TECHNOLOGY EXPLAINER
Wavelength Multiplexing and Microrings
— many colors on one fiber, and a fight against a 1 K temperature change
Sending many beams of light of different colors (wavelengths) through one optical fiber at the same time: that is wavelength division multiplexing (WDM). The "addresses" of the wavelengths are set by ITU-T: for long-haul use, 12.5 to 100 GHz spacing anchored at 193.1 THz; for low-cost use, 20 nm spacing. A leading candidate for sorting wavelengths on a chip is a ring a few micrometers across, the microring resonator. But silicon rings are said to stop working with a temperature change of more than 1 K.
- What wavelength multiplexing and microrings are (the short version)
- Why send light on separate wavelengths
- Assigning wavelength "addresses": the ITU-T grids
- Wavelengths in the data center: from 4 to 32
- A materials engineer's view (1): the spacing was set by temperature
- How a microring resonator works
- Temperature, the biggest enemy
- Fighting temperature: cancel it with materials, or chase it with heaters
- A materials engineer's view (2): negative thermo-optic coefficients as homework for materials
- Open problems and unconfirmed points
- Glossary / References / Claim-to-source audit
Sourced = stated in a standard, an MSA specification or a peer-reviewed paper (link given)
Our calculation = a figure this article derived from assumptions it spells out
Not yet confirmed = a plan or target with no confirmed track record yet
Structural readings and materials-design interpretations are marked separately as commentary.
Values specified in standards (ITU-T), industry specifications (MSAs) and experimental values from papers are different in kind, and the text says which is which each time.
1. What wavelength multiplexing and microrings are (the short version)
Wavelength division multiplexing (WDM) is a technique that puts a separate signal on each of several wavelengths of light and carries them together through one fiber. A microring resonator is a tiny ring of light that can draw in, or block, light of one particular wavelength only.
- How wavelengths are assigned: ITU-T G.694.1 defines the frequency grid for DWDM (dense WDM) anchored at 193.1 THz, and G.694.2 defines the wavelength grid for CWDM (coarse WDM) with 20 nm spacingSourced
- Their job on a chip: microrings act as filters, switches and modulators, and lining up several rings with slightly different resonances along one waveguide allows WDM transmission and receptionSourced
- Their biggest weakness: the refractive index of silicon changes strongly with temperature (a thermo-optic coefficient of 1.86×10⁻⁴ K⁻¹), and for typical applications a temperature deviation of more than 1 K renders the ring inoperableSourced
Loss and dispersion in the fiber itself are covered in our optical fiber explainer, and silicon waveguides and the four building blocks in our silicon photonics explainer. This article takes the ring temperature problem raised in that explainer's open problem 2, "temperature moves the wavelength of light", and works through the numbers and the countermeasures. Modulation formats that switch light on and off with an electrical signal are covered in our explainer on optical modulation, and waveguide materials that are less sensitive to temperature in our explainer on silicon nitride waveguides (our own division of topics).
2. Why send light on separate wavelengths
There are broadly three ways to increase how much one optical fiber carries: make each signal faster, add more fibers (or cores), or put several wavelengths on one fiber. WDM is the third, and its greatest advantage is that capacity grows without laying new fiber (our commentary).
3. Assigning wavelength "addresses": the ITU-T grids
For transmitters and receivers from different companies to work over the same fiber, which wavelengths to use has to be agreed in common. Two ITU-T Recommendations set this outSourced.
| Item | DWDM: ITU-T G.694.1 (October 2020 edition) | CWDM: ITU-T G.694.2 (December 2003 edition) |
|---|---|---|
| How the reference is set | By frequency, with 193.1 THz as the anchor | By wavelength |
| Spacing | Fixed grids of 12.5, 25, 50 and 100 GHz (and wider spacings in integer multiples of 100 GHz). Each channel sits at 193.1 + n × spacing | 20 nm |
| Flexibility | Flexible grid: center frequencies at 193.1 + n × 0.00625 THz and slot widths of 12.5 × m GHz. Any combination is allowed as long as slots do not overlap | 18 wavelengths from 1271 to 1611 nm (the values at either end of the table are given as examples) |
| Aim | — | Low cost through uncooled lasers, relaxed wavelength selection and wide-passband filters |
| Rationale for the spacing | — | Total source wavelength variation of about ±6 to 7 nm, with a guard band of one third of the minimum spacing judged sufficient; 20 nm chosen to maximize the channel count |
All Sourced (ITU-T G.694.1 (10/2020) [Ref. 1]; ITU-T G.694.2 (12/2003), main text and Appendix I [Ref. 2]).
