High-performance coherent optical modulators based on thin-film lithium niobate platform

The coherent transmission technology using digital signal processing and advanced modulation formats, is bringing networks closer to the theoretical capacity limit of optical fibres, the Shannon limit. The in-phase/quadrature electro-optic modulator that encodes information on both the amplitude and the phase of light, is one of the underpinning devices for the coherent transmission technology. Ideally, such modulator should feature a low loss, low drive voltage, large bandwidth, low chirp and compact footprint. However, these requirements have been only met on separate occasions. Here, we demonstrate integrated thin-film lithium niobate in-phase/quadrature modulators that fulfil these requirements simultaneously. The presented devices exhibit greatly improved overall performance (half-wave voltage, bandwidth and optical loss) over traditional lithium niobate counterparts, and support modulation data rate up to 320 Gbit s−1. Our devices pave new routes for future high-speed, energy-efficient, and cost-effective communication networks.

compared on voltage alone. The current must also be considered in order to assess the power draw.
I also do not agree with the authors about the 'LN is well-known for its drifts in the DC bias point upon the application of a static electric field, which is a phenomenon that originates from the piezoelectric nature of the material". The authors must provide a reference for this statement. Bias control is necessary for interferometric lithium niobate devices due to the pyroelectric effect (changes in temperature cause a change in surface charge which in turn changes the refractive index). I know of many lithium niobae devices that do utilise an electro-optic bias and are able to maintain stability. Perhaps the authors are effectively controlling the temperature of the LNOI film with their TO heaters? I do not believe that their analysis of the instability of their own device when driven electro-optically provides any scientific insight and hence, I would encourage presenting the thermo-optic switches in a more technical journal where it can be the focus and referencing this in a higher profile paper describing the 'world first' IQ measurement.
Overall, the LNOI technology presented here does not, in my view, represent any significant advancement in the state of the art; however, the implementatino of a IQ modulator may indeed be a world first, and perhaps this is significant enough to warrant publicatino in a high profile journal such as Nature Communications. However, in order to achieve this level, I think the focus needs to be placed more on the system demonstration that harnesses the IQ modulator (rather than the technology used to realise the chip itself) and it would need to be demonstrated that a record breaking transmission characteristic was achieved (such as modulation density, or power efficiency?).
I also note a number of typographical errors throughout the manuscript (for example "both branches of the 13-mm and 7.5-mm devices as a function of the applied voltage, showing Vπ of 1.9 and 7.5 mm, respectively." -the last number should be the Vpi of the 7.5mm electrode, not its length). There are several others.
Reviewer #3 (Remarks to the Author): Lithium niobate is one of the most important materials for optical modulators. With the LNOI (lithium niobate on insulator) platform, new types of optical modulators were introduced in recent years. This kind of modulators has the advantages of small volume and high performance, such as low driving power, high bandwidth, etc. However, LNOI based modulators still face several important questions. 1. Can this modulator be used on short reach links, such as metro and data-center interconnects? Modulators for short reach links account for the most amount of the modulator industry production. In such application, the basic requirements for the modulator are small volume, low optical loss, low power consumption, and low cost. Currently the standard choice is the semiconductor (InP, for example) based modulators. InP modulator has the drawbacks of high optical loss and non-linearity. LNOI waveguide showed an ultra-low loss down to 0.03 dB/cm (Optica 4, 1536(Optica 4, , 2017, and conventional LN modulator showed a good linearity. Therefore, LNOI modulator has a potential to be a strong competitor with InP modulators. 2. For LNOI modulator, which fabrication routine will be adopted? In other words, what is the future technology roadmap? Two complementary fabrication routine were used to make LNOI based modulators: monolithic and hybrid. Monolithic means direct etching of LN to form the waveguides (Nature 562, 101, 2018). This kind of modulator theoretically has a better device performance. An example of hybrid is SOI-LNOI modulator (Nat. Photon. 13, 359, 2019). This modulator has the advantage of the mature fabrication technology, and easiness to integrate with photo detectors (such as epitaxy of SiGe detector). Which fabrication routine will be the mainstream? This question is still not clear to the academy and the industry. The people in industry especially care about this question, because they concern the R&D inputs. 