1

Ultra-Broadband Interleaver for ExtremeWavelength Scaling in Silicon Photonic Links

Anthony Rizzo, Student Member, IEEE, Qixiang Cheng, Member, IEEE, Stuart Daudlin,and Keren Bergman, Fellow, IEEE

Abstract—We demonstrate an ultra-broadband silicon pho-tonic interleaver capable of interleaving and de-interleavingfrequency comb lines over a 125 nm bandwidth in the extended C-and L-bands. We use a ring-assisted asymmetric Mach Zehnderinterferometer to achieve a flat-top passband response whilemaintaining a compact device footprint. The device has a 400GHz free spectral range to divide an optical frequency comb with200 GHz channel spacing into two output groups, each with achannel spacing of 400 GHz, yielding a potential capacity of 78total wavelength-division multiplexed channels between 1525 nmand 1650 nm. This device represents an important step towardsrealizing highly parallel integrated optical links with broadbandfrequency comb sources within the silicon photonics platform.

Index Terms—Silicon photonics, optical frequency combs,wavelength division multiplexing.

I. INTRODUCTION

THE explosive growth of data-intensive applications suchas machine learning and artificial intelligence has placed

a great strain on interconnection bandwidths in data centerand high performance computing systems. Furthermore, theCOVID-19 pandemic has caused a rapid and unprecedentedshift towards a digital world which requires millions of si-multaneous high-definition video streams across the globe fortelecommuting. This extreme bandwidth pressure has led toconservative estimates of 8% of the total global electricitybeing consumed by data centers alone by 2030, with cur-rent trends indicating as high as 21% is possible withoutsignificant intervention [1]. Thus, optical interconnects willbecome ubiquitous in the coming years due to their reducedenergy consumption and large bandwidth potential [2]. Inparticular, silicon photonics is poised to revolutionize datacenter interconnects due to its extreme device density arisingfrom the large index contrast between Si and SiO2 and itspotential to leverage the same economies of scale as the elec-tronics industry by sharing the tremendous global investmentin silicon processing through CMOS infrastructure [3].

This work was supported in part by the U.S. Advanced Research ProjectsAgency–Energy under ENLITENED Grant DE-AR000843 and in part bythe U.S. Defense Advanced Research Projects Agency under PIPES GrantHR00111920014. The work of A. Rizzo was supported by the Department ofDefense Science, Math, and Research for Transformation (SMART) Scholar-ship.

A. Rizzo, S. Daudlin, and K. Bergman are with the Lightwave ResearchLaboratory, Department of Electrical Engineering, Columbia University, NewYork, NY 10027 USA. (Corresponding author: Anthony Rizzo, e-mail:anthony.rizzo@columbia.edu).

Q. Cheng was with the Lightwave Research Laboratory, Department ofElectrical Engineering, Columbia University, New York, NY 10027 USA.He is currently with the Electrical Engineering Division, Department ofEngineering, University of Cambridge, Cambridge, CB3 0FA, U.K.

A major advantage afforded by optical interconnects is theability to send many parallel data streams through a single fiberusing wavelength division multiplexing (WDM). Currently,the optical channels for WDM systems are generated bylaser arrays where each laser must be individually tuned tomaintain the desired channel spacing. Recent advances inmicroresonator-based optical frequency combs have given riseto a new paradigm for WDM sources where many wavelengthchannels are generated on-chip with precise, intrinsic channelspacing and correlated drift [4]–[7]. Futhermore, significantprogress has been made in normal group velocity dispersion(GVD) Kerr combs, which have shown better promise thananomalous GVD soliton combs for communication applica-tions due to fundamentally better conversion efficiency, higheroptical power per line, and more flat spectral shape [8]. Suchcomb sources can have bandwidths spanning hundreds ofnanometers and can consist of hundreds of distinct wavelengthtones. In order to utilize the full comb bandwidth, all deviceson the silicon photonic chip must have a broadband design.

De-interleaving and interleaving are essential operationsfor spectrally separating groups of comb lines into differentbus waveguides and recombining them on a single bus, re-spectively. Traditional approaches for silicon photonic linkswith frequency comb sources, which use cascaded micro-ring resonators (MRRs), have a fundamental restriction onthe number of wavelength channels on a single bus due tothe free spectral range (FSR) of the MRRs and crosstalk forincreasingly smaller channel spacing [9]–[12]. By using anon-chip interleaver, these limitations can be circumvented bydividing the comb into multiple buses, effectively halving thenumber of required resonators on a single bus while alsodoubling the channel spacing.

Here, we demonstrate an ultra-broadband silicon photonicinterleaver capable of interleaving and de-interleaving fre-quency comb lines over a 125 nm bandwidth from 1525–1650nm, spanning the full C- and L-bands. The device consistsof a ring-assisted Mach Zehnder interferometer (RMZI) withbroadband coupling to the ring, enabling consistent perfor-mance over the full bandwidth of interest. To the best ofthe authors’ knowledge, this device represents the broadestoperational bandwidth to date for an on-chip interleaver.

