1. Introduction

Carrier-depletion silicon photonic microresonator modulators are used as compact, energy-efficient electro-optic modulators for data communications, optical switching, and RF photonics [1–5]. One limitation of such devices at present is their optical power-handling capacity. Integrated semiconductor lasers can now efficiently generate and deliver more power in the bus waveguide than the sub-milliwatts of optical power that are typically used with silicon microresonator modulators [6,7]. Such devices lose the ability to adequately modulate light at higher power levels because of a combination of effects including two-photon absorption (TPA), free-carrier generation (FCA), and a thermo-optic shift of the refractive index [8,9]. Impaired performance at higher power levels under-utilizes critical resources in an optical communication system and decreases the link transmission distance.

For the popular NRZ (non-return to zero) modulation format, the OMA (Optical Modulation Amplitude) is defined as the difference (P1-P0) between the average power level for the 1 and 0 bits. Sometimes, the normalized OMA is reported as (P1-P0)/P_in where P1 is the average high bit value, P0 is the average low bit value, and P_in is the input power before the modulator. OMA is a key parameter used in short-distance serial communications [10,11], and is used as the figure-of-merit in this report. OMA is related to other quantities such as bit error rate (BER) and optical extinction ratio (ER) which are used to describe communication systems [12,13].

As the input optical power level to a silicon microresonator modulator increases, the OMA first increases, and then rapidly decreases because of the inability of the microresonator to modulate high optical powers. As shown in Fig. 1, the output light is modulated by the resonator because of the voltage-driven interference between the fraction of the input which does not couple into the resonator, and the fraction of the input which does couple into the resonator and experience an amplitude and phase shift before effectively coupling back out into the bus waveguide. At higher optical power, further amplitude and phase changes caused by nonlinear effects giving rise to optical waveform distortions have to be balanced by changing the fraction of input light that is coupled into it, i.e., the coupling coefficient or, equivalently, the bias condition. Note that in the devices studied here, the optical intensity in the resonator is significantly larger than in the bus waveguide, since the value of finesse (F, defined as the ratio of the free spectral range to the full width at half maximum) is as high as 245 at low power and 186 at the higher power level associated with the highest measured OMA. Thus, we still operate in a regime where resonant effects are significant, and substantially (by approximately a factor of F/2) lower the electrical power consumption of a microresonator modulator relative to a traveling-wave Mach-Zehnder modulator, as has been studied extensively elsewhere [14].

 

Fig. 1. (a) Schematic of a microdisc modulator coupled to two bus waveguides. (b)-(e) Transmission spectrum as the on-chip optical power is increased. The blue and red lines show the measurements when scanning the laser in opposite directions. (f) 20 Gbit/s eye diagram measured on an optical sampling oscilloscope; y-axis is the optical power in the bus waveguide after the modulator.

Download Full Size | PDF

We experimentally study the trends observed at a data rate of 20 Gbit/s NRZ format digital modulation using a silicon photonic microdisc modulator operated near 1550 nm. We compare the trends at high data rates with slower 0.2 Gbit/s data rates, covering the different time scales with respect to the carrier dynamics in silicon [15]. An OMA of 4.1 mW was obtained at an optical power level of 13.5 mW (11 dBm) without using pre-emphasis or frequency shaping on the electrical driving waveform. Our results demonstrate a much higher OMA for electro-optic data modulation using silicon microdisc resonators at 1550 nm, which are usually operated with lower optical power levels (see Table 1). Given a certain data modulation rate and detection sensitivity (minimum detector power required to achieve a certain bit error rate), a higher OMA directly translates into increased distance between the source and the destination. This can be critically helpful in improving the reach short-distance communications without using optical amplifiers or repeaters, or providing additional link margin for incorporation of optical switches or other devices. A small increase of the OMA from 0.3 mW [16] to 0.38 mW, for example, can result in increasing the transmission distance by 5 km before dispersion impairments, based on a propagation loss coefficient in fiber of 0.2 dB/km, and a free space optical link can also benefit in a similar way. If the need for optical amplifiers can be avoided, the bandwidth span of wavelength-division multiplexed (WDM) communications can be increased. This is a problem for short-distance optical communications such as unamplified ultrawideband WDM links which anticipate using a wider spectral range than the gain bandwidth of amplifiers [17]. The results shown here on increased OMA for high-speed silicon photonic microdisc modulators may benefit unamplified wideband WDM optical networks, among other applications.

