Multimode and single-mode transmission over universal fiber for data center applications

Optical Fiber Technology xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Optical Fiber Technology journal homepage: www.elsevier.com/locate/yofte

Invited Papers

Multimode and single-mode transmission over universal fiber for data center applications ⁎

Xin Chen , Jason E. Hurley, Aramais R. Zakharian, Jeffery S. Stone, William A. Wood, Bruce Chow, Doug Coleman, Ming-Jun Li Corning Research and Development Corporation, Corning, NY 14831, USA

A B S T R A C T Universal fiber is a multimode fiber that has an LP01 mode field diameter approximately matched to that of standard single-mode fiber. It can transmit both multimode and single-mode signals using transceivers designed for either multimode fiber or single-mode fiber. By using universal fibers, one can bridge the needs for both single-mode and multimode transmission through a uniform and simplified cable infrastructure to accommodate the full distance range needed, while having upgradability from 10G to 40G to 100G and to even higher data rates. We build on our previous work that analyzed the dependence of LP01 mode-field diameter on fiber parameters, and the coupling loss for both single-mode and multi-mode transmission, with a further study of mode coupling and multiple path interference associated with single-mode operation. We demonstrate system performance in both single-mode and VCSEL-based multimode transmissions for a number of 100G transceiver types using a QSFP form factor − the preferred choice in data centers.

1. Introduction 1.1. Use of multimode fibers and single-mode fibers in the data center In data centers (DC), both multimode fibers (MMF) and single-mode fibers are used with VCSEL-based multimode transceivers and singlemode transceivers, respectively. With the emerging hyper-scale DC, single-mode transmission is deployed more frequently to meet the need of longer system reach. Enterprise data centers primarily use OM3/OM4 multimode fiber for data transmission as most channel lengths are less than 100 m, and the trend looks to continue since multimode (MM) transmission remains a cost-effective solution covering the majority of transmission distances [1]. Even in large scale DC, a portion of the optical transmission has short distances, less than tens of meters, where MM transmission is more cost-effective than single-mode (SM) transmission. At the same time, continued efforts are underway to develop novel approaches for using long wavelength MMF over longer system reaches [2–3]. On the other hand, single-mode connectivity is used in data centers both at short distances, similar to MMF, and at longer distances, above 100 m. Because SM transmission enjoys high system bandwidth and is capable of longer system reaches, mega- and hyper- data centers tend to predominantly adopt SM transmission. Even in such large scale data ⁎

centers, there is still a large percentage of SM transmission for distances less than 100 m, as reported in Ref. [4]. For very short distances in enterprise data centers, single-mode fiber is used for the carrier interface to provide linkage to the router and FICON (i.e. Fibre Connection) mainframe for storage applications [5]. For distances up to several hundred meters, both MM and SM transmission co-exist, depending on the data rate and the type of data centers. 1.2. Fundamental mode transmission in multimode fiber It is desirable to do single-mode transmission over conventional MMF, which has a 50 μm core diameter. Although it is possible to launch the light into only the fundamental mode of MMF using various complicated mode expansion techniques to bridge the difference of fundamental modes between 50 µm core MMF and standard single mode fibers [6–9], the solutions are too expensive for cost-sensitive data center applications as they involve complex experimental setups. An alternative and more promising solution is to alter the MMF design to allow the mode field of fundamental mode to match that of standard single mode closer [10–12]. In Ref. [10], we presented a design of MMF we referred to as universal fiber (UF) with a core having a relative refractive index delta of 1% and a core diameter of about 23 μm. This study also demonstrated 850 nm VCSEL based multimode transmission at 10 Gb/s and 25 Gb/s over 100 m and 50 m, and 1310 nm single-mode

Corresponding author. E-mail address: [email protected] (X. Chen).

https://doi.org/10.1016/j.yofte.2017.11.003 Received 28 April 2017; Received in revised form 27 September 2017; Accepted 6 November 2017 1068-5200/ © 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Chen, X., Optical Fiber Technology (2017), https://doi.org/10.1016/j.yofte.2017.11.003

