Open Access
Add to BibSonomyAbstract
We demonstrate how to optimize the performance of PAM-4 transmitters based on lumped Silicon Photonic Mach-Zehnder Modulators (MZMs) for short-reach optical links. Firstly, we analyze the trade-off that occurs between extinction ratio and modulation loss when driving an MZM with a voltage swing less than the MZM’s Vπ. This is important when driver circuits are realized in deep submicron CMOS process nodes. Next, a driving scheme based upon a switched capacitor approach is proposed to maximize the achievable bandwidth of the combined lumped MZM and CMOS driver chip. This scheme allows the use of lumped MZM for high speed optical links with reduced RF driver power consumption compared to the conventional approach of driving MZMs (with transmission line based electrodes) with a power amplifier. This is critical for upcoming short-reach link standards such as 400Gb/s 802.3 Ethernet. The driver chip was fabricated using a 65nm CMOS technology and flip-chipped on top of the Silicon Photonic chip (fabricated using IMEC’s ISIPP25G technology) that contains the MZM. Open eyes with 4dB extinction ratio for a 36Gb/s (18Gbaud) PAM-4 signal are experimentally demonstrated. The electronic driver chip has a core area of only 0.11mm2 and consumes 236mW from 1.2V and 2.4V supply voltages. This corresponds to an energy efficiency of 6.55pJ/bit including Gray encoder and retiming, or 5.37pJ/bit for the driver circuit only.
© 2017 Optical Society of America
1. Introduction
The continuously growing popularity of internet based software applications such as social media, e-commerce and search engines is putting the data centers which underpin all such applications under significant pressure. This is leading to a continuous drive to develop higher bandwidth solutions for the short-reach links that are used to connect the data center’s servers together. While state-of-the-art short-reach links now have a capacity of 100Gb/s using either 10x 10Gb/s or 4x 25Gb/s non-return to zero modulated links, data center operators have expressed a need for larger link capacities. Importantly, increased capacity will need to be delivered using approximately the same physical footprint for the transceivers at either end of these links. Indeed, although bandwidth requirements may have grown exponentially, the space required on the front-panels of the switches and servers that host these transceivers cannot be grown due to obvious space restrictions. This leads to obvious challenges in terms of the power consumption and integration density of future high capacity links. These challenges have spurred significant research and development into the use of multilevel modulation formats such as 4-level pulse amplitude modulation (PAM-4) for such links. PAM-4 halves the number of parallel lanes required to support a total capacity, or allows a doubling of the overall bitrates using the same optics as current links that use 4x 25Gb/s lanes. A 400Gb/s transceiver can then be realized using e.g. eight 50Gb/s (25Gbaud) PAM-4 modulated lanes.
In this work we analyze in detail the use of lumped (i.e. the electrodes are not realized as transmission lines) Silicon Photonic Mach Zehnder modulators (SiPh MZMs) for these applications. SiPh MZMs offer the potential of small physical footprint and can be modulated with a driving voltage of only a few Volts, significantly reducing power consumption required for the driver compared to drivers for e.g. LiNbO3 modulators [1–3]. By using a lumped modulator (which in our implementation from an electrical point-of-view is a reverse biased PN junction), it is possible to integrated the modulator’s capacitance into an on-chip driver network to generate PAM-4 as presented previously in [4]. One of the main advantages of Silicon Photonics technology is that it allows the realization of MZMs and other functions in a very small footprint, with electrode lengths of only a few millimeters [3]. Compared to [2], where a segmented lumped MZM combined with a CMOS driver is used, our approach reduces the number of high-speed IOs required to control the lumped MZM. This reduces power consumption and packaging complexity. Moreover our approach does not require careful synchronization between different segments of the lumped MZM. This is especially important at baudrates beyond 10Gbaud where even a few picoseconds of signal skew will result in eye closure. Here we consider several trade-offs regarding the use of Silicon Photonic MZMs combined with driver circuits realized in deep submicron CMOS processes, and explore the bandwidth limits of simple lumped MZMs compared to more conventional travelling wave MZMs (TW-MZM). This paper also shows how to select the MZM bias point when modulating with a voltage swing considerably smaller than the MZM’s Vπ, and still maintain best possible optical budget trading off between achievable extinction ratio and modulation loss.
