1. Introduction

Shortly after the prediction by Kao and Hockman that the optical loss in silica glass could be reduced to less than 20 dB/km at 0.6 µm if a sufficiently pure glass could be synthesized [1, 2], the outside vapor deposition (OVD) process was invented for fabricating high-purity silica preforms [3, 4]. OVD involves the deposition of layers of glass soot derived from reactant gases (typically O₂, SiCl₄ and GeCl₄) that are pyrolysed to ultra-pure glass particles and then consolidated to solid glass. Two further processes were subsequently developed based on soot deposition: modified chemical vapor deposition (MCVD), where a glass layer is deposited on the inside of a substrate tube, and vapor axial deposition (VAD), where a preform is built up axially by deposition of successive layers of soot. In the case of MCVD, the tube is afterwards heated to collapse it to a solid rod, which is then directly drawn to fiber. The use of dry raw materials, a dry atmosphere and electrical heating elements permit the reduction of water-related contamination (in the form of OH bonds in the glass matrix) to extremely low levels, which remain low during fiber drawing because humidity present outside the preform cannot diffuse into the core over the relevant time-scales.

The multi-stage stack-and-draw process used in manufacturing air-silica photonic crystal fibers (PCFs) presents a series of quite different challenges, exacerbated by the presence of hollow channels and exposed surfaces [5]. First, high-purity dry silica (<1 ppm OH) tubes and rods of ~2 cm outer diameter are heated and drawn down ~10 times in linear dimensions to capillaries and rods. If a single solid core is the aim, hexagonally stacked capillaries are used to form the cladding and a central rod the core. The stack is then inserted into a glass tube (to add mechanical strength and stability), heated and drawn down by another factor of ~10 to an intermediate preform or ‘cane’. Finally, the cane is inserted into another glass tube, heated and drawn down by a further linear factor of ~50, while pressurizing the hollow channels to control the structural geometry. The risk of contamination during stacking is reduced by using a clean-room environment, typically ISO 5, i.e., less than 10⁵ m^–3 particles of diameter greater than 100 nm. Another major contaminant is diffusion of OH into the glass. Since the diffusion constant follows an Arrhenius relationship with temperature ($D\propto \exp (-E_a/k_{\text{B}}T)$), this form of contamination is particularly critical during the heating-and-drawing steps when the temperature is high. As a result, even if one starts with very pure dry fused silica, OH in-diffusion during the repeated drawing steps can lead to a significant OH concentration in the core of the drawn fiber. Lowering the OH contamination is important for reducing loss in supercontinuum generation in solid-core fibers [6], and could also reduce surface capillary wave losses in hollow-core fibers by increasing the surface tension [7].

It has been reported that minimizing the time interval between stacking the preform and drawing the fiber reduces OH-related loss [8]. Further, it has been found that the loss is also reduced if the cane is annealed immediately before fiber drawing (by purging it with nitrogen from one end while passing it three times through the fiber-drawing furnace at 20 mm/min and 1880°C [9]). It has been suggested that this treatment works by relaxing the silica network and removing adsorbed OH from the surfaces inside the cane.

In this paper we present a systematic study of different methods of dehydrating the glass in all three main stages of PCF fabrication: after stacking, after cane drawing and during fiber drawing. The glass is treated either (a) with chlorine or oxygen at elevated temperature in a dedicated oven at 900°C or (b) with oxygen during fiber drawing (temperature in the range 1850 to 2000°C). We find that all the dehydration treatments are effective at lowering OH-related losses, and in particular that OH contamination, accumulated over more than two years of storage inside a cleanroom with relative humidity in the range ~20% to 50%, can be almost completely removed.

2. Methods

2.1 Estimation of relevant time-scales at different fabrication stages

The core diameter shrinks by some three orders of magnitude from the initial stack to the drawn fiber. The time it takes for OH groups to diffuse from the center of the core to a hollow channel (or in the opposite direction, depending on the concentration gradient) therefore varies significantly in the various fabrication steps. For an estimate of the relationship between core radius ρ₀ and time t_1% required to reduce the OH concentration to 1% of its initial value, we solved the diffusion equation in cylindrical coordinates, assuming a uniform concentration within the region ρ < ρ₀ at t = 0 and zero concentration for ρ > ρ₀ at all times. The analysis yields the approximate solution:

 

Fig. 1 Time t_1% and temperature T required for the OH-concentration to decay to 1% of its initial value at the center of silica strands of different radii (from the expression in Eq. (1)). The diffusion coefficient is given by D(T) = D⁰exp[–E_A/(RT)], where D⁰ = 3 × 10⁻⁶ cm²/s, E_A = 90 kJ/mol, and R is the universal gas constant [10]. The white dashed lines mark the two temperatures relevant to this study: the chlorine furnace treatment temperature (900°C) and the fiber drawing temperature (~1900°C).

