VCSEL Light Coupling to a Waveguide to Interconnect XPUs and HBMs on Interposer Chips

Author Response to Reviewer 1 Comments

Thank you very much for taking the time to review our manuscript. Please find the detailed responses below and the corresponding revisions and corrections highlighted in the re-submitted files.

Point-by-point response to Comments and Suggestions for Authors

The authors propose a novel method for coupling light from multimode VCSELs into both conventional and cavity-enhanced SiO₂/Si₃N₄ optical waveguides that they fabricate via a CMOS-compatible PECVD process. To this end, they develop a coupling technique using a GaP prism and polyimide micro-lenses. The technologies required to employ the prism coupling technique include an adhesive technology, a stand-alone polyimide microlens, and a packaging technology. The authors successfully develop these steps within their associated requirements. Developing such technologies is an important step towards more efficient photonic integration, which is expected to play a major role in future datacom applications.

The technology to make the micro-lenses seems interesting, fairly easy to implement, and efficient. Although the losses are still a bit high, the authors demonstrate improvement compared to previous demonstrations of single-mode VCSEL coupling into single-mode silicon waveguides. Moreover, the authors demonstrate through simulations the amount by which the coupling losses could be further reduced upon application of HR/AR coatings in combination with the prism.  Notably, their simulations also show that their design enables both TE and TM modes to be utilized via their prism coupling and thin film WDM filters, which is not generally the case in SiP. These results are promising and could lay the foundation for future work.

Overall, I believe the paper is of good quality and the topic makes a valuable contribution to the ongoing efforts toward efficient integration of laser light with passive waveguides and CMOS-compatible technologies.

The paper is generally well-written and clear, though there are a few instances where improvements could be made to enhance clarity. Therefore, I recommend publication in Photonics after the following improvements are implemented:

Comment 1: ) The information given about the prism is very limited. Since the prism plays a pivotal role in their approach, the authors should provide additional information about it. For example in Line 153: some more details would be interesting: How big is the prism, how was it made, does the polishing refer to the prism or the adhesive part/wafer? which one of the two parts was polished, and using which technique?

Response 1: First of all, we sincerely thank Reviewer 1 for carefully reading our manuscript and for the thorough understanding of its content. We also appreciate valuable Comment 1. In response, we have added a new paragraph (highlighted in red) below Line 169.

Comment 2: and similarly in line 159 it is not clear which are the two thermoset polyimide layers referred to. Overall I would suggest a rewriting of this paragraph to enhance clarity, especially for readers who are not already familiar with their techniques.

Response 2: We agree with the reviewer’s suggestion and have revised the manuscript accordingly by adding sentences between Lines 181 and 188. We hope that the added paragraphs meet the reviewer’s expectations.

Comment 3: I would also recommend that the authors add a few more references in the introduction for the sake of completeness.  

1. a) For example, in line 53- the authors describe that “the research focus has shifted to co-packaged transceivers, which interconnect switches to servers or to other switches in data centers using silicon photonics-based technologies since mid-2010s,” but no references are provided. For example, Ref. K. Papatryfonos, et al., "Co-Package Technology Platform for Low-Power and Low-Cost Data Centers," Appl. Sci. 11, 6098, (2021), outlines the evolution of co-packaged transceivers and presents the initial steps of a joint effort among several companies and universities to develop a common platform toward this goal. So, this and/or other references could serve to fill this gap.
2. b) Also, the authors only discuss VCSEL lasers in their introduction as a promising technology for silicon photonics and interposer chips. However, edge-emitting lasers directly grown on silicon also deserve a mention as they recently demonstrated very low threshold current, higher output power, and excellent temperature tolerance. (see K. Papatryfonos, et al., “Low-Defect Quantum Dot Lasers Directly Grown on Silicon Exhibiting Low Threshold Current and High Output Power at Elevated Temperatures,” Adv. Photonics Res. 2400082 (2024) and W. Q. Wei, A. He, B. Yang, et al., “Monolithic integration of embedded III-V lasers on SOI,” Light Sci Appl. 12, 84 (2023).) It is perfectly understandable that the authors may prefer VCSEL technology which has its advantages, but I would suggest adding a short comment to mention the alternative technology as well for the sake of completeness and a more balanced introduction.

