Silicon Nitride Waveguides Explained | Photonics

Back to Technology, Institution & Company Guides

TECHNOLOGY EXPLAINER

Silicon Nitride Waveguides
— a light path you can run for a metre is decided by deposition temperature, hydrogen and stress

Silicon nitride (SiN, Si₃N₄) is a familiar insulating and passivation film in semiconductor fabs. Make a light path from that same material and the loss can come down to around 1 dB per metre, a level other integration materials cannot reach. Getting there, though, means driving hydrogen out with heat treatment above 1,000 °C and preventing stress cracking in thick films. Both the source of the low loss and the cause of the difficulty lie in how the film is deposited.

Built from primary sources: the JePPIX roadmap, an EPFL paper in Nature Communications, a UCSB paper in Light: Science & Applications, a perspective in Nature Communications, and a TU Eindhoven introduction to InP integration / Last updated September 2026

Conceptual image of a dark round substrate on whose surface a thin, pale line wound many times into a spiral is faintly visible through differences in reflection — **Conceptual image (AI-generated).** An impression of a long waveguide coiled inside a small chip. **It does not show the real number, length or width of waveguides, or the design of any company or research institution.**

What this article covers

Silicon nitride waveguides in three points
Why SiN — five properties other materials lack
Thin cores and thick cores
Our calculation: how many orders of magnitude apart are the losses?
Deposition: LPCVD, PECVD and the thermal budget
A materials engineer's view (1): the loss is hydrogen
A materials engineer's view (2): thick films crack — how to relieve the stress
Applications — what it is used for
What is still hard
Glossary / References / Claim-to-source audit

How claims are labelled in this article

Sourced = stated in published material or a peer-reviewed paper (link given)
Our calculation = a figure this article derived, with the assumptions spelled out
Not yet confirmed = a plan or research-stage result with no confirmed production record
Structural readings and materials-design interpretations are marked separately as Commentary. Loss figures are kept apart as values from research papers (prototypes) or ranges given in a roadmap.

1. Silicon nitride waveguides in three points

A silicon nitride waveguide is a light path in which a narrow strip (core) of SiN is surrounded by SiO₂, confining light inside it and carrying it along. Liu and colleagues at EPFL (2021) give the material refractive index of Si₃N₄ as 2.0Sourced. That is higher than SiO₂ (about 1.44) and lower than silicon (about 3.47) — a medium-index material (the SiO₂ and Si values are from sources listed below).

What it does well: the JePPIX roadmap (2021) describes SiN as the material that can achieve the lowest optical loss of any integration platform, citing a broad wavelength range of about 400 nm to 2,500 nm and propagation losses from 0.1 dB/cm down to 0.1 dB/mSourced
What it cannot do: the same roadmap notes that active devices are not native to the SiN platform, and that lasers and high-speed detectors and modulators are added through hybrid integration with InP and similar materialsSourced
What makes it hard: low loss requires high-temperature heat treatment (to remove hydrogen) and measures against stress in thick films. Bose and colleagues at UCSB (2024) write that the conventional LPCVD process requires deposition at up to 850 °C and annealing at 1,150 °CSourced

Where this sits in the series

The use of SiN in silicon photonic circuits for couplers to the fibre and for handling high optical power was touched on in our explainer on silicon photonics. A comparison with InP, silicon and thin-film lithium niobate is in our explainer on photonic integrated circuits and indium phosphide. Wavelength demultiplexing and multiplexing with ring resonators are covered in our explainer on wavelength multiplexing and microrings, and LiDAR applications in our LiDAR explainer. This article focuses on how SiN films are made and how that sets the loss.

