Hybrid fiber preform fabrication using CO laser heating

Transkript

1 kth royal institute of technology Doctoral Thesis in Physics Hybrid fiber preform fabrication using CO laser heating TAR AS ORIEKHOV Stockholm, Sweden 2022

2 Hybrid fiber preform fabrication using CO laser heating TARAS ORIEKHOV Academic Dissertation which, with due permission of the KTH Royal Institute of Technology, is submitted for public defence for the Degree of Doctor of Philosophy on Wednesday the 25th of May 2022, at 1:00 p.m. in FA31, Roslagstullsbacken 21, Fysikcentrum, AlbaNova, Stockholm Doctoral Thesis in Physics KTH Royal Institute of Technology Stockholm, Sweden 2022

3 Taras Oriekhov ISBN TRITA-SCI-FOU 2022:16 Printed by: Universitetsservice US-AB, Sweden 2022

4 III Русский военный корабль, иди наx*й! Ukrainian border guards of Snake Island

5 IV

6 V Abstract This thesis describes the development of a new prototyping technique for specialty optical fibers and covers all the fabrication steps from preform to fiber. The technique allows to produce fibers of a custom core structure and material composition, mainly focusing on semiconductor core fibers. By combining the optoelectronic properties of semiconductors with the advantages of the glass optical fiber platform, such fibers become a promising candidate in applications that require a wider infrared transmission window or stronger non-linear response. In contrast to traditional optical fibers, semiconductor core fibers are not standard off-the-shelf components. They exist as research samples, typically short in length, limited in core size, and exhibit a high loss due to the challenging and expensive fabrication process. The proposed preform fabrication method utilizes a carbon monoxide laser as a heat source. Employing this laser ensures extremely effective heat transfer to the preform with low surface silica vaporization, a minimal thermal gradient across the preform cross-section, and short exposure time of the preform to high temperatures. This allows to reduce the manufacturing time of the preforms and improve their optical properties. The aim of this thesis work was to design and build a system to fabricate fiber preforms made of semiconductors or other crystalline core materials. The work was primarily focused on preforms with silicon cores, but germanium and sapphire cores were also demonstrated. The ability to achieve preform tapering was a vital part of the preform fabrication. The process was developed using silicon as a test core material due to its abundance and widespread applications. As a typical representative of hybrid core materials, the properties of silicon imposed some common challenges that had to be addressed during the preform fabrication process. This includes a drastic difference in thermal expansion and thermal conductivity of the core compared to the cladding. Combined with a rod-in-tube approach, this system was used throughout the project to create silicon core fiber preforms in a wide range of core-to-cladding ratios, covering the core sizes from 17 μm up to 1 mm for preforms of 6 mm in diameter. The silicon core fibers produced from these preforms showed record minimal loss values of 0.1 db/cm.

7 VI Additionally, glass additive manufacturing was applied for the first time in combination with the laser-based preform manufacturing technique to prototype specialty optical fibers of custom core composition and structure. In particular, the Laser Powder Deposition method was used to prototype fiber preforms with alumina, titania, and erbium-aluminum doped cores in concentrations not achievable by standard techniques. The drawn fibers showed losses as low as 3.2 db/m, which is the best result achieved for glass fibers produced using 3D printing. Furthermore, multicore fiber preforms made of multi-component glass using a filament-based glass 3D printer have been demonstrated, showing the potential of using additive manufacturing for specialty fiber fabrication. These silicon core and glass-doped preforms were pulled into hundreds of meter-long fibers of a standard 125 μm diameter and a core size in the range of 1 to 20 μm. This was achieved in a specially designed lab-sized fiber draw tower. To further utilize the benefits provided by the laser heating, the tower was also retrofitted with a carbon monoxide laser-based furnace. This allowed a very flexible operation of the tower, suitable for on-demand fiber prototyping of different types and experimental compositions.

8 VII Sammanfattning Denna avhandling beskriver utveckling av en ny teknik för prototyptillverkning av optiska specialfibrer och omfattar alla steg i processen, från preform till färdig optisk fiber. Tekniken gör det möjligt att producera fibrer med skräddarsydda kärnor, i både form och material sammansättning, främst med fokus på fibrer med kärna av halvledare. Genom att kombinera de optoelektroniska egenskaperna hos halvledare med fördelarna av den fiberoptiska plattformen, kan dessa fibrer anpassas för tillämpningar som kräver ett bredare transmissionsfönster i det infraröda våglängdsintervallet, eller för icke-linjära optiska effekter. Till skillnad mot traditionella optiska fibrer är fibrer med halvledarkärna inte standardkomponenter som finns att köpa på marknaden idag. De kan vara tillgängliga från olika forskargrupper, och då vanligtvis i korta längder, med begränsat val av kärnstorlek, samt med höga optiska förluster på grund av den komplicerade tillverkningsprocessen. Den framtagna metoden för preform-tillverkning använder en kolmonoxid-laser som värmekälla. Detta säkerställer en extremt effektiv värmeöverföring till preformen utan att orsaka hög förångning av glaset, en minimal termisk gradient över preformens tvärsnitt, samt minskar preformens exponeringstid vid höga temperaturer. Tillsammans möjliggör detta en minskad tillverkningstid för preformen vilket förbättrar de optiska egenskaperna. Syftet med avhandlingsarbetet var att designa och bygga ett system för att tillverka fiber-preformer med halvledare eller andra kristallina material i kärnan. Arbetet var främst inriktat på preformer med kiselkärnor, men även germanium- och safirbaserade kärnor tillverkades. Förmågan att utföra avsmalning av preformen var en viktig del-process av preformtillverkningen. Processutvecklingen började med kisel som test-material i kärnan, på grund av tillgänglighet samt dess många tillämpningsområden. Kisel användes som ett referensmaterial för dessa hybrid-preformer och de skilda materialegenskaperna, framförallt den stor skillnad i termisk expansion och värmeledningsförmåga mellan kärnan och manteln var utmanande, och tillverkningsprocessen fick anpassas därefter. Under hela projektets gång användes en stav-i-rör metod för tillverkning av kiselkärnepreformer, och en stor variation i kärndiameter utvärderades; från 17 μm upp till 1 mm med en preform diameter på 6 mm. Fiber dragna från dessa preformer visade rekordlåga förlustvärden, som lägst 0,1 db/cm.

9 VIII Utöver kristallina material har för första gången även additiv tillverkning använts i kombination med det laserbaserade systemet för prototyptillverkning av specialfibrer med anpassad materialsammansättning och geometrisk struktur. Här användes laserbaserad pulverdeponering för att tillverka preformer med glaskärnor dopade med aluminiumoxid, titanoxid och erbium-aluminiumoxid, i koncentrationer som inte kan uppnås med standardtekniker. Fiber dragna från dessa preformer hade optiska förluster så låga som 3,2 db/m, vilket är de lägsta som uppnåtts för optiska fibrer tillverkade genom additiv tillverkning. Dessutom har flerkärniga preformer tillverkats, sammansatta av flera olika typer av glas, med hjälp av en filamentbaserad 3D skrivare för glas, vilket visar potentialen för additiv tillverkning av specialfibrer. Från de tillverkade preformerna drogs hundratals meter av fiber med en ytterdiameter på 125 µm och en kärndiametrar från 1 µm till 20 µm. Detta gjordes i ett specialdesignat fiberdragtorn i labbet. För att utnyttja fördelarna med laserbaserad uppvärmning användes kolmonoxidlasern även i dragtornet. Detta möjliggjorde en mycket flexibel användning lämpligt för experimentell prototyptillverkning av optiska fibrer.

10 IX Preface The research presented in this thesis was carried out within the Laser Physics group at the Department of Applied Physics of the Royal Institute of Technology (KTH) in Stockholm, Sweden. This work has been funded by the Knut och Alice Wallenbergs Stiftelse ( ) and Stiftelsen för Strategisk Forskning (RMA ).

11 X List of papers Paper I T. Oriekhov, C. M. Harvey, and M. Fokine, CO laser-based specialty optical fiber preform manufacturing technique, Review of Scientific Instruments (2022) Submitted. Paper II T. Oriekhov, C. M. Harvey, K. Mühlberger, and M. Fokine, "Specialty optical fiber fabrication: preform manufacturing based on asymmetrical CO laser heating," Journal of the Optical Society of America B 38, F130-F137 (2021). Paper III C. M. Harvey, K. Mühlberger, T. Oriekhov, P. Maniewski, and M. Fokine, Specialty optical fiber fabrication: fiber draw tower based on a CO laser furnace, Journal of the Optical Society of America B 38, F122- F129 (2021). Paper IV P. Maniewski, C. M. Harvey, K. Mühlberger, T. Oriekhov, M. Brunzell, F. Laurell, and M. Fokine, Rapid prototyping of silica optical fibers, Optical Materials Express (2022) Submitted.

12 XI Author contribution Paper I I developed the system, wrote the code, designed the hardware, conducted the measurements, performed data analysis, and co-wrote the manuscript. Paper II I developed a general procedure for preform fabrication, developed the tapering process to create crack-free preforms with silicon core, carried out preform manufacturing, conducted the measurements, and co-wrote the manuscript. Paper III I provided a ready-to-use motion hardware solution for preform alignment, analyzed the linearity of the main feeding stage, implemented the computer control for a laser, produced all the preforms for the fiber draw, and co-wrote the manuscript. Paper IV I produced and analyzed the preforms and co-wrote the manuscript.

13 XII Additional publications C. Liu, T. Oriekhov, C. Lee, C. M. Harvey, and M. Fokine, Rapid Fabrication of Silica Microlens Arrays via Glass 3D Printing, 3D Printing and Additive Manufacturing (2022) Submitted. M. Fokine, T. Oriekhov, C. Liu, Method and apparatus for additive manufacturing, , P31019SE (2021). M. Fokine, T. Oriekhov, C. Liu, Method and apparatus for additive manufacturing, , P31020SE (2021). Conference contributions K. Mühlberger, T. Oriekhov, M. Fokine, F. Laurell, and U. J. Gibson, Processing of semiconductor core fibers with CO 2 laser beam irradiation, Optics and Photonics in Sweden, Stockholm, Sweden (2017). M. Fokine, K. Mühlberger, T. Oriekhov, S. Tjörnhammer, F. Laurell, and U. J. Gibson, Laser processing of semiconductor core fibers, XL Encontro Nacional de Física da Matéria Condensada, Búzios, Brazil, ID: 44 1 [02.4] (2017). K. Mühlberger, T. Oriekhov, M. Fokine, F. Laurell, and U. J. Gibson, Processing of semiconductor core fibers with CO 2 laser beam irradiation, P@K conference, M/S Mariella, Helsinki, Finland (2017). T. Oriekhov, C. Harvey, and M. Fokine, Preform Manufacturing Setup Based on CO Laser One-Beam Configuration, 4th International Conference on Optics, Photonics and Lasers, Corfu, Greece (2021). C. M. Harvey, K. Mühlberger, T. Oriekhov, and M. Fokine, CO laserbased furnace for optical fibre fabrication, 4th International Conference on Optics, Photonics and Lasers, Corfu, Greece (2021). P. Maniewski, C. M. Harvey, T. Oriekhov, F. Laurell, and M. Fokine, Rapid fabrication of 3D fused silica structures with customized glass composition, 4th International Conference on Optics, Photonics and Lasers, Corfu, Greece (2021).

14 XIII C. M. Harvey, K. Mühlberger, T. Oriekhov, and M. Fokine, Low-loss Silicon core Optical Fibre Fabrication Using a CO Laser-based Furnace without an Interface Layer, Frontiers in Optics + Laser Science, Washington D.C., USA, FW1C.2 (2021). C. Liu, T. Oriekhov, and M. Fokine, "Rapid Fabrication of Glass Microlens Array Using Laser Assisted 3D Printing," in Frontiers in Optics + Laser Science, Washington D.C., USA, FM3C.5. (2021). C. M. Harvey, K. Mühlberger, T. Oriekhov, and M. Fokine, Optimising draw parameters for the fabrication of low loss silicon-core optical fibre, SPIE Photonics Europe, Strasbourg, France (2022).

15 XIV Acknowledgments The work described in this thesis would not have been possible without the help and support of countless people, to whom I am deeply grateful. First of all, I would like to thank Prof. Michael Fokine for choosing me for this Ph.D. position, even though my master s was in biophotonics. If it wasn t for you, I would have been defending Mariupol instead of this thesis at this time. Because of your approach of being closer to your students, I have never thought of you as my boss, but rather as a mentor and a great friend. I really appreciate all that trust and freedom you provided me to work on this project and am very grateful that you always had time to show your support and give wise advice when I needed it the most. It is thanks to your guidance and great supervision I was able to finish this project. I want to thank Prof. Fredrik Laurell for accepting me to the Laser Physics group and for all the help you have provided within and outside the work, especially recently. I am sorry for temporarily taking over your title of the First Racket within Laser Physics. I would also like to thank Clarissa Harvey, for always having time to answer my questions, making additional simulations for me, and drawing a countless number of fibers. The effort and dedication you put into this project is unmeasurable! I want to thank Chunxin Liu, for being a great friend, for your ability to understand me, and for the patience you demonstrated while we worked on the combined projects. I had a very good time playing some ping-pong with you time-to-time. I would not have won that LPTT Cup without your friendly coaching! Thank you, Korbi, for sharing the office and the lab with me for all those years. I have always valued your effort to keep the lab clean and organized. Thank you, Pawel, for introducing me to the world of 3D printing and all those tips and tricks you ve taught me. Thank you, Andrius, for your support, technical advice, and wiliness to help me measure or cut my samples, even though it sometimes meant solving 10 computer problems before the measurements and going home late. I also want to thank each and every one of you, who supported me in light of recent events in my country. It is because of you, guys, and your support I have never felt left alone to face this burden or felt abandoned here in Sweden.

