Low-Loss, Multi-Reticle-Stitched SiN Waveguides for 300 mm Wafer-Level Optical Interconnects

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors The authors report on  300-mm-sized wafer reticle-stitched LPCVD SiN waveguides, demonstrating a multi-reticle stitched ~56-cm-long waveguides across 20 reticles with propagation loss of 0.13~0.15dB/cm at lambda equals to 1310nm, as well as a  <0.001~0.002 dB SiN waveguide stitch loss due to <5 nm high-precision reticle lithography offset. The manuscript is well-written/organized. The contributions are relevant and suitable for publication.

Minor: Authors should fix grammar and typo issues starting from the abstract.

Author Response

The authors report on 300-mm-sized wafer reticle-stitched LPCVD SiN waveguides, demonstrating a multi-reticle stitched ~56-cm-long waveguides across 20 reticles with propagation loss of 0.13~0.15dB/cm at lambda equals to 1310nm, as well as a <0.001~0.002 dB SiN waveguide stitch loss due to <5 nm high-precision reticle lithography offset. The manuscript is well-written/organized. The contributions are relevant and suitable for publication. 

Comments 1: Minor: Authors should fix grammar and typo issues starting from the abstract.

Response 1: Thank you for pointing this out. We agree with this comment. We have revised the grammar and typo issues in the abstract. 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This article reports the fabrication of a 300 mm-sized wafer reticle-stitched SiN waveguides. With the help of long lopback waveguide architecture in standered SiN platform, the authors have demonstrated low propagation loss and stitched losses between multiple reticles. The article seems promising, but before publication, it requires a few improvements as suggested below.

1. The authors should clearly represent the wafers that are leveled with D as well as reticle connections. Also, what are the SiN dimensions for mask A and mask B? This section seems confusing.
2. Authors should add a comparison anlysis wth respect to SOI-based reticle-stitched wafers. The pros and cons of two different platforms. 
3. For interconnects, we normally look for reconfigurable circuits. Please add a discussion on reconfigurability with the help of PCMs or TO effect (doi.org/10.1109/JLT.2025.3583068,
   doi.org/10.1364/OE.448614). How can it be integrated to create large-scale reconfigurable interconnections following the proposed approach? What are the performance parameters in that case, etc.?   
4. While realizing a ring resonator, why is spiral waveguide architecture preferred in this study? Elaborate on this.

Author Response

This article reports on the fabrication of a 300 mm-sized wafer reticle-stitched SiN waveguides. With the help of long loopback waveguide architecture in standard SiN platform, the authors have demonstrated low propagation loss and stitched losses between multiple reticles. The article seems promising, but before publication, it requires a few improvements as suggested below.

Comments 1: The authors should clearly represent the wafers that are leveled with D as well as reticle connections. Also, what are the SiN dimensions for mask A and mask B? This section seems confusing. 
Response 1: We thank the reviewer’s comments. In the revised manuscript, we have included necessary information in Appendix 1 regarding the photolithography masks and information of wafer process splits. We hope this addresses the reviewer’s concerns. 

Comments 2: Authors should add a comparison anlysis wth respect to SOI-based reticle-stitched wafers. The pros and cons of two different platforms.
Response 2: We thank the reviewer for the comment. SOI offers greater advantages for the fabrication of active silicon devices (modulators and photodetectors). However, SOI-based silicon waveguides offer higher optical confinement but come with higher cost, higher propagation loss and greater sensitivity to reticle offset than silicon nitride alternative, making them less suitable for our envisioned WL-OIO application scenario. We have added more discussion and comparison on SOI-based reticle stitching and SiN-based reticle stitching in the introduction section.

Comments 3: For interconnects, we normally look for reconfigurable circuits. Please add a discussion on reconfigurability with the help of PCMs or TO effect (doi.org/10.1109/JLT.2025.3583068, doi.org/10.1364/OE.448614). How can it be integrated to create large-scale reconfigurable interconnections following the proposed approach? What are the performance parameters in that case, etc.?   
Response 3: Thank reviewer for the comments. To our best understanding, it is feasible to fabricate both PCMs or TO effect large-scale reconfigurable circuits, provided the process is CMOS-compatible. In this work, we demonstrate reticle stitching on 300 mm wafers using SiN waveguides for future high edge-bandwidth density wafer-level optical I/O. The results indicate reticle offsets of <5 nm, enabled by advanced DUV lithography tools. Without advanced lithography tools, typical reticle offsets are normally around 100~200 nm. However, through design optimization at the stitching interface, it should be feasible to achieve large-scale PCMs or TO effect circuit stitching and interconnections.