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

25 Gb/s NRZ and 44 Gb/s PAM4 transmission over 2 km. The main limitation of that particular fiber design is high insertion loss of about 5 dB when it is coupled to either a VCSEL transmitter or a 50 μm core MMF. An improved UF design was subsequently reported in Ref. [11] with core delta at 1.2% and core diameter around 30 μm, the typical insertion loss to VCSEL was reduced to below 3 dB and improved system performance in a number of applications was demonstrated. These include 110 m MM transmission with 100G SR4, 150 m MM transmission using 40G SWDM4 and 2.7 km SM transmission using a 100G CWDM4 transceiver. We also note here that a gradient-index multimode fiber that is designed for both 850 nm VCSEL transmission and single-mode transmission was previously proposed for home network applications [12]. The core of this fiber has very high relative refractive index delta of 2.5–2.9% between the core and the cladding and a core diameter of 40 μm. The transmission experiments demonstrated single-mode transmission using single-mode laser at 10 Gb/s over 100 m and similar distance for 1.25 Gb/s MM transmission. The high delta of the core, which results in very high numerical aperture, facilitates the coupling of the light from transmitter into the fiber and reduces the macrobend loss, as needed for a home network, but makes it difficult to achieve high modal bandwidth. The UFs proposed in [11] have much lower delta and are close to that of existing high bandwidth MMFs. They can be used in DC, LAN and central office environments, which can go as long as 2 km. Although the core diameter of 30 μm is smaller than the 50 μm MMF, the coupling losses from typical VCSELs and between UFs are good enough for short reach transmissions. With improved coupling optical designs inside VCSELs and connector technology, the coupling losses can be reduced further for future high speed applications. Table 1 compares the attributes for different standardbased MMFs and single-mode fibers.

Table 2 Comparison of cost and power consumption for multimode and single-mode transceivers. For both the cost and power consumption, the values are shown relative to that of 10G MM transceivers. Fiber

MMF (OM1OM4) SMF

For the majority of small to large enterprise data centers operating between 1G and 40G, multimode OM3/OM4 fibers working with VCSEL-based transceivers remain the most cost-effective way to cover the moderate distance requirements. This is driven mostly by the large difference in prices between SM and MM transceivers [13–14], with upto-date information in Table 2. However, as speeds approach 100G and beyond, the difference in price is shrinking. In fact, at or beyond 100G, many data center operators are being advised and are expected to be deploying single-mode optics due to the price and capability requirements. Therefore, operators who would like to build an infrastructure that is capable of supporting the needs for today, yet be flexible enough to also support the future, are faced with a difficult choice between two options. The first option is to deploy MMF today to take advantage of Table 1 Comparison of attributes and system performance for different multimode and singlemode fibers. 40G standard based SM transceivers have 2 km, 10 km, and 40 km variants for distance specifications. 100G standard based SM transceivers have 10 km and 40 km variants, while 100G MSA based SM transceivers have 500 m distance for PSM4, and 2 km distance for CWDM4.

OM1 OM2 OM3 OM4 SMF

2 1 1 1 0.34

Core diameter (μm)

62.5 50 50 50 9

OFL BW (MHz.km)

EMB BW (MHz.km)

Link distance (@850 nm) (m)

850 nm

850 nm

10G

40G

100G

200 500 1500 3500 NA

N/A N/A 2000 4700 N/A

33 82 300 400 N/A

N/A N/A 100 150 500 m/ 2 km/ 10 km/ 40 km

N/A N/A 70 100

Relative power consumption

10G

40G

100G

10G

40G

100G

1

6

17

1

1.5

3.5

3

15–21

13–37

1.5

5.0

5.5

the lower cost of MM transceivers today, but then rip the cable infrastructure and replace with SM version within the first or second upgrades. The second option is to deploy single-mode fiber today and pay more for the SM transceivers with each upgrade until the SM transceiver price becomes comparable to the MM option. Therefore, selecting an appropriate fiber is both a financial and technological decision with consequences for today as well as for the future. Multimode and single-mode fiber each has their own advantages and drawbacks. The universal fiber provides the freedom to choose low cost MM transceivers for the current and future needs, but is ready for upgrade to SM transceivers in the future, when MM technology becomes limited by distance and when SM transceiver prices become more favorable. Another factor that has attracted increased attention is the power consumption in DC [13]. DC operations are highly sensitive to cost and power consumption − in particular for mega- and hyper data center operators. The rise of SM transmission is driven by the desire to have a simplified and uniform transmission medium and cable infrastructure. However single-mode systems consume more power, as shown in Table 2. Better understanding of link length distribution can shed more light on the power consumption consideration. For enterprise DC, according to the system length distribution for the OM4 MMF cable products manufactured by Corning between 2013 and 2015 [4], the average length was 48 m with about 2% of lengths greater than 250 m. For hyper-scale data centers, based on the system length distribution for the single-mode cable products manufactured in the same period of time, the average length was 152 m with only 2% of lengths greater than 350 m. These data indicate that MMF can support most distances required for enterprise and hyper-scale data center networks. However, for the percentage of link lengths that cannot be supported by MMF, single-mode fiber has to be deployed. For data center operations, installing two types of fibers, both MM and SM, will increase network and logistics complexity, as well as make fiber cable management more challenging. A UF that can accommodate both MM and SM transmission will bridge the gap in fiber cable deployments for new DC and offers flexibility for future system upgrades. In this paper, we review recent progress on UF and present new system testing results to illustrate the UF transmission properties and performance covering several new 100G QSFP transceiver variants. In Section 2, we present the fiber design, and study the multiple path interference (MPI) associated with the SM transmission. In Section 3 we show new system performance results for several 100G QSFP transmission. Finally, in Section 4, we present a brief conclusion.