Response 3: Thank you for the insightful comment and for suggesting the reference to support the sentence, “the research focus has shifted to co-packaged transceivers …[8].” We also agree with the achievements in light sources highlighted by the two references. In accordance with the comments, we have added all three references in the Introduction, along with the following sentence in Line 58: “In recent years, advancements in light sources for co-packaged optics (CPO) and silicon photonics have been reported [9, 10].”

Comment 4: Lines 279-281: How was this calibration made? This is not clear to me; was the VCSEL power calibrated previously?

Response 4: We agree that the sentence may cause confusion. For clarity, we have added the phrase: “which represents not absolute power but relative values to compare measurements in mW.” To measure the absolute power of the VCSEL in mW, we injected light from a calibrated source (a fiber-coupled 850 nm multimode commercial laser) into our photodiode and measured the output in mV. We found that 1100–1200 mV corresponds approximately to 1 mW; however, each photodiode exhibited slightly different values within this range. Since our calibration is not highly accurate, we present the results only as relative values in mW for comparison. We hope this clarification addresses the reviewer’s concern.

Comments 5: Fig. 6b has no caption, please add it.

Response 5: Thank you for the careful reading. We changed from “Figure 6. The scheme to measure powers emitted at the edge (a) and through the prism (b), (c) is the micrograph of PD.” to “The scheme to measure powers (a) emitted at the edge and (b) through the prism. (c) is the micrograph of PD.”

Comment 6: Table 4: both columns refer to structure 1 while in the text both structure 1 and 2 losses are discussed, so perhaps there is a typo to be corrected?

Response 6: Thank you very much for finding a wrong typing. The second column was denoted as Structure 2.

Comment 7: Line 312: I didn't understand how was the ratio 12/13 of collection efficiencies between prism and waveguide decided.

Response 7: From Table 5, the average values of transmitted light measured at the waveguide edge and through the prism are 0.012 mW and 0.013 mW, respectively. Therefore, the ratio is 12/13. We hope this addresses Comment 7 appropriately.

Comment 8: Line 317: Also, how was the propagation loss of 0.466 dB for the 10.5 mm waveguide derived? I suggest adding a comment to make these easier to understand.

Response 8: The average propagation loss was measured 0.444 dB/cm. The propagation loss of 0.466 dB for the 10.5 mm waveguide was calculated from 0.444 dB/cm X 10.5 mm. According to the comment, we added “The loss of 0.466 dB was calculated from the average propagation loss of 0.444 dB/cm and the waveguide length of 10.5 mm (0.444 dB/cm × 10.5 mm).” We wish this would be a right answer to the comment 8.

Comment 9: Line 342, please correct the typo.

Response 9: Thank you very much for finding a wrong typing. We corrected it.

Comment 10: Lines 437-440: The authors mention that “the primary advantage of VCSELs is that their light is emitted perpendicular to the wafer surface, enabling laser characterization at the water level without the need to dice the wafer into individual chips. This advantage similarly applies to the fabrication of photonic interposer chips utilizing VCSELs.” I would like to note that the reference I suggested above for inclusion in the introduction [K. Papatryfonos, Appl. Sci. 11, 6098 (2021)] also presents a technique that enables the use of edge-emitting lasers with photonic interposers without the need to dice the chips (the integration of grating couplers in each laser cell permits that). While it is true that perpendicular emission is potentially an advantage for VCSELs, I believe the authors should acknowledge this point, as their current wording may give the impression that only VCSELs are compatible with this feature.