Comments 4: While realizing a ring resonator, why is spiral waveguide architecture preferred in this study? Elaborate on this.
Response 4: Thank reviewer for the comments. Previously we used waveguide spirals for stitch loss extraction, however, this approach requires a 5 times larger footprint. To monitor silicon photonic tape-out process flow, a compact ring resonators method is recommended to reduce the footprint and optimize area usage. Besides, the measurement of SiN grating couplers involves 0.5~1.0 dB optical power inaccuracies, which can lead to loss errors when using the spiral waveguide methods, while ring resonators normalize the optical power during internal loss data processing, thereby eliminating inaccuracies caused by grating coupler to fibre misalignment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript presents a low-loss, high-precision photolithographic stitching technology for silicon nitride waveguides on 300-mm wafers. The work represents a notable contribution toward enabling wafer-scale optical interconnects and is both comprehensive and innovative. However, several critical issues require clarification and further discussion:

1. It is recommended to supplement the key process parameters of LPCVD SiN, such as deposition temperature and pressure. These parameters impact waveguide loss and are essential for reproducibility. Additionally, the rationale for the ring resonator design parameters (bending radius, coupling gap) should be briefly clarified to ensure the device operates in the under-coupled regime for accurate loss extraction.
2. The specific implementation details of the Cut-back method for loss measurement (e.g., optical fiber coupling method, number of repeated measurements) should be briefly described to enhance the credibility of the results.
3. As shown in Figure 2d, the waveguide loss increases significantly after 1320 nm, which the text attributes to absorption by Si-H and N-H bonds. However, as an LPCVD process, the hydrogen content is typically already low. Please clarify whether the process flow includes a high-temperature annealing step. If so, why is there still significant absorption? If not, please explain in the discussion section the potential for further reducing long-wavelength loss in the O-band through annealing.
4. It is recommended to specify how the ultra-small stitching lithography offset of <5 nm for cross-wafer waveguides is achieved in the manuscript, and focus on explaining the reasons for the low stitching loss from multiple perspectives.
5. In Figure 5a, the bus waveguides for the stitched ring appear significantly longer/more complex than those for the reference ring. While the text implies the rings have an identical layout, this visual difference might suggest uncontrolled variables to the reader. Could the authors explicitly confirm that the ring resonators themselves (geometry, length, number of bends) are identical? Please also clarify that the difference in bus waveguide length does not affect the extraction of the internal loss factor (Q-factor).
6. The authors attribute the higher loss in 300-nm-thick waveguides (compared to 400-nm-thick ones) in Fig. 6 primarily to radiative loss induced by the "50-um-radius sharp bends". While it is intuitive that thinner waveguides have lower confinement and thus higher bending loss, the argument that they should exhibit lower scattering loss solely due to "smaller sidewall area" is debatable. As the waveguide thickness decreases to 300 nm, the optical mode becomes less confined. This delocalization could increase the modal overlap with the sidewall roughness, potentially leading to higher scattering loss despite the reduced physical sidewall area.Therefore, the observed higher loss could be dominated by scattering rather than bending radiation. To clarify this, please provide mode simulations calculating the theoretical pure bending loss (radiative loss) for the 300-nm waveguide at a 50-µm radius.
7. A performance comparison table is missing. It is suggested to tabulate and compare parameters such as wafer size, stitching loss, propagation loss, alignment precision, and process complexity of this work with those of recent representative literatures, and discuss the comparison in the manuscript.
8. In Appendix Table A1, the authors provide a detailed comparison between the proposed 300mm platform (accuracy <5 nm) and the conventional 200mm platform (accuracy 100 nm). Since this paper emphasizes "300mm wafer-level" and "high edge bandwidth density" as its core contributions, the limitations of existing 100 nm accuracy technologies are the fundamental motivation for this work. Placing this crucial comparison in the appendix weakens the flow of the narrative and obscures the quantitative advantages of the proposed technology. The difference in waveguide spacing is key evidence confirming the "high density" claim. It is recommended that the key parameters in Table A1 (especially the comparison of waveguide spacing, crosstalk, and bending radius) be moved to the introduction or discussion section. Explicitly discussing these improvements in the main text will significantly highlight the technological leap brought about by <5 nm accuracy and provide a basis for the claimed bandwidth density advantage.
9. The validation experiment described in Appendix 4, where a 20 nm offset was intentionally induced to verify the measurement sensitivity, is a critical "sanity check." It proves that the reported ultra-low loss is genuine. Recommend moving this validation from the Appendix to the main Results section.
10. The formats of the references are inconsistent. It is recommended to unify and standardize them in accordance with the reference format requirements of the journal to ensure the correctness of the format.

