A perspective on optical microcomb distillation: A tool to break power barriers for tiny rainbows

Microcombs are optical frequency combs generated by micro-photonic structures. They have, in recent years, been demonstrated for applications as far ranging as astronomy, precision measurement, spectroscopy, and communications (e.g., see Refs. 1 and 2). Often, advances in technology for optical frequency combs have impact across multiple fields at the same time, and so addressing the challenges faced in one application area can enable progress in others.² 

In optical systems, generally, the optical power is a key parameter to overcome system noise. Using systems based of microcombs as examples, astrocombs^3,4 require enough power per line to calibrate astronomical spectrometers, spectroscopy and microscopy require a sufficient number of photons at the right wavelengths to tell background from the fingerprint of markers or absorption lines,^5,6 microwave photonics requires high optical power carriers to work cleanly with microwave signals,^7,8 and optical communication systems are often limited by the optical signal-to-noise ratio.^9–12 For microcomb systems, per line power is quite intuitively diluted by using lower pump powers and requiring wider spectra. However, broad bandwidth and low power consumption are desirable traits in all these systems! This provides a tension in designing optical microcombs for a wide array of applications and has been underscored by recent developments in microcomb technologies.

One of the key differences between individual lasers and microcombs is that most microcomb types require a somewhat complex arrangement to initiate the microcomb state [e.g., see Ref. 13 for dissipative Kerr soliton (DKS) states and Ref. 12 for dark pulse states]. The reason for this is that as the input laser aligns with the target wavelength that is resonant with the micro-photonic resonator, thermo-optic effects due to the intense field built up within the resonator cause the resonance to shift, depending on the amount of power in the resonator.¹³ As a microcomb state is reached, quite often, this causes a large shift in the power within the resonator (the intracavity power), and that rapidly detunes the cavity away from the laser.^13,14 This has been overcome by locking together the input laser and the micro-photonic resonator together by self-injection locking^15,16 so that as the resonance undergoes a shift, the laser tracks this shift by being tied to the resonance through optical feedback. This then overcomes the need for complex initiation procedures and creates a true turnkey microcomb. One side advantage of this arrangement is that microcombs are able to be initiated with relatively low pump powers, which creates low energy consumption devices—to the point where they can be battery-powered.¹⁷ However, the flip side to this is that the generated microcombs have low per-line powers, which poses difficulties.

One intuitive route to beating this low power limitation is to use microcomb states that more efficiently turn pump light into comb lines. While DKS states are limited to a few percent conversion efficacy from pump to comb lines, soliton crystals,¹⁸ dark pulse states,¹⁹ combs based on coupled resonators,^20,21 and laser cavity solitons^22,23 can have much higher conversion efficiencies. However, there still remain difficulties in translating this to high per line powers: For example, both laser cavity solitons²² and coupled resonator bright states²¹ seem to use relatively low pump powers to operate in a single soliton state. Soliton crystals with single free-spectral range spacings^10,18 [as opposed to multiple free spectral range (FSR) spacings, e.g., Ref. 24], as required for many applications, boast high conversion efficiency, but their complex shape often means that many comb lines have low powers compared to the overall conversion efficiency (e.g., Ref. 10). As such, there seems to be more research needed before these novel microcomb states can translate into high power microcombs.

A wider optical frequency comb intuitively means that for a given amount of power pumping the comb (and for a given comb spacing), a wider comb will have less per-comb-line power. This has been analyzed well in the context of optical communications,²⁵ which nicely shows the limitations of scaling to higher comb bandwidths. For example, achieving high comb bandwidths to match the capabilities of optical communication systems, e.g., moving from C-band combs to C + L-band combs, would naively improve the power savings^9,11 enabled by moving from many optical sources to a single source. This increase in comb bandwidth is also desirable in other systems. However, we note that a single-soliton comb has a sech² shaped spectral profile, which has a limited roll-off. This means that for a family of comb lines that meet power requirements, there will be a number of comb lines with significant power that are not useful to the target system. These two factors indicate, again, that per-comb-line power is likely to drop in future wideband microcomb sources.