2. Lumped silicon photonic MZM
2.1 DC characteristics of silicon photonic MZM
The normalized electro-optical transfer function of an MZM is shown in Eq. (1), where V1 and V2 are RF signals applied to each MZM electrode respectively; VDCbias is the sum of the biasing voltages applied on both electrodes of the MZM and Vπ is the voltage required to change the phase in one modulator arm by 180 degrees, thereby causing the MZM to switch from full transmission to full extinction [5]. In this paper, the MZM is always driven in push-pull configuration. Conventionally, the MZM is biased at the so-called quadrature point of its transfer function. By applying push pull drive signals to both arms with a swing equal to half Vπ, the modulated extinction ratio (defined as the ratio of the highest power to the lower power after the MZM when applying the drive signal) can then in principle be as high as the DC extinction ratio which is typically of the order of 20dB. Although in practice, as explained below it is difficult to achieve modulated extinction ratios as high as this due to practical drive amplitude limitations.
For transceivers intended for short-reach optical links, low power consumption and process technology suitable for high volume production are important requirements. Therefore it is advantageous to realize the driver circuit in a deep submicron CMOS process node. However, the CMOS transistors in such a process node usually cannot withstand voltage swings beyond 1V to 2V without breaking down. Even when using more complex SiGe BiCMOS processes, careful consideration must be paid to the breakdown voltage of the high-speed heterojunction bipolar transistors (HBTs), which can be as low as 1.5V. Such low breakdown voltages are challenging when attempting to drive MZMs whose Vπ may exceed those voltages. For example, modulators available from the ISIPP25G technology can have a Vπ × L product of 12V × mm. A 1.5mm long MZM then has a Vπ of 8V [6]. Even when using circuit techniques to increase the available voltage swing from drivers realized using deep submicron CMOS processes, typically no more than a swing of a few Volts is available. Therefore, transmitters based upon Silicon Photonic MZMs which use drivers realized in deep submicron CMOS or SiGe BiCMOS typically have poor extinction ratios (~2.6dB) (~50Gb/s) [6, 7]. A push pull driving scheme whereby both arms are modulated in anti-phase helps by halving the required driver voltage swing for a given Vπ, however even then in our particular example a driver with 4V swing is required which is highly challenging for deep submicron CMOS and SiGe BiCMOS technologies.
Therefore here we consider the optical link performance when modulating the MZM with a swing less than Vπ. Figure 1 shows the transfer curve of an MZM with Vπ = 8V according to Eq. (1). We consider modulation of the MZM with a push-pull voltage swing of 2V, which is achievable using a deep submicron CMOS driver circuit [4]. Clearly, to improve the ER with this limited voltage swing we can choose the DC bias point of the Silicon Photonic MZM below its quadrature point. For example by biasing the MZM at 6V, 9dB ER can be achieved. However, two challenges arise.
The first challenge is how to provide the DC bias to the modulator. To compensate for random process and temperature variations in the MZM transfer curve, this DC bias must be actively controlled [8]. To realize dense integration, it is preferable to integrate such a control circuit together with the driver IC, which can then be flip-chipped on top of the silicon photonic IC [9]. However handling voltages as high as 6V is difficult with standard deep submicron CMOS technologies: for example while capacitive coupling could be used (see section 3 for an explanation), the breakdown voltage of on-chip capacitors is typically no higher than ~5V for 65nm CMOS process [10], leaving little to no room for the RF signal itself. The solution is to use an asymmetrical MZM as shown in Fig. 2, where a length difference is introduced between the two optical paths, which generates a phase difference equivalent to a half Vπ shift in the red symmetrical MZM transfer curve in Fig. 1. The new transfer function is Eq. (2), plotted as the black curve in Fig. 1. The quadrature point has been shifted to 0V thus reducing the required bias voltage for a given extinction ratio. For example the asymmetrical MZM only needs 2V DC bias voltage to achieve 9dB ER (black dotted curve in Fig. 1) instead of 6V for the symmetrical MZM (red curve in Fig. 1). Of course note that active bias control remains necessary due to unavoidable process variations, however with the π phase shift the maximum required bias voltage can be significantly reduced.