Download Full Size | PDF

We also investigate chlorine dehydration of a cane, which the analysis shows can be achieved within a reasonable time (~170 h at 900°C will reduce the OH concentration by 99% at the center of a 2ρ₀ = 300 µm diameter silica strand). Note that penetration of chlorine into the glass can be ignored, since its diffusivity at 900°C is 3 orders of magnitude lower than that of OH [11].

Finally, for the two core radii investigated in the experiments, the diffusion times at the fiber drawing temperature (~1900°C) are ~2 s for ρ₀ = 2.3 µm and ~5 s for ρ₀ = 3.7 µm. These are comparable with the time taken for the fiber to pass through the furnace hot-zone at the drawing speeds used (between 15 and 40 m/min), leading us to conclude that, if the OH concentration is higher on the inner surfaces of the hollow channels than at the center of the core, diffusion of OH into the core may be considerable during fiber drawing. This suggests that the dryness of the gas used to pressurize the hollow channels during drawing may be important. We investigated this by comparing N₂ (< 2 ppm H₂O) with O₂ (<0.5 ppm H₂O), neither of which gases react with the glass.

2.2 Fiber structure and confinement loss

A scanning electron micrograph of one of the solid-core PCFs studied in the experiments is shown in Fig. 2. All the fibers had a form-factor d/Λ ~0.6, where d is the hollow channel diameter and Λ the spacing. For the two core radii mentioned above, Λ ~3.3 µm and ∼5.4 µm. Calculations using the multipole method [12, 13] show that for these parameters the confinement loss at 1380 nm is less than 1 dB/km and can therefore be neglected.

 

Fig. 2 Upper: Chart summarizing the different treatments. The symbols, used in Fig. 4, indicate the different stack and fiber drawing treatments (square, triangle, six-pointed star and circle). The third and fourth yellow-shaded columns refer to the gas used during fiber drawing. The four right-hand columns give the background-corrected fiber loss α_c = (α_p – α_b) at 1380 nm in three cases when freshly drawn canes were immediately drawn to fiber. The standard error is also indicated—note that this can only be evaluated if the number of canes N is greater than 1. Lower left: sketch of a stack. Lower middle: optical micrograph of a cane. Lower right: scanning electron micrograph of a drawn fiber. The core radius ρ₀ is taken as half the minimum distance between the central glass-air interfaces.

Download Full Size | PDF

2.3 Chemistry of dehydration processes

Dehydration with nitrogen or oxygen works mainly by flushing away water vapor that appears when OH-groups diffuse to the glass surface where the concentration is lower. The water molecules are created via the following reaction:

2.4 Chlorine treatment of stacks

As mentioned in Section 2.1, the time needed for OH to diffuse out of the core-rod at the center of a stack is impracticably long at the maximum temperature (~900°C) at which we could operate the horizontal dehydration furnace. When the originally ultra-low OH silica rod is drawn down from 20 mm to 1.55 mm diameter, however, it spends only ~1 s in the hot-zone at ~1900°C, yielding a OH diffusion length of ~1 µm. The OH in this thin outer layer of contaminated glass should therefore be rapidly eliminated during stack treatment. To test this, F300 tubes and a Suprasil F300 rod (Heraeus GmbH, <1 ppm OH content) were first drawn down to 1.55 mm diameter. Sufficient numbers were drawn in one day to make three identical stacks (St1, St2 and St3) three days later. Each was treated differently:

An overview of the various treatments is shown in Fig. 2.

2.5 Chlorine treatment of canes

It is known that OH-related loss increases significantly with the time delay between stacking and fiber drawing [14]. Even if a cane is stored for a long time, it can yield low-loss fiber if it is thermally annealed and flushed with dry nitrogen on the tower just before drawing [9]. The fiber drawing furnace allows annealing to be carried out at very high temperature (~1900°C), resulting in rapid out-diffusion of OH-groups to the glass surface, where they join to form water molecules (Eq. (2)), which are then purged by the furnace flushing gas (Ar). A drawback of this procedure is that mounting the cane vertically on the drawing tower requires a careful balance between temperature and cane feed speed, so as to avoid tapering and deformation of the cane microstructure. It is therefore unsuitable for large-scale production of low-loss solid-core PCFs.

The use of a horizontal annealing furnace at ~900°C avoids any risk of structural deformation, while permitting treatment of several canes simultaneously. As with stack treatment, the hollow channels in the canes were flushed with chlorine gas for various durations up to a maximum of 48 h, and then drawn to fiber on the same day as the chlorine treatment finished. The time interval between drawing and chlorine treatment of the canes could be as long as a few years, during which period they were stored inside a closed (but not completely air-tight) tube inside an ISO 7 (<10⁷ particles of size ≥0.1 µm per m³) cleanroom without any special dry-air flushing of the storage tube. One goal of this study was to establish whether chlorine treatment could “rejuvenate” canes stored without special care over long periods.

