MENU
Advanced Materials Intelligence
  1. Home
  2. Optical Fiber Explained | Photonics

Optical Fiber Explained | Photonics

Back to Technology, Institution & Company Guides

TECHNOLOGY EXPLAINER

Optical Fiber
— a strand of glass 125 µm across that carries light for 100 km

Every component in optical communications ends up connected to this thin strand of glass. The international standard ITU-T G.652 caps the loss of standard single-mode fiber at 0.30 dB/km in the C-band. At the research frontier, two numbers now sit side by side: the silica-glass record of 0.1397 dB/km, and 0.091 dB/km for a hollow-core fiber that guides light through air instead of glass. This article works out where that gap comes from, from the side of materials and manufacturing.

Built from primary sources: ITU-T Recommendations, the IEICE Knowledge Base, published material from NTT, Sumitomo Electric Industries and NICT, and a paper in Nature Photonics / Last updated September 2026

Conceptual image of a transparent fiber as thin as a hair tracing a gentle curve against a dark background, with faint light travelling inside it
Conceptual image (AI-generated). An impression of light travelling through a thin glass fiber. It does not show a real product, real dimensions or how the light actually looks. Light at telecom wavelengths is invisible to the eye.
What this article covers
  1. What an optical fiber is (the short version)
  2. Structure: the refractive index is made with dopants
  3. Single-mode and multimode
  4. Loss: where the light goes
  5. Dispersion: different colors arrive at different times
  6. A materials engineer's view (1): how the glass is made
  7. Many paths in one strand: multicore fiber
  8. Taking the glass out of the path: hollow-core fiber
  9. A materials engineer's view (2): perfect the glass, or abandon it
  10. Open problems and unconfirmed points
  11. Glossary / References / Claim-to-source audit
How claims are labelled in this article

Sourced = stated in a standard, paper or published document (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 a standard (ITU-T), record values from peer-reviewed papers and conference talks, and typical values for commercial products are different kinds of number, and the text says which is which each time.

1. What an optical fiber is (the short version)

An optical fiber is a thin strand in which higher-index glass (the core) is wrapped in lower-index glass (the cladding), trapping light and carrying it a long way. The outer diameter of silica fiber for telecommunications is fixed by international standard at 125 µmSourced.

  • What it is made of: mostly silicon dioxide (SiO2). Germanium dioxide (GeO2) is added to raise the refractive index and fluorine (F) to lower itSourced
  • What sets its performance: how much the light weakens (loss) and how much a light pulse spreads out (dispersion). ITU-T G.652.D, the standard for ordinary single-mode fiber, caps cable loss at 0.30 dB/km between 1530 and 1565 nmSourced
  • What is changing now: multicore fiber, which puts several cores inside the same 125 µm, and hollow-core fiber, which makes the core out of air. The first has been announced for a transatlantic submarine cable; for the second, a paper has reported loss below the silica-glass recordSourced
How this article fits with the rest of the series

This article covers the fiber itself: the glass, its structure, how it is made and how it transmits light. How fibers are joined to each other and to chips is the subject of our explainer on optical connectors and fiber attach; carrying several wavelengths on one fiber is covered in our explainer on wavelength multiplexing and microrings; on-chip waveguides and fiber couplers are covered in our silicon photonics explainer; and putting optics inside a switch package is covered in our explainer on co-packaged optics (our own division of topics).

2. Structure: the refractive index is made with dopants

At the center of an optical fiber is the core, through which light travels, and around it is the cladding. Because the core has a higher refractive index than the cladding, light moves along the core by repeated total internal reflection at the boundarySourced. The IEICE Knowledge Base, a reference work published by Japan's Institute of Electronics, Information and Communication Engineers, explains how the index is set as follows.

What the IEICE Knowledge Base says (summary)
  • Silica optical fiber is mostly SiO2, with GeO2 and F used as dopantsSourced
  • Adding GeO2 to pure silica raises the refractive index; adding F lowers itSourced
  • In the GeO2-doped core type, the core is GeO2-doped silica and the cladding is pure silicaSourced
  • In the pure-silica core type, the core is pure silica and the cladding is F-doped silicaSourced
  • The index difference obtainable with ordinary doping is 1% or lessSourced
Fiber cross-section, and two ways of setting the refractive index (schematic) Not to scale. The core is drawn enlarged so that it can be seen. Coating (resin) Cladding Core Glass diameter 125 µm With coating, e.g. 250 µm GeO2-doped core type Pure-silica core type Raised with GeO2 Core is pure silica Lowered by F Lowered by F Pure silica Pure silica Radial position → Radial position → Cladding = index of pure silica Dotted line = pure silica index Note: dopant effects follow the IEICE Knowledge Base [Ref. 3]. Vertical axis = refractive index (up = higher); steps are schematic.
Fig. 1 Conceptual diagram (vector drawing). That GeO2 raises and F lowers the refractive index, and the make-up of the two types, follow the IEICE Knowledge Base [Ref. 3]; the 125 µm cladding diameter follows ITU-T G.652 [Ref. 1]; the 250 µm coating diameter example follows Sumitomo Electric [Ref. 10]. Circle sizes, index steps and coating thickness are schematic and are not real proportions.