The second challenge is the trade-off between the extinction ratio on one hand and modulation loss on the other hand, as a function of the employed RF swing. The modulation loss is defined as:where PCW is the power at the input to the modulator (i.e. the power from the laser minus the coupling and waveguide losses). Phigh and Plow are the maximum respectively minimum optical powers after the modulator when the RF signal is applied. This trade-off is shown for an asymmetrical MZM in Fig. 3 for three different voltage swings (Vπ = 8V). The curves show the achieved modulation loss and extinction ratio for a particular bias setting. Note that the trade-off between modulation loss and extinction ratio is especially severe for RF swings that are small compared to Vπ. For example for an RF swing of 1.0V (the voltage applied to a single arm in a push pull drive scheme), 10dB extinction ratio incurs a modulation loss of ~8dB. Reduced modulation loss can be achieved at the expense of a significantly degraded extinction ratio, for example 2dB modulation loss would result in an extinction ratio of 4dB. Obviously, such a trade-off has an impact on the achievable optical budget.
Fig. 3 The trade-off between modulation loss of the modulator and extinction ratio for different applied voltage swings (VRF is the voltage applied to a single arm in a push pull scheme) (Vπ = 8V).
To analyze this impact and determine the optimum bias point, optical budget calculations were performed assuming a link consisting of a laser (launch power from the laser is assumed to be + 10dBm), coupled via a grating coupler and routing waveguide to the Silicon Photonic MZM, and back out using a second grating coupler. The total loss of the two grating couplers, the routing waveguides and the modulator waveguides was assumed to be 11.9dB (corresponding to the losses in the actually fabricated device) [6]. The receiver was modeled using a fourth order Bessel-Thomson filter with a 3dB bandwidth of 20GHz; its total input referred noise current was 2.9µArms. Figure 4 shows the available optical budget (defined as the difference between the power launched after the second grating coupler into the fibre and the receiver sensitivity for the realized extinction ratio, at a bit-error ratio of 10−3) for PAM-4 modulation format based on the asymmetrical MZM: each curve was generated by sweeping the bias voltage of the modulator, calculating the resulting additional modulation loss and extinction ratio and then determining the achievable optical budget for a bit-error rate of 10−3. The extinction ratio is defined as the ratio of the power of the highest PAM-4 symbol to the lowest PAM-4 symbol. Equidistant PAM-4 symbols are also assumed, which can be achieved by pre-compensating the driver voltage levels (see also Section 3). The bias voltage was swept such that each time the full S-curve of the modulator is covered.
Fig. 4 Available optical budget (Vπ = 8V) for different applied voltage swing (VRF is the voltage applied to a single arm in a push pull scheme). The locus of optimum bias voltages is shown as well.
Figure 4 shows the achievable optical budget for three different voltage swings as a function of bias voltage. It can be seen how even at reduced RF swings, the optimum bias point to maximize the optical budget remains close to the MZM quadrature point. For bias voltages less than this optimum bias point, the extinction ratio improves, however this is more than offset by the worse modulation loss reducing the overall achievable optical budget. Conversely, for bias voltages higher than this optimum bias point, the modulation loss is improved, which then however is offset by a worsening extinction ratio thus again reducing optical budget compared to the maximum achievable optical budget. This trade-off is significant for low RF swings with ~5dB difference between the optimum bias point and the worst-case bias. For increasing RF swings this trade-off becomes less severe. It is also important to point out that sufficient optical budget is available to support a link with 5.5dB optical budget at an RF swing of 1V, which corresponds to the achieved value in our proposed PAM-4 transmitter, see Section 3.
