Dissertation/scifi.tex

\chapter{The Scintillating Fibre Tracker}
\label{ch:scifi}

Up to the end of the LHC Run 2 in 2018, two different detector technologies have achieved an excellent tracking performance within the \lhcbexperiment{}~\autocite{ot-performance,ot-performance-improved,it-performance}.
The Scintillating Fibre (SciFi) Tracker is going to replace the tracking stations T1-T3 downstream of the magnet implementing a single technology based on scintillating fibres.

The layout and working principle of the detector is outlined in this chapter, along with the motivation that led to the upgrade of the tracking stations and its requirements.
If not stated otherwise, the provided information is taken from the Technical Design Report (TDR) of the upgraded tracking system~\autocite{scifi-tdr}.


\section{Motivation for the Tracker Upgrade}
\label{sec:upgrade-motivation}
The Inner (IT) and Outer Tracker (OT), which have formed the downstream tracking stations of the initial \lhcbexperiment{}, were designed to provide a high precision estimation of the momentum of charged particles.
This was achieved by operating at high hit efficiency as well as high resolution in the bending plane of the magnetic field.

The OT was a gaseous detector covering an area of approximately \qtyproduct{6 x 5}{\meter} per layer and consisted of \qty{2.4}{\meter} long straw tubes with an inner diameter of \qty{4.9}{\milli\meter}~\autocite{ot-tdr}.
The tubes were filled with a mixture or argon, carbon dioxide and oxygen allowing for fast drift times below \qty{50}{\nano\second}.
The detector modules were composed of two staggered layers of 64 straw tubes each and were arranged in three tracking stations (T1-T3).
Each station consisted of four module layers following the so-called \textit{x-u-v-x} geometry:
Modules in the \textit{x} layers were oriented vertically, while modules in the \textit{u} and \textit{v} layers were tilted by \ang[retain-explicit-plus]{+5} and \ang{-5}, respectively.
Consequently, a total of 24 straw tube layers were arranged along the direction of the beam pipe.
The average hit efficiency was estimated to be \qty{99.2}{\percent} and the position resolution was measured to be close to the design value of \qty{200}{\micro\meter}~\autocite{ot-performance}.
Utilising novel real-time calibration methods as described in Ref.~\autocite{ot-performance-improved}, the resolution could be further improved to \qty{171}{\micro\meter} during LHC Run 2.

A plus-shaped area of approximately \qtyproduct{1.2 x 0.4}{\meter} in the centre of each layer was not covered by the OT and instead housed the IT.
The IT was a silicon microstrip detector using \qty{320}{\micro\meter} and \qty{410}{\micro\meter} thick single-sided $p^+$-on-$n$ sensors with a strip pitch of about \qty{200}{\micro\meter}~\autocite{it-tdr}.
Individual silicon sensors were \qty{7.8}{\centi\meter} in width and \qty{11}{\centi\meter} in height and were contained within a total of 12 detector boxes.
Each tracking station was composed of four boxes that were located above, below, and on both sides of the beam pipe, thus forming the aforementioned plus-shape.
Each detector box was made up of four layers following the same \textit{x-u-v-x} geometry as described above for the OT.
Despite only covering about \qty{1.3}{\percent} of the geometrical acceptance, about \qty{20}{\percent} of all charged particles that were produced close to the interaction point were passing through the active area of the IT.
The hit efficiency was estimated to be \qty{99.7}{\percent} and the resolution was measured to be \qty{58}{\micro\meter}~\autocite{it-performance}.

The initial tracking system was designed to limit the maximum occupancy in the OT to \qty{10}{\percent} at nominal instantaneous luminosities ${\luminosity = \qty{2e32}{\per\square\centi\meter\per\second}}$.
It has been shown that it is possible to go beyond that design limit and efficiently find tracks at occupancies up to \qty{25}{\percent} in the hottest regions of the OT~\autocite{upgrade-loi}.
The necessary improvements in the track reconstruction algorithms thereby allowed data collection at instantaneous luminosities ${\luminosity = \qty{4e32}{\per\square\centi\meter\per\second}}$.
However, going beyond ${\luminosity = \qty{e33}{\per\square\centi\meter\per\second}}$ has been found to require detectors of higher granularity.
Different scenarios including an increased area of the IT along with shorter OT modules in the central part were considered.
In the end, the decision towards a scintillating fibre tracker as a cost-effective, light and uniform detector covering the area of \qty{30}{\square\meter} per layer has been made.