The point to hold on to is that the index difference is only 1% or lessSourced. Core and cladding are almost the same glass. Because the light is held in place by a tiny difference in composition, any fluctuation in composition or roughness at the boundary shows up directly in transmission performance (our commentary).

3. Single-mode and multimode

Optical fibers fall into two groups according to the number of "modes", or ways the light can travel through them. A fiber that supports only one mode is single-mode fiber; one that supports two or more is multimode fiber. The Knowledge Base gives core diameters of about 8 to 10 µm for single-mode and 50 or 62.5 µm for multimodeSourced.

Single-mode and multimode (schematic) Values are those specified in ITU-T G.652.D (single-mode) and G.651.1 (multimode). Single-mode (G.652.D) Multimode (G.651.1) Mode field diameter 8.6-9.2 µm (1310 nm) Zero-dispersion wavelength 1300-1324 nm Cable cut-off wavelength max. 1260 nm Loss max. 0.30 dB/km (1530-1565 nm) Carrier networks, links between data centers Core diameter 50 µm / NA 0.20 Modal bandwidth at least 500 MHz km 550 m at 1 Gbit/s (850 nm) Arrival time differs from mode to mode Cabling inside buildings, for example Note: G.652.D values [Ref. 1]; G.651.1 values [Ref. 2]. G.651.1 is OM2-equivalent; higher-bandwidth OM grades are set by IEC etc. Note: the light paths (straight and wavy) are schematic and are not the shapes of real propagation modes.
Fig. 2 Conceptual diagram (vector drawing). Single-mode values are those specified in ITU-T G.652.D [Ref. 1], multimode values those specified in ITU-T G.651.1 [Ref. 2]. The ratio of core widths and the light paths are schematic and do not show real dimensions or the shapes of propagation modes. The split by application is our own framing.
ItemSingle-mode (ITU-T G.652.D)Multimode (ITU-T G.651.1)
Where the light travelsMode field diameter 8.6 to 9.2 µm (range of nominal values, 1310 nm)Core diameter 50 µm (tolerance ±2.5 µm)
Cladding diameter125.0 µm (tolerance 0.7 µm)125 µm (tolerance 1 µm)
Main transmission properties specifiedLoss, chromatic dispersion, polarization mode dispersionDifference in arrival time between modes (modal bandwidth of at least 500 MHz·km at 850 nm and 1300 nm)
Intended useOptimized for the 1310 nm region, usable in the 1550 nm region too1 Gbit/s over 550 m with 850 nm transceivers (1000BASE-SX)
Tensile guarantee (proof stress)Min. 0.69 GPaMin. 0.69 GPa

All Sourced (ITU-T G.652 [Ref. 1], ITU-T G.651.1 [Ref. 2]). G.651.1 itself explains that multimode fiber is no longer common in telecommunications networks, but remains widely used for Ethernet cabling inside buildings at speeds above 10 Gbit/s.

Multimode fiber has a wide core that makes it easy to get light in, but each mode travels a different path, so the pulse breaks up more the further it goes. That is why G.651.1 specifies performance in MHz·km (bandwidth times distance)Sourced. The workhorse for long distances is single-mode fiber, which has only one mode (our commentary).

4. Loss: where the light goes

Loss is expressed in dB/km (how many decibels are lost per kilometer). 3 dB roughly halves the light intensity, 10 dB cuts it to a tenth and 20 dB to a hundredth. The cable loss limits specified in ITU-T G.652.D are as followsSourced.

WavelengthG.652.D cable loss (max.)Notes in the standard
1310 to 1625 nm0.40 dB/kmIf the range is extended down to 1260 nm, add 0.07 dB/km for Rayleigh scattering to the 1310 nm value
1383 nm (after hydrogen ageing)0.40 dB/kmHydrogen ageing is done as a type test
1530 to 1565 nm (C-band)0.30 dB/kmThe band with the lowest loss
Short cables such as indoor and drop cablesThe values above do not applyFor example, indoor cable at 1.0 dB/km or less (IEC 60794-2-11)

All Sourced (ITU-T G.652 (08/2024), Table 2 [Ref. 1]). An appendix to the same Recommendation gives representative values for concatenated links of 0.275 dB/km in the C-band and 0.5 dB/km at 1260 to 1360 nm.

The notes in the standard hint at what the loss is made of. Simply shortening the wavelength from 1310 nm to 1260 nm means adding 0.07 dB/km for Rayleigh scatteringSourced. Rayleigh scattering is light being scattered by microscopic fluctuations in the density and composition of the glass, and it gets stronger at shorter wavelengths (our commentary). The region around 1383 nm is known as the wavelength where absorption by hydroxyl groups (OH) in the glass appears, and G.652.D sets a limit there too, after hydrogen ageingSourced.