II. DEVICE DESIGN AND OPERATING PRINCIPLE

A. Fundamental Characteristics of RMZI Interleavers

RMZIs blend two of the fundamental building blocks ofsilicon photonics: micro-ring resonators (MRRs) and Mach

2

Fig. 1. Device schematic and principle of operation, illustrating an incidentfrequency comb spectrum with line spacing ∆f and two output groups eachwith spacing 2∆f.

Zehnder interferometers (MZIs). RMZIs have found use aswavelength-and-space selective switches [13], athermal MRRs[14], and wavelength interleavers [15], [16].

When operating as a wavelength interleaver, an input fre-quency comb with line spacing ∆f is first incident on a 1× 2splitter which splits the power into the two MZI arms (Fig.1). The top arm is coupled to a MRR while the bottom armimparts a relative phase shift between the arms from the pathlength difference ∆d. The two arms are then recombinedon a 2 × 2 combiner and output in two groups, one with“even” wavelengths and one with “odd” wavelengths, bothwith line spacing 2∆f (Fig. 1). As a linear optical device,this operation is bidirectional, meaning that it can also beused to combine two separate frequency combs with spacing2∆f into a single comb with spacing ∆f assuming that thetwo combs are originally offset by ∆f. This (de-)interleavingoperation can also be achieved by a simple asymmetric MZIwith a path length imbalance between the arms, which exhibitsa sinusoidal transmission response as a function of wavelengthwith a periodicity depending on the magnitude of the im-balance. However, such a device is undesirable for filteringapplications since any channel misalignment with the centerof the passbands will lead to severe penalties in both crosstalkand insertion loss. By coupling one arm of the asymmetricMZI to a resonator, the passbands can be engineered to changethe shape closer to an ideal box-like response which has agreater tolerance to channel misalignment.

To achieve flat-top passbands compared to the sinusoidalpassbands of a standard asymmetric MZI, the coupled MRRis used as a wavelength-selective phase shifter to alter theRMZI output at particular wavelengths. The free spectral range(FSR) of the MRR determines the periodicity of these phaseshifts in the spectral domain, while the coupling coefficient(κ) determines the shape of the imparted phase as a functionof detuning from resonance [17]. By aligning the resonancesof the ring on the edges of the MZI passbands and choosingthe proper value of κ, the phase response can flatten the MZIoutput to give a sharper, box-like spectral response (Fig. 2).The required total path length of the ring to achieve a particular

Fig. 2. (a) The phase shift imparted by an ideal lossless resonator for variouscoupling coefficients as a function of detuning from resonance. Since theresonator is assumed to be lossless, all of these curves are in the overcoupledregime and thus have a continuous phase response. (b) Single passbandtransmission for an asymmetric MZI as a function of detuning from thepassband center. (c) Phase shift imparted by the resonator for the idealcoupling condition (κ = 0.89). (d) Output transmission for the RMZI system,demonstrating a flattened passband with sharper roll-off. For (a)-(d), theplots were simulated using the fundamental TE mode for a 500 × 220 nmwaveguide, with Lring = 558 µm and ∆d = 279 µm.

FSR is given by

Lring =2c

ngFSR(1)

where c is the speed of light in vacuum, ng is the group indexof the fundamental transverse electric (TE) mode, and FSRis the desired free spectral range. Furthermore, to positionthe MRR resonances properly such that the phase shifts areimparted at the band edges of the MZI response, Lring andthe path length imbalance of the MZI, ∆d, are related throughthe equality ∆d = Lring/2, assuming that ng is the same for

3

Fig. 3. Optical microscope image of the fabricated device.

all waveguides in both the MZI and MRR. These relationshold assuming that the phase shift imparted by the MRR-MZIcoupler is negligible (point coupler), but for couplers withlonger interaction lengths, this phase must be included in thecalculation of Lring and ∆d.

B. Requirement of Broadband Coupling to the Ring

Figure 2 illustrates the principle of passband flattening fora single channel, with the ideal MRR-MZI coupling occuringat κ = 0.89 [16]. It is well known that standard evanescent-based directional couplers are highly dispersive, meaning thatthe coupling coefficient is a strong function of wavelength.For these standard directional couplers, it is clear that sinceκ varies with wavelength, the shape of the phase responsewill also vary with wavelength and thus the flat-top passbandswill not be maintained over a wide bandwidth. Therefore, forbroadband performance, it is critical to take great care in thedesign of the MRR-MZI coupler to ensure that: (i) the couplingcoefficient is uniform across the bandwidth of interest, and(ii) the coupling coefficient is as close to the ideal value aspossible such that the phase response at all wavelengths leadsto the correct flattened passband shape.