A frequency spacing Δf converts to a wavelength spacing Δλ as Δλ ≈ λ² × Δf ÷ cOur calculation.
- The 193.1 THz anchor is about 1552.52 nm in wavelength
- Near there, 100 GHz = about 0.80 nm, 50 GHz = about 0.40 nm and 12.5 GHz = about 0.10 nm
- CWDM's 20 nm is about 50 times the 50 GHz (about 0.4 nm) spacing of DWDM
Assumptions and limits: approximate values converted with c = 299,792,458 m/s and λ = 1552.52 nm. The conversion changes with wavelength.
4. Wavelengths in the data center: from 4 to 32
Optical transceivers in data centers also borrow their wavelengths from these two grids. The CW-WDM MSA, an industry group, describes the lineage as followsSourced.
- ITU-T set the foundations in G.694.1 (DWDM) and G.694.2 (CWDM), and IEEE carved out parts of them for high-volume data center use
- LWDM derives from G.694.1 and was first used in 100GBASE-LR4. IEEE later doubled it for 400GBASE-LR8
- CWDM4 derives from G.694.2 and was first used in 40GBASE-LR4. The CWDM4 MSA reused it for 100G, and IEEE reused it for 200G FR4 and 400G FR4
Then in June 2021 the CW-WDM MSA published a specification (Rev 1.0) for light sources with 8, 16 and 32 wavelengthsSourced. The MSA notes that existing IEEE and MSA standards specify one or four wavelengths, and expects dense optics based on silicon photonics, such as co-packaged optics, to move to 8, 16 and 32 wavelengthsSourced.
| Wavelength span | Channel count | Channel spacing | Channel bandwidth |
|---|---|---|---|
| 9 nm | 8+1 / 16+1 | 200 GHz / 100 GHz | 100 GHz / 50 GHz |
| 18 nm | 8+1 / 16+1 / 32+1 | 400 / 200 / 100 GHz | 200 / 100 / 50 GHz |
| 36 nm | 8+1 / 16+1 / 32+1 | 800 / 400 / 200 GHz | 400 / 200 / 100 GHz |
All Sourced (CW-WDM MSA Technical Specifications Rev 1.0, 4 June 2021, Table 2-5 [Ref. 3]). The nominal center wavelength is 1300.05 nm (O-band). The shortest "+1" wavelength is optional. Founding members include Arista Networks, Ayar Labs, II-VI, imec, Intel, Lumentum, MACOM and Sumitomo Electric [Ref. 4].
What stands out is that only the light source is specified. The CW-WDM MSA explains that it specifies the laser source only, not the full communication link, so that developers are free to optimize their linksSourced. This is one reason microrings are seen as a strong candidate for the transmit and receive components that line up many wavelengths (our commentary).
5. A materials engineer's view (1): the spacing was set by temperature
An appendix to G.694.2 explains why the CWDM spacing became 20 nm. Building CWDM with uncooled lasers and wide-passband filters requires a spacing of 20 nm or more. The total wavelength variation of the source is about ±6 to 7 nm, and a guard band of one third of the minimum spacing is sufficient; the result is 20 nmSourced.
It names two main causes of wavelength variation: the wavelength spread allowed in manufacturing (to raise yield), and the change of an uncooled laser's wavelength with temperature over the operating rangeSourced.