2.2 Velocity mismatch of travelling-wave MZM (TW-MZM)
Conventionally, the electrodes of a high speed MZM are implemented as transmission lines (Travelling-Wave). This distributes the capacitance and inductance over the entire length of the device, which enhances the modulation efficiency over a large bandwidth [11]. In addition, optimized microwave and optical impedance matching is required to reduce reflections and the microwave attenuation of the TW electrode and the optical loss should be minimized. To achieve the optimal performance for a TW-MZM, the microwave group velocity should be matched with the optical group velocity [12, 13]. Any mismatch would introduce attenuation of the optical signal. Any walk-off between the RF and optical signals reduces the effective bandwidth of the modulator. As Silicon Photonic modulators have a length of only a few millimeters, the penalty due to velocity mismatch can be reduced.
Based on the references [12, 13] and [5], we can derive the expression for the optical power transfer function of a TW-MZM as a function of modulation frequency f as Eq. (3) by assuming no microwave attenuation of the electrodes, perfect impedance matching and no optical loss in the waveguide. In addition, it is assumed that the TW-MZM is biased at its Vπ point with 2Vπ peak-to-peak voltage swing of one electrode arm and the second electrode grounded. n0 is the optical refractive index, nRF is the equivalent microwave refractive index which can be calculated as nRF = 1/(εµ)0.5 where ε is dielectric constant and µ is the relative permeability of the electrode transmission line. L is the length of the electrode and c is the speed of light in the vacuum. Note how the mismatch between nRF and n0 is multiplied by the length of electrode (L). If the microwave signal and optical signal match perfectly (nRF = n0), the optical transfer would be one. The length of Silicon Photonic MZMs can be as short as a few millimeters, allowing significant mismatch before attenuation of the modulated optical signal. Figure 5 shows the optical power transfer as a function of modulation frequency for a modulator with length 1.5mm. The optical power transfer has been plotted for three mismatch (n0-nRF) cases. Even for a large 50% mismatch (n0-nRF = ± 1.5), the −1dB bandwidth (purely due to velocity mismatch, hence not taking into account any RC limitations) is over 35GHz. Normally, the dielectric constant of silicon is around 11.68, therefore, the microwave refractive index is 3.42 which is close to the optical index of silicon. So, the −1dB bandwidth due to the mismatch is beyond 60GHz.
2.3 Effective electrical length of the MZM electrodes
As the length of the electrodes of a Silicon Photonic TW-MZM can be as short as a few millimeters, it is interesting to know to what modulation frequency the electrodes can be considered to be electrically “short” or lumped devices and hence do not need to be implemented as transmission lines. A “long” electrode is generally considered to be one where the source’s AC signal waveform completed at least a quarter-cycle (90 degree phase rotation) before the propagating signal reaches the end of the electrode, which means the signal at the start of the electrode is out of phase compared to the signal at the end of the electrode at a given time. In this case, the electrode of the MZM needs to be treated as a transmission line. The frequency where this happens is called the boundary frequency and can be calculated from [14, pages 59-61]:
Based on the discussion above, the boundary frequency can be derived as Eq. (5), where c is the speed of light in vacuum, L is the electrical length of electrode, and nRF is the microwave refractive index of the electrode. In Fig. 6, we assume that a total electrode length for the MZM is 1.5mm. If we locate the signal source in the centre of the electrode, the effective length to be taken into account halves to 750µm. Based on Eq. (4), the boundary frequency for the Lumped-MZM in Fig. 6 is 29.2GHz; and the boundary frequency for the TW-MZM in Fig. 6 is 14.6GHz, assuming the effective dielectric constant nRF is 3.42. Therefore, as long as the Silicon Photonic MZM with a 1.5mm electrode length is operated within a bandwidth ranging from DC to 29.3GHz, the electrode is electrically equivalent to a capacitor instead of a transmission line.