\section{Performance Requirements}
\label{sec:performance-requirements}
The main objective of the tracking detectors within the \lhcbexperiment{} is the accurate determination of the momentum of charged particles.
This information can be used to precisely measure the mass and lifetime of decayed particles.
In order to achieve that, high hit efficiencies, as well as a good position resolution are necessary.
In addition, the material budget has to be kept low to minimise multiple scattering, which is the main limiting factor for the momentum resolution of low momentum particles.
Taking these considerations into account, the following requirements for the \scifitracker{} were formulated in the TDR~\autocite{scifi-tdr}:
\begin{itemize}
  \item Reaching high hit efficiencies at the level of \qty{99}{\percent}, while keeping the rate of reconstructed noise clusters low ($<\qty{10}{\percent}$ of the recorded hits) at any location.
  \item The single hit position resolution must be \qty{100}{\micro\meter} or better in the bending plane of the magnet.
  \item Minimise the effect of multiple scattering by keeping the material budget low.
        Achieving radiation lengths ${X/X_0 < \qty{1}{\percent}}$ per detection layer is sufficient to not be the limiting factor compared to material upstream of the magnet.
  \item The readout electronics needs to operate and output data at the full LHC clock speed of \qty{40}{\mega\hertz}.
        Inefficiencies due to dead time should be minimised by short recovery times.
  \item The aforementioned requirements should be met over the full lifetime of the upgraded LHCb detector up to an integrated luminosity of \qty{50}{\per\femto\barn}.
\end{itemize}


\section{Detector Design and Operating Principle}
\label{sec:scifi-design}
The \scifitracker{} fully replaces the OT and IT between the dipole magnet and RICH2 in the \lhcbexperiment{}.
It is designed to follow the identical layout as its predecessors consisting of three stations T1-T3 with four detection layers each.
The layers are arranged in the same \textit{x-u-v-x} geometry with the two inner layers being tilted by \ang[retain-explicit-plus]{+5} and \ang{-5} from the vertical axis.
Each layer covers an area of approximately \qtyproduct{6 x 5}{\meter} and consists of a minimum of 10 fibre modules.
The tracking station T3, which is furthest away from the interaction point, is equipped with two additional modules per layer to not limit the geometrical acceptance of the experiment.
\Vref{fig:T3} shows the schematic layout of the tracking station T3.

\begin{figure}
  \centering
  \includegraphics[width=0.9\textwidth]{figures/scifi/T3_with_dummy}
  \caption{Schematic side (left) and front (right) view of the SciFi tracking station T3 with a \qty{1.86}{\meter} tall human dummy for scale. Image adapted from Ref.~\autocite{scifi-conference-material}.}
  \label{fig:T3}
\end{figure}

Each detector module is composed of a total of eight fibre mats that are read out by multichannel arrays of silicon photomultipliers (SiPMs) located at the top and bottom of the detector.
The fibre mats itself consist of six staggered scintillating fibre layers with \qty{250}{\micro\meter} in diameter.

The operating principle is illustrated in \vref{fig:working_principle}.
Charged particles that traverse the layers of scintillating fibres dissipate energy as ionisation and excitation.
Subsequently, the absorbed energy is transformed into luminescence emission in the scintillation medium~\autocite{scintillation-process}.
Photons that are emitted along the fibres are trapped inside by total reflections at the fibre boundaries.
They follow the course of the fibres until they reach the SiPM channels at the fibre ends.

The amplitude of the resulting electrical signal is proportional to the number of pixels that are hit by photons within each channel.
The average light yield, i.e. the number of pixels hit, per traversing particle is in the order of 20 depending on the incident angle and distance from the SiPM array.
However, irradiating the fibres with expected doses after \qty{50}{\per\femto\barn} results in a loss of light yield by about \qty{40}{\percent}.
The position of the corresponding hit can be defined as the location of involved SiPM channels weighted by their signal amplitudes.

\begin{figure}
  \centering
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/Working_Principle}
    \caption{Operating principle of the LHCb Scintillating Fibre Tracker. Image adapted from Ref.~\autocite{scifi-tdr}.}
    \label{fig:working_principle}
  \end{minipage}
  \hfill
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/Captured_Photons}
    \caption{Longitudinal (top) and cross (bottom) section of a multi-cladding scintillating fibre with exemplary trajectories of trapped photons. Image taken from Ref.~\autocite{gavardi-diss}.}
    \label{fig:captured_photons}
  \end{minipage}
\end{figure}

