Optical fibers are made by depositing ultra-pure glass soot onto a rotating rod, consolidating it into a solid preform, and then drawing that preform in a tower at roughly 2,000 C into hair-thin filament at rates up to 3,000 meters per minute. The drawn fiber is coated, cured, and spooled the moment it leaves the furnace. Every step exists to keep one property intact: light loss measured in tenths of a decibel per kilometer.

That one-sentence answer hides an enormous amount of engineering. The glass must be purer than almost any material humans manufacture. A single transition-metal impurity at parts-per-billion levels can double the attenuation of the finished fiber. This explainer walks the full production chain, from vapor chemistry to the final spool, and explains why each stage looks the way it does.

The two-stage logic of fiber manufacturing

Fiber production splits into two distinct operations, and conflating them is the most common source of confusion. Stage one builds a preform, a solid glass cylinder typically 1 to 10 centimeters in diameter and 1 to 2 meters long. Stage two draws that preform into kilometers of fiber while preserving its cross-sectional geometry at a scale thousands of times thinner.

The reason for the split is economic and physical. You cannot build glass at fiber dimensions directly with any useful precision. Instead you build it at a large, workable scale where chemistry is controllable, then stretch it. Drawing a 100-millimeter preform down to 125 micrometers is a reduction of roughly 800 times in diameter. That reduction multiplies the preform's length by about 640,000, so a 1-meter preform yields hundreds of kilometers of fiber. Geometry shrinks uniformly, which is why the core-to-cladding ratio set in the preform survives the draw.

The purity requirement drives the chemistry choice. A conventional glass melt in a crucible contaminates the batch with iron, copper, and hydroxyl ions from the container and the furnace atmosphere. Those impurities absorb light in the telecom windows. So fiber makers abandon melting entirely and build the glass molecule by molecule from high-purity vapors, a family of processes known collectively as chemical vapor deposition.

Building the preform: soot, consolidation, and vapor chemistry

The dominant industrial process is outside vapor deposition, or OVD, alongside a close relative called vapor axial deposition and a modified form called modified chemical vapor deposition used heavily for specialty fiber. In OVD, a lathe rotates a target rod while a burner traverses its length. The burner feeds silicon tetrachloride and germanium tetrachloride vapors plus oxygen into a flame.

The chemistry is oxidation. Silicon tetrachloride plus oxygen yields silicon dioxide plus chlorine gas, and the germanium compound oxidizes in parallel. What lands on the rod is not glass but a white soot, a fluffy agglomeration of nanoscale oxide particles, exactly like the silica in fumed silica. Layer after layer deposits, with the germanium concentration varied layer by layer to build the refractive index profile. Germanium raises the index, so germania-rich layers become the light-guiding core.

The soot body is porous and mechanically fragile. Consolidation is the step that turns it into glass. The soot is moved into a furnace at roughly 1,300 to 1,500 C and treated with a drying gas, classically chlorine or a chlorine-bearing compound, which strips hydroxyl ions out of the pores. Then the furnace temperature rises toward 1,500 to 1,700 C, the porous body sinters and collapses into a clear solid rod. The drying step is where the OH problem is solved. Hydroxyl absorbs strongly near 1,385 nanometers, right inside a telecom band, and even a few parts per million cause unacceptable loss.

One counterintuitive detail: the fiber core is not pure silica. Pure silica would work as a cladding but needs an index-raising dopant for the core. Germanium is the classic choice in OVD, though some makers use other dopants or a pure-silica core with a depressed-index cladding made by fluorine doping. Which design wins depends on the target application, not on process superiority.

Quantified context: attenuation, dimensions, and draw speeds

Standard single-mode telecom fiber carries a 9-micrometer core inside a 125-micrometer cladding, with the whole strand protected by coatings that bring the outer diameter to about 245 micrometers. Multimode fiber for short reach uses 50 or 62.5-micrometer cores in the same 125-micrometer body. These numbers are standards, not manufacturer choices, so connectors and splices built by anyone fit anyone.

