1. Introduction

High-bandwidth modulators are essential components for various application such as optical interconnects, microwave photonics and optical frequency comb generation [1–4]. They are particularly relevant for short-reach optical interconnects (<2 km reach) that further require low costs, low power consumption, and small footprints [5].

Optical interconnects featuring data rates of 400 Gbit/s are expected in the near future [6, 7]. In general, there are two ways to scale data rates in such systems. The first being to aggregate multiple channels at lower speed to create a virtual higher speed link. The second approach is to increase capacity by scaling the data rate on the individual physical channel through higher symbol rates and advanced modulation formats such as four-level pulse-amplitude modulation (PAM-4). To go beyond, with total link capacities exceeding 1 Tbit/s, per-channel data rates need to be increased to >100 Gbit/s and low cost implementations will be a major challenge for the datacom industry [5]. Traditionally, vertical cavity surface emitting lasers (VCSELs) have been adopted for short-reach communications [8]. Yet, due to the bandwidth limitations of VCSELs, higher data rate links can only be implemented with more power hungry signal processing/equalization as well as forward error correction (FEC) along with associated latency [5, 9]. Conversely, external modulator solutions exhibit a higher bandwidth and can be scaled to higher data rates with less effort. Promising high-bandwidth modulators on the silicon platform are based on the free-carrier dispersion effect, the electroabsorption or electro-optic effect in graphene and the linear electro-optic effect in organic materials. Among the modulators relying on the free-carrier dispersion, those based on reverse biased pn junctions as well as metal-oxide semiconductor (MOS) structures are particularly attractive [10–14]. Typically, these devices require lengths of several hundreds of micrometers [15] and feature rather large bandwidths. For example a bandwidth of 35 GHz has been demonstrated in a reversed biased pn junction and even 55 GHz are predicted for a device with a voltage-length product of 2.4 Vmm [11]. Similar, MOS modulators with a bandwidth of 25 GHz have been achieved, but with a voltage-length product of 5 Vmm [13]. Moreover, modulators based on graphene have attracted attention [16–18] and only recently, broadband operation of 35 GHz has been achieved with a 30 µm long electroabsoprtion modulator [16]. Still, the modulation depth was only 2 dB at a voltage swing of 25 V. Alternative approaches in photonic and plasmonic configurations relying on organic electro-optic materials have already shown bandwidth exceeding 100 GHz [1, 2, 19–21]. Among these modulators the plasmonic approach is of particular interest as it has not only shown highest bandwidths but also largest light-matter interactions with voltage-length products of only 0.06 Vmm [22]. With plasmonic devices, even higher bandwidths are anticipated, since limiting RC time constants are minimized due to low resistance electrical contacts and smallest dimensions [20, 22]. Yet, experimental proof supporting the claim for largest bandwidths is still missing.

In this work, we report for the first time a plasmonic modulator showing an operation bandwidth beyond 170 GHz. The modulator is based on an organic electro-optic material (DLD164 [23]). The generation of the electrical drive signal at such high frequencies is made possible by exploiting the higher harmonics produced by a Schottky diode. The response was measured with an optical spectrum analyzer. The applicability of the plasmonic modulator for short-reach optical interconnects was then confirmed in a high-speed data modulation experiment. We use a direct detection scheme that allows for a technically simple low-cost implementation. The electrical drive signals were generated using a high-speed digital to analog converter (DAC) with a sampling rate of 100 GSa/s [24]. We show modulation at 100 GBd NRZ with a bit-error-ratio (BER) <1.7· 10⁻⁵ and 60 GBd PAM-4 (BER: 9.2 · 10⁻³). This is exceeding performance of previous experiments in which we showed 40 Gbit/s with direct detection and a BER 6 · 10⁻⁴ [25] as well as 54 GBd PAM-4 with coherent detection and a BER of 1.8 · 10⁻² [26]. This is the first example of a plasmonic modulator operating at 100 GBd NRZ.

2. Plasmonic Modulator

The plasmonic modulator consists of an imbalanced silicon photonic Mach-Zehnder interferometer (MZI) with plasmonic phase modulators embedded in each arm (see Fig. 1). The phase modulators [27] exploit the Pockels effect in an organic EO material (DLD164 [23]) to modulate the phase of surface plasmon polaritons (SPPs). The devices are similar to a previously published design and were fabricated in-house [26, 28].

 

Fig. 1 Colorized optical microscopy image of the imbalanced silicon photonic MZI with MMIs to split and combine the light. Each arm of the MZI contains a plasmonic phase modulator [27]. The zoom-in shows a colorized scanning electron microscopy image of a typical plasmonic phase modulator. The phase modulator consists of two gold electrodes separated by a 75 nm wide and 20 µm long slot. The slot is filled with an organic EO optical material (not shown).