1.3. Benefit of universal fiber for both single-mode and multimode transmission

Core Δ (%)

Relative transceiver cost

2. Fiber design, multiple path interference and coupling into standard single mode fiber UFs are designed for both SM and MM transmissions. Various considerations have been taken into account so that the fiber performs sufficiently well for both types of operations [10–11]. In this section, in addition to the coupling loss for MM operation and SM operation from Ref. [1], we present new results on understanding the effect of MPI related to SM operation, both through numerical modeling and experimental study. 2

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

main misalignment in the connector. Since we expect UF to be used for both SM and MM transmission, the connector quality should follow that for SM transmission. For modeling, we take maximum offsets around δmax = 0.7–1 μm as representative values inferred from 0.1–0.2 dB maximum insertion loss in standard SMF-to-SMF connectors at 1310 nm wavelength, assuming the insertion loss is due to the transverse offset only. Given an optical field at the connector input, the amplitude coupling coefficients for the modes of the fiber at the connector output are computed using mode projection (overlap) integrals [15]. We use numerical solutions of the scalar wave-equation to compute the LPmn modes of the fiber, and all guided mode groups of the UF found by the solver are kept in the analysis, without differential mode attenuation. For a nominal UF design, the propagation length required for 2π phase accumulation by higher order modes with respect to the LP01 mode, is less than 1 mm, and therefore we employ a random phase model for higher order modes in order to account for phase dispersion and modal noise. The resulting total optical field coupled into the UF is, therefore, a superposition of UF modes with random phases and magnitudes governed by the connector misalignment. For a given link realization, defined as a fixed set of random misalignments at the connectors, multiple instances of random phase realizations are considered to evaluate transmission through UF. Specifically, the LP01 mode of the launch SMF is coupled through the first connector into UF for Nphase different realizations of the resulting optical field in UF, due to random phases of modes. The optical field is propagated and detected coherently at the receiving SMF, resulting in a distribution of received power for Nphase realizations. This yields the statistical average of optical power, Pavg, over mode phase realizations. The insertion, or coupling, loss is defined based on the ratio of this average power to the input power, Pin:

2.1. Fiber design The key consideration of the UF design is for the fiber to be able to operate both for multimode transmission around 850 nm and singlemode transmission around 1300 nm and 1550 nm. The core of the UF has a simple alpha gradient-index refractive index profile similar to conventional OM3 and OM4 MMFs, as described by the following equation:

n (r ) = n 0 · 1−2Δ(r / a)α

(1)

where n0 is the refractive index in the center of the core, a is the core radius, and Δ = (n02−n12)/(2n02) is the relative refractive index delta, where n1 is the refractive index of the cladding. In order to accommodate the single-mode transmission it is necessary for the mode field diameter of the fundamental LP01 mode to approximately match that of a standard single-mode fiber, which is around 9.2 μm at 1310 nm. For multimode coupling from a 50 μm core fiber to a UF, the insertion loss is caused by the smaller core diameter of UF. For singlemode coupling, the insertion loss is affected by the mismatch of MFD. It is found in Ref. [11] that the MFD of LP01 mode increases with the core diameter and decreases with the core refractive index delta. Therefore one can search within the parameter space to identify a proper balance for both SM and MM transmission. Indeed, detailed analysis to this end has been presented in [11]. 2.2. Multi-path interference and coupling into single-mode fiber In addition to insertion loss, another phenomenon of interest is multi-path interference (MPI) effects. In conventional SM transmission, MPI results from the multiple reflections at the connector interfaces. For UF, the MPI results from the slight amount of light launched into higher order modes and coupled back to the fundamental mode. In the SM transmission regime, optical power coupling between the fundamental mode and higher order modes of UF can occur at SMF-to-UF or UF-toUF connectors. This can potentially lead to an MPI in the received optical signal, as well as an associated MPI power penalty. As discussed above, minimization of the insertion loss between standard SMF and UF by fiber core parameter optimization allows one to limit the maximum amount of power that may be coupled from a transceiver SMF patchcord into higher-order modes of UF, thus mitigating the MPI effect. To quantify the impact of insertion loss and MPI on the single-mode transmission at 1310 nm wavelength, we consider a model of the optical link, comprised of a standard single-mode launch fiber, connected to a UF under test, followed by a receiving SMF, as shown in Fig. 1. In the following analysis, we take into account multiple paths of optical signal due to single-pass transmission of higher order modes. Multiple paths due to signal reflections at the connectors and a round-trip over the fiber are not considered, since these require at least two reflections, which attenuate the optical power that contributes to the interference. To evaluate the worst-case performance, the coherence length of the source is assumed to be longer than the link lengths of interest, and continuous-wave excitation is used to model the time dependence of the optical signal. At each connector, misalignments due to transverse and longitudinal offsets, as well as angular tilt, can contribute to the coupling loss and higher order mode excitation. The latter two, however, have a smaller effect in practice, hence in our analysis, we retain only the random transverse offsets in the plane normal to the fiber axis, as the

