Kishore Padmaraju, Xiaoliang Zhu, Long Chen, Michal Lipson, Fellow, IEEE,and Keren Bergman, Fellow, IEEE
Abstract— We elaborate on and experimentally characterizethe intermodulation crosstalk properties of a 10-Gb/s siliconmicroring modulator. Bit-error-rate measurements and eyediagrams are used to discern the degradation in signal qualitydue to intermodulation crosstalk. Evaluation of the powerpenalties with varying channel spacing are used to supportwavelength-division-multiplexed cascaded microring modulatorchannel spacings as dense as 100 GHz with negligible expectedintermodulation crosstalk.
Index Terms— Intersymbol interference, microring modulator,optical modulation, wavelength division multiplexing.
I. INTRODUCTION
GROWING bandwidth needs in the microelectronicsworld have merited the need for optical interconnects
on a scale previously unprecedented. These local opticalinterconnects will enable the high-bandwidth interconnectsnecessary for the growing fields of high-performance comput-ing, data centers, and ultimately, multi-core processors [1]–[3].However, in order to meet the demands of existing micro-electronic environments such optical interconnects have to bemanifested in a manner that is compatible with existing CMOScircuitry, feasible in a small footprint, and above all energy-efficient enough to merit its introduction over traditionalelectrical interconnects. For the aforementioned reasons, thesilicon photonic platform has received attention for beingable to resolve the bandwidth constraints of microelectron-ics while satisfying the stated criteria. Within the siliconphotonic platform, microring-based structures in particularhave received noted recognition because of their high-energyefficiency, small footprint, and ability to be easily cascadedfor wavelength-division-multiplexed (WDM) operation [3].
The WDM advantage of microring structures arises fromtheir inherent wavelength selectivity. A microring has an opti-cal resonance which is dictated by the radius of the microring.
Manuscript received December 20, 2013; revised May 5, 2014; acceptedMay 16, 2014. Date of publication June 3, 2014; date of current versionJune 30, 2014. This work was supported by the National Science FoundationEngineering Research Center on Integrated Access Networks underGrant EEC-0812072. The work of K. Padmaraju was supported by anIBM/Semiconductor Research Corporation Ph.D. Fellowship.
L. Chen and M. Lipson are with the School of Electrical and Com-puter Engineering, Cornell University, Ithaca, NY 14853 USA (e-mail:[email protected]; [email protected]).
Color versions of one or more of the figures in this letter are availableonline at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/LPT.2014.2326621
Fig. 1. (a) Configuration that cascades microrings of varying radius along asingle waveguide bus to generate WDM optical signals from parallel electricaldata streams. (b) Transmission spectrum of the two bit-states (solid anddashed) of a single microring modulator. The position of the wavelengthis indicated with a vertical line. (c) Aggregate transmission spectrum ofthe cascaded microring modulators when using a coarse channel spacing.(c) Aggregate transmission spectrum of the cascaded microring modulatorswhen using a dense channel spacing
Hence, functionality of the microring structure will only occurfor wavelengths of light for which the radius has been tunedfor, while adjacent wavelengths remain unperturbed. One ofthe most common applications of silicon microring structuresare in the creation of high-speed, small-footprint, energy-efficient microring modulators, with demonstrated speeds ofup to 40 Gb/s [4]. Fig. 1(a) illustrates the common WDM con-figuration of microring modulators, whereby the radius of eachmicroring is tuned to an unique wavelength. As each sourcewavelength propagates along the bus waveguide, it is opticallymodulated by the microring modulator corresponding to itswavelength. By this arrangement, parallelized high-speed elec-trical signals are imprinted on the WDM source wavelengths.
PADMARAJU et al.: INTERMODULATION CROSSTALK CHARACTERISTICS OF WDM SILICON MICRORING MODULATORS 1479
Fig. 2. Experimental setup used to measure and characterize intermodulation crosstalk.
With the ability to independently modulate the optical signalsalong the same bus waveguide, the bandwidth density of thephotonic system is drastically improved, helping to meet thefootprint constraints of microelectronic environments [5].
