field of micro-combs [64], [106], which are defined as spontaneously and collectively ordered ensembles of co-propagating solitons that are regularly distributed in a resonator. The co-propagating solitons interact with each other and “crystallize” by arranging themselves into a self-organized sequence. Soliton crystal micro-combs have a distinctive ‘fingerprint’ optical spectrum arising from spectral interference between interacting solitons. Since the solitons are tightly packed within the resonator, the intra-cavity optical power of the soliton crystals is very close to that of spatiotemporal chaotic states. There is very little change in intra-cavity power during the transition from chaotic states to soliton crystal states, therefore, soliton crystals can be stably formed using slow tuning techniques without the need for complex dynamic pumping and stabilization schemes. Soliton crystal micro-combs have the unique advantages of intrinsic stability and high pump-to-comb conversion efficiency (due to large overlap between the CW pump and the tightly packed multiple solitons), although the overall spectrum is unevenly shaped with line-by-line intensity variations.

Figure 11A illustrates the single temporal defect crystal state used in a recent experiment [92]. To generate the soliton crystal micro-comb, the pump laser is automatically tuned to a pre-set wavelength. The primary comb (Turing pattern) is initially generated as the pump laser is tuned into resonance, and then the soliton crystals are formed in the micro-cavity with appropriate mode crossing, and a characteristic spectrum, as shown in Figure 11B. Figure 11C demonstrates the stability and repeatability of the soliton crystal micro-comb generation by showing a variation in individual tone powers of <±0.9 dB over 10 different generation instances through wavelength sweeping. The generated soliton crystal micro-comb in the study by He et al. [88] has a frequency spacing of 48.9 GHz and a bandwidth >80 nm, with a high pump-to-comb conversion efficiency of 42%. Eighty comb lines were selected over the telecom C band and used as the WDM source in an optical data transmission experiment. A line rate of 44.2 Tbit/s and a high spectral-efficiency was achieved using the modulation format 64-QAM [92].

Figure 11:

(A) Illustration of soliton crystal micro-comb state, which is a single temporal pulse defect across the ring. (B) Optical spectrum of the soliton crystal micro-comb. (C) The comb line power difference for 10 independent soliton crystal generation instances is within ±0.9 dB, indicating reliable generation of the soliton crystal micro-comb. Reprinted with the permission from Ref. [92], Copyright 2020, Springer Nature.