C. Device Design and Simulation

The group index ng for a 500 nm × 220 nm waveguidewas simulated using a finite difference eigenmode solver(Lumerical MODE [18]) over the wavelength range of interestwhich was then used to calculate Lring and ∆d from theprevious equations. The full device was then simulated usingan S-parameter-based circuit simulator (Lumerical INTER-CONNECT) to fine-tune Lring and ∆d to achieve the desiredspectral response. The broadband coupler was designed usinga balanced MZI with phase-control-based 50-50 directionalcouplers as the splitters and combiners [19]. By using a bal-anced MZI design as the coupler rather than a fixed couplingelement, the coupling coefficient can be tuned by adjustingthe relative phase shift between the paths. The wavelengthdependence and interferometric visibility of this device aredetermined by the splitter and combiner, which necessitatesthe use of broadband couplers for both of these elements.

Fig. 4. Comparison of simulation with the experimentally measured device(after thermal tuning to correct for fabrication variations) for (a) single pass-band bandwidth and (b) free spectral range.

III. EXPERIMENTAL RESULTS

The device was fabricated through Advanced MicroFoundry’s (AMF) 200 mm multi-project wafer (MPW) service.An optical microscope image of the fabricated device is shownin Fig. 3, demonstrating a compact device footprint of 200 µm× 800 µm. The thermo-optic phase shifters were implementedusing a titanium nitride layer as a resistive heater over thesilicon waveguide layer. By snaking the waveguide under theheating element to maximize the overlap between the gener-ated heat and the optical path, an ultra-efficient Pπ of 14 mW(Vπ = 0.8 V) was measured without the need for advancedfabrication techniques such as undercutting. Test structures ofindividual phase-control-based 3 dB couplers confirmed betterthan 1 dB of uniformity with minimal insertion loss over a100 nm bandwidth. After tuning to correct for fabricationvariations, the single passband bandwidth and FSR of theRMZI device very closely match the simulated design (Fig.4). Furthermore, these properties are maintained over the fullbandwidth of interest from 1525 – 1650 nm, demonstrating theuniform performance of the broadband MRR-MZI coupling(Fig. 5).

4

Fig. 5. (a) Experimentally measured spectrum of the tuned interleaver showing broadband performance over 125 nm of optical bandwidth from 1525 – 1650nm. (b) Zoomed section of the spectrum from 1545 – 1570 nm showing the maintained flat-top response of the pass- and stop-bands.

IV. CONCLUSION

We have demonstrated an ultra-broadband silicon photonicinterleaver capable of interleaving and de-interleaving fre-quency comb lines over a 125 nm bandwidth, which repre-sents a 55 nm improvement over previously reported devices[16]. The passbands exhibit an exceptionally uniform flat-topresponse, with a worst-case crosstalk suppression ratio of 10dB over the full 125 nm and typical crosstalk suppressionratio around 15 dB. This key metric can be improved to >20 dB extinction through further optimization of the couplerto the ring. The envelope of the spectrum decays only 1.5dB from the peak value between 1525 – 1650 nm, indicatingthat continued scaling in wavelength will not be insertion losslimited but rather dispersion and crosstalk limited. Throughthe use of advanced design techniques such as dispersionengineering for the waveguides and inverse design for thecouplers, continued scaling to bandwidths over 150 nm withlittle to no degradation in performance appears possible. Inaddition to the primary use-case of optical communications,the demonstrated device is promising for any application thatwould benefit from on-chip processing of frequency combs,such as spectroscopy and metrology.

V. ACKNOWLEDGEMENTS

The authors thank Advanced Micro Foundry (AMF) fordevice fabrication and Hao Yang for helpful discussions.

REFERENCES

[1] N. Jones, “The information factories,” Nature, vol. 561, pp. 163 – 166,Sept 2018.

[2] Q. Cheng, M. Bahadori, M. Glick, S. Rumley, and K. Bergman, “Recentadvances in optical technologies for data centers: a review,” Optica,vol. 5, no. 11, pp. 1354–1370, Nov 2018.

[3] D. A. B. Miller, “Device requirements for optical interconnects to siliconchips,” Proceedings of the IEEE, vol. 97, no. 7, pp. 1166–1185, 2009.

[4] J. Pfeifle, V. Brasch, M. Lauermann, Y. Yu, D. Wegner, T. Herr,K. Hartinger, P. Schindler, J. Li, D. Hillerkuss, R. Schmogrow,C. Weimann, R. Holzwarth, W. Freude, J. Leuthold, T. J. Kippenberg,and C. Koos, “Coherent terabit communications with microresonatorkerr frequency combs,” Nature Photonics, vol. 8, no. 5, pp. 375–380,May 2014.