Fig. 1(b) illustrates the functionality of an individual micror-ing modulator. Displayed in solid is the passive opticalresonance of the microring modulator. Through injection ofcarriers the resonance is shifted (displayed dashed), allowinga ‘1’ bit to be imprinted on the aligned optical wavelength.Schematically depicted is the decrease in the quality factor(Q), caused by the increased loss due to free carrier absorptionfrom the injected carriers [6]. It should be noted that depletionmode modulators work in a similar fashion, utilizing anexpansion of the depletion zone instead of carrier injection,resulting in a reversed but identical modulation mechanism [4].
Fig. 1(c) illustrates the scenario of microring modulatorscascaded in a WDM configuration. Here, the wavelengthspacing between microring resonances is kept sufficientlycoarse such that the modulated resonances do not overlap withneighboring optical signals. However, as the density of theWDM signals increases (the channel spacing is decreased),these resonances will begin to overlap, as shown in Fig. 1(d).As a consequence, each wavelength is not only modulated bythe corresponding microring modulator, but adjacent modula-tors as well [7]. This intermodulation (IM) crosstalk servesas a limit on the WDM channel spacing density. In thisdemonstration we show the deleterious effect of IM crosstalkon the quality of the microring-modulated signal. Additionally,we provide in our analysis the limitations on channel spacingWDM cascaded microring modulators.
II. MEASUREMENTS AND RESULTS
A. Experimental Setup
The goal of our experimental characterization was to verifyand measure the deleterious effects of the microring-inducedIM crosstalk using standard optical performance metrics. To dothis rigorously we created an experimental setup (Fig. 2)with two signals. The first signal, tuned near the resonantwavelength of the microring, was modulated by the microringmodulator to verify its functionality. The second signal emu-lated a signal spaced adjacently in the WDM grid. This secondsignal, modulated prior, experiences IM crosstalk from the
microring modulator as the wavelength spacing between thetwo signals is brought closer.
In our experimental setup, we used a 6-μm radius carrier-injection microring modulator (passive Q of ∼6000) fabricatedat the Cornell Nanofabrication Facility. Further fabricationdetails can be found in [8]. The modulator was electricallycontacted with (GGB Picoprobe) high-speed RF probe tips.To electrically drive the microring modulator, a pulsed-pattern-generator (PPG) was used to generate a 10-Gb/s non-return-to-zero (NRZ) 27 − 1 pseudo-random bit sequence (PRBS)electrical signal. The 1-Vpp signal was biased at 0.6 V andconditioned with a pre-emphasis circuit to enable high-speedoperation of the device [6]. A CW tunable laser was set toTE polarization and tuned in wavelength near the resonantwavelength of the microring (1546.7 nm) in order to producean optimal 10-Gb/s microring-modulated signal.
To generate the second signal, the one being afflictedby IM crosstalk, we used a second separate tunable laser,and a commercial LiNbO3 Mach-Zhender modulator (MZM).A second PPG, generating a 231−1 PRBS pattern, was used todrive the LiNbO3 MZM modulator. The use of two differentPPGs with two different PRBS patterns ensured that there wasno correlation in the modulation data of the two signals.
The laser power levels were set such that the power of thetwo signals coming off-chip was equivalent. Once off-chip,the two signals were amplified and then separated usinga wavelength-selective-switch (WSS). The WSS featureda 100-GHz pass-band with 1.25-GHz tuning resolution, aroll-off of ∼1 dB/GHz, and a 80-dB out-of-band extinctionratio. During the measurements, the WSS was tuned suchthat the microring-modulated signal was at the edge of theWSS pass-band, yielding an optimum filtering between thetwo signals.
The eye diagram of each signal was captured as thewavelength spacing between the two signals was decreased.Additionally, following reception of each signal on a PIN-TIAand limiting amplifier (LA), a bit-error-rate-tester (BERT) wasused to measure the bit-error-rate (BER) curves of each signal.
B. Eye Diagrams & BER Measurements
Figure 3 depicts the eye diagrams recorded as thewavelength spacing between the two signals is decreased.
Fig. 3. Eye diagrams when the MZM-modulated channel is spaced (a) belowin wavelength relative to the microring-modulated channel and (b) above inwavelength relative to the microring-modulated channel.
Specifically, the wavelength of the microring-modulated signalis fixed near the resonant wavelength of the microring. Thewavelength of the second signal (modulated by the LiNbO3MZM) was incremented closer to the wavelength of the firstsignal. This was done in two sweeps approached from bothhigher and lower wavelengths relative to the resonance ofthe microring. The motivation for this was to expose theasymmetry of the IM crosstalk relative to wavelength.