2. Why SiN — five properties other materials lack

Property	What the primary sources say
1 Extremely low loss	Liu et al. write that losses of around 1 dB/m or below have been demonstrated only in Si₃N₄ among all integration platformsSourced
2 A wide usable wavelength range	JePPIX gives about 400 to 2,500 nm, serving applications out of reach of silicon photonics and InPSourced
3 Tolerates intense light	Liu et al. cite the absence of two-photon absorption at telecommunication wavelengths and very high power handling. Shekhar et al. likewise write that SiN on the PIC can be used to handle high optical powerSourced
4 Small phase errors	JePPIX notes that the modest index contrast gives very small phase errors, making optical path performance predictableSourced
5 Already in the fab	Shekhar et al. note that, alongside Ge detectors, SiN has become a standard offering at several foundries, used to extend wavelength range and for high optical power, low loss and good phase control in interferometric devicesSourced

Sources: JePPIX roadmap [Reference 1], Liu et al. [Reference 2], Shekhar et al. [Reference 4]. The division into five is this article's own.

On the other hand, JePPIX describes modulation in SiN as thermal, at speeds of the order of kHz to MHz, with stress-driven modulators also availableSourced. Fast on-off switching is not a job for SiN — SiN is a material for carrying, storing and splitting light (our commentary).

3. Thin cores and thick cores

The same SiN behaves completely differently depending on the thickness of the core. Liu and colleagues note that thin cores under 100 nm thick have achieved ultralow loss but have a low effective index and large mode area, making them unsuited to nonlinear optics, while thick cores over 700 nm thick have negligible bend loss, engineerable dispersion and high nonlinearitySourced. JePPIX also explains that SiN waveguides can vary the strength of confinement within a circuit: strong confinement for nonlinear devices and sensors, and weak confinement (diluted waveguides) for high optical power and passive fibre attachment, joined on the same chip by three-dimensional tapersSourced.

Fig. 1 Conceptual diagram (vector drawing). The division into thin cores (under 100 nm) and thick cores (over 700 nm) and their features follow Liu et al. [Reference 2]; the ability to vary confinement within a circuit follows the JePPIX roadmap [Reference 1]. The drawing and proportions are this article's schematic.

4. Our calculation: how many orders of magnitude apart are the losses?

Fig. 2 Drawing that includes our calculation (vector drawing). The values are from Smit et al. [Reference 5], the JePPIX roadmap [Reference 1], Liu et al. [Reference 2] and Bose et al. [Reference 3]. The conversion to dB/m and the arrangement are this article's own, and this is not a like-for-like comparison.

Our calculation: how much light does a 1 m waveguide keep, and how long does it delay it?

Our calculation

Light remaining: in Liu et al.'s 1 m spiral waveguide (2.4 dB/m), 10^−0.24 ≈ about 58% of the light remains after 1 m
Comparison: at InP's roughly 2 dB/cm (= 200 dB/m), the same 2.4 dB is lost in 1.2 cm
Delay: using the group index of 2.1 given by Liu et al., the time to cross 1 m is 2.1 × 1 m ÷ the speed of light (3×10⁸ m/s) ≈ 7 ns

Assumptions and limits: the group index is the value Liu et al. used in analysing their resonators, and is not stated as the value for the 1 m spiral. Liu et al. say the spiral fitted on a 5 mm × 5 mm chipSourced. Being able to fit a path that takes light several nanoseconds into a few square millimetres is the basis for applications in delay lines and high-Q resonators (our commentary).

5. Deposition: LPCVD, PECVD and the thermal budget

SiN films are made mainly by LPCVD (low-pressure chemical vapour deposition) or PECVD (plasma-enhanced chemical vapour deposition). The difference between them lies in the deposition temperature and the hydrogen left in the film (our framing).