For ease of testing, TW-MZMs typically use transmission lines with a characteristic impedance of 50Ω for the electrodes. To deliver the signal from the input of the transmission line to the end with maximum power transfer, the signal source impedance and termination impedance should then also be 50Ω [14]. The required matching networks need to have a large bandwidth (e.g. >20GHz for 25Gbaud optical links), which necessitates the use of resistive low-Q matching. Significant power is required to drive such a 50Ω termination transmission line [14].
3. Driving scheme for lumped-MZM
As pointed out in the previous section, a Silicon Photonic MZM with an electrode length up to ~1.5mm (assuming it is driven from the midpoint of each electrode) can be considered a capacitor from the electrical point-of-view. When driving such a lumped MZM from the conventional 50Ω signal source its bandwidth will be severely limited due to the low-pass filter formed by the 50Ω signal source and termination resistors. An alternative driving scheme with significantly larger bandwidth is presented here.
3.1 Bandwidth enhancement using reduced source resistance and series capacitor
The equivalent circuit model of a lumped MZM driven from a 50Ω signal source impedance and a 50Ω termination resistor is shown in Fig. 7(a). It is assumed that the inductance of the interconnect between the driver and lumped MZM is negligible and the parasitic resistance is 10Ω. The double termination halves the effective available voltage swing over the lumped MZM. For a lumped MZM with a capacitance of 884fF (corresponding to the load of a fabricated device, see section 5), the bandwidth in the double terminated scheme is 5.1GHz. The obvious bandwidth limitations stemming from the 50Ω double matching can be overcome by driving the lumped MZM with a much lower signal source resistance and eliminating the termination resistor. The bandwidth can be further enhanced by including a series capacitor with the lumped MZM, as shown in Fig. 7(b). If the value of this series capacitor is chosen to be the same as the lumped MZM, the effective voltage across the lumped MZM is the same as the 50Ω matching method (assuming the same signal source voltage), however now the bandwidth is 24GHz, which is more than sufficient for 25Gbaud signaling.
3.2 Extension of proposed driving method to multilevel modulation formats
Another advantage of the proposed driving scheme using a series capacitor is that it can be readily extended to multilevel modulation formats such as PAM4. The proposed PAM4 driving scheme for the silicon photonic Lumped-MZM is shown in Fig. 8 [15]. First, two non-return to zero encoded input bit streams A and B are thermometer and Gray coded into three signals X, Y and Z (taking on a value of 0V for a digital ‘0’ and a value VDD for a digital ‘1’) as shown in Table 1. In turn, the signals X, Y and Z drive three capacitors CX, CY and CZ. The capacitance CARM of the reverse biased PN junction of the silicon photonic Lumped-MZM is combined with these on-chip capacitors CX, CY, and CZ, to form a switched capacitor bank. As a result, a PAM-4 signal is developed across CARM. The voltage of each PAM level is determined by the voltage swing of X, Y and Z and the ratio between the on-chip capacitors (CX, CY, and CZ) and capacitance CARM. To achieve push-pull operation, the logic inverse of X, Y and Z are applied via a second set of capacitors to the second arm of the lumped MZM.
Table 1. Encoding Scheme
DC bias voltages Vbias1 and Vbias2 can be easily applied via sufficiently large bias resistors, which can also be implemented on-chip thus minimizing component count compared to using bias-tees.
The signals X, Y and Z can be generated using transistor based switches which either connect to 0V or to a voltage source with value VDD. The achievable symbol rate of this circuit is then limited by the resistance of these transistor based switches and the value of the capacitors CX, CY and CZ. High-speed can be achieved by selecting sufficiently small on-chip capacitors CX, CY and CZ and using wide transistors with small (a few ohms) on-resistance. However, the smaller CX, CY and CZ, the smaller the effective drive voltage developed across the lumped MZM. An optimum choice was found to dimension the capacitors CX, CY and CZ such that the switched capacitor bank realizes a voltage division by a factor of 2. A circuit which implements this scheme using a 65nm CMOS technology is presented in [4].