Response 10: The reviewer introduced us to a high-quality research article that we had not read prior to preparing this revised manuscript. We fully agree that the grating couplers in Figure 5 enable wafer-level characterization of edge-emitting DFB lasers without the need to dice the wafer into individual chips. Accordingly, we have added the phrase suggested in the comment: “or specially arranged grating couplers for edge-emitting lasers [8].” In Line 472

Comment 11: In Fig. 5, what are the spots that appear throughout the structure? Are they important for affecting the losses? If so, please add a short comment to discuss them.

Response 11: We believe that the spots on the SEM images of cavity-type waveguides came from the preparing process of samples. The facets of the samples (i.e., pieces of wafer) were polished after dicing the wafer and subsequently treated with BOE solution. We suspect that polishing powders remained on the samples despite multiple cleaning and rinsing steps. We sincerely thank the reviewer for the careful examination and thoughtful review.

Author Response to Reviewer 2 Comments

Thank you very much for taking the time to review our manuscript. Please find the detailed responses below and the corresponding revisions and corrections highlighted in the re-submitted files.

Point-by-point response to Comments and Suggestions for Authors

The manuscript ID photonics-3798816 has been devoted to mainly present a study coupling of multimode VCSEL light to a non-silicon waveguide prepared by using a CMOS-compatible process. Different VCSEL-based interconnections for signals modulation were analyzed. Please see below a list of comments to the author:

Comment 1: The polarization of the system can be neglected in the performance of the modulation?

Response 1: We thank the reviewer for the insightful comment regarding the polarization issue affecting the performance of the modulated signal. Modal dispersion of TE and TM modes can occur in our system because the waveguide is asymmetric between the horizontal and vertical directions. In future work, the impact of modal dispersion on high-speed transmission through waveguides of tens of centimeters in length should be analyzed both experimentally and theoretically. If it is found to be detrimental, modal dispersion can be avoided by designing the waveguide to be symmetric between horizontal and vertical directions by tapering the waveguide width in the horizontal direction.

In the present study, we focused on the coupling of multimode VCSEL light into a non-silicon waveguide. We would greatly appreciate it if the reviewer could evaluate the results of our study, which, to the best of our knowledge, demonstrate for the first time the injection of multimode VCSEL light into a non-silicon waveguide fabricated using a CMOS-compatible process, with coupling and transmission losses relevant for the realization of commercial photonic interposer chips.

In response to this comment, we have added a paragraph in the discussion (Lines 533–539), highlighted in red.

Comment 2: Please comment about the influence of the size and shape of the VCSEL over the main findings.

Response 2: We agree that Comment 2 highlights a point that readers may frequently question. In response, we have added a paragraph in the introduction (Lines 78–82), highlighted in red: “Considering the typical dimensions of a VCSEL and a photodiode (250 µm × 250 µm in area and 150 µm in height), 64 channels require 64 VCSELs and 64 photodiodes, corresponding to a total area of approximately 8 mm². This is much smaller than the area of a typical processor chip (>100 mm²). Since a prism coupler can be fabricated as small as a VCSEL, the area required for photonic chips should not pose a limitation.”

Comment 3: A citation is missing for the data reported in table 3?

Response 3: We thank the reviewer for the careful examination. All data in Table 3, except for the coefficient of thermal expansion, were obtained from our measurements. The coefficient of thermal expansion was taken from the product specification provided by the polyimide manufacturer. Therefore, we believe that no additional references are required for Table 3.    

Comment 4: Table 8 presents numerical results; experimental data are requested.
Response 4: Since there is no Table 8 in the manuscript, the reference should be to Table 6. In future work, we plan to apply AR coatings to the prisms and measure the transmission through the waveguide. In the present manuscript, our aim is to demonstrate the theoretical feasibility of the prism. We hope the reviewer understands this objective.     

Comment 5: The use of machine learning and neural networks could be considered for improving the design of the system proposed? Please see for instance:  https://doi.org/10.1002/lpor.202401636
Response 5: We thank the reviewer for recommending a useful reference, which will be valuable for designing photonic circuits on an interposer chip. In future work, we plan to make effective use of transformer neural networks. This paper is cited in our manuscript as: “VCSELs and their properties can be designed and predicted by using machine learning-based transformer neural networks [26].”