[5] P. Marin-Palomo, J. N. Kemal, M. Karpov, A. Kordts, J. Pfeifle, M. H. P.Pfeiffer, P. Trocha, S. Wolf, V. Brasch, M. H. Anderson, R. Rosenberger,K. Vijayan, W. Freude, T. J. Kippenberg, and C. Koos, “Microresonator-based solitons for massively parallel coherent optical communications,”Nature, vol. 546, no. 7657, pp. 274–279, Jun 2017.

[6] B. Stern, X. Ji, Y. Okawachi, A. L. Gaeta, and M. Lipson, “Battery-operated integrated frequency comb generator,” Nature, vol. 562, no.7727, pp. 401–405, Oct 2018.

[7] A. Rizzo, Y. London, G. Kurczveil, T. Van Vaerenbergh, M. Fiorentino,A. Seyedi, D. Livshits, R. G. Beausoleil, and K. Bergman, “Energyefficiency analysis of frequency comb sources for silicon photonicinterconnects,” in 2019 IEEE Optical Interconnects Conference (OI),2019, pp. 1–2.

[8] B. Y. Kim, Y. Okawachi, J. K. Jang, M. Yu, X. Ji, Y. Zhao, C. Joshi,M. Lipson, and A. L. Gaeta, “Turn-key, high-efficiency kerr combsource,” Opt. Lett., vol. 44, no. 18, pp. 4475–4478, Sep 2019.

[9] Y. London, T. V. Vaerenbergh, L. Ramini, A. J. Rizzo, P. Sun,G. Kurczveil, A. Seyedi, J. Rhim, M. Fiorentino, and K. Bergman,“Performance requirements for terabit-class silicon photonic links basedon cascaded microring resonators,” J. Lightwave Technol., vol. 38,no. 13, pp. 3469–3477, Jul 2020.

[10] Y. London, T. Van Vaerenbergh, A. J. Rizzo, P. Sun, J. Hulme, G. Kur-czveil, A. Seyedi, B. Wang, X. Zeng, Z. Huang, J. Rhim, M. Fiorentino,and K. Bergman, “Energy efficiency analysis of comb source carrier-injection ring-based silicon photonic link,” IEEE Journal of SelectedTopics in Quantum Electronics, vol. 26, no. 2, pp. 1–13, 2020.

[11] D. Liang, G. Kurczveil, Z. Huang, B. Wang, A. Descos, S. Srinivasan,Y. Hu, X. Zeng, W. V. Sorin, S. Cheung, S. Liu, P. Sun, T. V. Vaeren-bergh, M. Fiorentino, J. E. Bowers, and R. G. Beausoleil, “Integratedgreen dwdm photonics for next-gen high-performance computing,” inOptical Fiber Communication Conference (OFC) 2020. Optical Societyof America, 2020, p. Th1E.2.

[12] A. Rizzo, L. Y. Dai, X. Meng, Y. London, J. Patel, M. Glick, andK. Bergman, “Ultra-low power consumption silicon photonic link designanalysis in the AIM PDK,” in Optical Interconnects XIX, vol. 10924,International Society for Optics and Photonics. SPIE, 2019, pp. 31 –36.

[13] Y. Huang, Q. Cheng, A. Rizzo, and K. Bergman, “Push—pull microring-assisted space-and-wavelength selective switch,” Opt. Lett., vol. 45,no. 10, pp. 2696–2699, May 2020.

[14] B. Guha, B. B. C. Kyotoku, and M. Lipson, “Cmos-compatible athermalsilicon microring resonators,” Opt. Express, vol. 18, no. 4, pp. 3487–3493, Feb 2010.

[15] A. Rizzo, Q. Cheng, S. Daudlin, and K. Bergman, “Ultra-broadbandsilicon photonic interleaver for massive channel count frequency combs,”in Conference on Lasers and Electro-Optics. Optical Society ofAmerica, 2020, p. STh1O.1.

[16] L.-W. Luo, S. Ibrahim, A. Nitkowski, Z. Ding, C. B. Poitras, S. J. B. Yoo,and M. Lipson, “High bandwidth on-chip silicon photonic interleaver,”Opt. Express, vol. 18, no. 22, pp. 23 079–23 087, Oct 2010.

[17] W. Bogaerts, P. De Heyn, T. Van Vaerenbergh, K. De Vos, S. Ku-mar Selvaraja, T. Claes, P. Dumon, P. Bienstman, D. Van Thourhout, andR. Baets, “Silicon microring resonators,” Laser & Photonics Reviews,vol. 6, no. 1, pp. 47–73, 2012.

[18] https://www.lumerical.com/products.[19] Z. Lu, H. Yun, Y. Wang, Z. Chen, F. Zhang, N. A. F. Jaeger, and

L. Chrostowski, “Broadband silicon photonic directional coupler usingasymmetric-waveguide based phase control,” Opt. Express, vol. 23,no. 3, pp. 3795–3808, Feb 2015.