Pavg ⎞ InsertionLoss = −10·log10 ⎛ ⎝ Pin ⎠ ⎜

⎟

(2)

Since MPI power penalty results from the reduction in the detected optical power due to interference, the ratio of the lowest received power, Plow, detected from the distribution, to the average power, defines the MPI loss for a given link realization:

P MPILoss = −10·log10 ⎜⎛ low ⎞⎟ ⎝ Pavg ⎠

(3)

The MPI loss definition here is consistent with the MPI loss terminology used in 100G CWDM4 specifications for MPI related penalties [16]. In traditional SM transmission, the MPI arises more often from multiple reflections at the connector interface, which creates multiple paths for the signals. However, the nature of MPI is the same as discussed for UF due to connector offsets, as illustrated in Fig. 1. A Monte-Carlo simulation of random misalignments at the connectors then can be performed by considering a total of NMC random link realizations to compute the coupling and MPI loss distributions for a desired maximum connector offset, δmax. The resulting highest MPI loss can be considered as a measure of worst case link performance in terms of MPI power penalty. We consider first the case of nominally aligned connectors by setting all offsets to zero, δmax = 0. In such a link, the modal mismatch between SMF mode and LP01 mode of UF is the sole contributor to the insertion loss, and MPI loss can be minimized for closely matched modes. Fig. 2 shows the insertion and MPI loss computed for a range of UF core index Δ and diameters. For a core with Δcore = 1.2% and Fig. 1. Schematic of a UF connected to a launch and receiving SMFs, and illustration of the MPI effect in the singlemode transmission regime due to Higher Order Modes (HOMs) of the UF. The optical power PHOM coupled from HOMs into the receiving SMF can interfere coherently with the signal transmitted through the LP01 mode, resulting in

MPI power penalty.

3

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

Fig. 2. Insertion loss and MPI loss computed for a link with nominally aligned connectors, δmax = 0.

Fig. 3. Insertion and MPI loss computed for links with the maximum allowed connector offsets of δmax = 0.7 μm (top row) and 1.0 μm (bottom row).

diameter Dcore = 30 μm both the insertion loss and the MPI loss are seen to be ∼0.2 dB, and there is an approximately linear relation between Δcore and Dcore for a given level of MPI loss. If we assume that the maximum offset of δmax = 0.7 μm is allowed at the connectors, the MPI loss increases to ∼0.3 dB for UFs with designs around Δcore = 1.2% and core diameter of 30 μm, as shown in Fig. 3. With even larger δmax=1.0 μm, the MPI loss is < 0.5 dB for variations in Δcore from 1.1% to 1.3% and diameter Dcore from 29 to 31 μm. The coupling loss for the same range is computed to be at most 0.55 dB. Note that δmax=0.7 μm corresponds to 0.1 dB standard singlemode fiber connector loss due purely to the offsets. A set of UF samples with nominally the same Δcore = 1.2% and core diameters in the range from 26 μm to 38 μm have been characterized experimentally with respect to the insertion loss and MPI loss. The experimental setup for MPI measurement is similar to that in Fig. 1. A 5 m segment of UF is sandwiched between two standard single-mode fibers. The standard single-mode fiber and the UF are mated with LC connectors. The optical transmitter is a 1310 nm SM transceiver designed for 10G transmission over distances up to 10 km. The received optical power is measured over time. The standard single-mode fiber near the optical transmitter is shaken manually to accelerate the processes that cause the received optical power to vary over the range that would otherwise take a much longer time to realize. The shaking

introduces changes in the input state of polarization and other minor perturbation of the optical fiber. The difference between the average received optical power and lowest received optical power in dB units is recorded as MPI loss, as defined in Eq. (3). The corresponding model results were computed by assuming a nominal allowed maximum connector offset of δmax = 1.0 μm. Fig. 4 shows the comparison of the measured data with simulated maximum and average (denoted by < >) losses. The differences between the modeling results and the experimental results are relatively minor, which suggests a good agreement. Since both insertion loss and MPI loss increase linearly with the core diameters, a choice or trade-off can be taken to balance the performance between MM operation and SM operation. The fiber core diameter can go as large as possible as long as the insertion loss and MPI loss are contained with a reasonable upper limit.

3. System level testing and verification for major MM and SM applications In this section, we show the system level testing results for UF with several 100G transceivers of QSFP form factors. The types of transceivers reported here in combination with those used in [11] essentially cover all 100G QSFP-based transceivers for DC. We cover both MM 4

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

Fig. 4. Measured insertion loss (IL) and MPI loss for Δcore = 1.2% UFs of different core diameters, compared to the corresponding computed quantities (maximum and average), assuming largest allowed connector offset of δmax = 1.0 μm.