In other words, CWDM's 20 nm is a value worked backwards not from communications needs but from the temperature behavior of materials and manufacturing spread (our commentary). Closer spacing fits more wavelengths, but in exchange the temperature must be controlled or a material less sensitive to temperature used. DWDM's 0.4 nm spacing (50 GHz) is about one fiftieth of CWDM'sOur calculation.
On a chip, this trade-off between spacing and temperature appears in an even harsher form. That is the microring.
6. How a microring resonator works
A microring is a ring-shaped waveguide placed beside a straight waveguide (the bus waveguide)Sourced. According to a review by Padmaraju and Bergman in Nanophotonics, the high index contrast of silicon has allowed rings as small as 1.5 µm in radiusSourced.
At wavelengths where light that has gone once round the ring overlaps the light from the previous round exactly in phase, the light builds up inside the ring and can no longer pass straight along the bus waveguide. This is the "dip" in the transmission spectrum, the resonance. The resonance condition is that the optical length of one round trip is a whole number of wavelengths, so the resonant wavelength is set by the refractive index and dimensions of the ring, and resonances recur at a fixed interval, the free spectral range (FSR) (our commentary).
Padmaraju and Bergman summarize what rings do as followsSourced.
- Their basic roles are as filters, switches and modulators
- Placing several rings with slightly offset resonances along the same bus waveguide lets each handle one wavelength, enabling WDM modulation. The same arrangement can demultiplex at the receiver
- For this reason they are expected to deliver low-cost WDM communication with a small footprint and high energy efficiency
- Assumptions: FSR ≈ λ² ÷ (ng × L). A radius of 5 µm (circumference L ≈ 31.4 µm) and a group index ng of 4.2 are assumed
- At λ = 1300 nm, FSR ≈ about 12.8 nm; at λ = 1550 nm, about 18.2 nm Our calculation
Assumptions and limits: the radius and group index are assumptions made in this article and change with the waveguide cross-section. The FSR formula is used as a general approximation for resonators. The result is of the same order as the 9 nm and 18 nm spans of the CW-WDM MSA (Section 4).
7. Temperature, the biggest enemy
If the resonant wavelength is set by the ring's refractive index, then when the index changes with temperature, the resonance moves too. The thermo-optic coefficient of silicon (the change in refractive index per kelvin) is given in two primary sources as followsSourced.
| Material | Thermo-optic coefficient dn/dT | Source |
|---|---|---|
| Silicon | 1.86 × 10⁻⁴ K⁻¹ | Review by Padmaraju and Bergman [Ref. 5] |
| Silicon (1550 nm, 300 K) | 1.8 × 10⁻⁴ K⁻¹ | Measurement by Komma et al. (Applied Physics Letters, 2012) [Ref. 6] |
| Silicon dioxide (SiO2) | 1 × 10⁻⁵ K⁻¹ | Padmaraju and Bergman [Ref. 5] |
| Thermal expansion coefficient of the Si substrate (for reference) | 2.6 × 10⁻⁶ K⁻¹ | Padmaraju and Bergman [Ref. 5] |
All Sourced. Padmaraju and Bergman note that because the thermo-optic coefficient of SiO2 is an order of magnitude smaller than that of Si, and the substrate's thermal expansion two orders smaller, the resonance shift can be approximated as dλ/dT = (λ₀/ng) × ∂neff/∂T.
How much resonance shift can be tolerated depends on the ring's Q factor, but for typical applications, deviations in temperature > 1 K will render the microring-based device inoperable. This is not compatible with the temperature ranges typical of microelectronic environments. What is hazardous is not the absolute temperature but the relative changes in temperature during active operation of the optical linkSourced.
- Assumptions: ∂neff/∂T approximated by the silicon value of 1.86×10⁻⁴ K⁻¹ (assuming the light is almost entirely confined in silicon), and a group index ng of 4.2
- λ = 1300 nm: 1300 ÷ 4.2 × 1.86×10⁻⁴ ≈ about 0.058 nm/K (58 pm/K) Our calculation
- λ = 1550 nm: about 0.069 nm/K (69 pm/K)
- A 10 K change gives about 0.58 nm at 1300 nm, about half the 200 GHz spacing of the CW-WDM MSA (about 1.13 nm near 1300 nm)
- Assuming Q = 10,000, the resonance width is λ÷Q = about 0.13 nm, so about 2 K shifts the resonance by its own width
Assumptions and limits: ng and Q are assumptions made in this article. In practice, part of the light extends into the cladding, which reduces the shift, and the amount differs from design to design.