LPCVD: Bose and colleagues write that LPCVD SiN waveguides require deposition temperatures of up to 850 °C and annealing at 1,150 °C, and that thick-core waveguides generally need annealing at 1,050 °C together with stress-relief structures and CMPSourced
Why anneal so hot: Liu and colleagues anneal at 1,200 °C to drive out hydrogen impurities remaining in the film, and also anneal the upper SiO₂ cladding at 1,200 °CSourced
PECVD: Bose and colleagues note that conventional PECVD SiN has large absorption loss due to hydrogen, and that sputtering and PECVD also suffer from scattering loss caused by particlesSourced
Why low temperature matters: the same paper says that bringing process temperatures below 400 °C would allow waveguides to be built directly on silicon electronics, silicon photonics, thin-film lithium niobate and III-V semiconductors, and that going down to 250 °C extends this to organic electronics, polyimide, quartz and othersSourced

Fig. 3 Conceptual diagram (vector drawing). Each temperature is as stated by Liu et al. [Reference 2] and Bose et al. [Reference 3]. Arranging the temperatures as a “thermal budget” is this article's own presentation and does not show any particular production process.

Bose and colleagues reported a process that makes both the SiN core and the SiO₂ cladding without annealing, using ICP-PECVD at 250 °C with deuterated silane (SiD₄) as the source gasSourced. According to the paper they obtained 1.77 dB/m with an 80 nm core and a best of 8.66 dB/m (median 13.9 dB/m) with an 800 nm core, and needed neither stress-relief structures nor polishingSourced (values for research-stage prototypes).

6. A materials engineer's view (1): the loss is hydrogen

Why this matters for materials engineers: hydrogen from the source gas absorbs light just short of the C band

Bose and colleagues attribute increased absorption loss to hydrogen, present in small amounts even in the deuterated silane source, citing as evidence that loss rises towards a wavelength of 1,520 nm, near the first overtone of SiN–H bond absorptionSourced. They attribute the rising loss on the 1,630 nm side to overtones of SiO–D bonds in the upper claddingSourced.

In other words, the vibration of chemical bonds inside the film creates absorption right next to the most important band for optical communications (around 1.55 µm). Annealing LPCVD films at 1,200 °C (Liu et al.) is done precisely to drive this hydrogen out of the filmSourced.

Our calculation: where does the absorption move if hydrogen is replaced by deuterium?Our calculation

In a simple diatomic harmonic oscillator, the vibrational frequency is inversely proportional to the square root of the reduced mass (our approximation)
Reduced mass of N–H: 14 × 1 ÷ 15 ≈ 0.93; of N–D: 14 × 2 ÷ 16 = 1.75 (atomic masses rounded to integers)
Frequency ratio: √(1.75 ÷ 0.93) ≈ 1.37 → the wavelength stretches by a factor of 1.37
The absorption at 1,520 nm is estimated to move to about 2,080 nm

Assumptions and limits: anharmonicity, and the fact that the real bond sits within a Si–N network, are ignored. This calculation shows the direction — that deuteration moves the absorption out of the C band — not the exact position of the absorption.

Two routes to low loss at low temperature follow from this (our framing): (1) drive the hydrogen out at high temperature (LPCVD plus anneal), or (2) replace hydrogen with deuterium from the start, shifting the absorption to a wavelength that is not used. The first eats into the thermal budget; the second costs more in source gas (an isotope). Both show that what sets the loss is the composition and impurities of the film.

Fig. 4 Drawing that includes our calculation (vector drawing). That the loss increase near 1,520 nm lies near the first overtone of SiN–H bonds is stated by Bose et al. [Reference 3]. The shift on deuteration (about 1.37 times, to about 2,080 nm) is this article's estimate from a simple harmonic-oscillator approximation, not a published value.

7. A materials engineer's view (2): thick films crack — how to relieve the stress

Thick cores (over 700 nm) are essential for nonlinear optics and similar uses, but thick LPCVD films carry stress from deposition. Liu and colleagues at EPFL solve this with the photonic damascene processSourced.

Fig. 5 Conceptual diagram (vector drawing). Patterning of trenches and filler patterns (248 nm DUV exposure), reflow at 1,250 °C, 1,000 nm of LPCVD SiN, etch-back planarisation including CMP, and annealing at 1,200 °C follow Liu et al. [Reference 2]. The four-panel layout and proportions are this article's schematic.