3. For conventional LN modulator, a major problem is DC drift. This phenomenon is attributed to the piezoelectric or pyroelectric properties of the LN material. People spent a lot of efforts to solve this problem. For LNOI modulator, because it is a new research field, there is very few data or public report on the DC drift problem. This manuscript reported the first monolithic IQ (in-phase/quadrature) LNOI modulator used for short reach links, solved the DC drift problem using TO (thermal-optic) phase shifters. The device dimension is small (15 mm), and the performance is excellent, such as low loss (1.45 dB), low VπL (2.5 V cm), high data rate (320 Gbit s−1), and large bandwidth (> 67 GHz). The drawback of this work is that TO phase shifters needed electric power to operate, which increased the power consumption. This work provided the strong evidences to answer the three questions mentioned above. It will be very interesting to the people working on the modulators. A publication of the manuscript is highly recommended. However, the authors should response the following comments before the publication. 1. The authors can give some comments on the comparison on the modulators fabricated by hybrid or monolithic. Since the authors had the experience and insight on SOI-LNOI hybrid modulator (Nat. Photon. 13, 359, 2019), and this manuscript was about monolithic modulator, the authors' comments would be very useful to the readers. 2. The authors can give some comments or data on the multimode interference couplers (MMIs) in this work. There is few report about MMI coupler on LNOI. Any information about it will be useful. 3. The authors used TO phase shifter to suppress the DC drift, but the TO effect would increase the power consumption of the modulator. Can EO effect be used for DC drift control in LNOI modulator? 4. The authors have to check the content of the manuscript carefully to avoid any miswritting, for example, in line 73, " Fig. 2b" should be " Fig. 1b".
The following response is for the referee reports for manuscript "High-performance Coherent Optical Modulators based on Thin-film Lithium Niobate Platform". Blue text signifies the reviewers' comments. Black text is our response. Black italic text is quotes from our revisions (and all figures included are also from our revisions).
Reviewer #1 (Remarks to the Author): This is a paper by the same authors who published a paper in nature photonics on a very similar device ("High-performance hybrid silicon and lithium niobate Mach-Zehnder modulators for 100 Gbit s −1 and beyond"). The improvement is that they only use the larger multi-mode LN waveguide instead of using evanescent coupled small LN waveguides, and the device shows a very low insertion loss. At the same time, the authors use coherent equipment to do the coherent modulation test at a higher baud rate (maximum 110 GBaud).
The results are pretty good. There are some technical questions about the manuscript, which follow below.
We are grateful to the Referee for their time and effort in reviewing our work.
What are the modal properties of the LN waveguide? It looks like a multimode waveguide.
Why didn't the authors choose a single mode waveguide? Only to have a lower optical loss?
Response: Thank you for your questions. In the phase modulation region, we adopt 4-μm-wide LN waveguide and this is indeed a multimode waveguide.
We choose 4-μm-wide LN waveguide not only to achieve low optical loss, but also to balance the trade-off between voltage-length product (V π L) and modulation bandwidth. In our simulation, we find that a single mode waveguide indeed support lower value of V π L, however, it will also lead to higher value of RF loss, which will degrade the modulation bandwidth. According to our simulation, a single mode waveguide is not an optimal choice for the overall performance of the modulator. We elaborate our design for the width of LN waveguide in phase modulation section in Supplementary Note 4| Design of phase modulation waveguides and travelling-wave electrodes. As depicted in Supplementary Fig. 5, we achieve the largest BW/ V π value at the width of the LN waveguide w= 4 μm and the optimized electrodes gap is 7 μm, and we adopt this design in our experiment.
Moreover, we design our device in such a way that only the fundamental mode is excited. We adopt single-mode waveguides for all the other parts of the device (including the access waveguides for MMI and all the bend waveguides) except the phase modulation parts. We use adiabatic waveguide tapers at both ends of the phase modulation parts to make sure that the no high order modes can be excited when the light propagates in the device. To better convey this information, we have added one section in the revised Supplementary Information for the detailed simulation results of the mode transition from the single mode waveguide to 4-μm-wide LN waveguide.
In the Supplementary Note 2: "In the practical devices, we adopt single-mode waveguides (width of 1 μm) for all the other parts of the device (including the access waveguides for MMI and all the bend waveguides) except the phase modulation parts. We use adiabatic waveguide tapers to connect 1-μm-wide single-mode waveguides and 4-μm-wide phase modulation waveguides, which does not incur any conversion from fundamental mode to high-order modes (Supplementary What do they give up by going multimode?