8. Fighting temperature: cancel it with materials, or chase it with heaters
Padmaraju and Bergman divide the countermeasures into two categoriesSourced: athermal solutions, which reduce the temperature dependence itself, and control-based solutions, which actively hold the ring's local temperature.
Approach 1, athermal: cancel it with a material that has a negative thermo-optic coefficient
Silicon's thermo-optic coefficient is positive. So a material with a negative thermo-optic coefficient is used as the cladding, bringing the temperature coefficient of the waveguide as a whole close to zeroSourced. According to the paper, polymer cladding achieved −5 pm/K (over a range of 50 K), and later work reported as low as 0.2 pm/KSourced. But because silicon's coefficient is large and its confinement strong, cancelling it requires a narrower, thinner waveguide that lets the light extend into the cladding, and the paper notes that this makes the dimensional requirements strictSourced.
Approach 2, control-based: hold the temperature with a heater
The mainstream approach today builds a resistive heater (nichrome, titanium, doped silicon and so on) right next to the ring, runs the ring at a slightly elevated temperature from the start ("run the microring hot"), and lowers the heater power when the surroundings warm up and raises it when they coolSourced. To avoid optical loss, the usual arrangement places the heater above about 1 µm of thermal oxide; efficiencies are mostly around 100 mW/FSR, with the best at about 42 mW/FSRSourced.
- Assumptions: 1300 nm, an FSR of 12.8 nm (our calculation in Section 6), a shift of 0.58 nm (10 K, our calculation in Section 7), and a heater efficiency of 100 mW/FSR (the value the paper gives for most results)
- 0.58 ÷ 12.8 × 100 ≈ about 4.5 mW per ring Our calculation
- If a 16-wavelength link has 16 rings each at the transmitter and receiver, 32 × 4.5 ≈ about 140 mW
Assumptions and limits: this is an order-of-magnitude estimate built entirely on this article's assumptions. In reality, manufacturing variation means rings are made with resonances away from their design values, and correcting that also takes power, so the value could be larger still. The number of rings, the efficiency and the range of temperature change vary greatly with design.
9. A materials engineer's view (2): negative thermo-optic coefficients as homework for materials
The two approaches in Section 8 amount to a choice: pay in power, or change the material. The paper gives the advantage of athermal solutions as requiring no active power consumption, and their disadvantage as difficult fabrication, through the incorporation of non-CMOS materials or additional photonic structuresSourced.
What it says about polymers reads as a concrete list of homework for materials developersSourced.
- They are vulnerable to degradation at the high temperatures of certain stages of a CMOS production cycle
- They suffer from chemical instability, UV aging and poor mechanical characteristics
- Their thermo-optic coefficient must be made equal in size and opposite in sign to silicon's, and matched precisely despite manufacturing variation
Attention therefore turned to titanium dioxide (TiO2). The paper describes TiO2 as one of the few CMOS-compatible materials with a negative thermo-optic coefficient (about −1.8×10⁻⁴ K⁻¹) on the same order as Si, and reports that TiO2-clad rings have shown less than 2 pm/K (over a range of 5 K)Sourced. Even so, it concludes that how to integrate it into a CMOS-compatible fabrication process remains a difficultySourced.
The reading here is that process compatibility, more than the optical constant itself, decides whether a material is adopted (our commentary). What is needed is not just a negative dn/dT but a material that simultaneously satisfies deposition temperature, tolerance of the thermal history of later process steps, reproducible refractive index, and long-term optical and thermal stability. The contradiction we saw in our silicon photonics explainer, "made with light, and made to withstand light", appears here in the form of a direction in which the material should change with heat, combined with properties that must not change with heat.