Comment 6: Perspectives can be added. The authors are invited to see for instance the potential use of optical nonlinearities for developing all-optical systems: https://doi.org/10.1007/978-3-031-10824-2
Response 6: We thank the reviewer once again for recommending this valuable reference. In future work, we plan to make effective use of the insights presented in this literature.

Comment 7: Advantages and disadvantages of the system proposed can be summarized in the discussion section.

Response 7: Please refer to the Discussion section for the advantages and disadvantages of our system. In the first three paragraphs, we describe the advantages of our VCSEL-based photonic interconnection system, while the last three paragraphs discuss its disadvantages. In response to the reviewer’s comment, we have added a paragraph in the Discussion (Lines 534–540) highlighting an additional disadvantage, in relation to Comment 1.We hope that our descriptions meet the reviewer’s expectations.

Comment 8: An important change required is to present as separated sections the discussions and the conclusions. The discussions should confront the main findings with updated publications in the topic, the conclusions should highlight the importance of the main findings.
Response 8: We agree the advice of comment 8. In response to the comment, we added a “5 Conclusion” separated from “4 Discussion”

Comment 9: Error bar in experimental data are mandatory..
Response 9: We agree with the reviewer’s opinion. In our manuscript, no graphs are drawn from experimental data. Figure 8 represents the theoretical prediction of WDM filter spectra, which are generally not shown with error bars. We hope the reviewer will take this point into consideration. 

Comment 10: In my opinion, the text could be improved if the citations presented in collective form could be Split in order to easily justify each reference selected for the presentation of the topic. 
Response 10: We agree with the reviewer’s suggestion. In response to this comment, the collectively presented citations have been split, with brief descriptions added for each individual reference.

Thank you very much for taking the time to review our manuscript. Please find the detailed responses below and the corresponding revisions and corrections highlighted in the re-submitted files.

Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Must be improved

Please refer to the revised manuscript.   We have made our best efforts to address the reviewers’ comments and improve the manuscript accordingly.

Is the research design appropriate?

Can be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the methods adequately described?

Must be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the results clearly presented?

Must be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the conclusions supported by the results?

Are all figures and tables clear and well presented?

Must be improved

Must be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Once the manuscript is accepted, they will be replaced by the original high-resolution figures.

Point-by-point response to Comments and Suggestions for Authors

Comments 0: The paper is poorly written in some parts, especially in the introduction and experiments sections where the authors make incorrect and vague statements.

Response 0: Thank you for the comment. If this introductory comment pertains to comments 1 through 5 below, we hope that our responses are appropriate and represent our best possible replies.

Comments 1: 

the statement "For a directly modulated VCSEL, butt coupling can only generatesignals at the edges of the chip. Therefore, it is impossible to generate signals at pointsthat are not located at the edges. " is incorrect. It's not clear what authors are trying to convey here.

Response 1: Thank you for the comment. Butt coupling refers to a configuration in which the top surface of the VCSEL is in direct contact with the end facet of an optical waveguide, enabling the injection of light from the VCSEL into the waveguide. Given the size of the VCSEL's top surface, this facet-to-facet contact is feasible only at the edge of the chip, where optical waveguides are formed on the photonic interposer chip. As a result, signals from a directly modulated VCSEL can be generated only at the chip edges, and it is not possible to generate such signals at locations that are not situated along the chip's periphery.

We have revised the manuscript accordingly by adding additional sentences from the line number 110 to 116. We wish that the added paragraphs are close to what the reviewer has expected.

Comments 2: The statement "For grating coupling, the difference in refractive index between silicon 
oxynitride and silicon dioxide is not only small (Δn = 0.1 - 0.55), which leads to a lowcoupling rate, but also creates statistical errors such as ring resonators". uses vague terminology. Please use "coupling efficiency" if the intent was to convey the loss due to low confinement.  

Response 2: We agree that “coupling efficiency” is a more appropriate term than “coupling rate,” and we have revised the text accordingly in response to the comment.  