Response: Thank you very much and this is an important question. We give up the device length for better modulation bandwidth, when we choose multimode waveguide instead of single mode. As mentioned above, the multimode waveguide for phase modulation is associated with larger value of V π L, which is a figure of merit for device length and drive voltage. The design of LN travelling wave modulator is to achieve a balance among drive voltage, modulation bandwidth and device length. In our design, we choose to optimize the value of BW/V π , which is a figure of merit for the modulation bandwidth and drive voltage, instead of V π L. Here we put more priority to modulation bandwidth compare to device length. For example, the V π L for 1-μm-wide waveguide and 4-μm-wide waveguide are 2.5 Vcm and 2.65 Vcm, respectively. However, the BW/V π for 1-μm-wide waveguide and 4-μm-wide are 17.2 GHz/V and 23.04 GHz/V.

What is the loss of the MMI?
Response: The measured excess loss of a LN MMI is ~ 0.054 dB. We added the following texts and The caption of Fig.2 is wrong. The EO line is green and the TO line is blue.
The Vpi description in Fig.3 caption is wrong. It should be 3.1 V instead of 7.5 mm. The wavelength data is (c) not (b).
Response: Thank you very much indeed, and sorry for the careless mistakes. We have corrected the caption of Fig.2 and Fig. 3 in the revised manuscript.
The intrinsic extinction ratio is around 25 dB. Is this a general phenomenon or a singular case? What is the limitation for this performance?

Response:
We have fabricated more than 10 devices on the same chip, and all of the measured extinction ratios are 24~28 dB. Therefore, we think this result is quite general.
The 25 dB extinction ratio we report in the main text is the value for the device which we used for the data transmission experiment.
The value of extinction ratio is limited by two factors: the power-splitting ratio of MMI and the loss imbalance between two arms of MZM. We have designed and fabricated another batch of device to measure the power splitting ratio of MMI, and the results confirmed that the power splitting ratio of the MMI and it is very close to 50:50 (Supplementary Note 2).
We believe the extinction ratio is currently limited by the unbalanced loss from two MZM arms. The lengths of the MZM arms are longer than 7.5 mm and 13 mm. Waveguides with lengths of this level are associated with unavoidable loss variations, resulting from fabrication imperfections like surface roughness and dimension fluctuations.
To clarify this, we add one sentence in the revised manuscript: "We have fabricated more than ten devices on the same chip, and the measured extinction ratios are between 24 to 28 dB." Regarding the s21 response, there are two large ripples at around 40 GHz and 56 GHz for both the 7.5-mm-long and 13-mm-long devices. What is the reason for this?

Response:
We believe the ripples around 40 GHz and 56 GHz are from the high-speed photodetector (Finisar 70 GHz XPDV3120), which we used for S21 measurement. We can always observe ripples around these two frequencies when we measured with this photodetector. The following figure shows the frequency response of at commercial modulator (purple line), which was measured with Finisar 70 GHz XPDV3120. The ripples appear at the frequency 38.5-44 GHz and 56-60 GHz (yellow regions).
In fig.5, it is mysterious that constellation diagram at 80 GBaud is better than that at 60 Gbaud, but the BER is worse? The constellation diagram is swapped or something else affects this?
Response: Thank you for your question. In the original manuscript, two different types of filters in the digital signal processing of optical modulation analyser (OMA) were used for 60 Gbaud (low-pass filtering) and 80 Gbaud (root raised cosine, RRC filtering) QPSK signals. The usage of RRC filtering for 80 Gbaud QPSK signal makes the constellation points appear more concentrated. This is the reason why the constellation diagram at 80 GBaud looks better than that at 60 Gbaud in the original manuscript.
In the revised manuscript, we use the same digital signal processing (RRC filters) for both 60 Gbaud and 80 Gbaud, and we update the results in Fig.5 in the revised manuscript.
The updated results of Fig.5b is as follows: For the 16-QAM measurement, the device is over driven from the constellation diagram.
Lower driving voltage is suggested to exploit the performance of the device, especially for Response: Thank you very much indeed for your very professional comment. We have fabricated a new batch of devices designed to achieve record low half-wave voltages while maintaining a high modulation bandwidth. The arm lengths of newly fabricated LNOI IQ modulator is 18 mm. The measured V π for the device is 1.25 V, corresponding to V π L of 2.25 Vcm, and the measured EO bandwidth is greater than 43 GHz. This performance metric is by far the best of its kind. We have updated the device information in the revised manuscript  The submitted paper quotes an optical insertion loss of 0.5dB/cm, while [1] quotes 0.5dB for the entire structure. The submitted paper does not report on the total insertion loss of the chip and in particular does not report the insertion loss due to propagation in proximity of the modulation or bias electrodes -it is important that these numbers be quoted as it is important that the electrodes do not interact strongly with the light for practical operation.