Our calculation: how much light is left after 100 km

Apply each loss value over 100 kmOur calculation. The fraction remaining is 10^(−loss in dB ÷ 10).

  • G.652.D limit of 0.40 dB/km: 40 dB → 0.01% (one ten-thousandth)
  • G.652.D C-band limit of 0.30 dB/km: 30 dB → 0.1% (one thousandth)
  • Sumitomo Electric's record of 0.1397 dB/km (1566 nm): about 14.0 dB → about 4.0%
  • Hollow-core fiber at 0.091 dB/km (1550 nm): 9.1 dB → about 12%

Assumptions and limits: these values simply stack up the loss of the fiber alone over 100 km and exclude splice, connector and bending losses. A limit in a standard, a record presented at a conference and a value measured in a peer-reviewed paper are different kinds of number, and this is not a comparison under the same conditions.

Loss (dB/km) and the fraction of light left after 100 km (our calculation) Bar length = loss (dB/km). Figure at right = fraction of light remaining after 100 km. G.652.D limit (1310-1625 nm) G.652.D limit (C-band) Silica fiber record Hollow-core fiber 0.40 0.30 0.1397 0.091 0.01% 0.1% about 4% about 12% A 0.1 dB/km gap becomes 10 dB, a factor of 10, over 100 km Note: 0.40 and 0.30 are ITU-T G.652.D limits [Ref. 1]; 0.1397 is Sumitomo Electric's OFC 2024 value (1566 nm) [Ref. 5]; 0.091 is the value measured in a Nature Photonics paper (1550 nm) [Ref. 7]. Measurement conditions are not the same. Note: fractions remaining were converted by this article as 10^(-dB/10). Splice and bending losses are excluded.
Fig. 3 Drawing that includes our calculation (vector drawing). Loss values follow ITU-T G.652.D [Ref. 1], Sumitomo Electric [Ref. 5] and Petrovich et al. (Nature Photonics, 2025) [Ref. 7]. The fractions remaining after 100 km were calculated by this article and are not published figures. The figure places a limit from a standard, a record presented at a conference and a value measured in a paper side by side; it is not a comparison under identical conditions.

5. Dispersion: different colors arrive at different times

A light pulse contains slightly different wavelengths, or colors. The refractive index of glass varies with wavelength, so each color travels at a different speed, and the pulse spreads the further it goes. This is chromatic dispersion (our commentary). In G.652.D fiber, the wavelength at which dispersion is zero (the zero-dispersion wavelength) lies between 1300 and 1324 nmSourced. Loss, however, is lowest in the 1550 nm region, and there dispersion is not zero. An appendix to G.652 gives a representative value at 1550 nm of 17 ps/(nm·km)Sourced.

Our calculation: how far a pulse spreads over 100 km
  • Assumptions: dispersion of 17 ps/(nm·km) (the representative value in the G.652 appendix), a distance of 100 km, and a signal spectral width taken to be 0.1 nm
  • 17 × 100 × 0.1 = 170 ps Our calculation
  • One bit at 10 Gbit/s lasts 100 ps. The spread works out longer than one bit

Assumptions and limits: the 0.1 nm spectral width is an assumption made in this article and varies widely with the modulation format and light source. Real systems deal with this through dispersion compensation or digital signal processing, and the G.652 appendix itself notes that some form of dispersion compensation is often used when the fiber is operated in the 1550 nm regionSourced. Compensation by digital signal processing is covered in our explainer on coherent optical communications and DSP.

So within one strand of glass, the wavelength of lowest loss and the wavelength of zero dispersion do not coincide. Designs such as dispersion-shifted fiber, which moves the zero-dispersion wavelength towards 1550 nm, were developed to resolve that mismatch through the design of the index profile. The Knowledge Base explains that controlling multilayer index changes within the core shifts the zero-dispersion wavelength towards longer wavelengthsSourced.

6. A materials engineer's view (1): how the glass is made

Conceptual image of the tip of a thick transparent glass rod glowing soft orange-red against a dark background, with an extremely fine thread extending downward from it
Fig. 4 Conceptual image (AI-generated). An impression of fiber drawing, in which a preform is heated and pulled out thin. It does not show real equipment, a real furnace, or the actual size or color of a preform.

Optical fiber is made by first producing a thick rod of glass (the preform), then heating it and drawing it out thin. The Knowledge Base divides manufacturing into three stages: making the soot preform, consolidating it into transparent glass, and drawingSourced.