Table 1. Optical Modulation Amplitude of resonant silicon photonic modulators

View Table | View all tables in this article

2. Experimental details

The modulator is a vertical PN junction microdisc, operated in carrier depletion mode under reverse bias, similar to [18], as pictured in Fig. 1(a). The modulator was partially doped so that dopants covered approximately one-half of the disc circumference. The microdisc diameter is 4 µm, giving a free spectral range (FSR) of about 59 nm, so that there is only one resonance in the C-band (and only one resonance in the L-band). Dopant levels were selected with sufficient bandwidth for 25 Gbit/s NRZ modulation, with the only difference among different test structures being the waveguide-to-resonator gap. The waveguide-to-resonator gaps were designed to be 280 nm for one design, while another design used 260 nm for a slightly larger waveguide-resonator coupling coefficient. The silicon photonic fabrication was performed in a multi-project wafer process at Sandia National Laboratories [19]. The V_πL of our vertical PN was measured to be 1.25 V-cm, extracted from the resonance shift relative to the FSR, in a method similar to the one performed in [20].

Figures 1(b)-(e) show optical transmission spectrums measured with high resolution for the disc with a coupling gap of 280 nm, from low optical powers where resonator spectrum asymmetry is absent (panel b), to high optical powers where there is strong nonlinearity and a large bistable regime (panel e). Based on the measured transmission at low powers, loaded quality factors of 6200 were inferred. To generate the plots in Fig. 1(c)-(e), a continuous wave (C.W.) laser was stepped from lower wavelength to higher wavelength at a tuning speed of 0.25 s per step (blue), and then stepped in reverse from higher wavelength to lower wavelength at the same speed (red).

For high-speed OMA measurements, the tunable C.W. laser was used again as the source. For on-chip optical powers (in the bus waveguide) less than -2 dBm, no external optical amplifier was used. A polarization maintaining erbium doped fiber amplifier (EDFA) was used to achieve on-chip optical powers greater than -2 dBm. Polarization-maintaining fiber arrays were used to couple light to the chip. At low powers, the total (fiber to fiber) insertion losses through the chip was measured to be 9.8 dB. A high-speed arbitrary waveform generator (Tektronix AWG70000 series) was used to generate the pseudo-random bit sequence (PRBS7) pattern which drives the modulator. The electrical signal from the AWG was amplified to 3.2 Vpp using a high-bandwidth RF amplifier (Keysight N4985A 50 GHz Microwave Amplifier), and a high-speed bias T was used to apply a reverse bias of -3.25 V. At the output, a fiber-coupled C band filter with bandwidth about 100 GHz was used to filter out the ASE noise before detection.

Optical transmission was measured by capturing the optical trace on the oscilloscope and analyzing the waveform using software (MATLAB). A section of a representative trace is shown in Fig. 2 (c), (f), (i), and (l). Since the optical oscilloscope was not equipped with a bit-error rate tester, the conventional sampling mode cannot determine when a bit error occurs, and so cannot distinguish a high “1” bit from a “0” that has been distorted to a high optical power level. Therefore, we post-processed the acquired data in MATLAB, by folding over the traces for over 1000 bit periods, such as shown in Fig. 2(c), and also obtaining an eye diagram as in Fig. 2(e). As a check, these post-processed eye diagrams were seen to be similar to those captured using the conventional sampling mode of the oscilloscope (Fig. 1(f) for 20 Gbit/s NRZ modulation).

 

Fig. 2. (a) OMA vs. optical power in the input bus waveguide, at 20 Gbit/s NRZ modulation while stepping the laser from lower wavelengths to higher wavelengths (blue points and line), and higher wavelength to lower wavelength (red markers and line). The labels “c”, “f”, “i”, and “l” indicate points which refer to subplots with the same label. (b) Wavelength at which the optimal OMA was achieved. (c), (f), (i), (l) A portion of the time-domain waveform corresponding to the points labeled ‘c’, ‘f’, ‘i’ and ‘l’, respectively, in panel (a), with 6.3 mW bus waveguide input power for (c) and (f), and 1.7 mW bus waveguide input power for (i) and (l). (d), (g), (j), (m) A binary-mode eye diagram was constructed by folding and overlaying the traces in panels (c), (f), (i) and (l), respectively, across each bit period. The regions shown in blue are the sampling points from which the OMA was calculated. (e), (h), (k), (n) Histograms for the logical “1” and logical “0” bits at the sampled points which used to calculate the OMA.