Response: Thank you very much indeed for your comment. The propagation loss of our LN waveguide was measured to be 0.3 dB/cm for single mode waveguide and 0.15 dB/cm for 4-um-wide waveguide. The data was obtained by measuring the Q factor of the micro-ring resonator co-fabricated on the same chip with the modulator devices. The details of how we measure the propagating loss in the revised supplementary information.
We carefully designed the gap between the waveguide and the metal electrode. The calculated metal induced loss is around 0.04 dB/cm, which is negligible. In order to make sure that the modulation and bias electrodes does not indeed incur any unwanted insertion loss, we fabricated micro-ring resonator with and without metal electrode. For the micro-ring resonator with metal electrode, the gap between the waveguide and the electrode are designed to the same value as that in the modulator device. The measured Q factors are almost identical for both cases, which indicates that the metal electrode dose not incur any absorption loss. We include the details in the Supplementary Note 3 and Supplementary We also give the detailed loss metric for the devices: In The submitted manuscript devotes a significant portion of the paper to reporting on the thermo-optic phase controllers. After a quick scan of the literature, I did not find any other report of thermo-optic switches on LNOI, so this may be innovative. However, the analysis done by the authors is not convincing. While electro-optic behaviour can be assessed based on the voltage required, the thermo-optic switches cannot be compared on voltage alone. The current must also be considered in order to assess the power draw. In our case, the z-cut surfaces are exposed by dry etching process and the distance between the exposed z-cut surface and the electrode is very close, as depicted in Fig. R1(a). According to ref.
[Broadband Optical Modulators, CRC Press 2011, page 350], the drift effect tends to become larger in modulator designs with larger exposed area of z-cut surface or closer distance between exposed z-cut surface and a center electrode. Therefore, the DC drift effect tends to be stronger than that in conventional LN modulators.
To study the drift effect in our device, we applied sawtooth modulation voltages of various frequencies to the 13-mm long modulator, and Fig. R1(b) and Fig. R1(c) shows the measured results for 10 Hz and 10 KHz, respectively. For a slowly-changed applied voltage, that is 10 Hz case, the response of the modulator shows fluctuations when the voltage changes or jumps abruptly (see the part identified by the dashed circle). We attribute this phenomenon to DC drift in LiNbO 3 as descript above.
For a fast-changed applied voltage, that is 10 KHz case, the fluctuations in the response curve disappear and influence of the DC drifts is suppressed. These results, together with the results we put in the manuscript (Fig.2 b), confirms that DC drift effect is there when applying a DC voltage or a slowly changed voltage. We would like thank you very much for your suggestions about writing a detailed paper about the thermal optic switches, and we are doing that exactly.
In the revised manuscript, we add three reference for the DC drift in LN modulators: Overall, the LNOI technology presented here does not, in my view, represent any significant advancement in the state of the art; however, the implementation of a IQ modulator may indeed be a world first, and perhaps this is significant enough to warrant publication in a high profile journal such as Nature Communications. However, in order to achieve this level, I think the focus needs to be placed more on the system demonstration that harnesses the IQ modulator (rather than the technology used to realise the chip itself) and it would need to be demonstrated that a record breaking transmission characteristic was achieved (such as modulation density, or power efficiency?).
Response: Thank you very much indeed for your insightful comment. We have added the following work in the revised manuscript in order to increase the impact of our paper.
1. We fabricated a new batch of devices with a longer modulation length (18mm) to achieve record low half-wave voltages while maintaining a high modulation bandwidth. The V π is 1.25V and the EO bandwidth is 43GHz, which surpass the value reported in Loncar's group.
Please find the details in the above text.

2.
In order to better compare TO phase shifter and EO phase shifter, we fabricated an LNOI-based IQ modulator with EO phase shifters for DC bias voltage control. The TO and EO phase shifter has both advantages and disadvantages. The TO phase shifter generates extra power consumption compared to the EO phase shifter, but the size is much compact and the bias control is more stabilized.  Figure 6 c-d) Supplementary Fig. 7)." I also note a number of typographical errors throughout the manuscript (for example "both branches of the 13-mm and 7.5-mm devices as a function of the applied voltage, showing Vπ of 1.9 and 7.5 mm, respectively." -the last number should be the Vpi of the 7.5mm electrode, not its length). There are several others.