Given the close spacing of the signals, a critical aspectof this experiment was to verify that observed crosstalk wasdue to the microring-induced IM crosstalk, and not filteringcrosstalk from the separation of the signals. The consistenteye diagrams of the microring-modulated signal indicate thatthe filtering was sufficient to completely separate the signals.The BER measurements in Fig. 4(a) & 4(b) further affirm thisfact, showing near identical BER curves for the microring-modulated signal at all channel spacings.
C. Power Penalties vs. Channel Spacing
The BER measurements in Fig. 4 are used to assess thepower penalties as the channel spacing is reduced. Trend lineswere established by exponentially fitting the BER measure-ments [9]. Figure 5 plots the power penalties from these BERcurves against the channel spacing.
We develop a simple model to provide a general expressionfor the crosstalk-induced power penalty vs. channel spacing.First, we note that a power penalty (pp) can commonly beattributed to the eye opening of the signal, that is, the powerin the ‘1’ bit (P1) and ‘0’ bit (P0) in the data signal [10], alsoillustrated in Fig. 6(a).
pp = 10 log
(P1 − P0
P1 + P0
)(1)
The main deleterious mechanism of IM crosstalk is tosuppress the power in the ‘1’ bit and ‘0’ bit of the adjacent
Fig. 4. BER measurements when the MZM-modulated channel is spaced(a) below in wavelength relative to the microring-modulated channel and(b) above in wavelength relative to the microring-modulated channel.
Fig. 5. Measured power penalties vs. relative channel spacing, fitted(separately for lower and higher wavelength spacings) to (3).
channel. Fig. 6(b) illustrates the corresponding decrease in eyeopening. To determine what this eye closure is we modelthe microring transmission spectrum using as a Lorentzianfunction [11].
T (λ, λ0, Q) =(
1 +(
2Q(λ − λ0)
λ0
)2)−1
(2)
PADMARAJU et al.: INTERMODULATION CROSSTALK CHARACTERISTICS OF WDM SILICON MICRORING MODULATORS 1481
Fig. 6. Eye opening (E.O.) under (a) normal operation, and (b) when thedata channel is suffering from intermodulation crosstalk. (c) Illustration of thetwo bit-states comprising microring modulation.
Fig. 7. Schematic depicting the use of DSP at the receive end of the opticallink to mitigate intermodulation crosstalk and allow denser channel spacing.
As seen in the diagram of Fig. 6(c), the two bit-statescomprising microring modulation can be described with para-meters described by the Q and resonant wavelength (λ0) ofthe microring. During modulation, λ0 changes from the freecarrier dispersion effect, and Q changes due to free carrierabsorption [6]. Equating the reduction in the ‘1’ bit to thetransmission of the microring, the relative incurred powerpenalty, �pp = pp – pp’, is then
�pp = 10 log
((P1 − P0
P1 + P0
) (T (λ, λ0, Q)P1 + P0
T (λ, λ0, Q)P1 − P0
)). (3)
Fig. 5 shows the measured power penalties as fitted to (3),using separate values of Q and λ0 for the two measurementsweeps. Large increases in power penalty are seen as thechannel spacing is progressively decreased from either side.
III. DISCUSSION
Our fitted measurements of power penalties show theexpected features of IM crosstalk. Mainly, that it increaseswith decreased channel spacing, and secondly, that it is asym-metric (attributable to the unique modulation mechanism ofthe microring). For a system using a microring modulator withthese Q characteristics, power penalties of 0.1 dB and 0.7 dBwere measured for channel spacings of 100 GHz and 50 GHz,respectively [Fig. 5], providing evidence that microring mod-ulators would be suitable for use in current telecom WDMsystems, which are routinely spaced at 100 GHz [10].
The Q characteristics of the microring modulator can betailored as necessary for applications requiring even denserchannel spacing. However, increasing the Q of the microringincreases the photon lifetime, which puts a limitation on themodulation bandwidth of the microring [4]. An interestingalternative for increasing the channel density is to use elec-tronic digital signal processing (DSP) at the receive end of theoptical link [12], as depicted in Fig. 7. The deterministic natureof IM crosstalk especially lends itself towards mitigationusing DSP. For example, a simple relation for IM crosstalk
cancellation (referencing Fig. 7) can be expressed as
zn = yn + αzn−1 (4)
where due to the asymmetry of IM crosstalk, only the con-tribution from the lower channel is relevant for the crosstalkcancellation (hence, z1 = y1).