Download Full Size | PDF

2.1 Calculation of on-chip OMA

To find the OMA of the disc modulator, the optical power level in the bus waveguide after the device was calculated. The insertion loss from the output coupling, fibers (1.26 dB), and filter (1.34 dB) was subtracted from the raw measurements to give the OMA values shown in Fig. 2. The oscilloscope data from each eye diagram was processed in MATLAB to sample the center of the eye, i.e., collecting all the datapoints shown in blue in Fig. 2 (d), (g), (j), and (m). To determine if the correct sequence of “1” and “0” bits was detected, a pattern match to the PRBS7 waveform was performed in MATLAB. The sampled measurement points were assigned to their respective bit as logical “1”s or “0”s. All sampled optical power levels corresponding to a logical “1” were histogrammed and contributed to the “1” level power used to calculate the OMA, and all sampled optical power levels corresponding to a logical “0” contributed to the “0” level power for OMA calculation, as shown in Fig. 2 (e), (h), (k), and (n). An average was taken for each histogram and the OMA was calculated by subtracting the mean optical power value of the logical “0” bits from the mean optical power value of the logical “1” bits.

This process was repeated for each input optical power level in the range of 0.2 mW to 15.9 mW, and at each wavelength of the laser scans. To find the best OMA that could be achieved at a given optical input power, the tunable laser was scanned from a wavelength that was a few nanometers lower than the cold-cavity resonance wavelength of the disc by increments of 0.1 nm until near resonance, and then incremented by 0.025nm until the laser wavelength is past the resonance to a few nanometers further. For each step, the ASE noise filter was programmed to follow the wavelength of the laser. The laser wavelength was then stepped backwards (from high wavelengths to lower wavelengths) to the starting point, and the oscilloscope data was saved and processed. The highest OMA across the wavelength scans was chosen for the plot shown in Fig. 2(a). At the same time, the biasing wavelength was plotted in Fig. 2(b). We verified that waveforms where the logical “0”s (or, equivalently, “1”s) are distorted to the point that they cross the middle threshold of the eye and would be incorrectly sampled as “1”s are not used to calculate the optimal OMAs, since these waveforms cannot contribute to error-free communications [20].

3. Results and discussion

3.1 OMA comparison between low-speed and high-speed NRZ data modulation

From Fig. 2(a), we see that the OMA increases with increasing optical power in a sub-linear fashion, up to a certain maximum value, and then decreases. At lower data rates (0.5 Gbit/s), such behavior has been attributed to the combined effects of two-photon absorption, free carrier absorption, and free carrier dispersion growing nonlinearly stronger at increasing optical power [15] and similar effects are also present at 20 Gbit/s but are manifest differently due to the different relative time scales of the data modulation and free carrier dynamics. To compare lower speed dynamics with 20 Gbit/s behavior, we performed a slower speed experiment to extract the OMA behavior versus optical power at data rates of 200 Mbit/s. Interestingly, the OMA of the slower 200 Mbit/s data rate was limited to below 1 mW. Typical free carrier lifetimes in silicon waveguides can be on the order of hundreds of picoseconds to a few nanoseconds whereas the thermal time constants are usually on the order of a few hundred nanoseconds to microseconds [21,22]. At the higher data rates, the effects of nonlinearities at higher power are pattern-dependent, and the distortions from ideal “1” and “0” levels will become most evident for a long string of bits. In principle, this can be mitigated by run-length encoding, although no coding was performed in the experiments reported here. At the slower data rates, the distortions from bit to bit are themselves stronger and cannot be mitigated by coding.

To demonstrate this difference in behavior of eye closure due to high optical power at 20 Gbit/s versus 200 Mbit/s, we plot the time domain waveforms in Fig. 3 (c), (f), (i), and (l) for the identical bit sequence over a longer span of bits. For 20 Gbit/s, Fig. 3 (f) shows the onset of eye closure, and the difference between it and Fig. 3 (c) (an optimal OMA point) is clearly due to pattern dependence, over a time span of a few nanoseconds, denoted by Δf1. In contrast to 20 Gbit/s data, Fig. 3 (l) shows that the onset of eye closure for 200 Mbit/s data is very different for each bit, but also on the order of a few nanoseconds, denoted by Δl1.