16 Finally, I want to thank my father Pavel, mother Halyna, and my sister Ira, for their never-ending love and support, but also for sacrificing their ways of life and coming to stay with me in Sweden at this difficult time for all Ukrainian people. XV

17 XVI Acronyms CAD CO CO 2 CW DGLD TM FBG GF HPCVD ID IR LPD MCD MCVD MeOH OD OVD ppm V PWM RI SMF VAD Computer-aided design Carbon monoxide Carbon dioxide Continuous wave Direct glass laser deposition, trademarked by Nobula3D Fiber Bragg grating Goodfellow, a glass manufacturer High-pressure chemical vapor deposition Inner diameter Infrared Laser powder deposition Molten core drawing Modified chemical vapor deposition Methanol Outer diameter Outside vapor deposition Parts per million by volume Pulse-width modulation Refractive index Single mode fiber Vapor-phase axial deposition

18 XVII Table of Contents Abstract... V Sammanfattning... VII Preface... IX List of papers... X Author contribution... XI Additional publications... XII Conference contributions... XII Acknowledgments... XIV Acronyms... XVI 1 Introduction Introduction Motivation Outline... 4 Part I: CO laser-assisted preform manufacturing CO Laser History of invention Inversion mechanisms of CO 2 and CO lasers Water vapor absorption at the CO laser wavelength Comparison of a CO and a CO 2 laser for glass processing CO laser characterization Fiber manufacturing techniques Optical fiber fabrication for telecommunication Optical fibers Outside vapor deposition (OVD) Vapor-phase axial deposition (VAD) Modified chemical vapor deposition (MCVD)... 19

19 XVIII 3.2 Hybrid optical fiber manufacturing Hybrid optical fibers High pressure chemical vapor deposition (HPCVD) Molten core drawing (MCD) Laser-based molten core drawing Post-processing methods CO laser-based fiber manufacturing Preform system Draw tower system Preform cleaning procedure Preform tube cleaning Inserts cleaning Hybrid fiber fabrication: an overview of the challenges based on example of silicon-core fiber fabrication Oxygen diffusion Interface layer Rayleigh-Plateau instability Thermal expansion during initial tapering Si-core fiber preform Preform preparation and fiber draw Preform geometry Purity of the used silica glass Other hybrid fiber preforms Part II: Specialty optical fiber preform fabrication using glass additive manufacturing Glass additive manufacturing techniques Direct glass laser deposition (DGLD) Laser powder deposition (LPD)... 54

20 XIX 7 Preform fabrication for all-glass specialty optical fiber Multicore optical fiber preform Doped-silica optical fiber preforms Conclusions and outlook References... 65

21 XX

22 1 1 Introduction 1.1 Introduction Today, all-silica fibers are manufactured at a very low cost and at incredible rates. These fibers, made for telecommunication purposes, exhibit ultra-low transmission losses and remain the major fiber type for most applications. However, despite its strengths, silica does have limitations when it comes to the functionality that it provides. Thus, fibers made of this material are predominately used in passive optical guidance and data transfer, with signal generation and manipulation happening outside of this platform. The first attempt to functionalize silica fibers was performed by employing active rare-earth dopants into the core. This led to new applications, like light detection and amplification, accessible in the convenient in-fiber platform [1 3]. In recent years, hybrid fibers, and semiconductor core fibers in particular, have emerged as promising candidates to complement all-silica-based fiber platform. The semiconductor core of these fibers has enabled applications, which require a wider IR transmission window [4] and stronger non-linear properties [5]. Additionally, semiconductor core fibers combine the advantages of the glass fiber platform, such as light guiding, operational safety, and mechanical flexibility, with the exceptional optoelectronic properties of a semiconductor. These fibers, however, are very challenging to fabricate due to the drastically different material properties of the core and cladding materials during processing. The main effort of this work was therefore directed toward developing a suitable technique for hybrid fiber preform fabrication. The chosen method was based on carbon monoxide laser heating of a rotating preform in a modified glass lathe. The preforms used in this method are typically assembled via a rod-in-tube technique with their structure defined by the cladding tube layout and inserted materials.

23 2 Further functionalization of these fibers, such as embedded in-fiber devices provided, for example, by multiple cores of different geometries or compositions [6], requires very specialized equipment, and is not possible using standard preform fabrication techniques. Fabrication of such fibers is, therefore, typically both expensive and time-consuming. Over the recent years, additive manufacturing (AM) has become a more promising option for rapid optical fiber preform fabrication and prototyping made of glass [7 10]. However, due to layer-by-layer deposition of cladding, the core-cladding interface of such 3D printed preforms is typically of poor optical quality, which further reflects in higher scattering losses. At the same time, losses associated with absorption are also present due to not fully debinded polymer in the manufactured glass bulk [7], or gradually increase when the melt-in-crucible glass printing approach is used [8 10]. Therefore, high demand for noncontact glass manufacturing using lasers exists today, to be able to reach the high temperatures required to melt pure silica glass and to avoid additional extrusion contamination upon melting. In this work, two such glass AM methods, free of these limitations, were applied to assist the specialty fiber preform fabrication process. 1.2 Motivation Among all elements, silicon stands out for its broad prevalence in modern electronics, acceptable electrical properties, and low manufacturing cost [11]. Integrating silicon in a fiber package enables the optical fiber to access some of the functionality of the more traditional silicon-on-insulator platform. Not only does this include accessing the electrical properties of the core, but also nonlinear optical properties, which can provide a four-orders of magnitude greater nonlinear response, compared to the results achieved in standard silica optical fiber [12]. This opens up a broad scope of applications, such as new types of in-fiber lasers [13], mid-infrared supercontinuum generation [14], and other nonlinear optical devices [15], provides a unique foundation for THz transmission and IR sensing [16,17], allows for an efficient mean of light harvesting [18], or direct signal transferring between integrated silicon-based optoelectronic devices. The myriad of applications and the potential to turn silicon into a single universal platform combining the realms of optics and electronics has resulted in granting silicon a benchmarking status in the scope of hybrid

24 fiber manufacturing. Thus, the first hybrid fibers produced with various methods were using silicon as core material [19,20]. An additional benchmarking claim is ensured by the material properties of silicon itself, as it provides several key challenges that had to be addressed during the manufacturing process, as characteristic of most hybrid materials. The two most pronounced challenges are the volume expansion of silicon during crystallization and the significantly higher thermal conductivity, compared to a silica cladding. On the other hand, silicon is known for its ability to nucleate, i.e., self-crystallize upon solidification [21]. This results in the crystalline structure of the core being achieved automatically without a need to engineer additional processing conditions during fiber preform manufacturing. However, despite the significant attention that has been paid to silicon as the core material, the typical attenuation in these fibers exceeds 1 db/cm, effectively limiting the light transmission range to length of 1 meter. The most recent results showed losses as low as 0.2 db/cm [22], which is only an order of magnitude higher compared to the numbers achieved in state-of-the-art integrated silicon waveguides [23]. Although such improvement is impressive, it is still far from reaching a potential 20 db/km loss measured in the bulk [24]. This thesis work was therefore focused on investigating an alternative technique based on CO laser heating to fabricate hybrid fibers, starting from silicon. The idea of this project was born at the same time as the first highpower carbon monoxide (CO) laser was brought to the market. Compared to CO 2, a CO laser has a shorter emission wavelength, centered near 5.5 μm. At this wavelength, most glasses have an order of magnitude lower absorption coefficient compared to CO 2 lasers operating at 10.6 µm. A deeper penetration depth of a CO laser provides an improved energy transfer efficiency to hybrid preforms of small to medium sizes with a very low silica vaporization rate. Moreover, using laser heating by itself for hybrid fiber fabrication carries many advantages over traditional fabrication methods. One such advantage is a localized hot zone that minimizes exposure of the material under process to high temperatures. This reduces effects causing core contamination, i.e., interdiffusion, dissociation, and oxidization [25]. 3

25 4 1.3 Outline The thesis is structurally divided into two parts. The Part I covers hybrid fiber preform manufacturing employing CO laser heating and Part II focuses on specialty fiber manufacturing. Firstly, Section 2 introduces the CO laser technology, discusses its working principle, and the benefits that open up in applications such as glass processing. The discussion about the performance of the CO laser system used in this project concludes this section. Section 3 provides an overview within the scope of optical fiber and fiber preform manufacturing, from more habitual telecom to hybrid and specialty optical fibers. The section ends with a description of both, the developed CO laserbased preform manufacturing system and the experimental draw tower. These systems are central points of Papers I and III. While Section 4 only focuses on the cleaning of the preforms, the final Section 5 of the first part of the thesis concerns directly the hybrid fiber fabrication. The scope of this section is focused on the challenges that arose during hybrid fiber preform fabrication using silicon as a benchmarking material. The overview of these challenges includes aspects like core-cladding oxygen interdiffusion, the use of interface modification layers, instability effects, and stress incorporations into the preform cladding due to the thermal expansion of silicon. Moreover, Section 5 contains the results of the process development for Si-core preform fabrication that was conducted and optimized during the project. A condensed version of this process development was published in Paper II, demonstrating the largest and the smallest achieved core sizes in the preforms. The second part of this thesis shows the combined effort of applying the state-of-the-art glass additive manufacturing techniques and the asdescribed CO laser-based fiber fabrication system to achieve rapid prototyping of specialty optical fibers. Section 6 briefly reviews these glass 3D printing methods while Section 7 discusses the quality of the resulting preforms. Although the results achieved using the Laser Powder Deposition technique are discussed in Paper IV, the filament-based glass 3D printing solution was not yet published. The preform manufacturing aspect of fiber fabrication is traditionally omitted or only briefly mentioned in scientific publications. Therefore, the focus of this part is shifted towards the preform fabrication and characterization. Finally, Section 8 holds combined conclusions from both parts of the thesis.

26 5 Part I: CO laser-assisted preform manufacturing 2 CO Laser 2.1 History of invention Since the mid-1970 s when the CO and CO 2 lasers were invented by Kumar Patel [26,27], a substantial amount of studies was performed to reveal the potential of these laser technologies. This was the golden age of carbon monoxide laser research and development [28 34]. During this time multi-kw CO lasers working in a continuous wave (CW) regime with up to 44 % quantum efficiency were demonstrated [35], which already reached the theoretical maximum of the competing carbon dioxide laser technology [36]. Apart from the improved energy consumption compared to the CO 2 laser, the broad-band spectrum emission of the carbon monoxide laser was in a shorter wavelength range of 5 to 6 um. This offered superior material absorption capabilities for materials like metal, plastic, glass, or ceramic due to deeper penetration depth. This proved to be extremely beneficial in many industrial applications, like material processing and cutting [35]. Although the potential of the technology was undeniable, the technological progress of that time made the real-live realization of the CO laser setup unpractical. These room-sized setups were able to achieve high powers only at cryogenic operation temperatures and for a very short time, due to a limited sealed-off gas lifetime measured in hours. A high atmospheric water vapor absorption has played a substantial role as well, causing high laser power attenuation and beam distortion. Though attempts were made on laser modes selection with high atmospheric transmission [37,38], the low

27 6 power output reported in these studies quickly proved this technology unusable in high power and military applications [39]. Thus, by the late 1990s, only three groups around the world were still doing research and development of CO laser technology, which were summarized in a review by Maisenhalder [35]. By that time, the CO 2 laser had completely taken over the market niche in gas laser material processing and many industrial applications due to the absence of crucial limitations a unique transparency window for atmospheric conditions and its aptitude to process a vast range of materials, including plastic, metal, ceramic, and glass. At the present time, only a few CO lasers are commercially available on the market with only one in the mid- to high-power output range. 2.2 Inversion mechanisms of CO2 and CO lasers To simplify the explanation process for the inversion mechanism in the CO molecule let s briefly recall the working principle of its famous brother - the CO 2 laser. Being a linear triatomic molecule carbon dioxide has three normal vibrational modes: bending, symmetrical and asymmetrical stretching. Lasing takes place between two of those vibrational states starting from the (001) level and either (100) with a central wavelength of 10.6 μm or (020) state with an emission around 9.6 μm. Figure 2.1 illustrates a simplified energy level diagram for the electronic ground state of the CO 2 and N 2 molecules. Fig The lowest vibrational levels of the ground electronic state of CO2 and N2 molecules. Adapted from [36]. Today, the most popular way of reaching inversion is done by virtue of arc discharge or heating plasma electrons by a modulated RF field. The

28 7 pumping of the (001) state in the first CO 2 lasers was reached via direct electron collision (e-v pumping) as follows: CO 2(000) + e* CO 2(001) + e [26]. Here, the carbon dioxide molecule dissociated on oxygen and carbon monoxide, where the latter played the role of a metastable level. Inspired by this fact, it was quickly discovered that adding N 2 to the gas mixture makes the excitation process extremely efficient [40]. Now, instead of directly exciting CO 2, the electron collision process targets the N 2 molecule, which excited state V = 1 is metastable due to the fact that the V = 1 V = 0 transition is electric-dipole forbidden by the symmetry [36]. The variety of excited N 2 molecules acts as an energy pool from where the transition to CO 2 is satisfied via the resonant energy transfer mechanism. This process is very efficient because the energy difference between the excited states of two molecules is very small ( 18 cm -1 ). The V-V relaxation mechanism is responsible for coupling energy from levels (100) or (020) to a (010) in the following way: CO 2(100/020) + CO 2(000) 2 CO 2(010) + ΔE, where ΔE is much smaller than kt. It ensures thermal equilibrium between these levels, which, in turn, will reduce population inversion if the relaxation time of the (010) level is not fast enough. In fact, the relaxation from the (010) can only happen by virtue of V-T relaxation (vibrational energy is transferred to the translation energy of the colliding partners). To mitigate depopulation of the aforementioned level a light atomic gas (usually helium) is introduced to the gas mixture. Fig The relative population of the rotational levels of the excited vibrational state of a CO2 molecule. Adapted from [36].