Thank you very much indeed, and sorry for the careless mistakes. We have thoroughly proofread our manuscript.
[ showed a good linearity. Therefore, LNOI modulator has a potential to be a strong competitor with InP modulators.
2. For LNOI modulator, which fabrication routine will be adopted? This manuscript reported the first monolithic IQ (in-phase/quadrature) LNOI modulator used for short reach links, solved the DC drift problem using TO (thermal-optic) phase shifters.
The drawback of this work is that TO phase shifters needed electric power to operate, which increased the power consumption.
Response: Thank you very much indeed for the Referee's very professional comment. We totally agree. The advantages of the TO phase shifters are shorter device length and more stable DC bias, but the drawback is that it consumes power and it will age when subjected to extreme temperatures. We believe TO phase shifter and EO phase shifter both have their owe advantages and disadvantages. They can be useful in different application scenarios. For example, the TO shifter can save the device length for several millimeters, which is a great if people want to operate the IQ modulator in a very compact module, say, QSFP-DD. The EO shifter could be useful in scenarios where the device length is not a priority. In order to clarify this, we revised the manuscript in following way: 1. The figure 2.a is changed into a plot of transmission against power consumption of the TO heater (the original figure 2.a is a plot of transmission against applied voltage).
2. We add one sentence in the manuscript to compare the pros and cons of TO phase shifter and EO phase shifter.
3. We fabricated new IQ modulators with EO phase shifter and we put the details of the device and measurement results in the Supplementary Note 5 to provide a better comparison. We also added the sentences in the maintext to clarify this information.
In This work provided the strong evidences to answer the three questions mentioned above. It will be very interesting to the people working on the modulators.
A publication of the manuscript is highly recommended. However, the authors should response the following comments before the publication.
1. The authors can give some comments on the comparison on the modulators fabricated by hybrid or monolithic. Since the authors had the experience and insight on SOI-LNOI hybrid modulator (Nat. Photon. 13, 359, 2019), and this manuscript was about monolithic modulator, the authors' comments would be very useful to the readers.
Response: Thank you very much indeed for your suggestions.
The monolithic LNOI device can be regard as the next generation of the widely used conventional lithium niobate modulators. The performance of the LNOI devices is much better than the conventional counterpart, and more importantly, the footprint of the LNOI can fulfil the requirement of certain emerging application scenarios. Moreover, the manufacture of LNOI devices does not involve any complicated processes. Therefore, we believe LNOI device can find practical applications in the near future. In contrast, SOI-LNOI hybrid device is a technology for longer term. Silicon photonics are promised to create a radically new landscape for photonic integrated circuits. By harnessing the tool sets and process flows in CMOS foundries, the technology offers advantages of low-cost, high-volume and reliable manufacturing. However, the performance of silicon optical modulator is limited by the lossy and nonlinear carrier effect. This is where the SOI-LNOI hybrid modulators come in. By combining LNOI with silicon photonics, the hybrid platform allows for the combination of 'best-in-breed' active and passive components, offering the solutions for high-performance silicon based modulator. Therefore, the SOI-LNOI hybrid device is aimed for improving the silicon photonic platform. In the future, SOI-LNOI hybrid device has to find its way to the compatibility of CMOS process in the foundry before the its practical applications. In the future, this hybrid technology can combine 'best-in-breed' components, including thin-film lithium niobate modulators, III-V semiconductor, and silicon circuits, which is very important to address the cost and energy crunch.
To clarify this, we have added the following texts in the supplementary information for the revised manuscript:  Figure 1a).
Three identical 50-μm-long linear tapers are used as adiabatic tapers and connected to single-mode waveguides. Supplementary Figure 1b  3. The authors used TO phase shifter to suppress the DC drift, but the TO effect would increase the power consumption of the modulator. Can EO effect be used for DC drift control in LNOI modulator?
Response: Thank you for your question. The TO heater does increase the power consumption of the modulator. But low power consumption can also be realized in the TO heaters through removing substrate to provide thermal isolation. We can apply this technique to the LNOI platform.
Besides, we fabricated another chip with IQ modulator, and all DC biases were controlled by EO effects on the modulator. The bias voltages were applied to I and Q branches via bias-T, and the 3-mm-long EO phase shifter was used to introduce a π/2 phase difference between two sub-MZMs. It turns out that the EO effect can be used to control the LNOI modulator but we need to use bias control board to maintain a stable operation.  Figure 6 c-d). It