\section{Scintillating Fibres}
\label{sec:scintillating_fibres}
The \scifitracker{} employs scintillating fibres with a round cross section of \qty{250}{\micro\meter} in diameter produced by Kuraray\footnote{Kuraray Co., Ltd., Tokyo, Japan}.
Fibres of type SCSF-78MJ were chosen for achieving a fast decay time of \qty{2.4}{\nano\second} and a large attenuation length of about \qty{3.5}{\meter}~\autocite{fibre-qa}.
A high light yield is achieved by utilising a multi-cladding structure with descending refractive indices towards the surface of the fibre, while also being mechanically more robust.
The inner cladding is made of Polymethylmethacrylate (PMMA, $n = \num{1.49}$) and surrounds the Polystyrene (PS, $n = \num{1.59}$) core, while Fluorinated polymer (FP, $n = \num{1.42}$) is used as the outer cladding.
Each cladding is about \qty{4}{\micro\meter} in thickness~\autocite{kuraray}.

Following Snell's law about the angle dependence of light passing the boundary of two isotropic media, there exists a critical angle of incidence
\begin{equation}
  \theta_\text{crit} = \arcsin{\frac{n_2}{n_1}}
\end{equation}
above which total internal reflection at the boundary occurs.
For the transition between the inner ($n_1 = 1.49$) and outer cladding ($n_2 = 1.42$) the critical angle can be calculated as $\theta_{\text{crit, PMMA}\rightarrow\text{FP}} = \ang{72.4}$.
This translates to a maximum opening angle of $2\times\ang{26.7}$ at the centre of the fibre as illustrated in \vref{fig:captured_photons}.
Assuming an isotropic photon emission, the trapping efficiency is thereby given as \qty{5.3}{\percent}.
Following similar considerations, having a single-cladding fibre with a PS core and PMMA cladding only results in a \qty{3.1}{\percent} trapping efficiency.

Pure Polystyrene has a relatively poor quantum efficiency and long relaxation time.
Therefore, a small amount (about \qty{1}{\percent} in weight) of p-Terphenyl is added to the PS core.
This organic fluorescent dye rapidly (below \qty{1}{ns}) absorbs the excitation energy of the PS base via a non-radiative dipole-dipole interaction known as Förster Transfer~\autocite{foerster-transfer}.
Subsequently, the absorbed energy is released by the emission of a photon.
The fluorescent dye has been chosen for its high quantum efficiency $>\qty{95}{\percent}$ and fast decay times below a few ns~\autocite{scifi-tdr}.

A secondary dye is added to the PS core in a very low concentration (below \qty{1}{\permille} in weight).
It acts as a wavelength shifter by absorbing the emission of the primary dye and re-emitting photons at a longer wavelength at which re-absorption in the fibre occurs less frequently~\autocite{scifi-reabsorption}.
Tetraphenyl butadiene (TPB) is used as the secondary dye in the scintillating fibres of type SCSF-78MJ and enables the required large attenuation length.
\Vref{fig:emission-spectrum} shows the resulting photon emission spectrum of the scintillating fibres, which peaks between \qty{450}{\nano\meter} and \qty{500}{\nano\meter} depending on the distance to the point of excitation.

\begin{figure}
  \centering
  \includegraphics[width=0.8\textwidth]{figures/scifi/Scintillation_Spectrum}
  \caption{
    Emission spectrum of a scintillating fibre of type SCSF-78MJ at different distances to the point of excitation.
    For short distances, the maximum is located at around \qty{450}{\nano\meter} and shifts towards larger wavelengths with increasing distance due to self-absorption of the inserted wavelength shifter.
    Image taken from Ref.~\autocite{janine-diss}.
  }
  \label{fig:emission-spectrum}
\end{figure}

More than \qty{10000}{\kilo\meter} of fibres are used for the \scifitracker{} and were delivered to CERN in 48 individual shipments between 2016 and 2018.
Every batch of fibres was subject to a detailed test procedure~\autocite{fibre-qa}:
Samples were taken to evaluate the attenuation length before and after being exposed to X-ray irradiation in order to reveal impurities in the used materials.
The ionisation light yield was measured using energy filtered \qty{1}{MeV} electrons from a Sr-90 radioactive source.
In addition to that, for every \unit{\kilo\meter} of fibre, the quality of the cladding and uniformity of the diameter was verified.
Fibre bumps up to \qty{500}{\micro\meter} in diameter were detected and removed by a hot shrinking procedure in a fully automatic manner~\autocite{hot-shrinking}, while larger bumps were cut out manually.
The conducted quality assurance measures ensured that only high quality fibres were assembled into the fibre modules as described in the following.