2.6 Measuring the transmission loss

A supercontinuum light source was used to measure the loss spectrum. The light was delivered via ~30 m of single-mode step-index fiber with an LP₁₁ cut-off at ~920 nm (Thorlabs XP1060); this converted the star-shaped mode from the supercontinuum PCF into a circular shape. The output from the single-mode fiber was then coupled into the core of the fiber-under-test, the output of which was coupled into another XP1060 fiber connected to an optical spectrum analyzer. Each spectrum was recorded several times, starting typically with a 400 m length and making at least 3 cut-backs of 40 to 100 m each. Using robust linear regression to the measured power-position data at each sampled wavelength we were then able to calculate the loss α and estimate the measurement error σ (single cut-back measurements provide no information about the error) [15].

3. Results

All the loss measurements were made within one week of drawing, so as to avoid significant ingress of water vapor into the hollow channels through the fiber ends, which is a known problem when fibers are stored for several weeks under normally humid ambient conditions [16]. Although it can be eliminated by cutting several meters off each fiber end before measuring the loss, this represents an undesirable waste of fiber.

A typical measured loss spectrum is shown in Fig. 3. To distinguish the OH-related absorption peak at 1380 nm from other loss factors (e.g., Rayleigh scattering), we took the minimum loss in the range 1350 to 1426 nm to be background loss (α_b ± σ_b) and subtracted it from the height of the peak (α_p ± σ_p) at 1380 nm. The background-corrected loss α_c = (α_p − α_b) then has an error of σ_c = (σ_p² + σ_b²)^1/2 for a single sample. When measurements from N > 1 identically-treated samples were available, we instead calculated the standard mean error${\sigma }_{\text{M}}=\text{SD}/\sqrt{N}$, where SD is the sample standard deviation of the N measurements of α_c.

 

Fig. 3 An example of a loss spectrum measured from a PCF used in this study. The dark blue curve shows the measured loss α and the shaded area indicates the error range ± σ_p at each wavelength. The lower dashed horizontal line indicates the minimum background loss α_b between 1350 and 1426 nm. The OH-related loss at 1380 nm, corrected for background loss, is then α_c = α_p – α_b.

Download Full Size | PDF

3.1 Chlorine treatment of stacks

To verify repeatability of the results for equal treatments, we repeated the procedures for a total of seven stacks treated in three different ways. Each stack was drawn to cane as described in Section 2.4 and then immediately to fibers with the two specified core radii (2.3 µm and ~3.7 µm). The measured fiber losses are tabulated in Fig. 2 and show a clear trend, with chlorine treatment of the stack offering the most benefit, followed by oxygen treatment. As expected the smaller core PCF experiences a larger drop in loss, although the errors are such that this may not be statistically significant.

3.2 Chlorine rejuvenation of stored canes

Following the procedure described in Section 2.5 we treated several long-term stored canes with chlorine gas at 900°C and measured the loss after drawing to fiber. The canes originated from all three types of stack treatment, as indicated in Fig. 4(a) (note that the measurements discussed in section 3.1 are not included, since these were made on fibers drawn immediately after cane drawing). The results show that there is a clear reduction in OH-related loss after 4 hours of chlorine-treatment, longer treatments offering only minimal benefit. It is also seen that, except for the untreated canes, there is no clear difference in loss between canes originating from differently treated stacks. This shows that stack treatment can be omitted if the canes are anyway to be chlorine treated before fiber drawing. Nor is there any clear difference in loss between the 2.3 µm (blue symbols) and 3.7 µm (red symbols) core-radii fibers drawn from chlorine-treated canes.

 

Fig. 4 Background-corrected peak fiber loss σ_c at 1380 nm for different types of treatment. The squares, triangles and circles (color-coded for core radius, as indicated in the inset) indicate canes originating from stacks that were respectively untreated (St1), oxygen treated (St2) and chlorine and oxygen treated (St3). The six-pointed stars indicate the use of low-humidity oxygen (<0.5 ppm H₂O) instead of nitrogen (<2 ppm H₂O) for pressurization during fiber drawing. The green error-bars indicate canes that were chlorine treated just before drawing to fiber. (a) Fiber loss versus chlorine treatment time of the canes at 900°C. Note that the plot does not include any of the fibers that were drawn immediately after cane-drawing (see Fig. 2 and Section 3.1). (b) Fiber loss versus cane storage time. Data-points with less than 1 month difference in storage time and otherwise equal treatment are averaged together, the error bars indicating the standard error of the mean.