Download Full Size | PDF

Light is coupled to and from the MZI via silicon photonic strip waveguides (450 nm x 220 nm) with grating couplers at each end. Silicon waveguide bends of 90° and with a radius of 20 µm were applied for routing. The light is split to the two arms of the MZI by a multimode interference (MMI) coupler (7.8 µm x 3.01 µm). In each arm of the MZI the light experiences a phase shift proportional to the applied electrical driving field. After that, it is combined in a second MMI which translates the phase modulation to an intensity modulation. The arms of the MZI have an imbalance of 80 µm to set the operation point by sweeping the laser wavelength.

The plasmonic phase modulator is based on a metal-insulator-metal waveguide consisting of two gold electrodes (150 nm height) separated by a narrow slot. The slot is filled with the EO material (inset of Fig. 1). The active section of the plasmonic modulator used for the frequency response measurement was 12.5 µm long with a 100 nm wide slot. For the data experiment we used a more recent device generation of 20 µm length and with a 75 nm wide plasmonic slot. To convert between the photonic and plasmonic mode a tapering section is employed [29]. The electrical signal is applied to the electrodes via RF probes. The contact pads are arranged in a ground-signal-ground (GSG) configuration. The total modulator footprint including contact pads is 287 µm x 270 µm. Note that the overall size is only limited by the contact pads needed to address the device with RF probes.

3. Frequency Response

The frequency response of the plasmonic modulator was measured from 75 MHz to 170 GHz. To this end, we modulated CW laser light at ~1550 nm with sinusoidal RF signals of various frequencies $f_m$. In order to cover the aforementioned spectrum, five different setups were required to generate the RF signals for different frequency ranges (75 MHz…12 GHz, 15…70 GHz, 70…95 GHz, 115…140 GHz, 141...170 GHz).

In the lowest frequency range (75 MHz…12 GHz), we used an HP71400C lightwave signal analyzer. The lightwave signal analyzer combines an RF spectrum analyzer with a calibrated optical receiver. The RF signals $f_m$ were swept in steps of 15 MHz to determine the EO transfer function of the plasmonic modulator.

For the frequency range from 15 GHz to 170 GHz, the modulated signal was recorded with an optical spectrum analyzer, while sweeping the frequencies in steps of 1 GHz. We determined the modulation amplitude by the power ratio between the optical carrier $f_c\gg f_m$ and the modulation sidebands around the carrier at frequencies $f_c+f_m$ and$f_c-f_m$. The RF signal in the mentioned range was generated in several steps. For the frequency range 15…70 GHz, an RF synthesizer provided the signal directly. For higher frequencies, the nonlinearities of a Schottky diode (Virginia Diodes) were exploited to produce higher harmonics. Subsequently, a hollow waveguide filtered undesired harmonics. For the range 70…95 GHz, the frequencies were multiplied by a factor of six. In the frequency range 95…114 GHz, the unwanted harmonics could not be suppressed in our setup, which prohibited a proper calibration. For the range 115…170 GHz, we used a frequency tripler. Since the undesired harmonics were separated by a larger frequency gap, they could be filtered out efficiently. This way, the frequency bands 115…140 GHz and 141…170 GHz were addressed.

Due to the extremely broad frequency range the microwave probe setup needed to be adapted for the different frequency bands. Coaxial cables and waveguide inputs were chosen accordingly. Further, an RF amplifier was used. To keep the power at the device constant within each frequency band, we calibrated the output power of the synthesizer by determining the power at the input of the RF probe with a broadband power meter. Power levels were set to −3 dBm, −3 dBm, −1 dBm, and 0 dBm for the frequency bands 15…70 GHz, 70…95 GHz, 115…140 GHz, and 141…170 GHz, respectively. The optical spectrum for a RF frequency $f_m$ = 170 GHz is displayed in Fig. 2 as an example.

 

Fig. 2 Optical spectrum for an RF frequency of 170 GHz.

Download Full Size | PDF

After the measurement, we subtracted the losses of the RF probes given in the datasheets. Subtracting the RF probe losses is permitted, since the modulation depends linearly on the applied RF power. For 115…170 GHz there was no datasheet available. Here, the RF probe was assumed constant across the measurement range. In a last step, the determined modulation amplitude was normalized to the mean of all data points of each frequency range.

The measured frequency response is shown in Fig. 3. There is no indication for a bandwidth limitation over the whole frequency range. As discussed in [20], an even larger bandwidth is predicted in theory. The frequency response measured here is in agreement with our published results across different frequency bands (below 110 GHz) with multiple devices from various fabrication batches [20, 22, 26, 28]. The device here can thus be considered to be representative for the plasmonic organic hybrid modulators.

 

Fig. 3 Measured electro-optic response up to 170 GHz.

Download Full Size | PDF

4. Data Modulation Experiment

To investigate the application of plasmonic modulators in optical interconnects, we performed a 2-ary and a 4-ary pulse amplitude modulation (PAM) experiment. Figure 4 shows a block diagram of the experimental setup.