In addition to aiming for wider combs and turnkey operation, the high pump power needed to generate microcombs in early micro-photonic resonator devices has prompted a push toward the use of high quality factor (Q) resonators. A higher quality factor results in a greater concentration of energy in the resonator, as this is related to the propagation loss in the ring—in short, a higher Q means a higher number of round trips a photon can make in a given resonator.^26,27 Broadly speaking, this lowers the threshold for the generation of microcomb states (e.g., see the review²⁸ and papers within). A key aspect of this is that a microcomb often cannot be scaled up in power purely by increasing the pump power. Many of the microcombs used are solitonic, meaning that their properties are linked both to device parameters and to pumping power.^13,18,19,28 This means that at a certain point, increasing pump power changes the microcomb properties, often in undesirable ways (e.g., initiating multi-soliton states with the complex spectra). So, again, this sets limits on the achievable per-comb-line power, which may decrease with a decreased initiation threshold.

From this picture of progress in microcombs, there is a clear tension between creating “better” microcombs with properties that are clearly generally desirable, with the need for a high per-comb-line power. While microcombs seem to happily provide lines that are stable in phase, frequency, and power, while also being low noise to begin with, the requirement to boost their power up would seem to suggest that optical amplification is a good solution. The obvious cost with this is the addition of optical noise.

However, a suite of approaches has noted that the there is a clear difference between the light within the comb and the optical noise added by the amplifier. While both noise and comb lines cover a broad bandwidth, the light in the comb is located at very well-defined, narrow frequency bins. The noise from optical amplification is spread out, at both the sparse locations of the comb lines, but also between the lines. What this has led to is a number of approaches that use this fact to, in various ways, filter out the noise between the comb lines while enabling amplification of the desired comb. The broad concept is illustrated in Fig. 1, where a noisy input comb is put through a device with a narrow transfer function, which reduces noise out-of-band with the comb lines. This approach, which is dubbed variously as comb recovery, comb filtering, or similar, we term comb distillation—the idea is that one is able to distil out the wanted comb lines from the unwanted polluting by-products of optical noise.

Here, we present our perspective on comb distillation methods and their potential utility to break the power barriers for microcombs, to push these micro-sources toward the performance of benchtop frequency combs. We review the state-of-the-art in comb distillation, covering active comb filtering, passive comb filtering, and comb cloning, and attempt to understand whether each of these approaches has advantages and disadvantages. We then look to the future challenges of this approach and try to understand what alternative technologies may be explored and where (or, indeed, whether) comb distillation might have an impact in future microcomb-based optical communication systems. More broadly, the aim of this Perspective is to provide a basis for categorizing a set of unique recent techniques demonstrated in research under the heading of “comb distillation,” as they (from our perspective) are trying to achieve the same goal. We hope that by presenting these comb distillation approaches and analyzing them as a cohesive whole, this will help other research teams address the issues of microcomb power scaling to enable further growth of the applications of microcombs. We believe that comb distillation may be a key step to transitioning microcombs out of research labs and into various applications in the field and hope that this Perspective may help others decide if this approach is valuable to their work.

At the time of writing, the approaches to comb distillation can be broken down into three main categories and are schematically represented in Fig. 2. Active filtering [Figs. 2(a) and 2(b)] is where a gain medium is used to provide selective optical amplification of comb lines in a narrow band around those lines, which can be used either to selectively amplify comb lines over noise or to replace broadband amplification. Comb cloning [Fig. 2(d)] involves taking light from an incoming comb and using that to seed the generation of new combs that mirror the properties of the original incoming (and, possibly, noisy) comb. Notably, this can be combined with coherent combination to produce high powered combs. Passive filtering [Fig. 2(c)] uses a resonant cavity, not dissimilar to those that may be used to generate microcombs, as a periodic, fine-toothed filter—light at the comb line frequencies is passed through, but light between the comb lines is not. Each of these would seem to have its own advantages and limitations, compared to the other methods.