Etch the trenches first: the waveguides and stress-relieving filler patterns are written directly into SiO₂ by DUV (248 nm KrF) exposure. Using DUV rather than electron-beam lithography greatly improves throughput, stability and reproducibilitySourced
Relieving the stress: the filler pattern is an array of horizontal and vertical bars laid uniformly across the whole wafer, which greatly relieves the stress in as-deposited LPCVD Si₃N₄ films and prevents cracks. It is also needed for uniform etching and CMPSourced
Result: with the same filler pattern, more than 30 wafers were processed with zero cracks, and a 100% yield of usable crack-free chips is reportedSourced (as reported by the group)
Planarisation: an etch-back combining resist coating, dry etching and CMP controls the polishing depth, giving sub-nanometre surface roughness and height variation below 3% across the waferSourced
Performance: a mean resonator Q above 30 million across a 4-inch wafer (equivalent to a loss of 1.0 dB/m), and 2.4 dB/m for a 1 m spiral waveguide fitted onto a 5 mm × 5 mm chipSourced

Conceptual image, seen at an angle, of a flat pale grey surface crossed by a single smooth narrow groove, with small square recesses laid out in a regular grid on both sides — **Fig. 6 Conceptual image (AI-generated).** An impression of the idea of tiling small stress-relieving patterns around a waveguide. **It does not show the real shape, dimensions or layout of a filler pattern, or any research institution's design.**

Why this matters for materials engineers: the stress is relieved by the pattern, not the material

When a film cracks under stress, the first thing a materials engineer usually thinks of is changing the deposition conditions (composition, temperature, pressure) to lower the stress itself. Liu and colleagues' method instead leaves the film composition alone and relieves the stress by using a pattern to break up the area over which the film is continuousSourced.

The reading is a deliberate trade-off: to protect the stoichiometric Si₃N₄ (LPCVD) that underpins the low loss, the stress problem was solved on the layout and process side. And the statement that the filler pattern is also needed for uniform etching and CMP reflects the same idea as dummy patterns, familiar from semiconductor interconnect processing (our commentary).

Bose and colleagues, meanwhile, say their 250 °C process needed neither stress-relief structures nor polishingSourced, so a materials-side solution (low temperature and deuteration) sits alongside a process-side solution (high-temperature LPCVD plus damascene). For now the high-temperature route gives lower loss (1.0 dB/m against 8.66 dB/m, both thick cores; the first is inferred from resonator Q and the second measured on straight waveguides, so the methods differ), while the low-temperature route wins on thermal budget — which one becomes the mainstream in volume production is not settledNot yet confirmed.

8. Applications — what it is used for

Application	What the primary sources say
Optical frequency combs (microcombs)	Liu et al. say Si₃N₄ has become the leading material for soliton microcombs, listing system demonstrations in coherent communications, astronomical spectrometer calibration, ultrafast ranging, parallel coherent LiDAR, frequency synthesisers and atomic clocksSourced
Narrow-linewidth lasers and stabilisation	Liu et al. say low-loss PICs and resonators have enabled narrow-linewidth lasers; Bose et al. report a demonstration that cut laser frequency noise by more than four orders of magnitude with a thin-core reference resonatorSourced
Optical beamforming (hybrid with InP)	JePPIX describes a module for satellite communications and 5G in which three InP PICs (a laser, a detector array and a modulator array) are assembled onto a TriPleX SiN PIC containing true time delay elementsSourced
High-power light and gyroscopes	Shekhar et al. say SiN can be used to handle high optical power, and note that SiN waveguide loss for gyroscopes has come down to 0.5 dB/m. They also write that moving to a wavelength of 1,050 nm requires SiN PICsSourced

Sources: Liu et al. [Reference 2], Bose et al. [Reference 3], JePPIX roadmap [Reference 1], Shekhar et al. [Reference 4]. Many of these are research-stage demonstrations and do not indicate adoption in products.