Attenuation is the headline performance number. Silica glass sets a fundamental floor through Rayleigh scattering, which scales as the inverse fourth power of wavelength and cannot be doped away. Commercial single-mode fiber at 1,550 nanometers typically achieves attenuation near 0.18 to 0.22 decibels per kilometer, against a theoretical Rayleigh floor in the vicinity of 0.15 decibels per kilometer at that wavelength. At 1,310 nanometers, attenuation runs higher, around 0.3 to 0.35 decibels per kilometer. Loss budget arithmetic makes these decimals matter: at 0.2 decibels per kilometer, an unrepeated link can span tens of kilometers before an amplifier is needed, and a 0.02 decibels-per-kilometer improvement translates to meaningful reach gains across a network.

Draw speed is the other quantified scale. Tower draw rates range from tens of meters per minute for large-core specialty fiber up to 2,000 to 3,000 meters per minute for standard telecom fiber. The preform is fed into the furnace slowly while the fiber is pulled fast, and the controller regulates the ratio so the final diameter stays within less than a micrometer of target along the entire spool.

The draw tower, step by step

The draw sequence is continuous and self-regulating, and it is worth walking through in order.

The preform is loaded into a feed mechanism at the top of a tower that can stand 10 to 20 meters tall. The furnace, usually a zirconia induction furnace or a graphite resistance furnace, heats the preform tip to roughly 1,900 to 2,200 C, where silica softens enough to flow. A gob forms and drops, and the tip is captured onto a starting rod.

Once pulling begins, the diameter measurement stage takes over. A laser micrometer measures the bare fiber continuously, and a feedback loop adjusts either the draw speed or the preform feed rate or both to hold the diameter. The furnace temperature, the draw tension, and the fiber take-up speed interact; a small tension change produces a small diameter change, which is exactly the lever the controller uses.

Coating comes next, and it happens while the fiber is still pristine. Bare silica fiber is strong only when it is flawless at the surface. Any microscopic handling damage becomes a crack that propagates under stress, a behavior called stress corrosion. Coating is applied through a die within a second or so of the fiber leaving the furnace, before it can touch anything. Dual-layer acrylic coatings, a soft inner layer and a harder outer layer, dominate, and the material is cured with ultraviolet light immediately after application.

The cured fiber then passes through a capstan that sets tension and speed, and onto a take-up spool. Some lines add a proof test, running the fiber under controlled strain over a set of pulleys to screen out sections with flaws below a strength threshold.

Inline monitoring runs throughout. Diameter measurement catches geometry drift, and a small number of lines measure coating concentricity and cure quality as the fiber flies past. A spool that fails any check is cut out rather than shipped, because a single weak section can fail in a duct years later.

Comparison: preform routes and when each is used

Route How the glass forms Typical use Key trade-off
OVD (outside vapor deposition) Soot deposited on outside of rotating rod, then consolidated High-volume standard single-mode and multimode Mature and scalable; large soot body handling
VAD (vapor axial deposition) Soot grows axially on end of a seed rod High-volume telecom, continuous preform growth Very continuous process; geometry control is demanding
MCVD (modified chemical vapor deposition) Layers deposited inside a tube, which is then collapsed Specialty fiber, low-volume and small preforms Precise profiles, limited preform size
PCVD (plasma chemical vapor deposition) Microwave plasma deposits inside a tube, then collapse Specialty fiber, steep index profiles Excellent layer precision, slower throughput

The pattern to notice is that process choice tracks volume and profile complexity. Telecom needs big preforms and low cost per kilometer, which favors OVD and VAD. Specialty fiber such as erbium-doped amplifier fiber or polarization-maintaining fiber needs unusual profiles in modest quantities, which favors MCVD and PCVD. None of these routes is obsolete, and several producers run more than one.