For active filtering of comb lines, two approaches have been demonstrated. Stimulated Brillouin scattering provides narrowband (≈100 MHz) optical amplification in standard optical fibers, and the optical frequency at which this occurs can be parametrically tuned through tuning the pump light. Optical injection locking can take low power light from comb lines, when closely enough matched to the free-running laser frequency, and lock to the phase and frequency of the incoming comb line light, effectively cloning the comb line and boosting its power to the original output power of the laser.

The use of stimulated Brillouin scattering (SBS) for comb distillation has several intriguing properties. As a parametric interaction, the location narrowband gain provided by SBS is set by the frequency of the pump light and the Stokes shift in the material used as a gain medium. As a nonlinear interaction, the gain exponentially scales with the power spectral density. These two properties mean that, if an incoming noisy comb is to be distilled, it can conceptually also be used as a pump. The comb power split off to be used as a pump just needs to be frequency-shifted by the Stokes shift (e.g., using a serrodyne frequency shifter²⁹) and amplified by a regular optical amplifier [see Fig. 2(a)]. Because the comb lines will typically have a much larger power spectral density than any broadband additive noise, this means that SBS will provide a selective gain around the comb lines.^30–33 There are several limitations to the use of this approach. The power needed to achieve significant SBS gain can be large, making the process less power efficient.³³ The dispersion of the Stokes shift can also cause a mismatch between the produced gain and the target comb.³⁴ However, there are also distinct advantages, in that this approach is agnostic to comb line spacing, is insensitive to precise comb line frequency, provides amplification, and can process multiple lines through a single device.

Optical injection locking similarly provides a targeted narrowband amplification effect. By seeding a laser with light from an external source, and given the correct conditions, the laser will lock to that external seeding light, effectively cloning its properties.³⁵ For frequency combs, this means that a low power comb line can be boosted to whatever power the locked laser would usually operate at.^36–38 This approach has been used to boost up individual comb lines, where a separate laser is used to injection lock each line. While this would then seem to run counter to the idea of using a frequency comb, what this method enables is the locking of individual lasers to a very precise frequency grid, without having to achieve fine control over the laser wavelength—as long as the laser and incoming comb line are within the locking range, the laser will lock to the incoming line. This then helps alleviate an amount of energy needed for the control of individual laser, which was one of the initially identified key advantages of using microcombs for optical communications. The drawback with this approach can be hardware complexity [see Fig. 2(b)]. Particularly, most optical injection setups rely on a circulator, which are generally magneto-optic and hard to integrate. If intermodulation between injected lines is present in the injected laser, then a wavelength demultiplexer is likely needed. Moreover, to really prove the utility of this approach, the reduction in potential energy consumption due to relaxed laser controls needs to be quantified.

New combs can be generated to replicate the properties of other remote or independent combs in a few different ways. This can be as simple as using a one or two incoming comb lines to pump the generation of a new microcomb, which will clone the central frequency and phase fluctuations of the incoming line. Alternately, this can operate more like optical injection locking, where a full or partial comb can seed a new comb, additionally locking the new comb in repetition rates. As the properties of the newly generated comb “clone” those of the remote or independent comb, we refer to this suite of techniques as “comb cloning” (inspired by Refs. 39 and 40), although these techniques are called many disparate names in the literature.

Cloning combs through comb line derived pumps can be achieved in different ways. Single or multiwavelength pumps are useful in different contexts and can achieve similar aims. A single comb-line-derived pump wavelength, when combined with a resonator with the same resonance spacing as the original comb, can provide a comb with the same spacing at the same wavelengths as the original comb (e.g., Ref. 39). However, precisely matching the resonance and comb spacing can be quite difficult. This can be overcome, by carefully tuning the single pump to fine-tune the spacing of the clone comb,^39,41 to reduce an error signal generated by beating one line of the cloned comb with a reference line from the original comb. Clearly, here, there is a feedback setup that is required, and so, this comes at the cost of device complexity. Alternately, pumping with two comb lines (in Ref. 42, pumping a non-resonant comb generator) can produce wavelength and repetition rate matched combs innately. The recovery and phase alignment of two individual comb lines (as needed for the cloning of the comb phase) may present some issues requiring additional device complexity, and this may not be an easy approach for compact resonator-based combs. While this produces a new comb, the combs produced will have a similar power to the original comb.