Download Full Size | PDF

On the other hand, there is a large loss difference between the two core sizes for non-chlorine-treated canes drawn to fiber using a gas with <2 ppm water (N₂) for pressurization (triangles and circles); this is caused by the slower drawing speed used for the larger-core fibers (15 m/min compared to 40 m/min for the smaller core fibers), allowing more time for OH contamination at the surface of the hollow channels to diffuse into the core. One would expect that for equal time spent in the hot-zone, larger-core fibers would have less OH contamination by diffusion into the center of the core due to the longer travel distance from the surface, but it is clear that the difference in time spent in the hot-zone is more important than the difference in diffusion length. Interestingly, when using a gas with <0.5 ppm water (O₂) for pressurization (the six-pointed stars) there is again negligible difference between blue and red points because the use of the drier gas causes less diffusion of OH into the core during fiber drawing.

Figure 4(a) also shows that without cane chlorine treatment, fibers originating from untreated stacks (St1, squares) have consistently higher loss than those drawn from oxygen treated stacks (St2, triangles), with chlorine-treated stacks yielding the lowest loss (St3, circles). This further supports the conclusions of Section 3.1. It is also interesting to note how chlorine-treatment of the canes removes the loss difference between fibers drawn to different core sizes and those originating from differently treated stacks.

To examine more closely the influence of cane storage, the measured OH-related losses are plotted versus cane storage time in Fig. 4(b), where the data includes canes drawn quickly to fiber without storage, as well as both treated and untreated canes. It is seen that fibers drawn from untreated canes already show markedly higher OH-related loss even after 1.4 months of cane storage. On the other hand, all the chlorine-treated canes (marked by green error bars in Fig. 4) resulted in fibers with lower OH-related loss, regardless of cane storage time, stack treatment method or core diameter.

3.3 Dryness of gas used during fiber drawing

Figure 4 also includes data from fibers drawn while using a drier gas (O₂ with <0.5 ppm H₂O instead of N₂ with <2 ppm H₂O) as pressurization gas (see end of Section 2.1). The data-points for these cases are marked by six-pointed stars, and the canes used all originated from an oxygen treated stack (St2). Using a drier pressurization gas is clearly effective in reducing OH-related fiber loss for canes that were not chlorine-treated. Combining the use of the drier gas during fiber drawing with 48 h of cane-treatment with chlorine produced the lowest OH-related losses overall (82 ± 5 dB/km, the uncertainty being in loss measurement itself since only one cane was used, i.e., N = 1), when the cane had been stored long-term, i.e., not immediately drawn to fiber. These values correspond to an OH concentration of 1.31 ± 0.08 ppm [17].

4. Discussion and conclusions

The lowest OH-related fiber loss in this study (62 ± 10 dB/km at 1380 nm, corresponding to an OH concentration of ~1.0 ± 0.2 ppm) was achieved starting with stacks that were chlorine-treated before drawing them to canes, followed immediately by drawing to fiber. As supplied, the glass (Heraeus F300HQ) is specified to have less than 1 ppm OH, with a typical value of 0.2 ppm. A judicious combination of the treatments described above makes it possible to recover this low OH concentration in the final fiber, despite the many manufacturing steps involved including handling, stacking, heating and drawing the glass multiple times.

Although long-term (more than a few weeks) storage of the canes quickly increases the OH-related fiber loss in untreated canes (also found in a previous study [8]), this effect can be largely reversed by chlorine treating the canes and using oxygen (or other suitable non-reactive gases with less than 1 ppm water content) for pressurization during fiber drawing. Even with this procedure, the low OH-related loss achieved when the fiber was drawn immediately after cane drawing (62 ± 10 dB/km for St3 stacks and N = 4) could not be recovered even when treating the stored canes with chlorine for 48 h, when loss values of 82 ± 5 dB/km were achieved, including using oxygen for pressurization. This suggests that not only does OH diffuse into the glass during cane storage, but that the long storage times allow the formation of a subset of bonds between OH and silica so strong that they cannot be broken by chlorine treatment. Measurement errors do not, however, allow us to claim this with certainty.

In this study our aim was not to produce the lowest loss PCF, but to explore the influence of different processing procedures on fiber loss and identify the trends. By further eliminating OH contamination, it is likely that much lower loss levels can be reached, approaching those of Kurokawa et al. [18], who mention that they used a dry pressurization gas and a “dehydration process”, without however giving any detailed information.

In conclusion, chlorine dehydration at 900°C provides a convenient means of removing OH contamination in PCF canes, without incurring any risk of microstructural deformation or tapering, and multiple canes can be readily treated at the same time. The method could therefore be important for future large-scale production of low-loss PCFs.