There is another way to look at it. The thermo-optic coefficient of SiO2 is 1×10⁻⁵ K⁻¹, an order of magnitude smaller than silicon'sSourced. That is why silica planar lightwave circuits are relatively insensitive to temperature, while silicon accepts the character of being small but sensitive to temperature (our commentary). Waveguide materials that are less sensitive to temperature are covered in our explainer on silicon nitride waveguides.
10. Open problems and unconfirmed points
(1) Wider use of 8 to 32 wavelengths is still ahead
The CW-WDM MSA has published light-source specifications for 8, 16 and 32 wavelengthsSourced, but no settled primary source could be found at the time of writing on which combination of wavelength count and spacing will become mainstream in volume productionNot yet confirmed. A statement on the MSA's own FAQ page puts it only as an outlook: "Initial products may use 8 wavelengths, and then move to 16 and 32 wavelengths"Not yet confirmed.
(2) Heater power grows with every ring
As the estimate in Section 8 shows, the more rings there are, the more temperature-control power piles upOur calculation. If athermal materials make it into the CMOS process, power will fall, but when that will become practical is not settledNot yet confirmed.
(3) What this article has not confirmed
Measured values for the thermal design (heater power, control scheme, operating temperature range) of microrings in any specific company's product are not stated because they could not be confirmed from primary sources within the scope of this article. Also, the review by Padmaraju and Bergman is a 2013 to 2014 paper, and this article does not cover improvements made since then.
- ITU-T assigns the wavelength addresses for WDM. DWDM uses 12.5 to 100 GHz spacing anchored at 193.1 THz; CWDM uses 20 nm spacing and 18 wavelengthsSourced
- CWDM's 20 nm was set by the temperature drift of uncooled lasers and manufacturing spreadSourced
- Data centers are moving from 4 wavelengths to 8, 16 and 32. The CW-WDM MSA has published an O-band light-source specificationSourced
- A silicon ring shifts by about 0.06 nm per kelvin, reaching about half a 200 GHz spacing with 10 KOur calculation
- The countermeasures are to cancel it with a negative thermo-optic material or to chase it with a heater. The barrier for materials answers such as TiO2 is integration into the CMOS processSourced
11. Glossary
- WDM (wavelength division multiplexing)
- A technique that puts separate signals on light of different wavelengths and sends them together through one fiber.
- DWDM
- Dense WDM. Spacing is set by frequency, with many wavelengths packed at narrow spacings such as 12.5 to 100 GHz.
- CWDM
- Coarse WDM. The wide 20 nm spacing allows uncooled lasers.
- Grid
- The list of permitted center wavelengths (frequencies), set by standards.
- Flexible grid
- A DWDM grid in which center frequencies and slot widths can be combined freely in fine steps.
- O-band
- The wavelength band around 1310 nm, widely used for data center optics.
- MSA
- Multi-Source Agreement. An industry specification agreed among several companies.
- Microring resonator
- A resonator made by forming a waveguide into a ring. It draws in or blocks one particular wavelength.
- Bus waveguide
- The straight waveguide running beside the ring, serving as the way in and out for light.
- FSR (free spectral range)
- The wavelength interval at which resonances recur. The smaller the ring, the wider it is.
- Q factor
- The sharpness of a resonance. The higher it is, the narrower the resonance and the more sensitive to small shifts.
- Thermo-optic coefficient (dn/dT)
- The change in refractive index per kelvin. About 1.8×10⁻⁴ K⁻¹ for silicon.
- Group index
- The effective refractive index that sets the speed at which a light pulse (a signal) travels.
- Athermal
- Cancelling or reducing temperature dependence through materials or structure.
- pm/K
- Shift in resonant wavelength per kelvin. 1 pm = 0.001 nm.
- mW/FSR
- The heater power needed to move the resonance by one FSR. A measure of heater efficiency.