Author Response

This manuscript presents a low-loss, high-precision photolithographic stitching technology for silicon nitride waveguides on 300-mm wafers. The work represents a notable contribution toward enabling wafer-scale optical interconnects and is both comprehensive and innovative. However, several critical issues require clarification and further discussion:
Comments 1: It is recommended to supplement the key process parameters of LPCVD SiN, such as deposition temperature and pressure. These parameters impact waveguide loss and are essential for reproducibility. Additionally, the rationale for the ring resonator design parameters (bending radius, coupling gap) should be briefly clarified to ensure the device operates in the under-coupled regime for accurate loss extraction.  
Response 1: Thank reviewer for the comments, the LPCVD SiN are deposited at 770℃ temperatures, with 235 mTorr low pressure, and gas flow DCS:NH3 ≈ 80:280. The SiN deposition recipe and ring resonator design parameters are shown in Fig. r1 and Fig. r2, respectively. This information is included in re-submitted manuscript Appendix 1, we hope this addresses the reviewer’s concerns. 
 
Fig.r1 The LPCVD SiN layer is deposited at 770 °C under a low pressure of 235 mTorr, using a gas flow ratio of DCS (dichlorosilane, SiH₂Cl₂) to NH₃ (ammonia) of approximately 80:280.
 
Fig.r2 The SiN ring resonators design parameters for internal loss extraction.   

Comments 2: The specific implementation details of the Cut-back method for loss measurement (e.g., optical fiber coupling method, number of repeated measurements) should be briefly described to enhance the credibility of the results. 
Response 2: Thank reviewer for the comments, the measurement methods are described as in Fig. r3. The measurements are performed using optical fibers with automatic alignment script on a Cascade probe station. By probing optical fibers at ①~⑤ different wafer locations, the corresponding optical losses for the 8.3 cm, 20.2 cm, 32.1 cm, 44.1 cm, and 56.0 cm multi-reticle stitched loopback waveguides can be measured. Detailed information is also provided in the revised manuscript.
 
Fig.r3 Example showing the use of the cutback method to measure cross-wafer SiN waveguide loss (right y-axis), by probing optical fibers at ①~⑤ different wafer locations (left y-axis).

Comments 3: As shown in Figure 2d, the waveguide loss increases significantly after 1320 nm, which the text attributes to absorption by Si-H and N-H bonds. However, as an LPCVD process, the hydrogen content is typically already low. Please clarify whether the process flow includes a high-temperature annealing step. If so, why is there still significant absorption? If not, please explain in the discussion section the potential for further reducing long-wavelength loss in the O-band through annealing.  
Response 3: Thank reviewer for the comments. We observed a slight increase in loss at longer >1320nm wavelengths, and Fourier transform infrared (FTIR) spectroscopy confirmed the presence of weak N–H absorption peaks (residual hydrogen atoms, in Fig. r4). This is mainly attributed to the relatively lower LPCVD deposition temperature of 770 °C. According to some studies, increasing the deposition temperature to above 1000 or 1100 °C can effectively reduce the residual hydrogen atoms in the dielectric and achieve lower absorption losses.
 
Fig.r4 FTIR spectrum inspection showing a slight N–H bonds were detected in the LPCVD SiN deposited at 770 °C.

Comments 4: It is recommended to specify how the ultra-small stitching lithography offset of <5 nm for cross-wafer waveguides is achieved in the manuscript, and focus on explaining the reasons for the low stitching loss from multiple perspectives. 