29 8 In practice, of course, each vibrational level consists of many closely spaced rotational states, whereas laser emission can occur on several equally spaced rotational-vibrational transitions from P- or R-branches. The oscillation with the highest gain will therefore occur from the level of the highest population, which according to the Boltzmann distribution in the case of the CO 2 laser has J = 21. This corresponds to the P(22) line (see Fig. 2.2) and the CO 2(001) CO 2(100) transition having a 10.6 μm wavelength. In the CO molecule, on the other hand, only one vibrational mode exists. Just like in the case of the CO 2 molecule, the initial pumping of the CO levels is performed by electron-impact collisions (e-v excitation), where states with the vibrational quantum number of up to V = 5 are excited (see Fig. 2.3(a)). As a result, the energy distributes at these levels according to the Boltzmann distribution. The pumping process is then taken over by near-resonant vibrational-vibrational (V-V) energy exchange collisions. This, in turn, is responsible for non-boltzmann population buildup on higher vibrational levels due to the slight anharmonic nature of the CO potential well. Primarily this happens due to the fact that the V-V relaxation process in the CO molecule is faster than the V-T relaxation. Such a process is usually referred to in the literature as anharmonic pumping [36]. Although this phenomenon does not allow to establish total population inversion, a so-called partial inversion can be sustained. The idea of such an inversion mechanism is illustrated in Fig. 2.3(b). The selection rules for this system allow transitions with ΔJ = ±1. This, and the anharmonic population nature of rotational states provide the possibility to find for a certain rotational state a state on a lower vibrational level with a lower density of states, even though the total populations of the two vibrational states are equal. This, for example, satisfies the transitions J i+1 = 5 J i = 6 and J i+1 = 4 J i = 5. Once occurred, such transitions depopulate a rotational state, providing partial inversion conditions for a sequential transition from a still higher vibrational state. As a result, the partial inversion is once again satisfied and the cascade lasing transitions occur. This property of the CO laser implies a theoretical maximum quantum efficiency to be near 100% [39]. The most perceptive reader may have already noticed that such partial inversion can mostly occur on P-branch transitions (ΔJ = +1), whereas all R- branch transitions (ΔJ = -1) have negative gain. Although it seems that partial inversion can also occur for two R-branch transitions, as showed in Fig. 2.3(b), in practice they have very little to no gain [27].

30 9 Fig (a) The vibrational/rotational energy level diagram of a CO molecule and (b) demonstration of the partial inversion between two vibrational transitions of the same total population. Red arrows correspond to cascade laser transitions. Adapted from [36]. 2.3 Water vapor absorption at the CO laser wavelength The non-equidistant energy separation of the lasing vibrational states in a CO molecule results in a spectrally broad comb-like output of the laser. The spectral output of modern CO lasers lies in the range of 5 6 μm at room temperature [39]. This wavelength range, however, has a considerable overlap with the water vapor absorption band (see Fig. 2.4), which attenuates the laser beam and causes beam shape deterioration [39,41]. This deterioration effect is caused by a localized change of the refractive index along the laser beam path, as a direct consequence of laser emission absorption in water molecules. Such an effect is called thermal blooming. The CO laser beam diameter deterioration follows a near-linear relationship with relatively low humidity below 2000 ppm V. The slope of this relationship is defined by the refractive index gradient across the laser beam path and therefore by the absolute laser power. In addition to a beam diameter, the overall laser power also experiences attenuation [41]. To minimize this absorption, CO lasers have to be operated at low humidity, which is achievable in a laboratory environment by enclosing the beam path and purging it with dry air. However, at very low humidity, typically below 60 ppm V of water concentration, the encased system may experience

31 10 Fig The CO laser emission and the water vapor absorption spectral overlap [39]. Scaled corresponding to 1 m of laser optical path length at a dew point of 7 C. Adapted with the permission of the Society of Photo-Optical Instrumentation Engineers (SPIE). unstable beam thermal blooming. This may be caused by uneven airflow across the laser beam path or humidity variation, caused by moving components inside the enclosure. This effect is discussed in more detail in Section III of Paper I Comparison of a CO and a CO2 laser for glass processing Silica glass has a very high absorption in mid-ir, specifically in the operating ranges of both CO and CO 2 lasers. In the case of a CO laser, however, the absorption is an order of magnitude lower [42], which gives this technology an advantage in terms of glass processing. Lower absorption allows laser radiation to penetrate deeper into the glass bulk, providing a more even and efficient energy transfer [43]. In the case of CO 2 laser, the penetration depth at 10.6 μm wavelength is approximately 4 μm at 1800 C [44]. This results in a very high energy density at the surface of the glass, which combined with a low thermal conductivity of glass, leads to a steep thermal gradient between the surface and the bulk. At the same time, the energy loss at the surface is the highest due to convection, surface emission, and silica vaporization. The consequence of this thermodynamic equilibrium is a limited temperature that can be achieved at a certain depth from the glass surface. This principle also applies to the CO laser emission, but

32 11 due to a greater penetration, the temperature achievable in the bulk is higher. Moreover, more efficient energy conduction within the glass bulk enables a faster energy deposition rate, resulting in faster heating. This implies that glass processing can be conducted at a faster rate with lower silica vaporization when using a CO laser. This effect of a different thermal response of a 6 mm silica rod, while heated by both lasers, was modeled in Paper III. In addition, the thermal response, as well as the highest reachable temperature of the rods of different diameters, was simulated under irradiation of a CO laser, while the rods were rotating (Paper II) CO laser characterization In this work, a CO laser was used to process glass samples primarily of a cylindrical symmetry. The processing included sample tapering, splicing, and cutting. While the last two procedures do not require the utmost performance in terms of power stability and power output, the tapering step does. Typically, tapering was conducted with a speed of up to 0.6 mm/s, and therefore required some time to be completed. Over this period of time, the tapering parameters have to be kept constant to reduce the additional influence on the newly established thermodynamic equilibrium of the glass sample under process. An example of such an influence is demonstrated in Fig. 2.5(a), where the resultant taper shows a substantial diameter variation in the section, where the laser power experienced fluctuations. Here, the silica rod of 10 mm in diameter was tapered down to 5 mm, without any laser power stabilization. Another important parameter of the laser, apart from the stability, is the high power output. Tapering glass samples that contain other materials in the core will require more power, especially if this material has a high melting temperature [45]. Therefore, in order to achieve the best possible performance of the laser, it has to be characterized first. A Coherent J-3-5 high power CO laser was used. It was capable to generate up to 330 W of CW output with an M 2 of 1.2. The laser had a 4 mm beam diameter with a divergence of 2 mrad. Most of the spectral output of this CO laser model was within the µm wavelength range, as shown in Fig To minimize the effects associated with water vapor absorption, the characterization of the CO laser was performed in a sealed enclosure in the humidity range of ppm V (see Paper I). The laser beam was directed

33 12 onto the power meter (400 W Ophir FL400A) via three gold-coated mirrors with a total beam optical path length of 1.38 m. The loss on a single mirror was measured to be 5.1 %. All measurements were performed with the power meter set to µm wavelength sensitivity range. To allow for computer control of the laser a custom-made control interface was designed. It provided an external signal to a laser controller unit, which, in turn, generated a pulse-width modulated (PWM) signal. This PWM signal was driving the laser pumping process with a defined duty cycle and frequency. The duty cycle was responsible for the average laser pumping power. It was regulated in the range of % with a 0.1 % step. The signal was modulated in a radio-frequency range of 100 Hz 100 khz. Although the whole range up to 100 % was available, the duty cycle of only up to 60 % was used, in accordance with the laser operational manual. Such an external power regulation was performed in a constant pumping frequency state, however, allowing for manual frequency regulation with a nob, when necessary. Fig (a) A test taper of a 10 mm silica rod down to 5 mm with no laser preheating or stabilization. The power meter was partially shadowed by the rod, and in this case, mostly measured black body radiation from a heated glass. The laser power used was 300 W. (b) A typical CO laser power drift over the first 75 min of operation. To evaluate the laser power stability, the laser power was measured at a 60 % PWM duty cycle. Figure 2.5(b) shows the laser power drift over the first 75 minutes. After the first 5 min of lasing the laser power stability typically showed slow power fluctuations. After this initial stabilization period, the laser power fluctuated within 2 % of the total laser power. To achieve a more stable lasing within 1.5 % drift, the laser was usually preheated for a time of min. In contrast to the case shown in Fig. 2.5(a), the power drift after

34 13 30 min preheating time was found to have an insignificant influence on the tapered sample geometry, which is further discussed in Section Although tapering is typically performed at the highest power available, sample cutting, and splicing were performed in the range of W (see Section IV of Paper I). To succeed in these tasks, the dependence between the output power and the laser control signal was studied. The laser power linearity was measured with respect to a control signal with a duty cycle step of 2 %. The measurement was performed at the highest starting power to allow the laser preheating for 30 min. For each duty cycle step, the lasing was performed for 10 min. The data was then analyzed by averaging the measured power over the last minute for each duty cycle step. The measurements were performed for pumping frequencies of 1.11, 1.85, and 2.85 khz and are summarized in Fig. 2.6(a). The measured laser power output showed a nonlinear dependence from the pumping duty cycle, with lower output efficiency at high power. Additionally, in this frequency range, the higher pumping frequency has led to a higher laser threshold power. This resulted in a lower laser output at the same duty cycle, when operating the laser at low power. Changing laser control signal frequency manually mid-process allows the operator to fine-tune the output laser power if required instead of restarting the whole process with the new parameters. Fig (a) A CO laser power as a function of a control signal and (b) a pumping frequency response. The major part of glass processing was conducted at the highest laser power available. For this reason, the CO laser pumping frequency response at 60 % PWM duty cycle was evaluated. Figure 2.6(b) shows that the highest output power was achievable in the frequency range of 1.5 khz to 3 khz.

35 14 Selecting frequency in that range allowed fine-tuning of the laser power output to compensate for day-to-day absolute power inconsistency. Running the laser at high power gradually increased humidity inside the sealed enclosure. This effect is believed to be related to additional water evaporation from heated surfaces as the setup was being heated. This effect was typically apparent at powers above 30 W. As a result, in high-power experiments, such as those, shown in Fig. 2.5 and 2.6, humidity levels below 24 ppm V ( -54 C) were not achievable.