IV. CONCLUSION
We have measured and extrapolated power penalties fromIM crosstalk for a typical microring modulator operating at10 Gb/s. We expect similar results at the higher data ratesthat microring modulators are trending towards [4]. Our powerpenalty measurements show credibility towards the use of thenormative WDM channel spacings of 100 GHz and 50 GHz.Our proposed use of DSP may help push this channel spacingfurther.
However, it should be noted that in a completely microring-based link, IM crosstalk is not the only factor constrainingchannel spacing density. Other relevant parameters includespectral crosstalk from the microring demultiplexing filters,aggregate (including electrical driving circuitry) energy effi-ciency when using different bit-rates, and non-linear opticaleffects within the silicon waveguides [3], [13]. Such studiespoint towards either a 50 GHz or 100 GHz WDM channelspacing as preferred for system-optimized microring-basedoptical links.
REFERENCES
[1] A. V. Krishnamoorthy et al., “Computer systems based on siliconphotonic interconnects,” Proc. IEEE, vol. 97, no. 7, pp. 1337–1361,Jul. 2009.
[2] D. Brunina, X. Zhu, K. Padmaraju, L. Chen, M. Lipson, andK. Bergman, “10-Gb/s WDM optically-connected memory system usingsilicon microring modulators,” in Proc. 38th ECOC, Sep. 2012, pp. 1–3.
[3] N. Ophir, C. Mineo, D. Mountain, and K. Bergman, “Silicon photonicmicroring links for high-bandwidth-density, low-power chip I/O,” IEEEMicro, vol. 33, no. 1, pp. 54–67, Jan./Feb. 2013.
[4] G. Li et al., “Ring resonator modulators in silicon for interchip photoniclinks,” IEEE J. Sel. Topics Quantum Electron., vol. 19, no. 6, p. 3401819,Nov./Dec. 2013.
[5] S. Manipatruni, L. Chen, and M. Lipson, “Ultra high bandwidthWDM using silicon microring modulators,” Opt. Exp., vol. 18, no. 16,pp. 16858–16867, Aug. 2010.
[6] Q. Xu, S. Manipatruni, B. Schmidt, J. Shakya, and M. Lipson,“12.5 Gbit/s carrier-injection-based silicon micro-ring siliconmodulators,” Opt. Exp., vol. 15, no. 2, pp. 430–436, Jan. 2007.
[7] K. Padmaraju et al., “Intermodulation crosstalk from silicon microringmodulators in wavelength-parallel photonic networks-on-chip,” in Proc.23rd Annu. Meeting IEEE Photon. Soc., Nov. 2010, pp. 562–563.
[8] L. Chen, K. Preston, S. Manipatruni, and M. Lipson, “Integrated GHzsilicon photonic interconnect with micrometer-scale modulators anddetectors,” Opt. Exp., vol. 17, no. 17, pp. 15248–15256, Aug. 2009.
[9] C. Jeruchim, “Techniques for estimating the bit error rate in thesimulation of digital communication systems,” IEEE J. Sel. AreasCommun., vol. 2, no. 1, pp. 153–170, Jan. 1984.
[10] G. P. Agrawal, Fiber-Optic Communication Systems, 3rd ed. Hoboken,NJ, USA: Wiley, 2002.
[11] J. E. Heebner, “Nonlinear optical whispering gallery microresonators forphotonics,” Ph.D. dissertation, Inst. Opt., Univ. Rochester, Rochester,NY, USA, 2003.
[12] M. Glick, P. Watts, R. Waegemans, P. Bayvel, and R. I. Killey,“Electronic signal processing to improve system performance of opticalinterconnects,” in Proc. ICTON, Jul. 2007, pp. 298–301.
[13] K. Preston, N. Sherwood-Droz, J. S. Levy, and M. Lipson,“Performance guidelines for WDM interconnects based on siliconmicroring resonators,” in Proc. CLEO, May 2011, pp. 1–2.