Response 4: Thank reviewer for the comments. Achieving an accuracy of <5 nm requires advanced DUV lithography equipment, which is one of the disadvantages in the reticle stitching experiment. Since advanced DUV lithography equipment is capable of fabricating chips at the 14–7 nm technology nodes, both overlay accuracy and wafer stage repeatability are typically high. Therefore, achieving 5 nm accuracy is both feasible and demonstrable. Besides, advanced DUV lithography equipment helps improve the stability of phase-sensitive passive devices (such as Mach–Zehnder Interferometers), which are becoming increasingly common in 300 mm silicon photonics manufacturing. 
Comments 5: In Figure 5a, the bus waveguides for the stitched ring appear significantly longer/more complex than those for the reference ring. While the text implies the rings have an identical layout, this visual difference might suggest uncontrolled variables to the reader. Could the authors explicitly confirm that the ring resonators themselves (geometry, length, number of bends) are identical? Please also clarify that the difference in bus waveguide length does not affect the extraction of the internal loss factor (Q-factor) 
Response 5: We appreciate the reviewer’s feedback. The ref ring and stitch ring share an identical layout design; therefore, all specifications should be identical. We have added a design summary table and discussion in Appendix 1 to address the reviewer’s concerns. 
The difference in bus waveguide length does not affect the extraction of internal loss, because power normalization is applied during ring resonator data processing. So, after data processing, the loss in the bus waveguide does not contribute to the internal loss extraction.

Comments 6: The authors attribute the higher loss in 300-nm-thick waveguides (compared to 400-nm-thick ones) in Fig. 6 primarily to radiative loss induced by the "50-um-radius sharp bends". While it is intuitive that thinner waveguides have lower confinement and thus higher bending loss, the argument that they should exhibit lower scattering loss solely due to "smaller sidewall area" is debatable. As the waveguide thickness decreases to 300 nm, the optical mode becomes less confined. This delocalization could increase the modal overlap with the sidewall roughness, potentially leading to higher scattering loss despite the reduced physical sidewall area. Therefore, the observed higher loss could be dominated by scattering rather than bending radiation. To clarify this, please provide mode simulations calculating the theoretical pure bending loss (radiative loss) for the 300-nm waveguide at a 50-µm radius.  
Response 6: We performed several device tests to measure SiN waveguide bending loss with a 50 µm radius (includes both radiation loss and sidewall scattering loss), the bending loss is shown in Fig. r5. 300-nm-thick SiN waveguide bend loss is 0.008-0.015 dB, 400-nm-thick SiN waveguide bend loss is 0.004-0.008 dB. The 300 nm-thick SiN waveguide exhibits roughly twice the bend loss compared to the 400 nm-thick SiN waveguide. Therefore, when the number of bends becomes sufficiently large, this results in a higher internal loss in the 300-nm-thick SiN ring resonators.
 
Fig.r5 The measured 50-µm-radius 90-degree bend loss (includes both radiation loss and sidewall scattering loss) we measured from both 300-nm-thick and 400-nm-thick LPCVD SiN.

Comments 7: A performance comparison table is missing. It is suggested to tabulate and compare parameters such as wafer size, stitching loss, propagation loss, alignment precision, and process complexity of this work with those of recent representative literatures, and discuss the comparison in the manuscript.  
Response 7: We appreciate the reviewer’s comments. In the resubmitted manuscript, we have added a summary table in Appendix 1 that summaries all information related to our LPCVD SiN, we hope addresses the reviewer’s concerns.

Comments 8: In Appendix Table A1, the authors provide a detailed comparison between the proposed 300mm platform (accuracy <5 nm) and the conventional 200mm platform (accuracy 100 nm). Since this paper emphasizes "300mm wafer-level" and "high edge bandwidth density" as its core contributions, the limitations of existing 100 nm accuracy technologies are the fundamental motivation for this work. Placing this crucial comparison in the appendix weakens the flow of the narrative and obscures the quantitative advantages of the proposed technology. The difference in waveguide spacing is key evidence confirming the "high density" claim. It is recommended that the key parameters in Table A1 (especially the comparison of waveguide spacing, crosstalk, and bending radius) be moved to the introduction or discussion section. Explicitly discussing these improvements in the main text will significantly highlight the technological leap brought about by <5 nm accuracy and provide a basis for the claimed bandwidth density advantage.  
Response 8: Thank you for your comment. We acknowledge the suggestion and have relocated that appendix 1 to the main text.

Comments 9: The validation experiment described in Appendix 4, where a 20 nm offset was intentionally induced to verify the measurement sensitivity, is a critical "sanity check." It proves that the reported ultra-low loss is genuine. Recommend moving this validation from the Appendix to the main Results section. 
Response 9: Thank you for your comment. We acknowledge the suggestion and have relocated that appendix 4 to the main text.

Comments 10: The formats of the references are inconsistent. It is recommended to unify and standardize them in accordance with the reference format requirements of the journal to ensure the correctness of the format.  
Response 10: Thank reviewer for the comment. We have checked the formatting of the references and made effort to ensure they are correct and comply with the journal’s requirements.

Author Response File: Author Response.docx