36 15 3 Fiber manufacturing techniques 3.1. Optical fiber fabrication for telecommunication Optical fibers Optical fiber is a dielectric-based optical waveguide having a cylindrical symmetry [46]. Although for some specific applications the geometry of the optical fiber can deviate from this symmetry [47 49], it is fundamental to fiber design and fabrication. Fig (a) Schematic representation of light guidance in an optical fiber, and (b) a generic optical fiber design. The jacket (usually made of polymer) is used as a protective layer from mechanical influences and OH diffusion into the fiber core. The optical fiber working principle is based on total internal reflection phenomenon. The conditions for total internal reflection to occur within the fiber are met by introducing dopants to the core region of the fiber, i.e., the fiber core, hence increasing its refractive index (RI) compared to the rest of the fiber. The un-doped region of the fiber is called fiber cladding. Figure 3.1 illustrates the guiding property of the optical fiber. Here, n i is the RI of the surrounding media, n core and n clad are the refractive indices of the core and the

37 16 cladding, respectively. θ i is the incident angle and θ c is the internal reflection angle. To satisfy total internal reflection conditions two criteria have to be met: the RI of the core has to be greater than the RI of the cladding (n core > n clad) and the incident light angle has to be within the total fiber acceptance angle (θ a > θ i). When the latter is fulfilled, all light that is coupled into the fiber core is guided and outcoupling to the cladding according to Snell s law is not possible. Eq. (3.1) shows the relation between the total fiber acceptance angle (θ a) and the RI of the surrounding media, core, and cladding: NA = n i sin θ a = n2 2 core n clad. (3.1) Here, NA is the numerical aperture, which is defined as the maximum cone of light that can exit the fiber on the output and is usually in the range of 0.1 to 1. Optical fibers were first envisioned in the early 1960s with the invention of the laser. Originally, the idea of using optical fibers made of glass to avoid degradation of the optical signal by the atmosphere appeared as the replacement for the coaxial cable or carrier transmission system. The enormous potential bandwidth of the optical-based systems in the frequency range of 10 5 GHz far surpassed the bandwidth capability of the conductorbased wires (typically around 20 MHz with 5 to 10 db/km attenuation). The first optical fibers, however, exhibited very high losses of up to 1000 db/km. The start of fiber optics began with the famous paper that appeared in Proceedings of the IEE in July 1966, written by Charles K. Kao and George A. Hockham [50]. In this work the authors theorized that the attenuation in optical fibers can be reduced to 20 db/km, making long-distance propagation possible. Later the same year, a Standard Telecommunications Laboratory press release said that the single-mode fiber of 1-GHz bandwidth could carry 200 television channels or 200,000 telephone conversations Outside vapor deposition (OVD) The first fibers with sub 20 db/km losses were obtained using a technique called outside vapor deposition (OVD), developed by Hyde [51] (later published by Blankenship [52]) for Corning Inc. Using this technique, Corning produced the first Ge-doped optical fiber exhibiting loss of 20 db/km in They further improved this result to 4 db/km by 1972.

38 17 The OVD process is based on depositing soot particles onto a horizontally positioned rotating bait glass rod, generated by hydrolyzing the halide vapors in an oxygen-hydrogen flame. During this process, gas burners are moved back and forth along the rod to ensure homogeneous layer-bylayer buildup (see Fig. 3.2). Fig (a) Peter Schultz using OVD to make a Ge-doped multimode fiber preform (1972) [Courtesy of Corning Inc.] and (b) Schematic diagram of OVD working principle. In general, any vapor-phase deposition-based technique [52 60], such as OVD, VAD, or MCVD, is used to first form a fiber preform, which is then drawn into fiber in a conventional draw tower [56,57]. In these techniques the silicon soot particles (SiO 2) are formed in the same way by passing oxygen (or water vapor) through silicon tetrachloride (SiCl 4), which is vaporized, removing any impurities in the following chemical reactions [53]: SiCl 4 +H 2 O heat SiO 2 + 4HCl (3.2) SiCl 4 +O 2 heat SiO 2 + 2Cl 2 (3.3) The core of the preform is doped by adding GeCl 4 or TiCl 4 vapors either to the same gas mixture or with a separate torch. The doping concentration is therefore carried out by precisely controlling the flow of these dopants. Here, a reaction similar to Eq. (3.3) occurs: GeCl 4 +O 2 heat GeO 2 + 2Cl 2 (3.4) TiCl 4 +O 2 heat TiO 2 + 2Cl 2 (3.5) Due to the layered nature of the OVD process, the preform preparation is typically performed in three steps. First, the doped glass is deposited to form a core structure of the preform, sequentially followed by pure silica glass

39 18 cladding encapsulation. The bait rod is then removed from the resulting soot preform and sintered in a furnace at 1700 C, where it slowly consolidates vitrifying the glass. This process also purifies the preform from any gaseous impurities that may have been incorporated during deposition. To further reduce OH content below 50 ppm V chlorine gas is usually employed during sintering as a drying agent [52]. Such a preform is collapsed during sintering [52] or directly during fiber drawing in the preform neck-down region [53]. Despite the fact that OVD is considered to be one of the most efficient ways to produce optical fiber preforms to date (the common preform diameter is 150 mm), the method has several drawbacks, mainly related to stress accumulation upon removal of the bait rod and the inability to adapt the setup for production of more than one type of fiber. Combined with the very high production cost of the smaller preforms OVD, in general, is not suitable for scientific applications and fiber prototyping. Due to the high deposition rate in the range of g/min, today, OVD remains the most ordinary and cost-effective method of producing optical fibers for telecommunication and is often used in combination with other vapor-phase deposition techniques, such as vapor-phase axial deposition (VAD) or modified chemical vapor deposition (MCVD) to expand the silica cladding diameter of the produced preforms Vapor-phase axial deposition (VAD) The vapor-phase axial deposition technique was developed in 1977 and later published in 2000 by Izawa [54]. The working principle of the method was very similar to OVD. Here, the soot material was deposited on the vertically positioned bait rod with two or more rows of gas burners, each depositing different material at the same time in a single pass. As such, the core and the cladding were deposited simultaneously in a continuous process (with a lateral shift) as the bait rod was slowly pulled up with the preform growth rate. This enabled the fabrication of very long preforms (more than 2 m) of the same diameter range, as in OVD, already in late 1977 [54]. Almost a decade of improvement in this field has allowed the production of optical fibers with db/km attenuation at the 1550 nm wavelength [55]. Unlike OVD, the preform processing parameters, i.e., deposition rate, preform size, and fabrication time in this method are proprietary and not well

40 19 known. However, today it is considered the most efficient way to produce low-loss optical fibers of thousands of kilometers in length [55] Modified chemical vapor deposition (MCVD) Following the low-loss advancement trend in telecommunication, which started in the early 1970s [61], a new approach for fabrication of optical fiber preforms was discovered, modifying the commonly used chemical vapor deposition technique. As such, the modified chemical vapor deposition (MCVD) technique was reported in mid-1974 almost simultaneously by AT&T Bell Telephone Laboratories [58,59] and Southampton University [60]. In the MCVD process, the high-temperature oxidation of reagents is realized inside a rotating substrate tube, which is heated from the outside by an external heat source, e.g., an oxyhydrogen torch. This process is using the same type of chemicals as described in Eq. (3.3) (3.5), and is schematically shown in Fig. 3.3(a). Fig (a) Schematic diagram showing the MCVD deposition process and (b) the following three-step-process for preparation of the final optical fiber preform. Unlike in OVD, here, the material is deposited from the inside of the tube layer-by-layer. First, the cladding is extended to create a protection barrier from the diffusion of impurities, such as OH [56], as well as to deposit a lowloss material to the core-cladding interface (see Fig. 3.3(b)). The composition

41 20 of the vapor steam is then changed to meet the desired dopant concentration to deposit the preform core. The function of the heating torch in the MCVD process is twofold. First, the traversing hot zone initiates a chemical reaction, hence allowing the new material to deposit. Second, the newly deposited layer is sintered and vitrified. The material deposition process is followed by a preform collapsing (see Fig. 3.3(b)) and sleeving step to create the required preform core-to-cladding ratio. The typical size of the preform produced by MCVD is around 20 mm in diameter and up to 1 m in length. Today, the MCVD technique is considered quite expensive due to relatively slow manufacturing time (deposition rate is typically 2-4 g/min) and smaller preform sizes. However, it remains the most popular method of manufacturing ultra-low-loss optical fibers for non-commercial telecom use [62 64]. Although vapor-phase deposition techniques were among the first to fabricate a viable optical fiber, they continue to produce telecom-grade optical fibers up to this day. 3.2 Hybrid optical fiber manufacturing Hybrid optical fibers The techniques described in Section 3.1 are highly specialized in depositing glass in the form of oxides and therefore are used to produce allglass optical fibers. Although, despite a tremendous effort devoted to the optimization of the transmission performance of these fibers, their bandwidth is limited due to the intrinsic absorption of silica glass in the mid- IR spectral range [42]. To preserve the best transmission performance and keep the benefit of the widest achievable bandwidth of the optical fiber, a wavelength near 1560 nm is used in modern telecom industry [65]. One way to access wavelengths beyond 2.5 µm is to use drastically different materials for the fiber core and the cladding, such as chalcogenide-core polymerclad [66] or semiconductor-core glass-clad fibers [67]. Such optical fibers made of disparate families of materials are usually referred to as multimaterial or hybrid optical fibers.

42 21 With the advent of these hybrid fibers, a wide range of previously inaccessible properties has been brought to the optical fiber format, including in-fiber optoelectronics [17,68], photodetection [69], and health monitoring [70]. One of the most widely used semiconductors today, silicon, historically has attracted the greatest attention for fiber applications due to its transparency in the mid-ir, its non-linearity, and the ability to combine optical and electrical properties. Embedded in an optical fiber package, silicon alone opened a door to many applications, such as in-fiber lasers [13,17], nonlinear optical devices [15], THz transmissions [16], and energy harvesting [18]. The core of a hybrid optical fiber comprises a single- or multi-component material, which is not possible to fabricate using traditional techniques described earlier. For this reason, two major techniques are used to produce hybrid optical fibers today: high pressure-assisted chemical vapor deposition (HPCVD) and molten core drawing (MCD). The fibers produced with these techniques typically consist of a polycrystalline or amorphous core. To further enhance transparency and improve material quality, these as-produced fibers are usually post-processed using thermal annealing or laser processing High pressure chemical vapor deposition (HPCVD) In 2006 the first crystalline semiconductor-core optical fiber was produced as a result of a collaboration project between Pennsylvania State University and the Optoelectronics Research Centre at the University of Southampton [19]. In this work, a commonly known CVD process was adapted to grow a crystalline core structure of elemental group IV materials (Si and Ge) within a silica micro-capillary. This was achieved by applying the reactant gas, i.e., the precursor gas mixture (SiH 4 and GeH 4) with an inert carrier gas, at a very high pressure of bars. The silica capillary was not only chosen to act as a low refractive index cladding material, but it also provided a strong enough reaction chamber to withstand such high pressures during deposition. The reaction was thermally initiated by an externally positioned heater, maintaining the temperature at the required level of up to 500 C [19]. Just like in the MCVD process, described in Section 3.1.3, the deposition in HPCVD is performed from the inner side of the capillary, as schematically shown in Fig. 3.4(a). The final thickness of the deposited material is determined by the deposition time; the deposition is continued

43 22 until the required film thickness is achieved. The thickness can be controlled with an accuracy down to a few nanometers, making it possible to sequentially deposit different materials in concentric layers. Core structures with core diameters from hundreds of nanometers up to several micrometers can be achieved [19]. The HPCVD-produced fibers are typically amorphous in structure [71], although applying post-processing techniques, polycrystalline [72], as well as monocrystalline [73] semiconductor-core structures were also achievable. The lowest losses achieved with HPCVD in as-produced Si-core fiber were reported at 6 db/cm [74] while applying post-processing allowed lowering this number down to 0.47 db/cm [75]. The low deposition temperature of the HPCVD process enables a very wide material library [19,76], including materials that sublimes or have a drastically different thermal expansion coefficient mismatched with the cladding [19,67,77]. Although, despite being a very powerful technique, HPCVD has a significant disadvantage. Due to low temperature used, the reaction rate and, therefore, deposition yield are very slow. A few weeks of continuous deposition typically results in a centimeter-long fiber [67], rendering long fiber fabrication with the HPCVD technique unviable. Fig (a) Schematic diagram showing the HPCVD process and (b) the MCD fiber manufacturing methods Molten core drawing (MCD) In 2008, a new approach for fabricating hybrid optical fibers was discovered at Clemson University, in an effort led by Ballato [20]. This

44 research was born in response to a rising demands for longer silicon core fibers, as those were required in applications such as on-chip lasers and mid- IR transmission. This technique was named molten core method (MCM) or molten core drawing (MCD) [78] but is also sometimes referred to in literature as melt-in-tube (MIT) [79]. MCD is based on a conventional fiber drawing technique. The preform assembly consists of a cladding glass tube, which is filled with a core material either in the form of a rod or powder (see Fig 3.4(b)). To distinguish the two methods, they traditionally are referred to as rod-in-tube [20,79 83] or powder-in-tube [84 88]. Although, in general, applying the material in a form of a solid rod is preferable in order to minimize the number of impurities within the core [80,82,89], a combined approach was also used to form crystalline multicomponent core structures [89]. This preform assembly is then externally heated, typically using a furnace or an oxyhydrogen torch, to exceed both glass transition and core melting temperatures. At these temperatures, glass cladding becomes soft enough and acts as a crucible for molten core material to preserve the original cylindrical geometry during the fiber draw. Leaving the furnace, the as-drawn fiber experiences rapid cooling, which results in a highly polycrystalline structure of the fiber. The major advantage of such a drawing method is the ability to fabricate fibers in long lengths, typically reaching tens of meters [20,90,91]. The core of these hybrid fibers produced with the MCD technique can contain long monocrystalline grains in centimeter length [90], which after post-processing, in general, show lower transmission losses, compared to fibers obtained with HPCVD. The record attenuation using conventional MCD is db/cm loss in post-processed Si-core silica-clad optical fiber [92]. However, the presence of high temperatures during the MCD process also results in slow diffusion of contaminants, such as oxygen and water, from the cladding into the fiber core [20]. Apart from creating unwanted inclusions and defects due to oxygen diffusion, the fiber core also experiences partial or complete vitrification at the core-cladding interface. As a result, early drawn hybrid fibers were restricted to have large core dimensions, typically with a core-to-cladding ratio above 0.05 for a standard size fiber (125 µm) [67]. Another challenge of the MCD technique is related to limited material usage due to a mismatch in melting/transition temperatures and thermal expansion coefficients of the core and the cladding. This is usually mitigated by choosing 23