How optical fiber is made (process flow, using the VAD method as the example) Arranged by this article from the IEICE Knowledge Base and published material from NTT and others. 1 Gases and flame 2 Soot preform 3 Consolidation 4 Draw and coat SiCl4 and GeCl4 are fed into the flame of an oxyhydrogen burner SiCl4 + 2H2O -> SiO2 + 4HCl flame hydrolysis Fine glass particles build up on a seed rod A network structure with voids of about 0.1-1 µm Rotated while being pulled up along its axis Heated in an electric furnace, it sinters and shrinks; the voids close A fully transparent glass body = the fiber preform VAD makes drying easy Heated to about 2,000 °C in a carbon furnace Drawn at a few hundred m/min to 125 µm across Diameter is monitored and resin applied at once Note: the reaction, void size, about 2,000 °C and a few hundred m/min follow the IEICE Knowledge Base [Ref. 3]. Note: VAD steps and strengths follow the NTT, Furukawa Electric, Sumitomo Electric and Fujikura release [Ref. 4] and [Ref. 3]. Note: besides VAD, preforms are made by MCVD, OVD and PCVD, and steps 1 and 2 look different in each. Note: the division into four steps is this article's own.
Fig. 5 Conceptual diagram (vector drawing). The content of each step follows the IEICE Knowledge Base [Ref. 3] and the joint release from NTT and three other companies [Ref. 4]. The division into four steps and the way they are drawn are this article's own, and do not show the process at any particular plant.

There are several ways of making the preform, each with its own character. When the VAD method was recognized as an IEEE Milestone in 2015, NTT, Furukawa Electric, Sumitomo Electric and Fujikura compared the methods in their joint release as followsSourced.

MethodOriginWhat the four companies' release says
VAD (vapor-phase axial deposition)Invented in 1977 in the course of joint research by NTT (then the Nippon Telegraph and Telephone Public Corporation), Furukawa Electric, Sumitomo Electric and Fujikura (then Fujikura Cable Works)Porous glass is grown while being pulled up along a single axis, so large preforms can be made. Well suited to volume production
MCVD1974, AT&T Bell LaboratoriesLittle contamination by impurities, which suits low loss, but deposition takes place inside a glass tube, which limits the preform's diameter and length. The reaction is thermal oxidation and so is slow
OVDCorningDeposits around a mandrel. The preform can be made larger in diameter, but its length is limited by the mandrel. The mandrel has to be pulled out before consolidation, which tends to scratch the contact surface

All Sourced (NTT news release of 21 May 2015, joint release by the four companies [Ref. 4]). The Knowledge Base [Ref. 3] also lists the PCVD method, and describes VAD as a manufacturing method that makes dehydration, essential for low loss, easy to carry out.

Why this matters for materials engineers: this is powder processing and sintering

"Optical fiber" calls up the image of molten glass being stretched out. But the heart of preform making is a sequence of making fine particles by a gas-phase reaction, building them up into a porous body, and sintering it to full density. The Knowledge Base describes the soot preform as a network structure with voids of about 0.1 to 1 µm, and says that when the temperature in the electric furnace is raised, sintering proceeds, the volume shrinks and the voids in the preform collapse, giving transparent glassSourced.

That is a familiar string of words to anyone in ceramics or powder metallurgy. What differs is how demanding the requirements are. As Section 4 showed, loss has to stay below 0.3 dB per kilometer, and the Knowledge Base calls dehydration, drawing water out while the body is still porous, essential for low lossSourced. This is the step that removes the hydroxyl groups blamed for the absorption around 1383 nm mentioned in Section 4 (our commentary). A porous body has a large surface area and lets gas through to its interior, so the stage before densification is the only chance to get impurities out (our commentary).

The other point to look at is the resin coating applied after drawing. The Knowledge Base lists four roles for the coating: preventing surface flaws, relieving lateral pressure, removing unwanted modes and making the fiber easier to handleSourced. And when Sumitomo Electric broke the world loss record in 2017, it explained that it had done so by further improving its technologies for the glass and the resin coatingSourced. A resin that the light is not supposed to pass through is part of a loss record. One reading is that because the coating eases lateral pressure and suppresses microbending, it ends up affecting loss (our commentary).

7. Many paths in one strand: multicore fiber

There is a limit to how much information one fiber can carry. Multicore fiber (MCF) addresses this by placing several cores side by side within the same 125 µm of glass. NICT points out that with a standard-diameter fiber, existing manufacturing equipment can be used when it is made into cableSourced.