\section{Fibre Modules}
\label{sec:fibre-modules}
The fibre modules constitute the active part of the \scifitracker{}.
In total, 128 of these modules are used and are arranged in 12 layers and 3 stations as described in \cref{sec:scifi-design}.
Each fibre module is \qty{4.85}{\meter} long and \qty{0.52}{\meter} in width.
\Cref{fig:exploded-fibre-module} shows an exploded view of one fibre module with its key constituents.
The core of one module consists of eight fibre mats that are arranged in two rows with four mats lying side by side in each row.

\begin{figure}
  \centering
  \includegraphics[width=0.8\textwidth]{figures/scifi/Fibre_Module}
  \caption{
    Exploded view of one fibre module of the \scifitracker{}.
    Image adapted from Ref.~\autocite{fibre-modules-design-review}.
  }
  \label{fig:exploded-fibre-module}
\end{figure}

The width of each fibre mat is about \qty{0.13}{\meter} and was chosen as a compromise between covering a large area and still being easy to handle.
Following similar considerations, the length was chosen such that two mats are required to cover the full module length of \qty{4.85}{\meter}.
Even though this design introduces a small gap of about \qty{2}{\milli\meter} at the centre of the module between the two rows of fibre mats, it bears the additional advantage that the light yield of a single particle crossing is concentrated into fewer SiPM channels on only one side of the module.
Photons that are emitted towards the centre of the module are reflected back to the outside by a mirror foil that is glued to the fibre mat end.

The fibre mats are sandwiched between two half-panels that are made of a carbon fibre, honeycomb structure.
The half-panels provide the necessary mechanical stiffness to the modules while keeping the material budget low.



\subsubsection{Fibre Mats}
The main components of the modules are the fibre mats.
Each fibre mat consists of six layers of scintillating fibres as described in \cref{sec:scintillating_fibres} and measures about \qty{2.4}{\meter} in length and \qty{0.13}{\meter} in width.
\Vref{fig:fibre-mat-crosssection} shows the cross section of one fibre mat.

\begin{figure}
  \centering
  \includegraphics[width=0.6\textwidth]{figures/scifi/Fibre_Mat_Crosssection}
  \caption{
    Cross section of one fibre mat used in the \scifitracker{}.
    Image taken from Ref.~\autocite{janine-diss}.
  }
  \label{fig:fibre-mat-crosssection}
\end{figure}

A fibre pitch of \qty{275}{\micro\meter} is chosen to be \qty{10}{\percent} larger than the nominal fibre diameter to allow for small production tolerances.
Proper fibre positioning is achieved by winding the fibre mats on a threaded wheel with a diameter of about \qty{0.82}{\meter}~\autocite{fibre-modules-design-review}.
While the scintillating fibres in the first layer are guided by the grooves in the winding wheel, fibres in subsequent layers fall into the gaps of the underlying layer.
This results in the staggered arrangement of the layers as can be seen in \vref{fig:fibre-mat-crosssection}.

Epoxy glue loaded with $\mathrm{TiO_2}$ is added during the winding process to bond the fibres together, while also providing a thin protection coating on the surface.
Additionally, the glue gets filled into \qty{2}{mm} deep holes that are located in the winding wheel with a distance of about \qty{245}{mm} to each other around the circumference of the wheel.
After curing, the glue sticks to the fibre mat resulting in small pins that are used to align the mats with a precision of better than \qty{100}{\micro\meter} in the $x$-direction during the module production~\autocite{scifi-tdr}.
A close-up view of one alignment pin is shown in \vref{fig:alignment-pin}.
The machine used for the serial production of the fibre mats is shown in \vref{fig:winding-machine}.