45 24 the right glass material for each type of hybrid fiber [79,93]. Hence, not all materials can be drawn in fiber using MCD, especially if their melting temperature exceeds 2000 C. Silica remains the only viable option as a fiber cladding for these high-temperature materials, which at those temperatures becomes too soft to act as a crucible. An introduction of an interface layer between the core and the cladding partially addresses the issue of oxygen interdiffusion [94,95]. It also mitigates axial stress accumulation in the core, reducing the crack formation when thermal expansion coefficients of the used materials are different. Implementing this in the MCD process allowed producing Si-core fibers with core sizes down to 4 µm [77,96] Laser-based molten core drawing The idea to apply laser heating for fiber manufacturing process by itself is not new and first appeared in the late 1970s [97 99]. The modified MCD process featured with a laser-based furnace carries many advantages. One such advantage is the possibility to fabricate a small batch of highly customized specialty fiber preforms for fast prototyping, research, and development. Laser heating provides a high level of control over the processing temperature, material viscosity, and tapering dynamics, necessary for hybrid fiber fabrication. A very localized hot zone, provided by laser heating, ensures brief thermal interaction between the core and cladding during fiber fabrication, enabling minimal OH diffusion and oxidization [56]. Although the technique was known for a long time, a laser-modified MCD was only recently applied to draw hybrid optical fibers [45,100]. In both cases, a CO 2 laser was used as a heat source and a silica glass tube as the cladding for preform assembly. Materials, such as germanium [100] and silicon [45] were incorporated and drawn into fibers. A complicated optical arrangement was used to maximize the energy transfer efficiency from the laser to the preform with an adequate silica vaporization rate [45]. Although only a few centimeter-long sections of hybrid fibers were achieved, these studies demonstrate the feasibility of using a laser-based furnace for the MCD method.

46 Post-processing methods As it was mentioned before, post-processing of hybrid fibers is usually required to improve their optical transmission. This is achieved by selective or complete melting of the fiber core, allowing it to recrystallize upon cooling. This allows to reduce the number of grain boundaries within the core by improving fiber crystallinity [83]. There are a few types of post-processing techniques available, ranging from oven-annealing to laser-recrystallization. Despite being affordable and easy to use, furnace-based methods, including the full-fiber annealing [101] and movable ring-heaters [102] (Bridgman type processing), share similar drawbacks, such as being slow, not energy efficient, or unable to reach high temperatures to process silica-based fibers [25]. Laser-based techniques, on the other hand, are not limited in this regard and are routinely used to improve semiconductor-core fibers in a postdrawing procedure [75,82,83,89]. This approach is usually performed at a few millimeters-per-second rate while having a high degree of control over the recrystallization process and thus a low core-cladding interdiffusion. However, this comes at the cost of intricate system alignment and additional stress buildup in the fiber due to a difference in thermal expansion coefficients and non-symmetric laser heating. The energy can be transferred to the fiber core in two ways: by means of direct core melting [75] or through the thermal conduction to the core from the fiber cladding [82,83,89]. The latter is more popular since it does not require a powerful laser in the visible range or sub-micron precision to continuously focus the fiber core during lateral processing. The overall impact of post-processing on the fiber transmission losses is related to the fiber cooling rate [83]. The cooling rate is typically regulated by the annealing speed, while the fiber thermalization is mostly conducted through natural convection (no active cooling). A recent study shows the potential to improve the cooling dynamics during post-processing by increasing the annealing speed beyond 10 mm/s, revealing room for improvement, compared to stationary cooling dynamics of the fiber [103]. Although laser post-processing is an effective tool to improve the quality of the drawn hybrid fibers, today, the conventional fiber draw is performed at a much faster rate, i.e., 20-fold compared to the state-of-the-art post-

47 26 processing methods. It is, therefore, far from being a commercially viable option to produce long hybrid fibers. 3.3 CO laser-based fiber manufacturing Preform system The schematic representation of the preform fabrication system, based on a dual chuck glass lathe, is depicted in Fig The scanning stage, as well as the pulling stage (the lathe tailstock), were motorized using two stepper motors, allowing to achieve processing speeds in the range of 0.05 to 10 mm/s. Fig Top view of the schematic layout of the setup and the optical arrangement. The pink arrows represent directions of motion. A 300 W, quasi-cw carbon monoxide laser was used as the heat source. The laser beam was directed to a glass preform via three mirrors (M1-3) and shaped by a lens module (L) to achieve the desired power density and form a localized hot zone of the desired size. The reference laser power measurement was performed by tapping 1% of output power with a beam tap (BT). The glass preform was fixed in the lathe chucks, which rotate at a desired rate of up to 220 rpm. This provided homogeneous energy deposition as the sample processing was performed.

48 27 To ensure optimal CO laser operation, the entire system, including the lathe and the laser beam path, was covered in a sealed enclosure, as shown in Fig It was continuously purged with dry air to achieve a low and uniform humidity. Fig Side view of the preform manufacturing setup in the open state, allowing access to a glass preform. The presence of this sealed enclosure around the working environment imposed certain design and operational challenges, such as repurging the system after any manual interaction with a preform. To minimize the exposure time of the sealed part of the system to humid air, all essential alignments were performed remotely using motorized solutions. This was achieved by automating several sub-systems and linking them in a single communication network, thus allowing to control everything from a computer in a single interface. To fully automate the production process, the interface was also designed to support the execution of G-CODE instructions; a machine control language commonly used in manufacturing machines, like CNCs and 3D printers. The temperature of the preform during processing was measured using a fiber-coupled pyrometer (Fluke, ENDURANCE), in the range of C. A more detailed description of each sub-system is discussed in Section II of Paper I. The preform fabrication procedure consists of a few subsequent steps, which number depends on the preform dimension, core material, and target

49 28 core-to-cladding ratio. These processing steps are summarized in Fig First, the preform assembly is constructed. The preform tube is spliced to an exhaust tube, precut to the required length, and filled with the preform core material (Fig ). This assembly is then spliced to a handle from the open preform tube side and connected to flexible PVC tubing through an exhaust tube. This allows to conduct the following tapering while applying vacuum or under a rarefied nitrogen atmosphere. At this stage, the core material was prevented from moving within the tube using a custom-made motorized spring mechanism (see Section II of Paper I). The processes of splicing, cutting, and tapering are discussed in detail in Section IV of Paper I. Fig Typical specialty fiber preform preparations sequence, comprised of four steps: (1) assembly of the preform; (2) initial tapering; (3) second preform assembly; (4) final preform. This preform assembly is then tapered to fit in a sleeving tube (Fig ). The speed of this initial tapering step mostly depended on the taper core-tocladding ratio and was typically performed at mm/s. At this rate, a single point on the preform had experienced a high temperature (above the melting temperature of the core material) for seconds. Such brief interaction time allowed to minimize oxygen interdiffusion between the cladding and the molten core. To achieve a required core-to-cladding ratio of the final preform, this taper had to be sleeved in another cladding tube. It was therefore precut to a specified length and placed in the next preform assembly (Fig ). To avoid fracturing of the hybrid taper during cutting, associated with

50 29 incorporated stress during processing, a special laser cutting procedure was designed and discussed at the end of Section V in Paper II. From here, the new preform assembly may either be retapered into the same cane size, as in Fig to further reduce the core size or directly into the final preform, which was typically 6 mm in diameter (Fig ). Such a preform is then drawn into fiber in the experimental draw tower, based on CO laser heating Draw tower system Conventional draw towers usually take up space of a 3 5 story building and are operated by a dedicated team of several people. Such monumental equipment is typically highly optimized to effectively pull a specific type of optical fiber from meter-long preforms with a speed in the order of kilometers per minute. These towers, however, are not very practical when it comes to fiber prototyping or pulling small preforms of custom compositions with predefined speed. To meet these requirements, a lab-size draw tower was designed that can be conveniently operated by a single person (see Paper III). The draw tower heating furnace was also based on a CO laser. The tower architecture, as well as its furnace, are shown in Fig The preform was mounted on a feeding mechanism with two clamps. One clamp was responsible for holding the preform in place, while the other one regulated the preform tilt. The preform was then fed to the laser-based furnace, where it was heated from four sides, as shown in Fig. 3.8(b). The incoming CO laser beam was split by a diffraction grating (DG) into four beams equal in both power and shape. These beams were then redirected onto the preform using gold-coated mirrors M1-M4. To achieve homogeneous heating, the position of the incoming CO laser beam was modulated by a mirror mounted on a voice coil, providing a scanning movement of the laser beams across the preform, creating a neck-down region. The drawn fiber sequentially passed through a diameter gauge, coating cup, and a UV curing chamber, where the fiber coating was applied and cured. The coated fiber was then directed by a take-up wheel through a tension sensor to a capstan, where the fiber was reeled on a spool.

51 30 Fig (a) Illustration of the fiber draw tower layout and (b) top view schematic of the laser furnace (marked with the pink rectangle). Two closed-loop feedback systems were responsible for the drawing process. The first one was used to regulate the fiber diameter as close as

52 possible to a set diameter. To achieve this, a signal from a diameter gauge was used as feedback to tune the capstan spooling speed, hence adjusting the fiber diameter. A second feedback system enabled the fiber draw at a constant tension. This system monitored the tension and controlled the laser power, ensuring a constant temperature and glass viscosity in the hot zone. Such a feedback systems allowed for efficient compensation of any variation in the preform diameter or its composition, while keeping other draw parameters constant. A more detailed description of the draw tower is given in Paper III. Utilization of the laser allowed to keep a very localized neck-down region of the preform during the fiber draw. This enabled benefits previously mentioned in Section 3.2.4, such as low oxygen interdiffusion between the core and the cladding and minimal water diffusion to the fiber cladding. An additional advantage of a laser-based furnace compared to a traditional flame- or induction-based furnace, is the open design, which allowed direct observation of the drawing process using four cameras C1-C4, see Fig. 3.8(b). In the scope of this project, various types of specialty optical fibers have been pulled in this tower at speeds of up to 30 m/min. The absolute majority of these fibers had a silicon core and were typically drawn at 10 m/min. Such a draw speed is usually achieved by gradually ramping the speed in an initial all-silica region of the preform to provide adequate time for the feedback systems to stabilize the fiber diameter at the desired value. At such a drawing speed silicon core fibers automatically receive recrystallization processing, thanks to a steep thermal gradient in a laser-based furnace and fast draw dynamics. Analyzing the loss values of different sections of the same fiber drawn at different speeds showed an improved transmission when the fiber was drawn faster [104]. 31

53 32

54 33 4 Preform cleaning procedure 4.1 Preform tube cleaning After the preform tube was spliced to an exhaust tube, it had to be cut out from the rest of the glass tube. Although this cutting procedure was performed in a horizontal orientation, the process still induced additional fumed silica particle contamination of the preform tube assembly. To clean the preform assembly from any organic and inorganic contaminations the following cleaning procedure was implemented. The tube was submerged in methanol (MeOH) and placed in an ultrasonic bath for 5 minutes. After methanol evaporated the tube was etched in 40 % HF for another 5 minutes, rinsed in deionized water, and submerged in fresh % MeOH. The preform was then mounted in the preform fabrication setup, connected to the vacuum system, and filled with core inserts. The preform tube was additionally wiped with MeOH before the tapering to remove any dust particles. 4.2 Inserts cleaning A typical preform assembly is comprised of two types of inserts: a core material in the form of a rod and two glass rods. These rods were used to sandwich the core material inside the preform tube, acting as a seal and buffer material. The outer diameter of these rods was chosen as close as possible to the inner diameter of the preform tube. For this purpose, the Thorlabs FP1000ERT and FP1500ERT multimode silica fibers were used. The coating from these fibers was mechanically stripped and the polymer cladding was removed by wiping them with acetone. The rods were then precut to 15 mm and 40 mm in length. These inserts, together with the preform core rod, were sequentially cleaned in

55 34 acetone and MeOH ultrasonic baths for 5 minutes each. In some cases, plasma cleaning under the O 2 atmosphere was applied at this stage prior to the implementation of the inserts into the preform tube. Upon the retapering step, a tapered cane with a preform core material was used as the core for a new preform assembly (see Fig ). The glass surface of these canes was purified at a high temperature during the tapering process. In the case of a hybrid fiber preform, an ultrasonic bath was not performed, as these canes suffered from high internal stress and were typically very sensitive to bending and vibrations. Additional cleaning was therefore employed by wiping them in MeOH prior to insertion.