Edge cases and the failure modes that shape design

The hard edge case in fiber making is hydrogen. Hydrogen diffuses into silica and reacts at defect sites, creating hydroxyl and raising loss across a broad band, a mechanism called hydrogen aging. Fibers are tested for this, and coating chemistry and glass composition are tuned to resist it. This is why a fiber that performs perfectly on a test bench can degrade in a wet, hydrogen-rich environment over years.

Another edge case is bend loss. A fiber bent below a critical radius leaks light out of the core, and the effect scales badly as the bend tightens. Modern bend-insensitive designs add a trench of low-index glass around the core to reflect escaping modes back, which is why indoor and access cables can turn tight corners that would have been unthinkable with early single-mode fiber.

A professional caveat worth stating plainly: attenuation specifications are measured under defined launch conditions, and measured loss depends on how light is launched into the fiber. A multimode fiber measured with a restricted launch can appear to have lower loss than the same fiber measured with a full overfilled launch. Comparing datasheets without checking the measurement method is a classic procurement error.

Warnings to keep in mind when reading fiber specifications:

Amplifier reach claims assume a specific loss budget and include splice and connector losses, not just fiber attenuation; a generous reach figure can hide a tight budget. Minimum bend radius figures quoted for a cable differ from those for bare fiber, and violating the cable spec is what actually breaks installations. Coating strip force specifications matter for field termination; a fiber that strips inconsistently causes connector yield problems that never show up on a loss test. Proof-test strain levels indicate screening strength, not long-term reliability, and different grades are screened at different strain levels. Attenuation uniformity along a spool can matter more than the average for long coherent links, and only some vendors report it.

What buyers and designers should actually do

The practical actions here are about reading specifications correctly and matching fiber grade to installation conditions, since the manufacturing process itself is not something most users control.

Start by specifying attenuation at the actual operating wavelength with the stated measurement method, not a generic loss number. For a 1,550-nanometer link, ask for the 1,550-nanometer value and confirm the test standard behind it. Next, match bend-insensitivity grade to the tightest bend the route will actually see, including cable trays and wall penetrations, and leave margin. Then confirm coating type against the environment: standard dual acrylic is fine in ducts, but some harsh or high-temperature routes call for different coatings, and the wrong choice shows up as long-term reliability trouble rather than immediate loss.

For short-reach multimode, select the bandwidth grade against the link length and data rate you plan, and check whether the transceiver specification assumes a particular modal bandwidth. For specialty work such as sensing or amplifier fiber, buy from a supplier who publishes the full index profile and dopant information, because generic telecom fiber is not a substitute. Finally, ask for proof-test data and hydrogen-aging test results when the deployment is long-lived or in a wet environment; these are the two specifications most often omitted from a simple quote.

Limitations: what the process cannot promise

Fiber manufacturing is extraordinarily precise, but it cannot deliver perfection, and several limits are structural. Rayleigh scattering sets a floor on attenuation that no process improvement removes; it is physics, not manufacturing error. The draw process holds diameter to extreme tolerance, but the core-cladding concentricity is set in the preform and can only be preserved, not improved, during drawing, so preform quality caps fiber quality.

Cost and scale are bounded by the tower and the preform, not by demand alone, so specialty fiber remains expensive relative to telecom fiber because it is made in small volumes with slower processes. Availability of exotic dopants and coating materials can constrain some grades independently of production capacity. And the measurement problem never fully disappears: attenuation, bandwidth, and geometry are all specified under standardized conditions, and real deployments differ from those conditions in ways that can only be bounded, not eliminated. Understanding the manufacturing chain is ultimately about knowing which numbers are promises and which are conditions.

The fiber that arrives on a spool is the product of vapor chemistry, a controlled collapse into glass, and a tower that stretches and coats it in seconds. Each stage exists to protect a single-digit-decibel-per-kilometer budget and a 125-micrometer geometry. That is what makes the industry's precision claims meaningful, and what limits them.