AnnouncementContentNature
NICT (May 2022)Transmission experiment over a 4-core fiber of standard diameter (0.125 mm), using 801 wavelengths across 20 THz of bandwidth: 1.02 petabits per second over 51.7 km. A post-deadline paper at CLEO 2022Transmission experiment with a research fiber
Sumitomo Electric (September 2023)Announced volume production of "2C Z-PLUS Fiber ULL", a 2-core ultra-low-loss MCF. Loss 0.158 dB/km (1550 nm, typical), counter-propagating crosstalk of −43 dB or less, glass diameter 125 µm, coating diameter 250 µmProduct (typical values)
NTT (November 2024)Assembled a line-up of technologies for building, maintaining and operating 4-core MCF: fusion splicing with automatic rotational alignment from side-view images; a fan-in/fan-out (FIFO) device made of two stacked layers of silica PLC; and a cable of about 20 mm diameter holding up to 8,000 coresEnabling technologies established. Aims for practical deployment and international standardization around 2027
Meta, NEC and Sumitomo Electric (September 2026)2-core MCF adopted for "Petal", a transatlantic submarine cable of about 7,000 km between the United States and France. 24 fiber pairs (48 fibers), giving 48 core pairsScheduled to enter service in 2029

Based on announcements by NICT [Ref. 8], Sumitomo Electric [Refs. 10 and 11] and NTT [Ref. 9]. Performance values and configurations are Sourced. NTT's timing for practical deployment and international standardization, and Petal's entry into service, are Not yet confirmed. Wording such as "world first" is each company's own claim.

Changing the light path inside the same diameter (schematic cross-sections) All schematic. Core positions and sizes, and the hollow structure, differ from real designs. Conventional (1 core) 4-core MCF Hollow core 125 µm diameter Still 125 µm, four light paths Red dot = marker for core numbering Light travels in the central air Guided by thin glass around it Note: 4-core MCF and marker per NTT [Ref. 9]; hollow-core principle per Petrovich et al. [Ref. 7]. Shapes are our schematic.
Fig. 6 Conceptual diagram (vector drawing). The 4-core MCF layout and the marker used to number the cores follow NTT [Ref. 9]; that hollow-core fiber guides light by surrounding an air core with a precisely engineered glass microstructure follows Petrovich et al. [Ref. 7]. Core layout, diameters and the shape of the hollow structure are all schematic and are not real designs.

What catches a materials engineer's eye in NTT's announcement is that joining the fiber was the barrier to practical use. Because the cores sit away from the center, connecting two MCFs requires rotational alignmentSourced. Instead of viewing the end face directly, NTT says it derives the rotation angle from the brightness distribution in a side-view image of the fiber, so that the function can be built into an existing portable fusion splicerSourced. To connect to single-core fiber, it stacked planar lightwave circuits (PLCs) in two layers, each layer breaking out two coresSourced. The connector side of the problem is also touched on in our explainer on optical connectors and fiber attach.

8. Taking the glass out of the path: hollow-core fiber

In 2025 a research group including Petrovich and Poletti reported measurements of an optical fiber whose core is air in Nature PhotonicsSourced. The numbers given in the paper's abstract are as follows.

ItemAs stated in the abstract
State of the art for conventional silica fiberLoss of 0.14 dB/km, with 26 THz of bandwidth below 0.2 dB/km; little changed over the past 40 years
This hollow-core fiber0.091 dB/km at 1,550 nm, with 66 THz of bandwidth below 0.2 dB/km
SpeedLight travels 45% faster
StructureInstead of a solid glass core, an air core surrounded by a meticulously engineered glass microstructure

All Sourced (abstract of Petrovich et al., Nature Photonics 19, 1203–1208 (2025) [Ref. 7]). These are research measurements, not the specification of a mass-produced product.

Our calculation: what "45% faster" changes over 1,000 km
  • Assumptions: the speed of light in vacuum is about 300,000 km/s; the group index of silica fiber is taken to be 1.47
  • In air (index close to 1): about 3.3 µs per km
  • In silica fiber: 3.3 × 1.47 = about 4.9 µs per km Our calculation
  • The difference is about 1.6 µs per km, or about 1.6 ms over 1,000 km

Assumptions and limits: the group index of 1.47 is an assumption made in this article and varies with fiber type and wavelength. 4.9 ÷ 3.3 is about 1.47, broadly consistent with the paper's "45% faster". It is a number that matters in uses where latency counts for more than capacity, such as financial trading or synchronization between data centers (our commentary).

9. A materials engineer's view (2): perfect the glass, or abandon it

Why this matters for materials engineers: the same problem has two opposite answers

Efforts to cut loss are now heading in two directions.

  • Perfect the glass: Sumitomo Electric commercialized its pure-silica-core fiber, whose core is made of pure silica, in 1988, and presented 0.1419 dB/km at a conference in 2017 and 0.1397 dB/km (1566 nm) in 2024Sourced. According to the Knowledge Base, the pure-silica core type lowers the index by adding fluorine to the cladding insteadSourced. In other words, the design leaves the core, where most of the light travels, free of dopant and creates the index difference on the cladding side
  • Abandon the glass: hollow-core fiber sends most of the light through air. The paper reported 0.091 dB/km after noting that the loss and bandwidth of silica fiber had barely changed over the past 40 yearsSourced

Both are answers to the same question: how far can you reduce the material that causes loss in the place where the light travels? The first took the dopant out of the core; the second took the glass itself out of the core (our commentary).