12. References (primary sources)
- ITU-T "G.694.1 (10/20) Spectral grids for WDM applications: DWDM frequency grid" (the full Recommendation is freely available as a PDF) https://www.itu.int/rec/T-REC-G.694.1-202010-I/en
- ITU-T "G.694.2 (12/03) Spectral grids for WDM applications: CWDM wavelength grid" https://www.itu.int/rec/T-REC-G.694.2-200312-I/en
- CW-WDM MSA "CW-WDM MSA Technical Specifications Rev 1.0", 4 June 2021 (specifications page) https://cw-wdm.org/specifications/
- CW-WDM MSA "CW-WDM MSA Consortium Releases New Specification for Multi-Wavelength Optical Laser Sources", 8 June 2021, and FAQ https://cw-wdm.org/cw-wdm-msa-consortium-releases-new-specification-for-multi-wavelength-optical-laser-sources/
- K. Padmaraju, K. Bergman "Resolving the thermal challenges for silicon microring resonator devices", Nanophotonics 3(4-5), 269–281 (2014) https://doi.org/10.1515/nanoph-2013-0013
- J. Komma et al. "Thermo-optic coefficient of silicon at 1550 nm and cryogenic temperatures", Applied Physics Letters 101, 041905 (2012) https://doi.org/10.1063/1.4738989
- W. Bogaerts et al. "Silicon microring resonators", Laser & Photonics Reviews 6(1), 47–73 (2012) https://doi.org/10.1002/lpor.201100017
- CW-WDM MSA "Frequently Asked Questions" https://cw-wdm.org/faq/
13. Claim-to-source audit
| Claim in the text | Basis | Label |
|---|---|---|
| That G.694.1 defines the DWDM frequency grid anchored at 193.1 THz and supports spacings from 12.5 GHz to 100 GHz and above. The fixed-grid formula (193.1 + n × 0.0125 / 0.025 / 0.05 / 0.1 THz). For the flexible grid, center frequencies of 193.1 + n × 0.00625 THz, slot widths of 12.5 × m GHz, and that any combination is allowed as long as slots do not overlap | Reference 1 https://www.itu.int/rec/T-REC-G.694.1-202010-I/en | Sourced |
| That G.694.2 defines the CWDM wavelength grid with 20 nm spacing and shows 18 wavelengths from 1271 to 1611 nm (the ends given as examples). That low cost is achieved with uncooled lasers, relaxed wavelength selection and wide-passband filters. That a spacing of 20 nm or more is needed, that source wavelength variation is about ±6 to 7 nm, that a guard band of one third of the minimum spacing suffices, and that 20 nm was chosen. The two causes of wavelength variation (manufacturing spread and the temperature change of uncooled lasers) | Reference 2 https://www.itu.int/rec/T-REC-G.694.2-200312-I/en | Sourced |
| The grid of CW-WDM MSA Rev 1.0 (4 June 2021): 9 / 18 / 36 nm spans, 8+1 / 16+1 / 32+1 wavelengths, spacings such as 200 to 800 GHz and channel bandwidths; the nominal center wavelength of 1300.05 nm; and that the shortest wavelength is optional | Reference 3 https://cw-wdm.org/specifications/ | Sourced |
| That the light-source specification for 8, 16 and 32 wavelengths was published on 8 June 2021; uses such as co-packaged optics; the list of founding members | Reference 4 https://cw-wdm.org/cw-wdm-msa-consortium-releases-new-specification-for-multi-wavelength-optical-laser-sources/ | Sourced |
| That existing IEEE and MSA standards specify one and four wavelengths, and that SiPh-based optics are expected to move to 8, 16 and 32. The LWDM and CWDM4 lineage (100GBASE-LR4, 40GBASE-LR4, CWDM4 MSA, 200G/400G FR4, 400GBASE-LR8). That only the light source is specified, not the full link. The outlook that initial products may use 8 wavelengths and then move to 16 and 32 | Reference 8 https://cw-wdm.org/faq/ | Sourced |