BER

transmission and SM transmission. Traditional 40G/100G standardbased connectivity is based on eight fiber parallel optics using MPO type connectors. However, in recent years, optical transmission using duplex LC connectivity has become popular because the same data traffic can be transmitted with two fibers instead of eight fibers, due to the use of wavelength division multiplexing (WDM) for SM transmission from 1270 to 1330 nm and MM transmission from 850 nm to 950 nm. These Duplex LC based transceivers are typically supported by multi-source agreements (MSA) outside the standard bodies. Note that duplex SM LC 40/100G WDM variants in the Ethernet Standard are specified at long reaches of 10 km to 40 km, and are longer than distances used in DC. In Ref. [11] we have shown some 100G testing results using 100G eSR4 MM transceiver and a 100G CWDM4 SM transceiver along with 40G SWDM4 transmission. In the current work, we show new results on a more complete set of transceivers using the improved UF that further include 100G SWDM4 and 100G BiDi for MM duplex LC transmission, and 100G PSM4 for parallel single-mode transmission at 1310 nm or 1550 nm. The UF under test is an improved fiber as compared to Ref. [11], with a core having a 1.2% delta and a diameter of approximately 31 µm. The fiber tested has a modal bandwidth of 2.9 GHz.km at 850 nm, 3.43 GHz.km at 900 nm and 2.33 GHz.km at 950 nm, as shown in Fig. 5. The bandwidth is measured with a ModCon launch conditioner manufactured by Arden Photonics [17], which provides an encircled flux launch condition. Other properties of this fiber are similar to those reported in [11]. This improved fiber is used for MM transmission reported below, but makes little difference for SM transmission as compared to previously reported results.

10

-4

10

-5

10

-6

Back to Back 150m UF

-7

10 -8 10 -9 10-10 10-11 10-12 10-13 10

-15 -14 -13 -12 -11 -10

-9

-8

Received Optical Power (dBm)

-7

Fig. 6. The BER vs. received power for back to back and 150 m system configurations.

from a 50 μm core MMF based fan-out cable into the UF is 1.9 dB in the current experiment. Fig. 6 shows the measured BER as a function of received power. Similar to [11], the transceiver reaches bit error free performance at −11 dBm for back to back condition with 2 m of jumper fiber. With the use of the improved UF, we see that this fiber reaches bit error free performance at −9 dBm at 150 m length. This is much better than the optical power reported in [11] for 60 m and 110 m when approaching bit error free conditions. The optical eye diagrams are shown in Fig. 7 for both B2B and 150 m conditions. The eye opens widely even at 150 m. This indicates that the UF can have the potential to perform in extended reach regime longer than the 100 m limit defined by the standard using OM4. We note that to obtain the BER vs. received optical power curve, we inserted an multimode variable optical attenuator (VOA) near the receiver. The VOA itself has attenuation of around 2.7 dB even when the setting is at 0 dB. The fact we can achieve error free at −9 dBm with VOA suggests the system can handle additional loss for example due to connectors. In addition, we adopt a different instrument, Viavi optical network tester (ONT) [18] to conduct the

3.1. Multimode transmission with 100G eSR4 transceiver We tested the new UF at 850 nm with a 100G eSR4 VCSEL-based transceiver using the same setup as described in [11]. The transceiver is a commercial transceiver with QSFP form factor operating with four 25 Gb/s lanes in compliance with IEEE 802.3bm standard but with the enhancement capable of transmitting 200 m over OM3 and 300 m for OM4. Only one channel was used. The insertion loss of coupling light

Modal Bandwidth (GHz.km)

4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0

800

850

900

Wavelength (nm)

950

1000

Fig. 5. The modal bandwidth of the UF under testing.

Fig. 7. Optical eye diagrams for back to back (B2B) and 150 m system configurations.

5

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

OSA Signal (dBm)