From the materials side, the skills demanded are very different. The first is about control of composition and thermal history, pushing dopant distribution, dehydration, sintering, and draw temperature and tension to the limit. The second is about control of shape, drawing out the thin glass membranes that surround the air holes over many kilometers without their thickness and arrangement collapsing. As the paper puts it, a "meticulously engineered glass microstructure"Sourced. What makes the field interesting is that both end up in the same process: drawing glass (our commentary).

10. Open problems and unconfirmed points

(1) Standardization of multicore fiber is still ahead

For its 4-core MCF transmission line, NTT says it aims to pursue practical deployment and international standardization with around 2027 as the targetNot yet confirmed. ITU-T G.652 (08/2024), which this article draws on, is the Recommendation for conventional single-mode fiber. How standards for MCF splicing, fan-in/fan-out and cable will be settled is not yet fixed (our commentary).

(2) Use in submarine cable is still at the planning stage

Petal is scheduled to enter service in 2029Not yet confirmed. There is no track record yet of how reliably an ocean-crossing cable that uses 2-core MCF for every fiber performs in actual operation.

(3) No volume-production record for hollow-core fiber could be confirmed

0.091 dB/km is a value measured in a peer-reviewed paperSourced. Within the scope of this article, no primary source could be found stating that hollow-core fiber with this performance has been mass-produced and installed in a commercial network. Production lengths, splice loss and how to connect it to existing fiber are also outside the scope of this article.

(4) Records and product values are different things

0.1397 dB/km is a record value presented at a conference; 0.158 dB/km (the 2-core MCF) and the product values Sumitomo Electric gives are typical values; and G.652.D's 0.30 dB/km is a limit set by a standardSourced. Because the numbers are of different kinds, they cannot simply be lined up to decide which is better (our commentary).

The article in summary
  • Optical fiber traps light using glass with an index difference of 1% or less. The index is raised with GeO2 and lowered with FSourced
  • G.652.D caps loss at 0.30 dB/km in the C-band. After 100 km, a thousandth of the light is leftOur calculation
  • Preforms are made by powder processing: fine particles are formed in the gas phase, built into a porous body and sintered. Dehydration at the porous stage is the key to low lossSourced
  • The resin coating also affects loss. Sumitomo Electric cited its resin-coating technology as one of the means behind a new recordSourced
  • For multicore fiber, the joining technology is the key to practical use. A transatlantic cable using 2-core MCF is scheduled to enter service in 2029Not yet confirmed
  • Hollow-core fiber has been reported at 0.091 dB/km. Light travels through it 45% fasterSourced

11. Glossary

Core / cladding
The central region through which light travels (the core) and the lower-index region that surrounds it (the cladding).
Single-mode fiber
Fiber in which only one mode can propagate. Core diameter about 8 to 10 µm. Used for long distances.
Multimode fiber
Fiber that carries several modes. Core diameter 50 or 62.5 µm. Used for short distances.
Mode field diameter (MFD)
The diameter over which the light is effectively spread. Used to estimate splice loss.
dB/km
Loss per kilometer. 10 dB cuts the light intensity to a tenth.
Rayleigh scattering
Scattering caused by microscopic fluctuations in density and composition within the glass. Stronger at shorter wavelengths.
Chromatic dispersion
Pulse spreading caused by different wavelengths travelling at different speeds. Unit: ps/(nm·km).
Zero-dispersion wavelength
The wavelength at which chromatic dispersion is zero. 1300 to 1324 nm in G.652.
C-band
The 1530 to 1565 nm wavelength band, where silica fiber has its lowest loss.
Preform
The cylindrical block of glass from which optical fiber is made. It is heated and drawn.
Soot
Fine glass particles produced by flame hydrolysis, deposited to form a porous preform.
VAD (vapor-phase axial deposition)
A preform-making method in which soot is grown along the axis. Invented in Japan in 1977.
Drawing
The step in which a preform is heated in a furnace and pulled into a thin fiber, followed immediately by resin coating.
Multicore fiber (MCF)
A fiber with several cores inside a single strand.
Fan-in/fan-out (FIFO)
A component that connects each core of an MCF to its own ordinary fiber.
Hollow-core fiber
A fiber whose core is air, with light guided by the surrounding glass microstructure.
Proof stress
The tensile stress applied along the full length of the fiber during manufacture as a guarantee. At least 0.69 GPa in G.652.