4. Experiment results
4.1 Implementation of the realized PAM-4 transmitter
The Silicon Photonic die was fabricated using the IMEC ISIPP25G technology through their multiproject wafer service [6]. A schematic drawing of the MZM is shown in Fig. 9. As mentioned above, the signal pads are connected to the middle of each electrode to ensure that the MZM is a lumped device for the maximum frequency of operation. Ground and dummy pads were added on each side to form a GSGSG structure, which helps with on-die probing of the modulator. Additional pads were added around the modulator for flip-chip mounting of the driver chip on top of the Silicon Photonic die.
Grating couplers (each with a ~3dB loss [6]) are used to transfer the optical signal in and out of the modulator. The insertion loss of the modulator including waveguide routing is 6dB, and hence the zero bias insertion loss of the complete chip is 12dB. In addition, since the asymmetrical Lumped MZM is naturally operated at the quadrature point, an extra 3dB modulation loss should be added. Therefore, the total optical attenuation is around 15dB.
The driver chip, fabricated using 65nm CMOS, was flip-chipped on top of the Silicon Photonic die, as shown in Fig. 10. The connection between the Silicon Photonic die and the CMOS driver chip was realized using 50µm diameter solder balls, deposited using a laser assisted solder jetting process [9]. After flip-chipping, the 3D stacked chipset was mounted on a copper submount and wirebonded to an AlN ceramic and printed circuit board assembly with grounded coplanar waveguides for transferring the high-speed data and clock inputs.
Fig. 10 Driver micrograph (top left), Silicon photonic die (bottom left) and driver flip-chipped on top of SiPhotonic die (right).
4.2 Measurement results
The measured normalized DC transfer function of the lumped MZM integrated with the CMOS driver chip is shown in Fig. 11. The output optical power was measured by fixing the voltage of one arm and sweeping the DC voltage of the other arm. The maximum applied voltage was kept below 4V to avoid breakdown of the on-chip coupling capacitors. The measurement results are well matched with the transfer function Eq. (2).
The electro-optical bandwidth was measured on a second Lumped MZM sample which did not have any driver chip mounted. Bias voltages of 1V and 3V were selected respectively for each arm in order to achieve optimal optical budget. The normalized S21 measured through RF probing the device and a vector network analyzer is shown in Fig. 12. It can be seen that the –3dB bandwidth is 3GHz. Assuming a 10Ω contact resistance for the lumped MZM, an electrode capacitance of 884fF can then be estimated.
The PAM-4 optical signal generated by the driver circuit and lumped MZM is detected by the photoreceiver (XPRV2021A) and captured using a real-time oscilloscope. The resulting eye diagram is plotted in Fig. 13. The extinction ratio between the top and bottom PAM-levels (ER) is 4dB. A baudrate of 18Gbaud (36Gb/s) was achieved. This was limited mainly by the retiming D flipflops, which fail to align the gray coding outputs (X, Y, and Z) correctly for specific data patterns when the bitrate is beyond 18Gb/s. The bit-error rate was measured through error counting using offline processing and found to be 6 × 10−4 [4]. Note that the PAM-4 levels are not equidistant. The reason for this is that the capacitors CX, CY and CZ were sized in such a way that the generated PAM-4 voltage levels were precompensated for the cos2 non-linear transfer (see Eq. (3)) of the MZM. The selected size of the capacitors CX, CY and CZ depends on the Vπ and capacitance of the MZM, which turned out to be different on the actual device compared to the values assumed during the design process. In addition VDD can be tuned as well to match the actual Vπ. This can be easily corrected in future driver chips by resizing the capacitor bank.