 

Fig. 4 Experimental setup of the data modulation experiment. The electrical signal was generated using Micram’s DAC4 prototype. The modulated signal was received in a direct detection scheme by a photodiode and a real-time scope. At the receiver, post-equalization was applied to compensate for frequency limitations of the electronics. In the upper right hand corner we show eye diagrams of the electrical signal at 100 GBd NRZ measured with a digital sampling oscilloscope: (a) after the DAC and (b) after the RF amplifier.

Download Full Size | PDF

The electrical drive signal was generated with offline digital signal processing (DSP) and a high-speed DAC. No pre-emphasis was used in the transmitter DSP. The digital waveform was stored in the internal memory of a Micram VEGA DAC4 prototype [24] that generated the data signals. The DAC4 is a 6 bit digital-to-analog converter with a sampling rate of up to 100 GSa/s and a 3 dB bandwidth of >40 GHz. An RF amplifier with 55 GHz bandwidth amplified the signal to a peak-to-peak voltage of 4 V_pp for 50 Ω termination. The signal was applied to the plasmonic modulator with an RF probe having >67 GHz bandwidth.

Light from a tunable laser source at ~1540 nm wavelength was amplified before coupling to the chip by an erbium-doped fiber amplifier (EDFA) to 20 dBm and 25 dBm for PAM-2 and PAM-4, respectively. Such high powers were necessary to compensate for the high losses of the non-optimized silicon grating couplers and feeding waveguides (26 dB). Note that the insertion loss of the plasmonic MZ modulator itself was only ~8 dB. However, cut-back measurements of the silicon photonic devices indicate fiber-to-chip coupling losses of 11.8 dB per grating. The silicon photonic propagation loss for the 0.102 cm long feeding waveguides is ~2.7 dB. In fact, silicon photonic losses in this work were rather high and are attributed to fabrication imperfections.

The modulated light was detected in a direct detection receiver. The direct detection receiver consisted of an EDFA followed by a 2 nm pass-band filter and a 70 GHz PIN photodiode. A real-time oscilloscope (160 GSa/s, 63 GHz bandwidth) digitized the signal for offline DSP. In the offline DSP, we resampled the signal to two samples per symbol and performed timing recovery. We applied a feed forward equalizer with 61 taps to compensate for the frequency responses of the electronics. Finally we made a symbol decision and counted the BER.

In the first step, we generated a 100 GBd NRZ PAM-2 signal with a de Brujin bit sequence (DBBS) having a pattern length of 2¹⁵. To investigate the quality of the electrical test signal at the transmitter we connected a digital sampling oscilloscope to the output of the DAC, see Fig. 4(a), and to the output of the RF amplifier, see Fig. 4(b). Though the eye diagram after the DAC is clearly open, the electrical signal quality is degenerated after the RF amplifier due to bandwidth limitations of the electronics. These bandwidth limitations are compensated by a feed forward equalizer (FFE) in the receiver DSP. Figure 5(a) shows the 100 GBd NRZ eye diagram after direct detection and equalization. The obtained symbols were resampled with a raised cosine pulse-shape (roll-off factor α = 1) to model the low-pass characteristic of the system. Error free operation was obtained with 6 × 10⁶ evaluated bits (BER <1.7 × 10⁻⁵).

 

Fig. 5 Eye diagrams and BERs measured at (a) 100 GBd NRZ, (b) 50 GBd PAM-4 (line rate 100 Gbit/s) and (c) 60 GBd PAM-4 (line rate 120 Gbit/s). The eye diagrams were resampled after equalization with a raised cosine shape that resembles the low-pass characteristic of the system.

Download Full Size | PDF

In the second step, we investigated multi-level signals. We sent a 50 GBd PAM-4 signal with NRZ pulse shape and a DBBS of pattern length 2¹¹. This corresponds to a line rate of 100 Gbit/s. Furthermore, we sent a 60 GBd PAM-4 signal (120 Gbit/s) with a square-root-raised cosine pulse shape (roll off α = 0.05, DBBS with pattern length 2¹¹). The eye diagrams of the detected signals are shown in Figs. 5(b) and 5(c). The measured BER for the 50 GBd signal was 1.6 · 10⁻³. At 60 GBd we achieved a BER of 9.2 · 10⁻³. This effectively corresponds to a net data rate of 105 Gbit/s assuming FEC with 14.3% overhead [30]. The BER limitations were mainly due to the high coupling losses (26 dB) that led to fairly low power at the output of the modulator.

In the future, the BER may be improved by reducing the insertion losses that are mainly introduced by the non-optimized couplers. This would further allow for longer plasmonic waveguides which could increase the interaction length, so that driving voltages may go down.

5. Conclusion

We have shown operation of a plasmonic modulator with a flat electro-optic response from 75 MHz to 170 GHz. The plasmonic phase modulators are arranged in a silicon photonic MZI. They employ an organic EO material and are as short as 12.5 µm and 20 µm. Data experiments with 100 GBd NRZ and 60 GBd PAM-4 demonstrate the applicability of plasmonic modulators for next-generation optical interconnects.