56 35 5 Hybrid fiber fabrication: an overview of the challenges based on example of silicon-core fiber fabrication 5.1 Oxygen diffusion Oxygen contamination of the fiber core has a great influence on the optical transmission properties of the optical fibers. In the case of telecom optical fibers, the manufacturing process was thoroughly optimized for decades to avoid impurities associated with the manufacturing process, especially at the preform preparation stage. The manufacturing techniques to produce hybrid fibers typically do not even involve the preform fabrication and are performed directly in a draw tower with an open preform tube, as described in Section During the last decade, the two main mechanisms of such oxygen contamination were defined. The first one was related to the preform tube contamination by water vapor through the open end of the crucible glass tube produced by an oxyhydrogen flame [88]. Sealing the preform tube from one end and evacuating the air with a vacuum system from another while drawing the fiber allowed to drastically reduce the oxygen content in the core of the produced Si-core fiber, compared to previously demonstrated results [20]. Although the improvement was significant, manufacturing of Si-core fibers with a micrometer-sized core was still not achievable due to complete core vitrification. The process responsible for this effect was the oxygen interdiffusion from the preform glass cladding to the core during the preform drawing/manufacturing process [96]. This interdiffusion has an intrinsic nature and typically is a function of high-temperature exposure. The rate of this interdiffusion also depends on the purity of the used glass for cladding and worsens with higher OH content.

57 36 In this work, both of these issues were addressed. Manufacturing of a sealed Si-core fiber preform while applying vacuum to its core allowed to extract any air/water vapor that was generated during preform heating. The core-cladding diffusion was reduced by employing laser heating instead of a traditional induction furnace or open flame. This enabled very localized processing with an extremely high energy transfer rate, thus amplifying the processing rate and reducing the heating time of a single point on a preform down to several seconds. This allowed, for example, the formation of a 200 mm long Si-core fiber preform of 6 mm OD and 30 µm core diameter, manufactured in four sequential tapering steps (see Section VI of Paper II). A core size smaller than 17 µm was not attainable at the current preform scale due to the presence of a thermal gradient in the thick glass cladding and the Rayleigh-Plateau instability effect that caused silicon core separation on smaller sections. This effect will be discussed in Section 5.3. Although drawing such a preform without causing core separation was a challenge due to a very narrow draw temperature window, a fiber with a bigger core size of 1.3 µm was successfully pulled. The section of this fiber was further tapered in a power-stabilized CO 2 laser-based system, described in [103]. The 5 cm long section of the tapered fiber had a core size of 310 nm in diameter and showed no silicon core separation or vitrification [104,105]. The core integrity was confirmed by observing the core glow at areas where the silicon core was at high temperature, as shown in Fig This result sets an up-to-date world record of the smallest achievable core produced in Sicore fiber. Fig A Si-core fiber with a 1.3 µm core diameter tapered down to 31 µm OD. The orange glow corresponds to silicon emission. The contrast of the image was altered for better visibility.

58 Interface layer Traditionally, oxygen interdiffusion was mitigated by either including oxygen getters to the core material [95] or implementing an interface layer between the core and the cladding of the fiber preform that served as the oxygen diffusion barrier [94]. The latter quickly became a standard technique to counteract oxygen core contamination. The aforementioned studies have demonstrated the possibility to obtain a Si-core fiber with a core size down to 8 µm [67] using a standard powder-in-tube MCD technique. In later studies, the Si-core size was further reduced down to 4 µm [77]. The typical manufacturing process of Si-core fibers is accompanied by a large stress buildup. This stress is associated with a difference in thermal expansion of silicon core and silica cladding. This effect normally leads to a crack formation or even complete taper disintegration upon handling for greater core-to-cladding ratios. An additional implication of applying an interface layer to the preform tube prior to the fiber draw is a partial relief of this stress [92,94]. This is achieved by reducing the viscosity around the core area, during high-temperature processing [94]. The direct consequence of such viscosity change is an increased instability effect that introduces as fluctuation of the fiber core diameter during the draw. 5.3 Rayleigh-Plateau instability The first derivation of the linear stability of low viscosity liquid was made by Plateau in Two decades later, his work was extended by Lord Rayleigh by introducing viscosity to the equations. Hence, the effect is known nowadays as Rayleigh-Plateau instability; a fundamental limitation that states that a stream of fluid breaks up into smaller packages when its length becomes bigger than its radius. A more relevant effect to the fiber manufacturing case was studied by Tomotika [106], where one cylindrical viscous thread was surrounded by another. In this work, he used an analytical approach to investigate how the instability was defined by the parameters, such as the core fluid diameter, surface tension, and fluid viscosities. The following equation summarizes the general case of this system: ω = σ (1 2ɑµ k2 ɑ 2 )Φ(kɑ, µ ). (5.1) µ

59 38 Here, ω represents the instability growth rate, σ is the surface tension of the molten core material with the unperturbed radius ɑ, k is the perturbation wavenumber (defined as 2π/λ, λ is the perturbation wavelength), and µ and µ are the viscosities of the core and the cladding, correspondingly. A schematic representation of the scaled dependence of the right-hand side of Eq. (5.1) on kɑ is shown in Fig Here, the amplitude of the distribution corresponds to the susceptibility of the system to instability. The closer it is to 1, the greater the instability effect will be observed in the final preform. An example of such an unstable system is shown in Fig The position of the instability amplitude peak is mostly defined by the µ /µ ratio and only affects the observed instability wavelength, i.e., the period between the preform core beads in Fig The dependence in Fig. 5.2 was assumed by combining a partial case of the highly viscous external fluid, when µ /µ = 0, and the general case when µ /µ was close to 1. These two cases were presented and discussed in Ref. [106]. Fig Schematic representation of the instability growth rate ω as a function of kɑ, scaled by the factor of σ/2ɑµ, at a fixed and close to zero viscosity µ /µ ratio. Although the shape of the presented curve and its peak position do not represent the real distribution and are only qualitative in nature, it allows us to draw an important conclusion. According to the two aforementioned cases, the amplitude of the (1 k 2 ɑ 2 )Φ(kɑ, µ ) was varying from 1 for µ /µ = 0, to 0.08 for µ /µ = Considering that the viscosity ratio at a typical Si-core fiber draw temperature ( 1750 C) is in the order of [107,108], the amplitude at this viscosity ratio should remain close to 1 and only mildly change with µ /µ. This means that the instability growth rate ω mainly depends on the viscosity of the cladding µ, core size ɑ, and the surface tension σ of the molten silicon. Therefore, with unchanged core µ

60 39 parameters, the instability growth rate will decrease when the viscosity of the cladding µ increases. This can explain the elevated instability, for example when applying interface coating that causes a substantial decrease in viscosity of the cladding material, while the ratio of µ /µ was still small. Figure 5.3 demonstrates a retapering attempt of a 1.8 mm OD cane with 100 µm Sicore in a 4 mm OD tube with an ID of 2 mm at elevated processing temperature. This Si-core cane was produced by applying a thin interface coating layer of unknown composition. Although at a lower temperature a successful tapering was achieved, the core diameter variations were unavoidable with variations of up to 50 % and the same perturbation wavelength, as shown in Fig To avoid such core variations, it was decided to proceed to fiber preform manufacturing without using any interfacial modifications. For reference, the canes without an interface coating showed no trace of core diameter variations that could have been related to the Rayleigh-Plateau instability at the same or even higher processing temperatures. Fig Core beading is caused by the instability effect due to reduced viscosity of the surrounding core area of the preform cane. A viscosity reduction was performed by applying the interface layer to the preform tube prior to the draw. The instability wavelength in this case is measured to be 10.1 mm. Although in this case the core dimension was very stable during tapering, at very small core sizes the Rayleigh-Plateau instability was still observed. This was previously mentioned in Section 5.1. For smaller cores, the surface tension prevails over the product of viscosity and the core diameter at a given processing temperature, as shown in Eq. (5.1). To overcome this effect, the fiber draw/preform tapering has to be conducted at a lower temperature to increase the value of µ. This brings us to a predefined temperature window for such a process to be possible:

61 40 T gs + ΔT < T < T µ, (5.2) where T µ is the highest temperature at which the cladding viscosity still allows tapering without core beading, T gs is the silica glass softening point ( 1680 C), and ΔT is the thermal gradient, associated with a temperature difference between the glass surface and the core. In the case of CO laser processing, this thermal gradient is typically on the verge of 100 C for a 6 mm preform (see Fig. 4 in Paper III). Therefore, at even smaller core sizes when T µ T gs + ΔT retapering the cane further, preserving its core structure is not possible. It is, however, possible to taper it down, since the thermal gradient for thinner cladding will reduce, making some extra room in the aforementioned temperature window. Thus, the smallest Si-core of 17 µm size in the 6 mm OD preform was experimentally demonstrated. Such a preform would have produced a Si-core fiber with a core size of 330 nm, but the required high tension for the fiber drawing was not achieved due to frequent fiber fractures and a continuous Si-core was not obtained. 5.4 Thermal expansion during initial tapering Tapering dynamics of a hybrid fiber preform, and a Si-core preform in particular, is varying as it crosses different sections of the preform (see Fig. 5.4). These sections are: all silica region [points (a) and (e)], silicon-core region [point (c)], and transitional regions [points (b) and (d)]. Fig (a) Tapering sections of the Si-core fiber preform and (b) tapering process at point (c), showing a silicon melt zone. The emissivity of silicon changes with phase, resulting in darker lines being visible at both ends of the molten region. Adapted from Paper II. The tapering process is initiated at point (a), with the laser stage and the pulling stage moving to the right. The temperature at regions (a) and (c) is relatively constant and defined by the laser power and tapering speed. Here,

62 41 this process is typically performed at temperatures higher than the silica glass softening point of 1630 C measured at the core-cladding interface. At transitional regions (b) and (d) temperature dynamics changes mostly due to a two orders of magnitude higher thermal conductivity of silicon. Such an effect of temperature tapping along the core is demonstrated in Fig. 5.5(a), which shows a numerical simulation of heating a 6 mm preform with a 1 mm silicon core. The initial transition from the all-silica region to the bi-material region is accompanied by a decrease in temperature with a subsequent increase in viscosity of the preform tube. This viscosity change alters the tapering conditions, thus forming a locally enlarged region at the start of the silicon core at point (b). This change in preform diameter can be observed in Fig 5.6. Fig Si-core fiber preform during (a) heating and (b) passive cooling. The simulation is performed at a laser power of 250 W with a beam diameter of 14.6 mm and a preform rotation speed of 100 rpm. Lines emphasize regions of constant temperature. Another characteristic of silicon is a 10 % volume expansion during crystallization from the molten state [109]. This results in a large stress buildup in the Si-core taper during processing and often leads to crack formation. These cracks typically occurring at the material interfaces near points (b) and (d), as shown in Fig In general, the cracks that form in Si-core tapers can be divided into two types: transverse (T) and longitudinal (L) cracks. The T-cracks are typically starting at the core-cladding interface and strive toward the cladding surface. Two kinds of T-cracks were observed with respect to their propagation

63 42 nature. One is characterized by multiple short crack structures and typically have a long propagation distance that often can transform into a longitudinal or L-crack. An example of such a formation can be observed in Fig. 5.6(a). Another kind of a T-crack is a single crack formation of a larger size. An example can be seen in Fig. 5.6(b) and Fig. 5.6(c). When invoked at the right point, these kinds of cracks often remain localized and do not propagate along the taper. The L-crack can only start from a T-crack and is usually propagated in the cladding along the whole length of the preform at some distance from the core-cladding interface. These L-cracks have a crescent shape when observed from the cross-section and have been seen before in the first Sicore fibers ever produced [20]. While these types of cracks form in the cladding, they do not necessarily have a great impact on the resulting loss in the drawn fibers, as they are self-sealed during fiber drawing. However, there is always a risk that such an L-crack will transform into another type, reaching the core-cladding interface and forming a defect. The formation of any crack leads to a local stress release in the preform, which can be seen from the weakened glow in the cross-polarized light associated with the embedded stress in Fig. 5.6(a). The initial enlarged region of the taper holds a large amount of stress, which always spontaneously forms random kinds of cracks. For successful tapering of a long section of the cracks-free Si-core fiber preform, the accumulated stress in this enlargement has to be released in a controlled manner to form a favorable single self-sealing T-crack. This crack has to be formed on a downhill slope of the enlargement to turn the crack perpendicular to the cladding surface. This effect can be observed in Fig. 5.6(b) and the cover picture of the thesis. Such crack configuration at this point is favored by an embedded stress combination of the tensile stress at the interface and shear stress in the cladding (due to relatively cold tapering). To achieve this, a controlled fracture technique is implemented by introducing a thermal perturbation approximately 10 mm after point (b); shown with a pink arrow in Fig. 5.6(b). At this point, all translation stages were stopped, and the laser shutter closed for a duration of 2 s. The preform was then exposed to laser irradiation for a duration of 10 s prior to resuming the tapering process. Such a procedure first shrinks and then expands the melt zone (see Fig. 5.4(b)). The volume change associated with the solidification front movement and the rapid cooling and reheating of silicon was sufficient to induce a thermal perturbation. This perturbation has initiated an

64 43 intentional crack at the right point (see Fig. 5.6(b)), which released the embedded stress and prevented any new cracks from forming. If no enlargement is formed at the beginning of the preform tapering the embedded stress at the slope region is lower and the chance of forming a selfsealing T-crack is significantly reduced. In this case, multiple small T-cracks are usually formed, which, in turn, transform into a longitudinal fracture that propagates with the melt zone during processing. Normal tapering is resumed after this crack is successfully created. Fig Micrograph of the starting tapering point having (a) a detrimental propagating crack and (b) intentional crack formation in cross-polarized light; (c) the endpoint, showing silicon thinning. The white glow indicates the captured stress within the taper. The pink arrow depicts the laser position at which the perturbation technique is performed to form a self-sealing T-crack. Adapted from Paper II.