9. What is still hard

(1) No active devices

SiN has neither lasers nor high-speed modulatorsSourced. Combination with InP or thin-film lithium niobate is a given, and the process temperatures for that bonding collide with the thermal budget of section 5 (our commentary).

(2) Taking laboratory low loss into the fab

Liu and colleagues say that transferring their technology to commercial foundries could substantially raise the performance of integrated photonicsSourced, but the status of that transfer and loss values in volume production could not be confirmed in the primary sources within the scope of this articleNot yet confirmed. Shekhar and colleagues expect that multilayer SiN and Si structures will become commonly offered by commercial foundriesNot yet confirmed.

(3) Quantitative stress values could not be confirmed from public information

Values for residual stress in LPCVD films, and the maximum film thickness that stays crack-free, could not be confirmed as numbers in the primary sources this article consulted, and are not given.

The article in summary

SiN is the only integration material in which losses of around 1 dB/m or below have been demonstrated (Liu et al.), and it covers about 400 to 2,500 nmSourced
Compared with InP (about 200 dB/m), it is more than two orders of magnitude lower. After 1 m, about 58% of the light remains, delayed by about 7 nsOur calculation
The key to low loss is hydrogen. SiN–H absorption lies near 1,520 nm, and is escaped by high-temperature annealing or deuterationSourced
Thick films crack under stress. Filler patterns relieve the stress, with zero cracks reported across more than 30 wafersSourced
Best loss at high temperature, or easier integration at low temperature — the mainstream for volume production is not yet settledNot yet confirmed

10. Glossary

Silicon nitride (SiN, Si₃N₄): A compound of silicon and nitrogen, widely used as an insulating and passivation film in semiconductors. Refractive index about 2.0.

Waveguide: A path that confines and carries light by surrounding a high-index core with lower-index cladding.

LPCVD: Low-pressure chemical vapour deposition. Forms films by reacting source gases at high temperature. Gives dense films close to stoichiometry.

PECVD: Plasma-enhanced chemical vapour deposition. Uses a plasma to assist the reaction, forming films at low temperature. Tends to leave hydrogen behind.

ICP-PECVD: PECVD using an inductively coupled plasma. The high-density plasma can dissociate nitrogen.

Deuterated silane: A source gas (SiD₄) in which the hydrogen of silane (SiH₄) is replaced with deuterium.

Overtone: A weak absorption at roughly an integer multiple of the frequency (roughly an integer fraction of the wavelength) of a fundamental molecular vibration.

Anneal: Heat treatment. Drives impurities out of a film or reduces defects.

Thermal budget: The upper limit on temperature and time allowed across a whole process, set by the materials and devices underneath.

Photonic damascene process: Making waveguides by etching trenches first, filling them with SiN and polishing the surface.

Filler pattern: Small patterns tiled around the waveguide. Used to relieve film stress and even out polishing.

CMP: Chemical mechanical polishing. A step that flattens and smooths a surface with chemicals and abrasion.

Reflow: Softening a material at high temperature to smooth out surface roughness.

Q factor: A measure of how long a resonator can store light. The lower the loss, the higher it is.

Group index: The refractive index that sets the speed of a light pulse. Used to calculate delay time.

Two-photon absorption: Absorption of two photons at once. Causes loss under intense light, and occurs in silicon at telecommunication wavelengths.

Microcomb: Light at many evenly spaced wavelengths generated by nonlinear effects in a tiny resonator.

TriPleX: The name of LioniX International's SiN waveguide technology.