A performance summary and comparison with state-of-the-art Silicon Photonic transmitters is provided in Table 2. Compared to [2], the bitrate is increased by 80%, however, the energy efficiency is much worse. This is likely to be mainly due to the fact that the authors in [2] used a more advanced Silicon Photonic Lumped-MZM and a faster 40nm CMOS technology. References [7] and [16] use SiGe BiCMOS technology to achieve higher bitrates (but limited to non-return to zero modulation rather than the PAM-4 demonstrated in our work), at the expense of similar to worse energy efficiency. It is also noteworthy how the extinction ratio (here measured as the ratio of optical power of the highest to the lowest PAM-4 level) is 4dB, significantly better than [7] and [16]. The smallest circuit size has been achieved thanks to the use of 3D solenoids and the small active area of the switched capacitor approach [4].
Table 2. Performance Summary
Finally, we compare the performance of our hybrid (i.e. the photonics and electronics have been integrated into two different technologies) approach to two different monolithic implementations. In [17], an NRZ transmitter is reported which uses a 90nm CMOS technology equipped with photonic components, while in [18] a PAM-4 transmitter has been reported using the same technology. The modulator used in [18] is a travelling wave segmented (two segments used) MZM, terminated using sets of 50Ω resistors. The low VπxL product of the used MZM results in low power consumption and excellent energy efficiency of 5.4pJ/bit at 25Gbaud (50Gb/s) PAM-4. We anticipate that by using our proposed PAM-4 driving scheme in such monolithic technology, an even lower power consumption can be achieved. Indeed in a monolithic technology the capacitance associated with the bondpads can be completely eliminated. On the other hand, note that in the hybrid approach, the overall chip area can be reduced as electronics can be stacked on top of the photonics.
5. Conclusion
The performance limitations of Silicon Photonic lumped Mach-Zehnder modulators have been investigated. It is explained how the small length of Silicon Photonic MZMs allows effective modulation bandwidths in excess of 20GHz. An optimized driving scheme which overcomes the limitations imposed by the capacitance of the Silicon Photonic lumped MZM has been presented and experimentally demonstrated for a PAM-4 modulation format. The impact of limited voltage swing available from deep submicron CMOS driver circuits on the achievable optical budget has been investigated. It is shown how careful optimization of the MZM bias voltages allows a tradeoff between modulation loss and extinction ratio in order to maximize the achievable optical budget.
Funding
Science Foundation Ireland (grants 11/SIRG/12112, 12/IA/1270 and 12/RC/2276), Microelectronic Circuits Centre Ireland (MCCI, grant 2013-01).
References and links
1. M. Watts, W. Zortman, D. Trotter, R. Young, and A. Lentine, “Low-voltage, compact, depletion mode silicon Mach-Zehnder modulator,” IEEE J. Sel. Top. Quantum Electron. 16(1), 159–164 (2010). [CrossRef]
2. M. Webster, C. Appel, P. Gothoskar, S. Sunder, B. Dama, and K. Shastri, “Silicon photonic modulator based on a MOS-Capacitor and a CMOS driver,” IEEE Compound Semiconductor Integrated Circuit Symposium (CSICS), pp. 1–4, Oct. 2014. [CrossRef]
3. H. Park, M. Sysak, H. Chen, A. Fang, D. Liang, L. Liao, B. Koch, J. Bovington, Y. Tang, K. Wong, M. Jacob-Mitos, R. Jones, and J. Bowers, “Device and integration technology for silicon photonic transmitters,” IEEE J. Sel. Top. Quantum Electron. 17(3), 671–688 (2011). [CrossRef]
4. S. Zhou, H. Wu, K. Sadeghipour, C. Scarcella, C. Eason, M. Rensing, M. Power, C. Antony, P. O’Brien, P. Townsend, and P. Ossieur, “Driver circuit for a PAM-4 optical transmitter using 65 nm CMOS and silicon photonic technologies,” Electron. Lett. 52(23), 1939–1940 (2016). [CrossRef]
5. P. Winzer and R. Essiambre, “Advance optical modulation format,” Proc. IEEE 94(5), 952–985 (2006). [CrossRef]
6. “IMEC ISIPP25G technology handbook,” Confidential Documents from Interuniversity Microelectronics Center, July. 2014.