12. References (primary sources)

  1. ITU-T "G.652 (08/24) Characteristics of a single-mode optical fibre and cable" (the full Recommendation is freely available as a PDF) https://www.itu.int/rec/T-REC-G.652-202408-I/en
  2. ITU-T "G.651.1 (11/18) Characteristics of a 50/125 µm multimode graded index optical fibre cable for the optical access network" https://www.itu.int/rec/T-REC-G.651.1-201811-I/en
  3. IEICE Knowledge Base Group 5, Part 2, Chapter 3 "Optical fiber" (Kazuhide Nakajima and Masaharu Ohashi, ver. 1, 23 May 2017, PDF, in Japanese) https://www.ieice-hbkb.org/files/05/05gun_02hen_03.pdf
  4. NTT, Furukawa Electric, Sumitomo Electric and Fujikura "The VAD method, a manufacturing method for high-quality optical fiber well suited to volume production, recognized as a prestigious IEEE Milestone", 21 May 2015 (in Japanese) https://group.ntt/jp/newsrelease/2015/05/21/150521a.html
  5. Sumitomo Electric Industries "Conference presentation on lowering the loss of silica glass optical fiber", March 2024 (OFC 2024, "Record Low Loss Optical Fiber With 0.1397 dB/km"; release in Japanese) https://sumitomoelectric.com/jp/press/2024/03/prs036
  6. Sumitomo Electric Industries "World record for optical fiber transmission loss updated", March 2017 (OFC 2017 post-deadline paper; release in Japanese) https://sei.co.jp/company/press/2017/03/prs029.html
  7. M. Petrovich, E. Numkam Fokoua, …, F. Poletti (12 authors) "Broadband optical fibre with an attenuation lower than 0.1 decibel per kilometre", Nature Photonics 19, 1203–1208 (2025) https://doi.org/10.1038/s41566-025-01747-5
  8. NICT "World first: 1 petabit per second transmission achieved over 4-core optical fiber", 19 May 2022 (in Japanese) https://www.nict.go.jp/press/2022/05/19-1.html
  9. NTT "Line-up of construction, operation and maintenance technologies for multicore optical fiber, quadrupling the capacity of a single fiber", 15 November 2024 (in Japanese) https://group.ntt/jp/newsrelease/2024/11/15/241115a.html
  10. Sumitomo Electric Industries "World first: volume production of ultra-low-loss multicore optical fiber achieved", September 2023 (in Japanese) https://sumitomoelectric.com/jp/press/2023/09/prs115
  11. Meta, NEC and Sumitomo Electric Industries "Collaboration to build Petal, the world's first commercial petabit-class optical submarine cable system", September 2026 (in Japanese) https://sumitomoelectric.com/jp/press/2026/09/prs105