11. References (primary sources)

JePPIX “Unleashing the power of PIC — Technology Roadmap 2021-2025”, 8 March 2021 (PDF) — jeppix.eu
J. Liu, G. Huang, R. N. Wang et al., T. J. Kippenberg (EPFL) “High-yield, wafer-scale fabrication of ultralow-loss, dispersion-engineered silicon nitride photonic circuits”, Nature Communications 12, 2236 (2021, DOI: 10.1038/s41467-021-21973-z) — nature.com
D. Bose, M. W. Harrington, A. Isichenko et al., D. J. Blumenthal (UCSB and others) “Anneal-free ultra-low loss silicon nitride integrated photonics”, Light: Science & Applications 13, 156 (2024, DOI: 10.1038/s41377-024-01503-4) — nature.com
S. Shekhar, W. Bogaerts, L. Chrostowski, J. E. Bowers et al. “Roadmapping the next generation of silicon photonics”, Nature Communications 15, 751 (2024, DOI: 10.1038/s41467-024-44750-0) — nature.com
M. Smit et al. (TU Eindhoven and others) “An introduction to InP-based generic integration technology” (revised version of the 2014 paper, published by JePPIX as chapter 1 of a book, PDF) — jeppix.eu
D. Zhu et al. (Harvard University and others) “Integrated photonics on thin-film lithium niobate”, Advances in Optics and Photonics 13, 242 (2021, DOI: 10.1364/AOP.411024). Checked in the authors' version — arxiv.org