Comments 3: "statistical errors such as ring resonators" what is this sentence trying to convey? Fabry perot effect due to reflections?

Response 3: Thank you for the comment. Here, statistical errors refer to fabrication-induced variations in the resonance wavelengths of ring resonators. In the processes of photolithography and dry etching, the width and height of the waveguides may slightly deviate from their target or nominal values. These deviations exhibit statistical behavior depending on the fabrication process. Considering mass production using 200 or 300mm wafers, and given the involvement of photolithography and dry etching, deviations or variations in key parameters—such as resonance wavelength or coupling efficiency—are inevitable. These variations can occur both across a single wafer (from center to periphery) and between different wafers. Since the fabrication of grating couplers also involves photolithography and dry etching, fabrication-induced variations are likewise unavoidable.

However, to avoid confusion, we removed the phrase “statistical errors such as ring resonators” and revised the sentence to: “For grating coupling, the difference in refractive index between silicon oxynitride and silicon dioxide is small (Δn = 0.1–0.55), which leads to a low coupling efficiency.”

Comments 4: 

"Microlenses consisting of quartz and silicon materials arecommercially available, but their refractive indices are not suitable for current purposes." This sentence is not clear. Please mention the purpose to make it clearer for the readers.Response 4: We agree with reviewer’s comments. In response to the comment, we revised the sentence to explain why quartz and silicon materials are not suitable as below.

“Microlenses made of quartz and silicon are commercially available, but their refractive indices (n₁=1.45 for quartz and n₁=3.45 for silicon) are not suitable for the current application. When a commercial adhesive with a refractive index of n₂=1.4~1.5 is used, the focal length calculated by the equation f = R·n₁ / (n₁ – n₂), where R is the microlens radius (15 μm), becomes either too long (> 435 μm for quartz) or too short (< 26.6 μm for silicon) for practical use.”

Comments 5: THe text in figures 1a, 2a,2b, 4a is unreadable..please increase the font size or improve the image quality

Response 5: Thank you for the comment. Once the manuscript is accepted, they will be replaced by the original high-resolution figures. As far as we know, the PDF file sent to the reviewers is not the final version.

Comments 6: The plots in Fig 7 are also undicernable..... axis labels are not clear.

Response 6: Thank you for the comment. Once the manuscript is accepted, they will be replaced by the original high-resolution figures. As far as we know, the PDF file sent to the reviewers is not the final version.

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Please refer to the revised manuscript.   We have made our best efforts to address the reviewers’ comments and improve the manuscript accordingly.

Is the research design appropriate?

Can be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the methods adequately described?

Can be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the results clearly presented?

Can be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Are the conclusions supported by the results?

Are all figures and tables clear and well presented?

Can be improved

Must be improved

We have done our best to revise the manuscript in response to the reviewers’ comments.

Once the manuscript is accepted, they will be replaced by the original high-resolution figures.

Point-by-point response to Comments and Suggestions for Authors

Comments 1: The abstract mentions "The experimental coupling of multimode VCSEL light to a non-silicon waveguide fabricated using a CMOS-compatible process is demonstrated." However, the specific process and key steps of the experiment, such as the preparation method of the waveguide, the coupling method, etc., are not described in detail, and it is suggested to briefly mention this key information in the abstract so that readers can understand the core of the paper more quickly. 

Response 1: Thank you for pointing this out. We agree with this comment. According to the comment, we added a sentence in the abstract.

“The GaP prism was tested and adopted as a coupling method. Both conventional and cavity-type optical waveguides, fabricated from CMOS-compatible PECVD SiO₂, Si₃N₄, and SiO_xN_y materials, were evaluated.” We wish that this is close to what the reviewer has expected.

Comments 2: In the introduction, the development history of VCSEL is introduced in more detail, but the overall technical background of the optical interconnect chip is introduced relatively little, and the lack of in-depth analysis of the status and challenges of its application in data centers, high-performance computing and other fields fails to highlight the importance and urgency of this research.