| That a ring is a travelling-wave resonator placed beside a bus waveguide; radii down to 1.5 µm; roles as filter, switch and modulator; WDM by cascading several rings. Thermo-optic coefficients of 1.86×10⁻⁴ K⁻¹ for silicon and 1×10⁻⁵ K⁻¹ for SiO2, Si substrate thermal expansion of 2.6×10⁻⁶ K⁻¹, and dλ/dT = (λ₀/n_g)∂n_eff/∂T. That in typical applications a deviation of more than 1 K renders the device inoperable, and that relative temperature change is the problem. The athermal / control-based categories and their pros and cons. Polymer cladding −5 pm/K (50 K) and 0.2 pm/K; TiO2 (about −1.8×10⁻⁴ K⁻¹, CMOS-compatible) below 2 pm/K (5 K). Polymer degradation, chemical instability, UV aging and mechanical properties. Heater materials (nichrome, titanium, doped Si), about 1 µm of SiO2, about 100 mW/FSR, best about 42 mW/FSR at 14 µs, inner heater about 20 mW/FSR at about 1 µs, and "run the microring hot" | Reference 5 https://doi.org/10.1515/nanoph-2013-0013 | Sourced |
| That the thermo-optic coefficient of silicon is 1.8×10⁻⁴ K⁻¹ at 1550 nm and 300 K | Reference 6 https://doi.org/10.1063/1.4738989 | Sourced |
| That it is a review of the basic theory and applications (filters, delay lines, sensors, modulators) of silicon microrings | Reference 7 https://doi.org/10.1002/lpor.201100017 | Sourced |
| GHz-to-nm conversions (193.1 THz ≈ 1552.52 nm, 100 GHz ≈ 0.80 nm, 50 GHz ≈ 0.40 nm, 12.5 GHz ≈ 0.10 nm, 200 GHz ≈ 1.13 nm near 1300 nm). 20 nm being about 50 times 50 GHz. FSR (about 12.8 nm / 18.2 nm for a 5 µm radius). Resonance shift (about 58 pm/K and 69 pm/K; about 0.58 nm for 10 K) and a resonance width of about 0.13 nm at Q = 10,000. Heater power (about 4.5 mW per ring, about 140 mW for 32) | Our calculation. A group index of 4.2, a 5 µm radius, approximating ∂n_eff/∂T by the Si value, Q = 10,000, 100 mW/FSR and 32 rings are assumptions made in this article | Our calculation |
| The mainstream for volume production of 8 to 32 wavelengths; when athermal materials will become practical in the CMOS process | No settled primary source could be confirmed at the time of writing in this article. Stated as an outlook | Not yet confirmed |
| The principle of resonance (a round-trip optical length equal to a whole number of wavelengths) and the approximate FSR formula. The reading that 20 nm was worked back from materials' temperature behavior and manufacturing spread. The reading that process compatibility decides adoption. The contrast between temperature-insensitive silica and sensitive silicon. The summary of WDM's advantages | General optics explanation and this article's commentary based on published content. Not views expressed by the institutions | Commentary |
| Measured thermal-design values for microrings in specific companies' products | Not stated, because they could not be confirmed from primary sources within the scope of this article | Commentary |
| That Figs. 1, 2, 4 and 6 are explanatory drawings and Fig. 5 a drawing that includes our calculation, and that the hero image and Fig. 3 are AI-generated images | Our note | Commentary |
Last updated 26 September 2026. Sources are limited to primary material (ITU-T Recommendations, specifications and official announcements of an industry MSA, and peer-reviewed papers); market estimates from research firms are not used. Values specified in standards, MSA specifications and experimental values from papers are different in kind and are kept distinct. The mainstream wavelength count for volume production of 8 to 32 wavelengths, when athermal materials will become practical, and measured thermal-design values for rings in specific products are not stated because they could not be confirmed in published primary sources. All figures are for explanation. Figs. 1, 2, 4 and 6 are vector drawings, Fig. 5 is a vector drawing that includes our calculation, and the hero image and Fig. 3 are AI-generated images; none of them shows a real device, spectrum or product.