Seeding a pumped resonator with multiple comb lines can produce a synchronization between the incoming comb lines and the comb within the pumped resonator^43,44 [see Fig. 2(d)], similarly to optical injection locking. This has some similarities with pumping resonators with pulses,^45,46 although in the pulsed pumping case, the outcome is the generation of a spectrally broader comb from a powerful, but spectrally limited, comb. In the case where a pumped resonator is seeded, the output comb should ideally have a similar spectral profile to the input comb.⁴³ The key advantage to this approach is that the cloned combs here can be coherently combined so that their powers constructively add and then produce an overall more powerful comb. This has been shown off-chip⁴³ and in an integrated structure.⁴⁷ One key criterion to consider is that the resonators used need to have very closely matched resonance spacings to enable the seeding to lock the combs together.^43,47 While this is clearly advantaged by integration on the same chip, the fabrication process across the chip does need to be highly precise to enable device matching. This was clearly achievable with two resonators in Ref. 47, and it was shown that power could be coherently combined almost perfectly—convincingly showing the promise of this technique for power scaling of microcombs. However, it may be that scaling to many locked resonators provides further technical challenges.

Passive filtering, by using a resonator to remove noise between the comb lines [see Fig. 2(c)], has been employed successfully in high precision optical frequency comb generators.⁴⁸ In that work, the resonator was used to remove noise prior to nonlinear broadening. Our group has then looked at using this technique for optical communications, where the resonator is used after optical amplification up a low power comb, to remove noise before modulation. This has been shown to practically reduce the required optical carrier to noise ratio (OCNR—i.e., OSNR before modulation) by a factor of 10 dB,⁴⁹ with simulations showing that this benefit is close to, if not exactly, modulation format independent.⁵⁰ By analyzing the system, closed-form solutions can be developed that show that the benefit in terms of signal SNR is directly related to the width of the resonance used to selectively filter the comb lines,⁵⁰ which makes intuitive sense—the closer one can filter the comb lines, the greater the portion of broadband noise removed from the comb lines. One clear concern is that dispersion in the passive filtering resonator is likely to mean that the comb lines will walk off the resonances, affecting the comb line OCNR. This OCNR degradation can be shown to match the line shape of the resonance,⁵¹ which sets up a trade-off between the OCNR improvement due to using narrow resonances and OCNR degradation due to comb/resonator spectral walk-off. There are also some clear questions about fabrication tolerance, in a similar way to the issues facing comb cloning through seeding.⁴⁷ These are tied up with the question of nonlinearity in the “passive” resonator in this passive filtering approach: Can one reliably design a normal dispersion passive filtering device to suppress nonlinear interactions with a bright-pulse solitonic microcomb, or would parametric interactions in an anomalous dispersion ring be beneficial—something akin to comb cloning^43,44 or pulsed pumping?^45,46 Again, these are challenges in device complexity that need to be assessed in the context of other comb distillation approaches.

In optical communications, optical frequency combs have been used to support research for many years, from a few terabits in the early 2000s^52,53 to multiple petabits in recent years.^54,55 In the context of optical communications, optical frequency combs are generally used as a means to replace a large number of lasers, with each line of the comb used to support a separate wavelength division multiplexed (WDM) channel carrying independent data. The optical frequency combs used for communication experiments are often “benchtop” scale devices,^52–57 occupying several “rack units” of space. At each stage in the development of optical communication systems based on optical frequency combs, benchtop combs have provided the highest data carrying capacity as a single source,^52–57 which we understand is because of their combination of high bandwidth, OSNR, and per-line comb power compared to other comb technologies available at the time. This is especially true when considering results that set records for capacity,^54,55 which push the limits of what can be achieved with current technologies. The argument for the need for microcombs in future optical communication systems is that while benchtop combs may provide some power savings compared to using individual lasers (e.g., see the analyses in Refs. 9 and 11), a multiwavelength source that is integrable with multiple electro-optic interfaces is more compelling for future networks.⁵⁸ However, one issue that appears when trying to translate from benchtop to chip-scale frequency combs is the achievable optical power per comb line—which ideally should be similar to what current individual lasers can provide, for compatibility with existing optical system designs.