12. Claim-to-source audit

Claim in the text	Basis	Label
That SiN can achieve the lowest optical loss of any integration platform, with a wavelength range of about 400 to 2,500 nm, losses of 0.1 dB/cm to 0.1 dB/m, and applications out of reach of silicon photonics and InP; that the modest index contrast gives small phase errors; that confinement can be varied within a circuit, joining strong confinement (nonlinear devices, sensors) and weak confinement (high power, fibre attachment) with 3D tapers; thermal modulation at kHz to MHz and stress-driven modulators; that active devices are not native to SiN and are added by hybrid integration with InP and others; and the satellite communications and 5G module with three InP PICs on a TriPleX SiN PIC (with true time delay elements)	JePPIX Technology Roadmap 2021-2025, Reference 1 https://www.jeppix.eu/wp-content/uploads/2020/pilotline-files/JePPIX_Roadmap_2021_2025.pdf	Sourced
A material refractive index of 2.0 for Si₃N₄ and a group index of 2.1 (used in analysis); that losses of around 1 dB/m or below have been demonstrated only in Si₃N₄; the absence of two-photon absorption at telecom wavelengths, very high power handling and strong Kerr nonlinearity; the features of thin (under 100 nm) and thick (over 700 nm) cores; the photonic damascene process (trenches and filler patterns by 248 nm DUV, reflow at 1,250 °C, 1,000 nm LPCVD Si₃N₄, etch-back with resist, dry etching and CMP, sub-nm roughness and below 3% height variation, a 1,200 °C anneal to drive out hydrogen, and a 1,200 °C anneal of the upper SiO₂); that filler patterns relieve stress in LPCVD films and prevent cracks and are also needed for uniform etching and CMP, with no cracks over more than 30 wafers and 100% yield; a mean Q above 30 million across a 4-inch wafer (equivalent to 1.0 dB/m) and 2.4 dB/m for a 1 m spiral on a 5 mm × 5 mm chip; the list of microcomb system demonstrations and narrow-linewidth lasers; and the statement that transfer to commercial foundries could raise performance	Liu et al., Nature Communications 12, 2236 (2021), Reference 2 https://www.nature.com/articles/s41467-021-21973-z	Sourced
That LPCVD SiN waveguides require deposition at up to 850 °C and annealing at 1,150 °C, and thick cores generally a 1,050 °C anneal with stress-relief structures and CMP; hydrogen-related absorption in conventional PECVD and particle scattering in sputtering and PECVD; that below 400 °C waveguides can be built on electronics, Si photonics, TFLN and III-V, and at 250 °C also on organics, polyimide, quartz and others; ICP-PECVD with deuterated silane at 250 °C without annealing, with 1.77 dB/m for an 80 nm core, a best of 8.66 dB/m and median of 13.9 dB/m for an 800 nm core, and no stress relief or polishing needed; that hydrogen raises loss towards 1,520 nm near the first overtone of SiN–H, with the 1,630 nm side attributed to SiO–D overtones; and the reduction of frequency noise by more than four orders of magnitude with a reference resonator	Bose et al., Light: Science & Applications 13, 156 (2024), Reference 3 https://www.nature.com/articles/s41377-024-01503-4	Sourced
That SiN has become a standard offering at several foundries, used for extended wavelength range, high power, low loss and phase control; that SiN can be used to handle high optical power; that SiN waveguide loss for gyroscopes has come down to 0.5 dB/m and that 1,050 nm requires SiN PICs; and the refractive index of Si as 3.47 (1,550 nm)	Shekhar et al., Nature Communications 15, 751 (2024), Reference 4 https://www.nature.com/articles/s41467-024-44750-0	Sourced
The expectation that multilayer SiN and Si structures will become commonly offered by commercial foundries	An outlook by Shekhar et al. for the coming decade, not a result, Reference 4 https://www.nature.com/articles/s41467-024-44750-0	Not yet confirmed
Loss of about 2 dB/cm from the p-type cladding in InP	Smit et al. (book chapter 1 published by JePPIX), Reference 5 https://www.jeppix.eu/wp-content/uploads/2025/03/Chapter-1-Introduction.pdf	Sourced
The refractive index of SiO₂ as about 1.44 (1,550 nm)	Property table in Zhu et al., Advances in Optics and Photonics 13, 242 (2021), Reference 6 https://arxiv.org/abs/2102.11956	Sourced
The conversions to dB/m (InP 2 dB/cm = 200 dB/m, 0.1 dB/cm = 10 dB/m); about 58% remaining after 1 m at 2.4 dB, and the same loss in 1.2 cm of InP; a delay of about 7 ns over 1 m at a group index of 2.1; and the reduced masses of N–H and N–D (about 0.93 and 1.75), a frequency ratio of about 1.37, and the estimate of 1,520 nm moving to about 2,080 nm	Our calculation. The group index is borrowed from the resonator analysis. The deuterium estimate uses a diatomic harmonic-oscillator approximation with integer atomic masses, ignoring anharmonicity and the effect of the network	Our calculation
Whether high-temperature LPCVD plus damascene or low-temperature deuterated PECVD becomes the mainstream for volume production; and the status of transfer to commercial foundries and loss values in volume production	Not stated because no primary source could be confirmed as of this article's research (September 2026). This article's own judgement	Not yet confirmed
Residual stress values in LPCVD films, and the maximum crack-free film thickness	Not stated because they could not be confirmed as numbers in the primary sources this article consulted	Commentary
The framing of SiN as a material for carrying, storing and splitting light; the classification into five properties; organising deposition methods by temperature and hydrogen; the reading that film composition and impurities set the loss, and the two routes of driving hydrogen out or replacing it; the reading that stress was relieved by pattern rather than material, the analogy with dummy patterns, and the contrast between materials-side and process-side solutions; the point that the thermal budget collides with bonding; and the structure and drawing of the figures	Our own framing and commentary based on the published material. Not views expressed by the research institutions	Commentary
That Figs. 1 to 5 are explanatory drawings rather than real cross-sections or measured data, and that the hero image and Fig. 6 are AI-generated images	Our note	Commentary

Last updated 26 September 2026. Sources are limited to primary material (peer-reviewed papers and reviews, a published roadmap from a research consortium, and book chapters made available by universities). Because the article includes numerical conversions and materials-design readings, those are marked as “Our calculation” or “Commentary” and kept separate from sourced fact. Loss values from papers are for research-stage prototypes, not products or production processes. Stress values for LPCVD films, the maximum crack-free film thickness, and production loss at commercial foundries are not stated here because no published primary source could be confirmed. All figures are explanatory concept graphics. Figs. 1 to 5 are vector drawings, and the hero image and Fig. 6 are AI-generated images; none of them shows a real cross-section photograph, micrograph or physical product.