13. Claim-to-source audit

Claim in the textBasisLabel
That G.652 is the Recommendation for single-mode fiber with a zero-dispersion wavelength near 1310 nm, usable in the 1550 nm region too. For G.652.D: mode field diameter 8.6 to 9.2 µm (1310 nm); cladding diameter 125.0 µm (tolerance 0.7 µm); cable cut-off wavelength max. 1260 nm; proof stress min. 0.69 GPa; zero-dispersion wavelength 1300 to 1324 nm; cable loss max. 0.40 dB/km (1310 to 1625 nm), 0.40 dB/km (1383 nm, after hydrogen ageing) and 0.30 dB/km (1530 to 1565 nm). That 0.07 dB/km is added for Rayleigh scattering when extending to 1260 nm. That the values do not apply to short cables (example of 1.0 dB/km or less for indoor cable). The appendix's representative values (C-band 0.275 dB/km, 1260 to 1360 nm 0.5 dB/km, D1550 = 17 ps/(nm·km)). That dispersion compensation is often used in the 1550 nm regionReference 1 https://www.itu.int/rec/T-REC-G.652-202408-I/enSourced
For G.651.1: core diameter 50 µm (tolerance 2.5 µm), cladding diameter 125 µm (tolerance 1 µm), numerical aperture 0.20, modal bandwidth min. 500 MHz·km (850 nm and 1300 nm), proof stress 0.69 GPa, 550 m at 1 Gbit/s (1000BASE-SX), and OM2 equivalence. That multimode is no longer common in telecommunications networks but remains widely used for cabling inside buildingsReference 2 https://www.itu.int/rec/T-REC-G.651.1-201811-I/enSourced
That the main component is SiO2, with GeO2 raising and F lowering the index. The make-up of the GeO2-doped core and pure-silica core types. That the index difference from doping is 1% or less. That an outer diameter of 125 µm is standard. Core diameters (single-mode about 8 to 10 µm, multimode 50 or 62.5 µm). The three manufacturing stages, and the MCVD, OVD, PCVD and VAD methods. That VAD suits large preforms, volume production and dehydration. The flame hydrolysis reaction. That the soot preform is a network structure with voids of about 0.1 to 1 µm, and that it sinters, shrinks and becomes transparent in an electric furnace. Drawing at a few hundred m/min in a carbon furnace at about 2,000 °C, followed by coating. The four roles of the coating. That multilayer index changes shift the zero-dispersion wavelengthReference 3 https://www.ieice-hbkb.org/files/05/05gun_02hen_03.pdfSourced
That the VAD method was invented in 1977 in the course of joint research by the four companies and recognized as an IEEE Milestone in 2015. The comparison of VAD, MCVD (1974, AT&T Bell Laboratories) and OVD (Corning)Reference 4 https://group.ntt/jp/newsrelease/2015/05/21/150521a.htmlSourced
That Sumitomo Electric achieved 0.1397 dB/km at 1566 nm and presented it at OFC 2024. That it commercialized pure-silica-core fiber (Z Fiber) in 1988. The definition of pure-silica-core fiberReference 5 https://sumitomoelectric.com/jp/press/2024/03/prs036Sourced
That it achieved 0.1419 dB/km (1560 nm) in 2017, explaining that it did so by further improving its glass and resin-coating technologiesReference 6 https://sei.co.jp/company/press/2017/03/prs029.htmlSourced
That the hollow-core fiber showed 0.091 dB/km at 1,550 nm, 66 THz of bandwidth below 0.2 dB/km and 45% faster transmission. That the state of the art for conventional silica fiber is 0.14 dB/km and 26 THz, little changed over the past 40 years. That the structure is an air core surrounded by a meticulously engineered glass microstructure. Journal, volume, pages and yearReference 7 https://doi.org/10.1038/s41566-025-01747-5Sourced
That NICT achieved 1.02 petabits per second over 51.7 km with a standard-diameter 4-core fiber (801 wavelengths, 20 THz). That it was a post-deadline paper at CLEO 2022Reference 8 https://www.nict.go.jp/press/2022/05/19-1.htmlSourced
That NTT assembled a line-up of technologies for building, maintaining and operating 4-core MCF. That rotational alignment is required; automatic rotational alignment from the brightness distribution of side-view images; the marker; a FIFO with a two-layer silica PLC structure; a cable of about 20 mm diameter with up to 8,000 coresReference 9 https://group.ntt/jp/newsrelease/2024/11/15/241115a.htmlSourced
The announcement of volume production of the 2-core MCF "2C Z-PLUS Fiber ULL"; loss 0.158 dB/km (typical); counter-propagating crosstalk of −43 dB or less; glass diameter 125 µm and coating diameter 250 µmReference 10 https://sumitomoelectric.com/jp/press/2023/09/prs115Sourced
That Petal runs about 7,000 km between the United States and France and is configured with 2-core MCF, 24 fiber pairs (48 fibers) and 48 core pairsReference 11 https://sumitomoelectric.com/jp/press/2026/09/prs105Sourced
Petal's planned entry into service in 2029. NTT's practical deployment and international standardization of its 4-core transmission line, targeted for around 2027Company plans and outlooks, not results. Reference 11 https://sumitomoelectric.com/jp/press/2026/09/prs105 / Reference 9Not yet confirmed
The fraction of light remaining after 100 km (0.01%, 0.1%, about 4%, about 12%). The 170 ps of spreading from dispersion. The latency difference with hollow core (about 1.6 µs per km, about 1.6 ms over 1,000 km)Our calculation. Conversion by 10^(−dB/10), a spectral width of 0.1 nm, a group index of 1.47 and a speed of light of about 300,000 km/s are assumptions made in this article. Splice and bending losses are not includedOur calculation
That Rayleigh scattering is stronger at shorter wavelengths, and that the region around 1383 nm is known as the OH absorption wavelengthCommentary given as a general physical explanation. The basis in the standard is the notes in Reference 1 (the 0.07 dB/km addition and the 1383 nm limit)Commentary
The reading that the porous stage is the only chance to remove impurities. The reading that the resin coating affects loss by suppressing microbending. The contrast between perfecting and abandoning the glass. The explanation of why multimode is unsuited to long distancesThis article's commentary based on published content. Not views expressed by the companies or institutionsCommentary
A track record of volume production and commercial installation for hollow-core fiberNot stated, because no primary source could be confirmed within the scope of this articleCommentary
That Figs. 1, 2, 5 and 6 are explanatory drawings and Fig. 3 a drawing that includes our calculation, and that the hero image and Fig. 4 are AI-generated imagesOur noteCommentary

Last updated 25 September 2026. Sources are limited to primary material (ITU-T Recommendations, a learned society's knowledge base, peer-reviewed papers, and official announcements from companies and public research institutes); market estimates from research firms are not used. Values specified in standards (limits), record values presented at conferences, typical product values and values measured in peer-reviewed papers are different in kind and are kept distinct. A track record of volume production and commercial installation for hollow-core fiber, and when splicing standards for multicore fiber will be settled, are not stated because they could not be confirmed in published primary sources. All figures are for explanation. Figs. 1, 2, 5 and 6 are vector drawings, Fig. 3 is a vector drawing that includes our calculation, and the hero image and Fig. 4 are AI-generated images; none of them shows a real cross-section photograph, manufacturing equipment or physical product.

🌐 Japanese