7. G. Denoyer, C. Cole, A. Santipo, R. Russo, C. Robinson, L. Li, Y. Zhou, J. Chen, B. Park, F. Boeuf, S. Crémer, and N. Vulliet, “Hybrid silicon photonic circuits and transceiver for 50Gb/s NRZ transmission over single-mode fiber,” J. Lightwave Technol. 33(6), 1247–1254 (2015). [CrossRef]
8. A. Narasimha, B. Analui, Y. Liang, T. Sleboda, S. Abdalla, E. Balmater, S. Gloeckner, D. Guckenberger, M. Harrison, R. Koumans, D. Kucharski, A. Mekis, S. Mirsaidi, D. Song, and T. Pinguet, “A fully integrated 4 × 10-Gb/s DWDM optoelectronic transceiver implemented in a standard 0.13 μm CMOS SOI technology,” IEEE J. Solid-State Circuits 42(12), 2736–2744 (2007). [CrossRef]
9. N. Pavarelli, J. Lee, M. Rensing, C. Scarcella, S. Zhou, P. Ossieur, and P. O’Brien, “Optical and electronic packaging processes for silicon photonic systems,” J. Lightwave Technol. 33(5), 991–997 (2015). [CrossRef]
10. M. Ker, C. Lin, and Y. Hsiao, “Overview on ESD protection designs of low-parasitic capacitance for RF ICs in CMOS technologies,” IEEE Trans. Device Mater. Reliab. 11(2), 207–218 (2011). [CrossRef]
11. G. K. Gopalakrishnan, W. K. Burns, R. W. McElhanon, C. H. Bulmer, and A. S. Greenblatt, “Performance and modeling of broadband LiNb03 traveling wave optical intensity modulators,” J. Lightwave Technol. 12(10), 1807–1819 (1999). [CrossRef]
12. W. Leeb, A. Scholtz, and E. Bonek, “Measurement of velocity mismatch in traveling-wave electrooptic modulators,” IEEE J. Quantum Electron. 18(1), 14–16 (1982). [CrossRef]
13. R. Alferness, S. Korotky, and E. Marcatili, “Velocity-matching techniques for integrated optic traveling wave switch/modulators,” IEEE J. Quantum Electron. 20(3), 301–309 (1984). [CrossRef]
14. D. Pozar, Microwave Engineering (John Wiley and Sons, 2012).
15. S. Yoo, J. Walling, O. Degani, B. Jann, R. Sadhwani, J. Rudell, and D. Allstot, “A Class-G switched-capacitor RF power amplifier,” IEEE J. Solid-State Circuits 48(5), 1212–1224 (2013). [CrossRef]
16. E. Temporiti, G. Minoia, M. Repossi, D. Baldi, A. Ghilioni, and F. Svelto, “A 56Gb/s 300mW silicon-photonics transmitter in 3D-integrated PIC25G and 55nm BiCMOS technologies,” IEEE International Solid-State Circuits Conference (ISSCC), pp. 404–405, Feb. 2016. [CrossRef]
17. D. M. Gill, C. Xiong, J. E. Proesel, J. C. Rosenberg, J. Orcutt, M. Khater, J. Ellis-Monaghan, A. Stricker, E. Kiewra, Y. Martin, Y. Vlasov, W. Haensch, and W. M. J. Green, “Demonstration of error-free 32-Gb/s operation from monolithic CMOS nanophotonic transmitters,” IEEE Photonics Technol. Lett. 28(13), 1410–1413 (2016). [CrossRef]
18. C. Xiong, D. Gill, J. Proesel, J. Orcutt, W. Haensch, and W. M. J. Green, “A monolithic 56Gb/s CMOS integrated nanophotonic PAM-4 transmitter,” Optical Interconnects Conference, pp. 16–17, Apr. 2015.