Ch. 1

Ch. 3

Ch. 2

Ch. 4

3.3. MM transmission with 100G BiDi

-20

100G BiDi is an emerging transceiver that utilizes duplex LC connectivity and two wavelengths (850 nm/900 nm) similar to 40G BiDi transceiver. In addition to moving to higher baud rate, it also adopts PAM4 technology, which doubles the number of bits in serial data transmissions by increasing the number of levels of pulse-amplitude modulation and therefore doubles the data rate. 100G BiDi transceiver combines the host card four 25G NRZ electrical lanes into two 50G wavelengths (850 nm/900 nm) using a PAM4 ASIC inside the module. On receiving the 50G PAM4 signals, the transceiver module PAM4 ASIC converts the 50G PAM4 signals back to 25G NRZ, to interface to the host card four 25G lane CAUI electrical interface. FEC technology is included in the module ASIC to address the PAM4 insertion loss and to provide system performance at equal to, or better than, 10−12 BER. The 100G BiDi transceiver has the QSFP form factor and supports up to 70/ 100/150 m over OM3/OM4/OM5 multimode fibers. Since the 100G BiDi transceiver has built-in ASIC for FEC and expects Ethernet traffic, traditional BER testing cannot be performed. Instead the 100G system level testing can be done with Viavi ONT used in Section 3.1 and 3.2 without using FEC from the testing equipment. We were able to demonstrate 150 m and 200 m error free performance using the UF shown in Fig. 5 for over 24 hours’ time. We also measured the coupling loss into the UF as compared to OM3 pigtail fiber. For the transceiver we used, we observed 0.8 dB coupling loss for 850 nm channel and 2.1 dB coupling loss for the 900 nm channel.

-40 -60 -80 840

860

880

900

920

Wavelength (nm)

940

960

Fig. 8. The optical spectrum of 100G sWDM transceiver.

system testing. With Physical Layer, forward error correction (FEC), PCS and Mac/IP layers all turned on, we were able to show the error free performance over multiple days. 3.2. Multimode transmission with VCSEL based 100G sWDM transceivers Here, we tested the UF with a VCSEL-based 100G multimode transceiver that is designed for sWDM [19–20]. In the previous work [11], we reported the transmission results for 40G SWDM transceiver using UF at 150 m. We have since obtained a 100G SWDM transceiver. Similar to 40G SWDM4 transceiver, this transceiver utilizes four wavelengths as shown in Fig. 8, with each operating at 25 Gb/s for an aggregated 100G data rate within a single fiber. A UF fiber with the modal bandwidth reported in Fig. 5 and a length of 150 m was used in the sWDM transmission test. The UF was connectorized with LC connectors on each end. The BER versus received power curves are shown in Fig. 9 for a PRBS pattern with a 231-1 bit sequence. Here the transmission is based on 25 Gb/s for each channel, which is more challenging than 40G SWDM based on 10 Gb/s channels. We also observed VCSEL to UF coupling loss of 3–5 dB as compared to VCSEL coupling into a convention OM3 MMF as driven by the specific launch conditions of the transceiver under test. In our measurement of larger set of VCSEL transceivers, coupling loss for majority is below 3 dB. But some hot outside source can reach 5 dB. Although the requirement is to meet 5 × 10−5 BER before incurring forward error correction (FEC), we were able to get BER data to 10−10 or below in most cases for 100 m of UF. In addition, we have used Viavi ONT and similar configuration used in Section 3.1 to test the UF at 150 m length. The system performed error free for more than ten hours. The optical eye diagrams obtained after 150 m of UF are shown in Fig. 10. The eyes are open for all four channels.

3.4. Single-mode transmission using 100G PSM4 at 1310 nm and 1550 nm Another type of SM transmission that is enjoying wide use is parallel single-mode transmission [21]. It is designed for use in 100 Gigabit Ethernet links and utilizes a parallel single-mode fiber infrastructure to support reach of up to 500 m. These transceivers are compliant with the QSFP28 MSA, PSM4 MSA [21] and applicable portions of IEEE P802.3bm. The transmission of 100G PSM4 can be based on either 1310 nm lasers or 1550 nm lasers. We have obtained transceivers operated at both wavelengths for our testing. Although it is common for 100G PSM4 transceiver to be implemented at 1310 nm wavelength, the PSM4 can also be based on the 1550 nm wavelength, which is optimal for its underlining Silicon Photonics technology and can achieve up to 2 km system reach. The 1550 nm 100G PSM4 transceiver has full interoperability with most other industry 100G PSM4s based on 1310 nm wavelength, even though the wavelength is different, because it uses a wide band receiver that can detect both 1310 nm and 1550 nm wavelengths. The fundamental mode transmission of UF can be done in both 1310 nm and 1550 nm. If the LP01 MFD of the UF is similar to that of standard single-mode fiber at 1310 nm, and it also closely matched at 1550 nm [11]. The optical spectrum of the two transceivers are shown in Fig. 11, (a) for the 1310 nm transmitter and (b) for the 1550 nm transmitter. The wavelengths of the peak power for each laser are very close to 1310 nm and 1550 nm respectively. The two 100G PSM4 transceivers use an angle-polished MPO connector. We inserted a commercial fan out cable made from standard single-mode fiber with an angled MPO connector into the transceivers. The other end of the fan-out cable is a set of individual fiber pigtails with LC connectors, which allow us to access individual transmitters or receivers. We conducted the experiments to obtain the BER vs. received optical power curves. We used a SM VOA near the receiver to control the amount of optical power reaching the receiver. In Fig. 12, we show the BER vs. received optical curve for the 1310 nm transceiver in (a) and the 1550 nm transceiver in (b). In the experiment, we have tested two conditions, B2B condition and a condition with 2 km UF. The 2 km distance well exceeds the specified distance of 500 m for using 100G PSM4 with standard single-mode fiber. For the 1310 nm transceiver, it can be found that both B2B and transmission over 2 km of UF have very robust performance. They reach bit error free condition at a received

Fig. 9. BER vs. received power for four wavelength channels using 100 m UF.