 

Fig. 3. (a) OMA vs. optical power in the input bus waveguide, at 20 Gbit/s NRZ and 200 Mbit/s NRZ; the markers “c”, and “i”, correspond to panels (c) and (i), respectively. “f” and “l” represent measurements at wavelength biases which eye closure would result, due to “1” (“0”) bits being incorrectly sampled as “0” (“1”) for every PRBS7 sequence. They do not contribute to the error-free OMA, and are plotted in panels (f) and (l) to show the onset of eye closure. (b) Wavelength at which the optimal OMA was achieved. (c), (f) A portion of the time-domain waveform at the optimal bias wavelength when the optical power in the input bus waveguide was 9.7 mW [panel (c)] and 15.4 mW [panel(f)]. (i), (l) Plots similar to panels (c) and (f), for 200 Mbit/s modulation, at input power levels of 1.7 mW [panel(i)] and 6.3 mW [panel(l)]. (d), (g), (j), (m) A binary-mode eye diagram was constructed by folding and overlaying the traces in panels (c), (f), (i) and (l), respectively, across each bit period. The regions shown in blue are the sampling points from which the OMA was calculated. (e), (h), (k), (n) Histograms for the logical “1” and logical “0” bits at the sampled points which used to calculate the OMA. Different colors are used to show the overlap, if it occurs.

Download Full Size | PDF

Figures 3(a) and 3(b) summarize the major differences between higher data rate 20 Gbit/s waveforms and slower 200 Mbit/s with respect to increasing optical power. The 200 Mbit/s waveforms do not reach the higher OMA values achievable at 20 Gbit/s modulation. Because each bit and shorter (in lengths of bits) bit sequences of the slower waveforms distort strongly compared to the change in average power, one must detune the bias further (in the direction of decreasing wavelength) with respect to the shift in resonance wavelength caused by the increasing average power. This was observed in the models from [15], and is shown here in Fig. 3 (b): the black line denoting the optimal wavelength is not monotonic and decreases at the point where the OMA also starts to decrease. However, for the 20 Gbit/s modulation, the differences between unique segments of longer runs of pattern are smaller, and longer pattern difference cause eye closure. The difference between average power from distortion causing pattern segments is also less. The bias point does not need to be detuned as far to get to the optimal OMA, and so the optimal bias wavelength keeps increasing with the power as shown by the blue line in Fig. 3(b).

3.2 Biasing

To summarize the observations made in the previous section when discussing the OMA trends, at each optical power level in the nonlinear regime, the ideal biasing point of a microdisc modulator is (slightly) different in order to achieve the highest OMA. Figure 2(b) shows this behavior using a shift of the wavelength of the input laser. While keeping the modulator’s on-chip thermal shifter at a constant voltage and temperature, and keeping the chip mount thermoelectric-cooler (TEC) at a constant temperature, the laser was stepped from a wavelength shorter than the cold-cavity resonant wavelength to a wavelength longer than the resonant wavelength, and then stepped backward.

Figure 2(b) shows that the optimal biasing point for the highest OMA changes from 1549.5 nm to 1556.7 nm across a span of below 1 mW to near 10 mW of optical power in the input bus waveguide and grows to 1558.8 nm before the highest OMA is reached, a span of nearly 10 nm. The biasing trends for the slower data rate, Fig. 3 (b), are different. There is a decrease in optimal biasing wavelength when the peak OMA is reached and then starts to decrease with increased power, due to the response times of the nonlinearities giving rise to the distortions as noted in the prior section. In these cases, the laser needs to be detuned further away (i.e., remain at lower wavelength and not bias closer to resonance) from the resonant frequency since the impact of the single bits and shorter bit sequences on eye-closing distortion is large, while the slower thermal shift towards longer wavelength continues. This may result in an inability to reach the wavelength at which the strongest resonant peaking in the modulation response occurs (i.e. ${\omega _m} = {\omega _r} - {\omega _0}$, where ${\omega _m},\; \; {\omega _r},\; \; {\omega _0}$ are the modulation, resonator, and laser bias frequencies, respectively [23]). On the other hand, for the higher data rate modulation, one does not need to detune as much to reach the optimal OMA, since eye closure results from the cumulative effects of a much longer string of bits, which, by pseudorandom nature, has less variation from sequence to sequence. In other words, the timescale of change in optical power due to power dependent nonlinearities is much slower than timescale of bit “0” to bit “1” differences, so the bias wavelengths, at which the eye closes, tracks the resonance wavelength shifts due to increasing power more closely.