Response 2: Thank you for catching an important point in introduction. We agree that the point will make our statements more powerful. In response to the comment, we have added the paragraphs from line number 46 to 75 highlighted by red and removed the paragraph below to avoid duplication. We wish that the added paragraphs are close to what the reviewer has expected. 

Recently, co-packaged optics (CPO), which interconnect switches to servers or to other switches, have been actively developed by companies, and commercial deployment is expected to begin in 2026 [4]. The goal of optical interconnects is to enable XPU-to-XPU and XPU-to-HBM communication via photonic interposer chips, with commercialization expected between 2028 and 2030.

Comments 3: Although the paper describes the various components and techniques used in the experiments, it does not go deep enough in some details, for example, it lacks detailed elaboration on the preparation process and the performance optimization process of the PI microlens, and simply gives its parameters and performance, without sufficient explanation on how to achieve these performances and how to ensure its compatibility with VCSEL.

Response 3: Thank you for the insightful comment. We agree with the suggestion and have revised the manuscript accordingly by adding explanatory sentences from the line number 155 to 162 and from 171 to 186. We wish that the added paragraphs are close to what the reviewer has expected. 

Comments 4: There are no detailed instructions on how to accurately control and measure the alignment and coupling process between the VCSEL and the waveguide, and how to avoid errors and interferences in the experiment, which may affect the accuracy and reproducibility of the experimental results.gh

Response 4: We agree with the comment and have revised the manuscript accordingly by adding explanatory sentences from the line number 228 to 238 and from 277 to 280. We wish that the added paragraphs are close to what the reviewer has expected. 

Comments 5: In the modeling of WDM filters and AR coatings, although theoretical calculations and related graphs are given, the method of modeling, assumptions, and parameter selection are not sufficiently explained, and readers may have difficulty in understanding the accuracy and applicability of the model.

Response 5: We agree with the comment and have revised the manuscript accordingly by adding additional sentences from the line number 381 to 392 and from 411 to 412. We wish that the added paragraphs are close to what the reviewer has expected. 

Comments 6: Although the advantages of VCSEL optical coupling to non-silicon waveguides are explained in the discussion section, the problems and challenges that the technology may face in practical applications, such as cost, complexity of the manufacturing process, and compatibility with the integration of other chip components, are not analyzed and discussed in depth, which makes the discussion not comprehensive and in-depth enough.

Response 6: Thank you for the insightful comment. We agree with the comment and have revised the manuscript accordingly by adding additional sentences from the line number 497 to 513. We wish that the added paragraphs are close to what the reviewer has expected. 

Comments 7: There are some language expressions in the paper that are not accurate and fluent enough, such as "The goal of optical interconnects is to enable XPU-to-XPU and XPU-to-HBM communication via photonic The goal of optical interconnects is to enable XPU-to-XPU and XPU-to-HBM communication via photonic interposer chips, with commercialization expected between 2028 and 2030. between 2028 and 2030" in this sentence is not clear enough and can be easily misunderstood. It is recommended that the linguistic expression of the whole text be checked in detail.

Response 7: Thank you for reviewing the English. Please refer to the changes highlighted in blue in the revised manuscript. We have done our best to revise the manuscript in response to the reviewers’ comments. Prior to the first submission, the manuscript was proofread by a native English speaker who is not an expert in this field, and was further checked using ChatGPT. However, we acknowledge that the English was
 still insufficient. If necessary, we will seek professional English editing after acceptance.

Comments 8: All the pictures in the paper are too low in clarity, and the key text in the pictures is blurred, which seriously affects the readability. It is recommended that the pictures in the paper be replaced by high-definition pictures. In addition, there are two “Figure 6” in the thesis, and it is recommended to check them in detail.

Response 8: We deeply appreciate the reviewer for thoroughly reviewing not only the sentences but also the figures. Once the manuscript is accepted, they will be replaced by the original high-resolution figures. As far as we know, the PDF file sent to the reviewers is not the final version.