6

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

Fig. 10. Optical eye diagrams obtained from each wavelength channel after 150 m of UF.

0

OSA Signal (dBm)

-10

OSA Signal (dBm)

0

(a)

-20 -30 -40 -50 -60 -70 1310.5

1311.0

1311.5

1312.0

1312.5

-10 -20 -30 -40 -50 -60 -70

1549.5

-3

10

-4

10

(a) -5

10 -6 10 -7 10-8 10-9 10-10 10-11 10-12 10-13 10-14 -15 10 10

B2B 2km

-18 -17 -16 -15 -14 -13 -12 -11 -10

Receicved Optical Power (dBm)

BER at 1550nm

BER at 1300nm

Wavelength (nm)

10

Fig. 11. Optical spectrum of a) 1310 nm 100G PSM4 transceiver and b) 1550 nm 100G PSM4 transceiver.

(b)

1550.0

1550.5

1551.0

Wavelength (nm)

-4

10

-5

10

-6

(b)

Fig. 12. The BER vs. received optical power curves for a) 1300 nm 100G PSM4 transceiver and b) 1550 nm 100G PSM4 transceiver.

-7

10 -8 10 -9 10 -10 10-11 10-12 10-13 10

B2B 2km

-11

-10

-9

-8

-7

-6

Received Optical Power (dBm)

-5

We also obtained the optical eye diagrams for both transceivers at B2B and with 2 km of UF as shown in Fig. 13. The left side shows the results from the 1310 nm transceiver and the right side shows the results from the1550 nm transceiver. The quality of the optical eye diagrams is consistent with those obtained from received optical power vs. BER curves. The optical eyes from 1310 nm transmission have good quality, with little difference between the B2B condition and with 2 km UF in the link. For 1550 nm, the optical eye quality is not as good as at 1310 nm, but the eyes remain open with 2 km UF in place. In addition to the PSM type of SM transmission, we have demonstrated 2.7 km UF transmission using 100G CWDM transceiver in [11]. In general, the UF has a robust SM transmission performance, with the

optical power below -11 dBm, and the performance of the 2 km UF is very close to that of B2B condition. For 1550 nm and the B2B condition, the system reaches bit error free around -6.75 dBm while for 2 km of UF, the error free optical power is around -6 dBm. There is a slight power penalty due to the use of 2 km UF at 1550 nm. We believe this is due to the fact that at 1550 nm the higher chromatic dispersion of the fiber induces an additional impairment, but at 1310 nm there is minimal penalty from chromatic dispersion of the fiber, as its value is nearly zero. In addition to BER testing, we have also used ONT for the testing, the system using both 1310 nm and 1550 nm 100G PSM4 transceivers, and we found that 2 km UF can perform error free for more than ten hours. 7

Optical Fiber Technology xxx (xxxx) xxx–xxx

X. Chen et al.

1310nm Transceiver

1550nm Transceiver

B2B

B2B

2km

2km

only penalty being the small coupling loss and MPI, as reported in Section 2.2.

Fig. 13. Optical eye diagrams obtained at B2B condition and with 2000 m UF for 1310 nm 100G PSM4 transceiver on the left and 1550 nm 100G PSM4 transceiver on the right side.