3.3 Direction of bias scan

When searching for the optimum bias point in the nonlinear regime, it is important to note that shifts to higher and lower wavelengths (or equivalently, to lower or higher reverse-bias voltages) can have asymmetric effects. Figure 2(c)-(f) shows scans in wavelength from a lower wavelength to a higher wavelength), followed by a reverse-direction scan. These plots show the extent of wavelength tuning required to approach the optical power-dependent resonant wavelength in each direction. Figure 2(a) shows the maximum OMA that can be reached from the shorter-to-longer wavelength scan (shown by the blue left-to-right arrows), as well as the best OMA that can be reached returning to the initial wavelength with a decreasing-wavelength scan. As previously discussed [15,24], the optimal operating conditions in a regime that is characterized by the bistable behavior of the resonator are found by scanning the laser from lower to higher wavelength. Figure 2 shows how the bistable regime grows with increasing optical power. At 10 mW of input power, the region of bistability is wide, and requires nearly 6 nm of wavelength shift or an equivalently large tuning of the temperature controller. In the measurements reported here, the optimum bias point was found through post processing, after the measurements were performed. An automated controller would search for the optimum bias point by exploring both sides of the resonance using a systematic trajectory, rather than blind bidirectional searching. Abruptly turning on a laser, as opposed to a swept-initialization case, can result in two impairments: (i) the loss of the ability to track the resonance and achieve the highest OMA or ER, and (ii) may result in self-modulation due to the interplay of heating and free-carrier dispersion over a range of the bias conditions [15]. The development of a practical and robust bias controller in the high-power regime is an important future goal in this topic.

3.4 OMA improvements

As the optical power increases, the loss inside the resonator also increases from the increase of free carrier absorption induced by TPA. In a microdisc resonator that was designed to be at critical coupling for optical power levels where TPA is low, at higher optical power when the round trip loss is higher, the resonator becomes more under-coupled since the self-coupling (or through transmission) coefficient of the directional coupler between the bus waveguide of the resonator becomes larger than the roundtrip amplitude coefficient that describes the amplitude change in propagating through one round trip around the circumference of the microdisc [25]. The extinction ratio will then become smaller, and the OMA will decrease with increasing optical power, even if distortions from nonlinear power induced resonance shifts do not fully close the eye. This also means that a disc designed to be slightly over-coupled at low optical power may move closer to critical coupling when the optical losses in the resonator increase at higher optical power, and the extinction and OMA will improve.

We verify that a microdisc that is more over-coupled at low optical powers will have higher OMA at high power through a time-domain simulation in MATLAB following the methodology shown in [26,15] which captures the nonlinearity-induced bistability behavior of the microdisc modulator. Figure 4 shows this effect on two microdisc resonators that are identical except for the bus-waveguide to resonator coupling gap. Figure 4(a) shows the result of our simulation using a 20 Gbit/s PRBS7 NRZ modulation, where the OMA is plotted versus input optical power, relative to its value at 10 mW input power. We observe that the OMA of datapoint “a2” is higher than datapoint “a1” by about 0.4 mW. The plot in Fig. 4(b) shows the low-power static transmission response, with the initially over-coupled (at low power) disc having a power coupling coefficient, |κ₂|², 1.5 times larger than the coefficient of the critically coupled (at cold cavity) disc, |κ₁|². The power coupling coefficients are related to the time domain coupling coefficient, µ₁ and µ_2, of our time domain simulation as discussed in [27], and were fitted from the fabricated and measured devices, as discussed next.

Figures 4(c)-(d) shows the experimental results of two measured devices, identical except for the coupling gap. Figure 4(d) shows that the modulator with a 260 nm gap between the disc and the bus waveguide has a full-width-half-maximum (FWHM) of 42 GHz and Q of 4,500 and is slightly more over-coupled than the disc with a 280 nm gap, which has a FWHM of 31 GHz and a Q of 6,200. The extinction of the disc with 280 nm gap is also slightly larger. This results in the smaller gap disc showing a smaller change in transmission for a given voltage swing, or equivalently, for the same change in resonant wavelength, less change in transmission and a smaller modulation efficiency. This results in the smaller-gap, relatively over-coupled disc having smaller OMA at lower optical power. However, at higher power, the larger gap disc shows a larger decrease in extinction ratio at higher power and the OMA of the smaller-gap disc is higher than the OMA of the larger-gap disc at larger input optical powers, and is about 0.5 mW higher at the highest OMA points which are marked by the points “f” and “e” in Fig. 4(c). We measured 4.1 mW OMA at an input power of 13.5 mW. In agreement with the simulations, our measurements show that the disc with the smaller waveguide-resonator gap shows a more open eye diagram at higher optical powers. This device also shows a lower modulation loss (sum of insertion loss and excess modulation loss) of about 0.5 dB at point “f” (compared to about 1.5 dB of modulation loss for the 280 nm gap disk at point “e”), and a slightly higher extinction ratio of about 3.1 dB (compared to an extinction ratio of about 2.9 dB for point “e”). At lower optical powers, a silicon microring modulator operated at 25 Gbps was seen to have an excess modulation loss on the one level of approximately 3 dB, while achieving sub volt V_πL [20]—trading off loss for increased dopant concentration can be used to increase modulation efficiency.