Why do we consider per-comb-line power important? The reasons are the same as that individual laser power matters in optical communication systems—the ability to generate and transmit signals with a sufficient optical signal-to-noise ratio (OSNR) to support high data rates. To achieve high spectral efficiencies (10 b/s/Hz, or more), the electrical signal-to-noise ratio (SNR) for a signal needs to be about 20 dB or above. A back-of-the-envelope calculation using the “58” equation [e.g., see Ref. 59, Eq. (7.111)] can provide an estimate of the power required. Purely considering OSNR as the only limiter to signal SNR (i.e., assuming SNR = OSNR × Baud/12.5 GHz), if we consider a channel bandwidth of 25 GHz (close to the current experimental optimum^11,55,60), a transmission distance of 1000 km using 50 km spans of standard fiber (0.2 dB/km loss), and good quality optical amplifiers (e.g., EDFAs with 5 dB noise figure), in this ideal scenario, the per channel launch power required is greater than −7 dBm (0.2 mW). To cover 4 THz in the C-band, this requires a full comb power of 15 dBm, which is a figure that would seem achievable with most microcomb devices.^9–11,61 However, modulation for coherent optical communications is quite lossy, with figures around 20 dB common,^62,63 due to insertion loss to the modulator and the requirement to drive in the linear (or relatively linear^64,65) part of the sinusoidal transfer function of the device. If that loss is included, then we need 13 dBm (20 mW) per line, or 35 dBm (3 W) total comb power in this scenario. This then highlights the likely need for optical amplification in a modern optical communication system using a microcomb as a source.

This analysis assumes that comb OSNR (i.e., OCNR) matters for enabling ultra-high levels of parallelism in optical communications, but it is worth interrogating that idea. One of the interesting findings in analyzing the capacity of microcombs in the regime of unlimited spatial division multiplexing,¹¹ where the comb power could be split to support modulation on many spatial modes, was that the penalty for using lower power comb lines in terms of achievable capacity could be circumvented by merely increasing the number of spatial channels. This has interesting implications for using low OCNRs for cases where the amount of parallel spatial channels is effectively unlimited, and only low SNR signals are needed. In this situation, it may not make sense to improve the achievable comb line OCNR, but to merely tailor the system to achieve high-capacity links.

However, there may be situations where this approach may not make sense, as indicated by the gap between record results using high OCNR sources and those using microcombs. In systems where the spatial parallelism is limited and capacity is to be maximized, maximizing the spectral efficiency may still be an important step. This has clearly been a consideration in record high-capacity experiments in specialized fibers for space-division multiplexing (SDM), and while those are seemingly far away from implementation in the field, this consideration is likely to hold true for more modest SDM infrastructure (e.g., the Dunant cable⁶⁶).

In Fig. 3, we inspect how high the effective comb line OCNR has to be, in order to enable systems to run close to the capacity set by noise in that system. Here, the generalized mutual information (GMI—a measure of information carrying capacity useful for modern coherent communication systems⁶⁷) for a 64-QAM signal is plotted against the OCNR of the comb line being modulated, which in turn is degraded in OSNR by the link. This system is modeled by modulating a noisy field (the comb line) and then adding additional noise. Figure 3(a) shows the GMI, while Fig. 3(b) looks at the percentage of achievable GMI, which in each case is limited by the added link noise. This indicates that to achieve 95% of the potential link-noise-limited capacity, a comb line OCNR of a bit above 20 dB is required. Interestingly, when using this as a benchmark for comparison for different link OSNRs, as the link OSNR changes from 25 to 10 dB, the required comb line OCNR only drops by about 5 dB. This does then indicate that comb line OCNR is an important factor in achieving link-OSNR-limited capacity. Although simple, this simulation does suggest that comb distillation, where comb line OCNR can be significantly improved, can help for case where microcomb line power is limited.

In short, we think that there is potential for progress in applying metrics for comb distillation across the different techniques, interrogating whether improvements in microcomb generation techniques can be made to displace distillation and to re-evaluate whether microcombs are, indeed, the best technology to move different fields forward or whether alternative approaches exist.