\begin{figure}
  \centering
  \begin{minipage}{0.58\textwidth}
    \centering
    \includegraphics[width=1\textwidth]{figures/scifi/Winding_machine}
    \caption{
      Winding machine for the serial production of the fibre mats for the \scifitracker{}.
      The path of the wound fibre is highlighted in blue.
      Image taken from Ref.~\autocite{fibre-modules-design-review}.
    }
    \label{fig:winding-machine}
  \end{minipage}
  \hfill
  \begin{minipage}{0.38\textwidth}
    \centering
    \captionsetup[subfigure]{justification=centering}
    \begin{subfigure}{1\textwidth}
      \includegraphics[clip, trim=2cm 0cm 1cm 0cm, width=\linewidth]{figures/scifi/Fibre_Mat_Winding_OK.pdf}
      \caption{Nominal winding pattern.}
      \label{fig:winding-ok}
      \vspace{0.2cm}%
    \end{subfigure}
    \begin{subfigure}{1\textwidth}
      \includegraphics[clip, trim=2cm 0cm 1cm 0cm, width=\linewidth]{figures/scifi/Fibre_Mat_Winding_Err1.pdf}
      \caption{Scintillating fibre skipped one groove.}
      \label{fig:winding-err1}
      \vspace{0.2cm}%
    \end{subfigure}
    \begin{subfigure}{1\textwidth}
      \includegraphics[clip, trim=2cm 0cm 1cm 0cm, width=\linewidth]{figures/scifi/Fibre_Mat_Winding_Err2.pdf}
      \caption{Scintillating fibre jumped to a higher layer.}
      \label{fig:winding-err2}
    \end{subfigure}

    \caption{
      Nominal pattern (a) and different types of errors (b,c) that can occur during the winding of a fibre mat.
      Images adapted from Ref.~\autocite{fibre-modules-design-review}.
    }
    \label{fig:winding-errors}
  \end{minipage}
\end{figure}

A reliable winding of the fibre is enabled by applying a constant tension using a dancer roller system.
However, different types of errors can occur during the winding of the fibre mats and are illustrated in \vref{fig:winding-errors}.
The winding procedure is monitored with a monochromatic industrial camera that feeds its output to a Convolutional Neural Network (CNN)~\autocite{cnn}.
This approach allows for a semi-automatic winding procedure where manual intervention is only needed in case of detected defects~\autocite{janine-diss}.

%Due to wobbling motions of the winding wheel, the wound fibre can jump to an advanced position (see \vref{fig:winding-err1}), or to a preceding groove causing it to be stacked onto the current layer (see \vref{fig:winding-err2}).
%In either case, the winding is stopped and the defect corrected by manually turning back the wheel.
%The detection of winding errors is achieved by monitoring the currently wound fibre tangentially to the wheel with a monochromatic industrial camera.
%The camera output is fed to a Convolutional Neural Network (CNN)~\autocite{cnn} in order to classify each image for winding defects.
%The trained CNN reached an accuracy, i.e. correct classification of an image, of \qty{99.99}{\percent}~\autocite{janine-diss}.
%This approach allows for a semi-automatic winding procedure where manual intervention is only needed in case of detected defects.

%After the actual winding is performed, which is a process that takes about \qty{2.5}{\hour}, a \qty{25}{\micro\meter} Kapton foil is applied to the accessible side of the fibre mat.
After the actual winding is performed, a \qty{25}{\micro\meter} Kapton foil is applied to the accessible side of the fibre mat.
The foil acts as an additional protection layer while also shielding the scintillating fibres from external light sources.
%After curing of the used epoxy glue for about \qty{36}{\hour}, the fibre mat is cut at a predefined location on the winding wheel perpendicular to the fibre direction and taken off the wheel.
Afterwards, the fibre mat is cut at a predefined location on the winding wheel perpendicular to the fibre direction and taken off the wheel.
In the process, the fibre mat immediately shrinks in the order of \qty{1}{\centi\meter} due to the constant tension applied during the winding.
The removed mat is heated up to \qty{40}{\celsius} to flatten it.
Afterwards, the Kapton foil is applied from the remaining side, leaving out small cutouts for the alignment pins.

At this stage, the fibre mat is still slightly larger than the final dimensions.
The nominal length of \qty{2424}{\milli\meter}~\autocite{fibre-modules-design-review} is defined by so-called end pieces made from plastic that are glued to both sides on both ends of the fibre mat.
The excess parts of the mat are pre-cut using a saw blade and further processed using a diamond milling head.
On one side of the mat, the SiPM arrays will be pressed against the smooth surface that is obtained in this way.
On the other side, a mirror foil is applied to reflect back photons that are emitted away from the SiPMs in order to increase the overall light yield.
An aluminized Mylar foil is used for that purpose which proved to reach a reflectivity in the order of \qty{80}{\percent}~\autocite{fibre-mirrors}.