65 44 At the end of the tapering process, Fig. 5.4 point (d), the silicon core forms a sharply tapered tail, inducing an additional stress point to the glass cladding Fig. 5.6(c). To disperse the stress applied at that point, the glass around the expanding core has to be consolidated. This consolidation is performed by slowly cooling the sharp silicon tip down to 1000 C with a rate of approximately 20 C/s by gradually reducing the laser power. To properly conduct such cooling, the tapering process has to end with a melt zone at the Si-tip formation area (right side of Fig. 5.6(c)). This step proved to be vital for the prevention of catastrophic fracture formation and back-propagation along the ingrained stress lines spanning the entire length of the preform. The crack formation was also an issue when cutting off the Si-core taper from the enlarged area with controllably created fractures into suitable lengths. The cutting was performed using a specially designed laser-based technique. The glass was slowly ablated using a focused CO laser beam while rotating the preform. The laser cutting was terminated when the silicon core was reached. The final separation was performed by cleaving the silicon core after scoring it with a scriber. Although such a cutting technique allowed for tapered cane separation, mechanical end polishing of the edge was still causing fractures in the cladding. This fact hindered any attempts to study the transmission properties of short sections of the tapered canes. Heat conduction makes a significant contribution to stress allocation in glass cladding. As the preform cools down, a warmer silicon core transfers its heat to a surrounding cladding, which at that time is colder due to convection. As a result, the viscosity of the glass cladding halfway to the core is the highest. When the glass cladding with such a viscosity distribution faces the sudden core volume expansion, the majority of the stress will be embedded in between the core and the area with the higher viscosity. This effect can be seen from the temperature cooling simulation, which is demonstrated in Fig. 5.5(b). To a certain extent, the radial stress distribution during tapering can also be managed with processing temperature. Thus, processing Si-core fiber preforms at a temperature close to the silica softening point have led to broader stress distribution, as the viscosity of the glass cladding is higher at the point of solidification (the left edge of the melt zone in Fig. 5.4(b)). Compared to the first tapering step that used a 1 mm diameter Si-core, subsequently tapered preforms were found to be less susceptible to crack formation. This was attributed to the lower core-to-cladding volume ratio,

66 45 reducing the overall effect on the cladding from silicon expansion during solidification. Smaller core cross-sections also resulted in a lower temperature change during the tapering process at the transitional regions at points (b) and (d) (see Fig. 5.4(a)). The amount of tensile strain in the silicon core stays relatively the same regardless of the core-to-cladding ratio of the preform or the tapering speed. This can be seen from a Raman analysis performed in Section VI of Paper II. 5.5 Si-core fiber preform Preform preparation and fiber draw Many different configurations of Si-core fiber preforms have been manufactured during this project. Most of them were fabricated to aim for a 5 µm core size in a drawn fiber. Such a preform is fabricated in two subsequent steps, as described earlier in Section A 58 mm long, 6 mm OD, 1.1 mm ID tube was used as a preform tube. A fiber-based spring was used to hold the 20 mm long high-purity silicon rod and silica inserts in place with a 2 mm allowance (see Fig 5.7(a)). This allowed to compensate for the silicon volume redistribution due to the difference in silicon core size and tube inner diameter. This preform assembly was tapered down to a 1.8 ± 0.1 mm OD at a temperature ranging from 1600 to 1800 C to fit into the sleeving tube, showed in Fig. 5.7(b). The length of this taper is typically 220 mm long, with a 190 mm crack-free Si-core section. Prior to removing this taper from the system, it is cut out from the regions with cracks and also in the middle to reduce the length and prepare the cane for further retapering. To engineer the desired 5 µm core diameter, one of these sections is inserted into a sleeving silica tube of 8 mm OD and 2 mm ID preserving the tapering direction. At this stage, the core insert pretension allowance is set to 0.3 mm to avoid additional stress application to the tapered cane. This preform tube configuration is then tapered into a final preform with 6 mm OD, as shown in Fig 5.7(c). The usable length of the tapered preforms, defined by the length of a silicon insert was mm. Air extraction was permitted by applying a low pressure of mbar to the preform at each processing step. The preform rotation speed was set to 100 rpm with the laser

67 46 beam diameter on the preform tube equal to 14.6 mm. All processing was performed at a humidity in the range of ppm V. Fig (a) Si-core preform filled with 1 mm Si rod insert prior tapering and (b) after tapering down to 1.8 mm; (c) final 6 mm preform with 250 µm core made in twostage tapering process. This preform is drawn in a 125 µm fiber with 5 µm core. The white glow indicates the captured stress under cross-polarized illumination. The core appears larger due to the lensing effect. These preforms were drawn into several hundreds of meter long Si-core fibers using an experimental draw tower, which was described earlier in Section and Paper III. The fibers were drawn at speeds ranging from 1 to 10 m/min with a tension of 40 cn. These fibers have been characterized using a cut-back measurement technique, gradually reducing the length from a starting 150 cm long section to 80 cm in eleven cuts. The loss value was calculated using exponential fitting of the measured outcoupled power with respect to the fiber length. The fiber was cleaved using a ruby blade with a subsequent edge-polishing to maximize outcoupling efficiency and create equal conditions for each cut-back measurement. The analysis was

68 47 performed using a Thorlabs S122C power meter with a sensitivity window of nm, showing an average loss of 0.2 db/cm [110], with a 0.1 db/cm loss at 1550 nm [104], for a fiber pulled at 10 m/min drawing speed. For these measurements, a white-light source Fianium WhiteLase SC480 was used with a spectrum coverage from 500 to 2300 nm. Although these values were still a factor of 5000 higher, compared to bulk silicon of the same purity [24], they represent the lowest loss in Si-core fibers ever measured to date. Preforms of different core-to-cladding ratios were also fabricated and the fibers with Si-core diameters ranging from 1.3 to 30 µm were drawn. Figure 5.8 shows two examples of the drawn fibers. The configurations of these preforms are discussed in detail in Paper II and the drawn fibers in Paper III. Attenuation evaluation of these fibers was not performed due to timeconsuming measurements. Fig Polished end of Si-core fibers with (a) 20 µm and (b) 1.3 µm core sizes. Adapted from Paper III. In summary, during this project 52 Si-core tapered canes were fabricated with various core sizes, 36 of which were aimed to produce fibers with a core size of 5 µm. Half of the latter were made at different processing speeds, ranging from 0.18 and up to 0.6 mm/s at a compatible laser power of 270 ± 10 W. For this particular cane configuration, the most reliable results in terms of a success rate were achieved at a speed of 0.3 mm/s. At that point, the perturbation technique to induce the self-sealing crack at the right location in the taper was not yet discovered and the yield of the crack-free tapers was around 50 %. All tapered canes showed a similar radial stress

69 48 distribution. The number of final preforms drawn was 24, part of which was used to optimize drawing parameters Preform geometry When molten, silicon (or any crystalline material) takes the shape of the inner cavity of a silica crucible. Therefore, the used glass tubes must be straight and of a consistent wall thickness to preserve the same core-tocladding ratio along the preform length. To evaluate the consistency of the manufactured canes, 20 sections from 16 test canes were mechanically cut off in random places along the length and measured at the edge to avoid additional error due to the lensing effect. Such cutting resulted in glass fracturing around the Si-core, however, this did not affect the measurement. The measured tapers showed some variations in their diameters. Measured OD showed deviations of up to 4.2 % with a mean of 1767 µm, while the core measured 293 µm and remained within 6.5 % along the entire length of the preform. The OD variations were measured to have a period of approximately 62 mm along the cane length. This effect was associated with the mechanical limitations of the driving system of the preform manufacturing setup, as such periodicity matched the effective circumference of the driving gears, used for both scanning and pulling stages. In a final 6 mm OD preform, such slow OD variation was not noticeable. On the other hand, a faster OD change was more obvious here, with a distance of exactly 5 mm between the minimum and maximum OD. This corresponded to a distance between the teeth of the straight-tooth gear. Such faster OD variation was within 1.9 %. The ellipticity of the 6 mm preform was measured to be 0.5 % and is believed to be related to the lathe chucks centricity and overall glass tubes straightness. These variations in diameter size had minimal effect on the resulting fiber diameter thanks to a working active feedback system during the draw, as described previously in Section The enlargements of the tapered cane OD mostly correlated with the corresponding change of the core diameter. Any discrepancies were determined by the tube manufacturing quality. For example, standard Goodfellow tubes used during the project were produced with a tolerance of ±0.1 mm for the OD and a 10 % allowance for the wall thickness. For this reason, the 6.1 mm OD, 1.1 mm ID tubes were requested to have a tolerance

70 49 of ±0.05 mm for both diameters, which correlates with the indicated measurement results. Periodic core variations with a shorter period, that could have been associated as characteristic of Rayleigh Plateau instability, were not observed, contrary to the case shown in Fig Purity of the used silica glass Silica glass tubes of two types were used to fabricate specialty fiber preforms during this project: fused quartz, Goodfellow (GF), and fused silica, F300 Heraeus (F300). Almost all preforms discussed in the frame of this thesis used GF fused quartz tubes for preform manufacturing due to their availability in the required sizes. The F300 tubes, however, were used as well. These preforms are discussed for example in Paper II, where a fused silica tube with 4.1 mm OD and 1.1 mm ID was used as a preform tube with following re-tapering within a 10 mm OD GF tube. In Paper III both F300 tubes listed in Table 5.1 were stacked and drawn to achieve a 20 µm-core fiber (see Fig 5(a) in Paper III). Each tube from the list was typically used to perform a specific function. As such, for example, a fused quartz glass tube of 4 mm OD and 2 mm ID was primarily used as the part of the preform responsible for gas evacuating and creating a convenient connection interface of the preform to a gas channel. Table 5.1. Used glass tubes Glass F300 F300 GF GF GF GF ID OD Function Sleeving Cladding Cladding Exhaust Sleeving Sleeving To evaluate the purity of the used glass, transmission spectra of GF and F300 tubes were measured (see Fig. 5.9). The measurements were performed using two spectrophotometers to cover the transmission in the nm spectral range. The Varian Cary 50 was used to capture the signal in the UV and PerkinElmer Spectrum 400 FTIR for longer wavelengths.