[3] T. Kise, T. Suzuki, M. Funabashi, K. Nagashima, R. Lingle, D.S. Vaidya, R. Shubochkin, J.T. Kamino, X. Chen, S.R. Bickham, J.E. Hurley, M.-J. Li, and A.F. Evans, Development of 1060nm 25 -Gb/s VCSEL and demonstration of 300m and 500m system reach using MMFs and link optimized for 1060nm, OFC2014, paper Th4G.3, 2014. [4] D. Coleman, Optical Trends in the Data Center, in: Presented at BICSI Canadian conference, May 3, 2016. https://www.bicsi.org/uploadedfiles/bicsi_conferences/ canada/2016/presentations/GS_TUE_2.pdf. [5] FICON is the IBM proprietary name for the ANSI FC-SB-3 Single-Byte Command Code Sets-3 Mapping Protocol for Fibre Channel (FC) protocol. [6] D.H. Sim, Y. Takushima, Y.C. Chung, Transmission of 10-Gb/s and 40-Gb/s signals over 3.7 km of multimode fiber using mode-field matched center launching technique, paper OTuL3, OFC Technical Digest, 2007. [7] D.H. Sim, Y. Takushima, Y.C. Chung, High speed multimode fiber transmission by using mode-field matched center-launching technique, J. Lightwave Technol. 27 (8) (2009) 1018–1026. [8] W.V. Sorin, M.R. Tan, Interoperability of Single-Mode and Multimode Data Links for Data Center and Optical Backplane, paper OW1B.6, OFC Technical Digest, 2007. [9] W.V. Sorin, M.R. Tan, Converting a multimode fiber into a single-mode fiber, in: Photonics Society Summer Topical Meeting Series, 2013 IEEE, 256–257. [10] X. Chen, J.E. Hurley, J. Stone, J.D. Downie, I. Roudas, D. Coleman, M.-J. Li, Universal fiber for both short-reach VCSEL transmission at 850 nm and single-mode transmission at 1310 nm, Paper Th4E.4, OFC Technical Digest, 2016. [11] Xin. Chen, Jason E. Hurley, Jeffery S. Stone, Aramais R. Zakharian, Doug. Coleman, Mi.-Jun. Li, Design of universal fiber with demonstration of full system reaches over 100G SR4, 40G sWDM, and 100G CWDM4 transceivers, Opt. Express 24 (2016) 18492. [12] A. Fall, E. Le Cren, K. Lengle, C. Lepers, Y. Gottesman, M. Thual, L. Bramerie, D. Molin, P. Sansonetti, D. Van Ras, M. Gadonna, C. Populaire, G. Martin, L. Valencia, P. Guignard, Versatile graded-index multi-mode fiber for high capacity single- and multi-mode optical home network, ECOC’14, paper Th.1.4.6. [13] Bruce Chow, The future of data center optical interconnects: SM, MM, or both, presented at OFC 2017 workshop “Scaling Datacenter Bandwidth: Novel Optics, Advanced Electronics or New Architectures?” organized by Piero Gambini, MingJun Li, and Ilya Lyubomirsky. [14] Doug Coleman, Optical Trends in the Data Center, in: BICSI Conference presentation, Canada, May 2016. [15] A.W. Snyder, J.D. Love, Optical Waveguide Theory, Kluwer Academic Publishers, 2000. [16] http://www.cwdm4-msa.org/. [17] http://www.ardenphotonics.com/products/modcon-mode-controller-telecom/. [18] http://www.viavisolutions.com/en-us/products/ont-100g-test-solution-supporting40ge-100ge-otu4. [19] J.A. Tatum, D. Gazula, L.A. Graham, J.K. Guenter, R.H. Johnson, J. King, C. Kocot, G.D. Landry, I. Lyubomirsky, A.N. MacInnes, E.M. Shaw, K. Balemarthy, R. Shubochkin, D. Vaidya, M. Yan, F. Tang, VCSEL-based interconnects for current and future data centers, J. Lightwave Technol. 33 (4) (2015) 727–732. [20] http://investor.finisar.com/releasedetail.cfm?releaseid=933294. [21] http://psm4.org/.

4. Conclusions We have reviewed recent progress on UF and presented a set of new analysis and experimental results to understand the UF transmission properties, as well as new system performance characterization covering several new 100G QSFP applications for DC. In addition to the dependence of the MFD of LP01 mode on fiber parameters and the insertion loss from previous work in Ref. [11], we studied the coupling loss from UF to SMF and investigated the MPI penalty for SM transmission. We found that for SM level of alignment precision, the coupling loss is contained to less than 0.6 dB and the maximum MPI is less than ∼0.3 dB. We made an improved UF with higher modal bandwidth, 1.2% core delta and 31 μm core diameter. We presented updated and new system testing results on three types of MM transmissions, i.e. 100G SR4, 100G SWDM4, and 100G BiDi, as well as SM transmission at both 1310 nm and 1550 nm wavelengths using 100G PSM4 transceivers. In conclusion, the UF is a new fiber concept and through a series of experiments, we observed that UF performs well over a wide range of 100G transmission options, which suggests that the UF could be a potential viable transmission medium for current and future DC applications. Acknowledgement We thank Foxconn Interconnect Technology (FIT) for providing 100G eSR4 and 100G BiDi transceivers, Finisar for providing an early version of 100G SWDM and 100G PSM4 transceivers operating at 1310 nm, and Mellanox for providing 100G PSM4 transceiver operating at 1550 nm. References [1] D. Coleman, Optical trends in the data center, ICT TODAY 36 (5) (2015) 16–22. [2] X. Chen, S.R. Bickham, H.-F. Liu, O.I. Dosunmu, J.E. Hurley, M.-J. Li, 25 Gb/s Transmission over 820 m of MMF using a multimode launch from an integrated silicon photonics transceiver, ECOC2013, postdeadline paper pd-4-f-5, 2013.

8