 

Fig. 4. (a) Simulated OMA vs. optical power in the input bus waveguide for two discs, normalized to an input power of 10 mW. µ₁ and µ₂ are the waveguide-disc time-domain coupling coefficients of the two discs. The inset shows that the over-coupled disc (µ₂) has lower OMA at low power. The difference between the OMA at high power marked by “a1” and “a2” is about 0.4 mW. (b) Fitted transmission spectrum at low power for the two discs, where κ₁ and κ₂ are the field amplitude coupling coefficients. (c) Measured OMA vs. optical power in the input bus waveguide, at 20 Gbit/s NRZ of the two discs. The inset shows that the OMA of the 260 nm gap disc is lower at low power, in agreement with simulations. (d) Measured transmission at low power, sweeping the tunable laser from lower to higher wavelength. (e) Time-domain waveform measured on an oscilloscope of the 280 nm gap disc, folded by an integer number of the period in a post-processing step to show the extent of eye closure at 20 Gbit/s, at high optical power for the measurement point marked in Fig. 4(c). (f) Time-domain waveform of the 260 nm gap disk, at high optical power. The eye is more open and the difference between the minimum “1” and maximum “0”, labeled as “Δ Eye f”, is about twice as large as “Δ Eye e”. (g) A 25 Gbit/s NRZ eye diagram measured for the 260 nm gap disc at higher optical powers with 12,000 points per waveform over 1,050 captured traces. The disc with the 260 nm gap can support higher data rates when the optical power is increased. The optical power shown is the power in the output bus waveguide after the modulator, when the input power in the bus waveguide was 6.3mW, and the OMA was measured to be 2.2 mW.

Download Full Size | PDF

Higher OMA, extinction, and less eye closure at higher optical powers for the 260 nm gap disk results in a more open eye diagram at an increased data rate of 25 Gbit/s, shown in Fig. 4(g). The optical power shown is the power in the output bus waveguide after the modulator, when the input power in the bus waveguide is 6.3mW, and the OMA is measured to be 2.2 mW.

We study the OMA in absolute units of Watts rather than normalized to an input value, and a list of previously reported OMA’s in absolute values is given in Table 1. We prefer to report OMA in absolute rather than in normalized units, since it directly corresponds to link range in an unamplified communications link. However, in many reports, the OMA is absolute units is not available and reported OMAs are instead normalized to the input power, as (P1-P0)/P_in where P1 is the average high bit value, P0 is the average low bit value, and P_in is the input power before the modulator. A list of normalized OMA’s is given in Table 2. Previously at 20 Gbit/s, an OMA of 0.47 normalized to the input power was reported [28]—while the input power level was not stated in that reference, the normalized OMA’s from the OMA’s in absolute power shown in Fig. 4 (a) range from 0.38 at input powers up to 6.3 mW, to 0.27 at the higher OMA’s beyond 10 mW, and are comparable to state-of-the-art resonant modulators [24,29].

Table 2. Normalized Optical Modulation Amplitude of resonant silicon photonic modulators

View Table | View all tables in this article

4. Conclusion

We have shown that a silicon microdisc modulator can achieve an OMA of greater than 4 mW at 13.5 mW of optical power in the bus waveguide at a data rate of 20 Gbit/s (NRZ modulation format) at wavelengths near 1550 nm. Simulations were performed to support the explanations for the differences in OMA versus input power trends that were observed at low (200 Mbit/s) and high (20 Gbit/s) modulation speeds, as well as coupling differences. A substantial wavelength shift, or an equivalent shift of the bias point using thermal or bias voltage tuning, may be required in order to achieve the best OMA performance, which could be achieved with further substrate removal [35]. Further studies can focus on an investigation of long-term drift and stability, an understanding of thermal issues (including heat spreading, crosstalk and control dynamics), and waveguide-fiber couplers which can tolerate high power with low insertion loss. Improved OMA performance at the several-milliwatts levels improves the utilization of available laser power and improves the link transmission distance in an unamplified optical communication system, and lessens the need for optical amplification which can decrease the spectral span of wavelength division multiplexing and decrease the wall-plug energy efficiency.