71 50 Fig Transmission spectra of silica glass used in preform preparation. The fused quartz tube samples (GF) showed some traces of OH infusion, which increased the absorption in the nm spectral range, associated with the fundamental stretching of four OH vibrational modes [111]. These are usually accompanied by the neighboring absorption components in the nm range, which are the overtones of the Si- OH vibrations [112]. The first fundamental OH overtone in high-oh glasses can also usually be observed near nm [113]. In our case, however, this overtone is only barely distinguishable in the GF sample. The Heraeus fused silica tubes (F300) seem to have lower OH content with no visible absorption band near 2720 nm. In general, both glasses showed comparable transmittance within nm and nearly identical behavior in the nm spectral range. The F300 sample seems to outperform the GF in nm and nm wavelength regions. 5.6 Other hybrid fiber preforms In addition to silicon core preforms, taper attempts of other types of hybrid fiber preforms were also performed. The initial fiber drawing test was performed from a suspended sapphire fiber within a 6 mm OD silica tube and was covered in Paper III. The experiment showed that the temperature during processing has exceeded the required sapphire melting temperature of 2050 C, which resulted in almost complete diffusion of molten alumina into the glass cladding. A total of 14 optimization attempts were made to taper

72 51 ruby and sapphire rods of 1 mm in diameter within a 6 mm silica tube. One such taper that resulted in a relatively stable OD is shown in Fig. 5.10(a). It was achieved using 310 W of laser power with a beam diameter of 16.4 mm, 0.2 mm/s processing speed, and a rotation rate of 100 rpm. At these parameters, the core-cladding interface showed a temperature of C, measured by a pyrometer. At this temperature, the viscosity of the silica cladding was too low to support tapering under usual vacuum levels of mbar. To mitigate this, the vacuum system was turned off during tapering once sapphire started to melt. Tapering attempts at a higher laser power around 325 W showed the same temperature at the interface; however, the tapering itself was unsuccessful due to the inability of silica to effectively conduct heat to the core to keep it molten and act as a crucible at those temperatures. Therefore, the conclusion regarding the presence of a narrow temperature window has also been confirmed while processing sapphire within a glass tube on a preform scale. Fig (a) Sapphire-derived cane and (b) a relatively stable Ge-core taper in the middle and (c) its tail section in cross-polarized light. The red glow is a reflection from a preform holder. It can be seen from Fig. 5.5(a) that alumina did not crystallize back into sapphire, even though the tapering process has started from the center of the core region to providing a crystalline seed conditions to the following resolidification region. In a common Laser-Heated Pedestal-Growth technique,

73 52 which is usually conducted without a crucible, the standard growth rate for sapphire is approximately 0.03 mm/s [114]. Such a tapering speed was, unfortunately, not achievable in the current setup due to the chosen rack and pinion-based motion mechanism. The character of cracks in the cladding around solidified alumina and longitudinal cracks in the core suggested that alumina suffered some diffusion into the glass cladding and has shrunken upon solidification, pulling glass inward. Another tapering test was performed using germanium as a core material. A germanium rod of 2 mm in diameter was inserted in a silica tube of 8 mm OD and 2 mm ID and tapered down to 1.8 mm. Due to a huge coreto-cladding ratio of the preform and sub-optimal glass type used (a borosilicate glass is a better fit for Ge-core [115]), the taper suffered from instabilities and, as a result, massive diameter variations between 1.2 and 1.8 mm. A relatively straight section of the taper is demonstrated in Fig. 5.10(b). It is worth to note that even though germanium also experiences significant 5.5 % volume expansion upon solidification [116,117], there was no indication of stress visible in the cladding. One possible explanation of this effect is that germanium crystallizes at a lower temperature (compared to silicon), at which the silica cladding has a much higher viscosity and does not give, which in turn inflicts visible cracks in the Ge-core instead and releases some accumulated stress. These cracks can be seen in Fig. 5.10(b) and Fig. 5.10(c). These preforms were never drawn into fibers due to observed cracks in the core and expected poor optical properties.

74 53 Part II: Specialty optical fiber preform fabrication using glass additive manufacturing 6 Glass additive manufacturing techniques 6.1 Direct glass laser deposition (DGLD) A Nobula3D glass 3D printer Nobula TM Zero was used to print a multicore insert for a fiber preform. This glass 3D printer was based on the direct glass laser deposition (DGLD TM ) technology developed at KTH Royal Institute of Technology as a part of a research project. The CO 2 laser-based printing technique was similar to a previously reported method of glass additive manufacturing, developed at Missouri University of Science by Kinzel [118,119]. The key difference of the DGLD TM method is the use of four radially placed laser beams. These beams were formed by splitting an original laser beam and redirecting it onto a glass filament using gold-coated mirrors. The filament is fed to the formed hot zone perpendicular to the build plate that carries a mm 2 fused quartz substrate of a sub-millimeter thickness (see Fig. 6.1(a)). This build plate is translated by a three-dimensional stage so that the object is printed depositing the glass filament layer-by-layer, as shown in Fig. 6.1(b). The CAD model of the printed object was converted to a machine code using commercial slicing software. This system has been previously used to manufacture silica microlens arrays [120,121]. A detailed description of the system, the printing process, and the glass bonding mechanism are outside of the scope of this thesis and will be discussed in future studies.

75 54 Fig (a) Schematic representation of the direct glass laser deposition method and (b) an example of layer-by-layer glass deposition structure. The pink arrow represents the glass filament feeding direction. 6.2 Laser powder deposition (LPD) The laser powder deposition (LPD) technique was used to print silica rods of custom composition for specialty optical fiber manufacturing [122,123]. The setup for fumed silica particle deposition was based on using a CO 2 laser (ULR OEM 50 W, Universal Laser Systems), operating at a wavelength of 10.6 µm, with M 2 = 1.2. The laser emission was guided perpendicular to the build platform by a few silver-coated mirrors and shaped by an AR-coated ZnSe lens (40 mm focal length). The two-dimensional positioning of the printing process was performed by a translation stage with a mounted glass substrate, see Fig The print head is composed of three coaxial powder nozzles symmetrically positioned at 45 degrees relative to the incident laser beam, as shown in Fig The layer buildup was performed by vertical positioning of the print head together with the focusing lens, to keep the laser focus stationary relative to the printed layer. The coaxial nozzles had inner diameters of 1.04 mm and 3 mm. Fumed silica powder was fed through the inner tube while shaping dry air was purged through to the outer tube. The airflow rate was regulated separately using flow regulators. The glass powder was transported to the powder nozzle using a powder feeder (MARK XV, Powder Feed Dynamics Inc.) through grounded, antistatic polyurethane tubing to minimize electrostatic charging of silica particles. The powder

76 55 feeder was equipped with a heating blanket and kept at a temperature close to 100 C to keep the powders dry. The custom glass composition of the printed preform inserts was met by mixing silica particles with dopants in the predefined proportion. The low dopant concentration showed very little influence on the deposition process. Fig Additive manufacturing of the glass rod of custom composition for fiber fabrication made in laser powder deposition system. The red arrow represents CO2 laser irradiation. Preforms fabricated using both of these glass AM techniques are discussed in the next chapter.

77 56

78 57 7 Preform fabrication for all-glass specialty optical fiber 7.1 Multicore optical fiber preform Multicore fiber preforms were designed and manufactured by combining DGLD TM and stack-and-draw approaches, described in Sections 6.1 and 3.3.1, respectively. First, the preform core insert was 3D printed longitudinally in a hexagonal packing arrangement using a multicomponent glass fiber (see Fig. 7.1(a)). This insert was then placed in a silica preform tube assembly between two silica rods, as shown in Fig. 7.1(b). The preform assembly was sealed from one end and connected to a vacuum system, as described previously in Section A fiber-based spring was used to fix the inserts in place. Such a preform assembly was then tapered or collapsed into the final preform, as shown in Fig. 7.1(c) and Fig. 7.1(d). Two types of multicore fiber preforms were produced as a proof-ofconcept using two different glass printing filaments. For the first type, a Gedoped core silica clad optical fiber was used with a core size of 13.1 µm and a diameter of 200 µm. The hexagon preform insert comprised 68 extruded lines and was approximately 1.8 mm in diameter and 15.5 mm in length. A preform tube with an OD of 8 mm and ID of 2 mm was used to accommodate the insert (shown in Fig. 7.1(b)). This preform was tapered down to 6 mm using the following parameters. The laser power was set to 275 W with a beam diameter of approximately 14.6 mm. The preform was rotating at 100 rpm and a vacuum of 12 mbar was applied. The laser processing speed during tapering was set to 0.35 mm/s, while the pulling stage performed tapering at 0.15 mm/s to ensure the final preform OD of 6 mm. The humidity during processing was measured to be around 60 ppm V. In general, this type of all-silica glass preforms showed little dependence on the manufacturing parameters, such as tapering speed or laser power, as long

79 58 as the temperature was above the softening point of silica. The resulting preform had a good core-cladding interface fusion with a very few gas bubbles trapped inside (see Fig. 7.1(c)). The second multicore fiber preform insert was additively manufactured using a silica-core fluorine-doped silica-cladding optical fiber (Thorlabs FG200LEP). This time, a hexagonal structure of 16 lines was printed using the DGLD TM method. The fiber used as a filament for 3D printing had a 200 µm core with a 10 µm thick fluorine-doped cladding. Although it was possible to print with this type of filament, the drastically different glass transition temperature of these glasses required operating in a very narrow temperature window. For this reason, the cladding material evaporated quicker and produced bubbles when the printing temperature was outside of this allowable range. The 20 mm long 3D printed insert was 1.2 mm in diameter. For this insert, a custom-made preform tube was fabricated to maintain a bigger core-to-cladding ratio. A silica tube with 4 mm OD and 2 mm ID was tapered down to 2.4 mm OD and 1.3 mm ID. As before, the insert was cleaned and placed in the preform tube. The core and the cladding were fused by scanning the laser beam of 14.6 mm in diameter with 0.1 mm/s speed, while the preform was rotating at 150 rpm. In total, three attempts were made to optimize the fusion temperature. The best result was achieved with 141 W of laser power with a humidity of 40 ppm V, as shown in Fig. 7.1(d). The observed gas bubbles mainly formed during the 3D printing process and were trapped between the printed layers. Table 7.1 summarizes the most relevant parameters of the produced multicore preforms. Table 7.1. Parameters of the multicore preforms. Filament composition Coreclad Number of cores Single core [mm] Core size [mm] Preform diameter [mm] Proc. speed [mm/s] Laser power [W] Ge-doped core, silica clad Silica core, fluorinedoped clad

80 Fig (a) Example of a multicore preform insert made using the DGLD TM method. The insert comprises 19 Ge-doped fibers, deposited layer-by-layer in a hexagonal packing arrangement; (b) 68 Ge-doped cores insert prior to tapering; (c) 68 Ge-doped cores fiber preform, 6 mm in diameter; (d) 16 cores multimaterial glass fiber preform, 2.4 mm in diameter; drawn fibers from (e) Ge-doped preform and (f) multimaterial glass preform. 59

81 60 Both preforms were drawn into fibers with a diameter of 125 µm at 1 2 m/min speed. The preform with 68 Ge-doped cores was drawn into a 60 m long fiber, while the tiny multi-glass preform has only produced approximately 6.7 m. These fibers are shown in Fig. 7.1(e) and Fig. 7.1(f). These fibers were shown to guide light, although thorough measurements were not conducted due to complex light incoupling conditions and the crosstalk between the cores. 7.2 Doped-silica optical fiber preforms The preform core rods with different dopant concentrations were manufactured using the LPD technique (see Section 6.2). Five rods were printed using predefined powder compositions, as specified in Table 7.2. The rods were printed by vertical translation of the print head with simultaneous powder feeding at a rate of 0.5 g/min. With such a material deposition rate a printing speed of 1.1 mm 3 /s was reached. The as-printed rods were in a form of a green body with some silica particles remaining unsintered at the surface. Some rods were sintered using a CO 2 laser prior to tapering into a fiber preform. However, further experiments showed no improvement in the final preform quality when using processed rods, compared to unprocessed (non-transparent). Table 7.2. Rods composition and length. Name Ti-1 Ti-5 Ti-15 Al-2 Er-Al Composition TiO2- SiO2 TiO2- SiO2 TiO2- SiO2 Al2O3- SiO2 Er2O3-Al2O3- SiO2 Ratio [wt%] 1 : 99 5 : : : : 3.4 : 96.3 Rod length [mm] The preform configurations were chosen to target the core size to be approximately 5 µm in the final 125 µm fiber. To achieve this, a two-step tapering process was performed, following steps 1 4 in Fig A silica tube of 6 mm OD and 1 mm ID used as a preform tube was tapered into a 2 mm OD cane. This cane was placed in a new preform tube, based on a silica tube of 8 mm OD and 2 mm ID. Finally, this tube was tapered down to 6 mm in a

82 61 final preform with a 250 µm core. All tapers were conducted applying mbar vacuum to the preform cores to mitigate air extraction during glass fusion. At high dopant concentration, the core temperature was measurable with the Fluke pyrometer. Thus, the Ti-5 cane s core was tapered at 1800 C while both Ti-15 taper s cores experienced C. Other relevant tapering parameters are specified in Table 7.3. The tapering process is explicitly described in Section IV of Paper I. Fig (a) The zoomed-in core area of the tapered Er-Al rod and (b) Al-2 rod. The samples were tapered once down to 2 mm OD with a core size of 330 µm. (c) The final preform after a two-step tapering process made of the Al-15 sample rod under the top and bottom illumination ( 250 µm core). Fig. 7.2 shows the resulting tapered canes and one example of a final preform. A closer inspection of the manufactured preforms revealed the presence of small gas bubbles in all samples but Ti-15, where the bubbles were not observed probably due to the opaqueness of the core structure. The size of these bubbles was ranging from 5 to 35 µm in the tapered cane, which corresponds to a possible defect size of nm in a standard-sized 125 µm fiber. Such defects are expected to have a high impact on the scattering losses of the produced fibers. It is believed that air was incorporated within the 3D printed core rods mostly during the production stage. Each preform from Table 7.3 was drawn into fibers of several hundreds of meters in length. The transmission and power attenuation of these fibers was measured and discussed in Paper IV. No sign of gas bubbles was observed in the optical microscope on a fiber scale within the core upon cleaving. However, fiber made of the Ti-1 preform showed the presence of periodic air