The key metric for the performance of these approaches is the change in OCNR before, and after, the distillation step. This is referred to in various ways across the literature, e.g., as ΔCNR in Ref. 33 and as a parameter α in Ref. 50. In the cases of injection locking, either with single lasers or with comb cloning, this is a harder parameter to define. From our perspective, further work should be undertaken to provide a more unified framework for understanding the efficacy of various comb distillation approaches.

One key consideration that is not explicitly made in the current work on comb distillation is the effect of energy consumption, nor then the ability to compare the energy consumption of different distillation approaches. Given that the answer to many of the applications for comb distillation could effectively be “just use more combs,” there would seem to be an opportunity to compare approaches with a solid metric in mind. This would also then help to understand the further trade-offs between using lower or higher power combs.

Similarly, when assessing the need for comb distillation, this is predicated on the use of low power combs to begin with. Why not just use higher power combs? Clearly, there are limits to the comb line power one can generate on a given platform. If using solitonic effects, then there is a balance to be made between pump power and dispersion, in order to generate the wanted soliton state. Broadly speaking, the power required to induce a wanted soliton state is related to the resonator quality factor and the chromatic dispersion of the ring. A higher Q relates to a lower required pump power, while a lower dispersion corresponds to a microcomb with a broader spectrum (for DKS solitons, see the review in Ref. 28). There has been a strong focus on providing high quality factor microrings to reduce the required driving power, but this reduces the per-comb-line power as well. An opportunity for investigation is to find “Goldilocks” designs that enable high per-line-power (maybe by tailoring Qs to enable higher required pumping power) and, with dispersions, that confine most of the power in the wanted communication bands (maybe by using dispersions that are not so close to zero). Exploring the unique properties of various microcomb states^18–22 may also play a pivotal role.

A further question that could be asked is whether it is sensible to use a microcomb at all! As an example of an alternative approach that would seem to be suited to optical communication systems, self-injection locking to a microring^68,69 can enable the locking of individual lasers to the resonances of a microring. In this way, one may expect the energy cost of individual laser control (a key parameter in the energy analysis of Ref. 9) to go down as the laser is locked to the ring, for the lasers to only be enabled where you want laser lines, and for the per “line” power to be equal to the laser output power (or close to it). This would then seem to be a strong competitor to using microcombs. We suggest that, for this example, there are two factors to consider. The first factor is compactness. Lasers do take up some space and so would (on chip) beam combiners, which would likely increase the on-chip space. The second factor is the quality of the locking. It may be possible for the lasers to move around inside the resonance (a Q of 10⁶ is a resonance bandwidth of about 125 MHz). This may lead to less precise frequency spacing than is wanted for superchannel generation⁷⁰ and other schemes that utilize the comb for more than just a source of WDM light.^71–73 Whether these considerations are important remains to be seen.

Comb distillation provides a tool to enable low power microcombs to be significantly boosted up in power, while keeping relatively high optical signal to noise ratios. There are several schemes for this, each with unique advantages and challenges. It is difficult currently to compare the distillation schemes on an apples-to-apples basis given the current state of the literature, but an effort to understand the issue of energy consumption is likely a good place to start. Comb distillation provides a method to enable micro-sized optical frequency combs with performances that are truly comparable to their macro-scale benchtop counterparts. It remains to be fully understood if this is a key consideration of future optical communication systems, although we think signs point to “yes”—and that there remains considerable work to be done to prove this right or wrong.

Beyond optical communications, comb distillation may also have a significant impact, especially given its use in high precision combs already.⁷⁴ This tool may be of broader use for the plethora of microcomb applications under active study by research groups globally. If considerations of power and optical noise are limiters to the adoption of microcombs in industry, comb distillation may be the way to help get microcombs out of the lab and into the world.

We acknowledge the support from the Australian Research Council, through their Future Fellowships program (Grant No. FT220100835).

The authors have no conflicts to disclose.

Bill Corcoran: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Writing – original draft (equal); Writing – review & editing (equal). Chawaphon Prayoonyong: Conceptualization (equal); Formal analysis (equal); Investigation (equal); Methodology (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.