All the steps listed previously were performed in four different locations: TU Dortmund, RWTH Aachen, EPFL in Lausanne and Kurchatov Institute in Moscow.
A total of 1024 fibre mats that are needed for the complete detector plus spares were produced in these winding centres.
A finished fibre mat is shown in \vref{fig:finished-mat} and is already cut to its nominal length.
However, when leaving the winding centres, the width is still slightly larger than their nominal value and they feature excess Kapton foil to both edges.
This allows for a safe transport to the module assembly centres, where the mats are embedded into the mechanical support structure of the fibre modules.

\begin{figure}
  \centering
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/Alignment_Pin}
    \caption{Close-up view on a wound fibre mat with alignment pin. Image taken from Ref.~\autocite{janine-diss}.}
    \label{fig:alignment-pin}
  \end{minipage}
  \hfill
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/Fibre_Mat_finished.pdf}
    \caption{Finished and laminated fibre mat with plastic end piece. Image taken from Ref.~\autocite{janine-diss}.}
    \label{fig:finished-mat}
  \end{minipage}
\end{figure}


\subsubsection{Mechanical Support Structure}
The wound fibre mats that are produced in the winding centres are shipped to two different locations in order to get assembled to full-size detector modules.
In these so-called module assembly centres, which are at Heidelberg University and Nikhef in Amsterdam, the fibre mats are embedded into a support structure that provides mechanical stiffness and protection to the modules.
Before that can happen, the excess Kapton foil needs to be removed and the fibre mats need to be cut to their nominal width of \qty{130.65}{\milli\meter}.
%This is done with the help of two parallel circular saws.
%They perform the cuts with a precision of better than \qty{150}{\micro\meter} to ensure a minimal loss of acceptance between two neighbouring fibre mats~\autocite{fibre-modules-design-review}.

Afterwards, eight fibre mats are laid out in the final \numproduct{2 x 4} arrangement as shown in~\vref{fig:exploded-fibre-module} onto a high precision template made from a single aluminium plate.
The alignment pins of the mats are facing down and are guided by four longitudinal grooves in the template.
The grooves have a centre-to-centre distance of \qty{130.8}{\milli\meter} leaving a gap of \qtyrange[range-phrase=--,range-units=single]{0.15}{0.20}{\milli\meter} in-between two neighbouring fibre mats to account for tolerances of the longitudinal cut.
The positioning follows the geometry of the SiPM arrays (see \cref{sec:sipm}) such that no additional loss of acceptance is introduced by the gaps.

In the next step, glue is applied on the surface of the fibre mats that bonds the mats to the \qty{2}{\centi\meter} thick support structure made from a light core material (Nomex honeycomb).
At the ends of the modules, the honeycomb is cut out to provide space for four individual pieces of aluminium, the so-called end plugs.
The end plugs act as an anchor point to mount and align the fibre modules in the detector.
Additionally, they house the light injection system (see \cref{sec:lis}) that is used for calibrating the individual channels.
The outer shell of a fibre module consists of a carbon fibre reinforced polymer (CFRP) skin that is \qty{200}{\micro\meter} thick and provides the necessary stiffness to the module.
%The CFRP skin is attached to the end plugs and honeycomb using another layer of glue.
%Pressure is applied using a vacuum pump and foil while the glue cures for eight hours.

The partially assembled module is turned over to the other side and placed onto another, flat aluminium template.
In this position, the placement and straightness of the fibre mats within the module is verified using a test setup consisting of a laser and a beam camera:
The laser is positioned on one side of the module pointing along the centre of two facing fibre mats.
The beam camera is successively positioned on top of the total of 16 alignment pins on the two mats and the position of the beam projection is recorded.
The camera is mounted on an aluminium block with the same kind of groove as in the aluminium template to allow for an accurate placement on top of the alignment pins.
Like this, the straightness of the fibre mats in the module can be evaluated by the deviation of the recorded beam projections to a straight line in both the horizontal and vertical direction with respect to the setup on the assembly table, corresponding to the $x$ and $z$ direction within the LHCb coordinate system.
\Vref{fig:laser-test} shows the exemplary result of two facing fibre mats demonstrating a straightness of better than the required \qty{100}{\micro\meter} along all pins.

\begin{figure}
  \centering
  \includegraphics[width=1.0\textwidth]{figures/scifi/FSM00073_LaserTest}
  \caption{
    Evaluation of the straightness of two facing fibre mats within a fibre module using a test setup consisting of a laser and a beam camera.
    The deviations from a straight line along the alignment pins are shown in both the $x$ (top) and $z$ (bottom plot) direction within the LHCb coordinate system.
  }
  \label{fig:laser-test}
\end{figure}

After the straightness of the fibre mats is verified, the module is closed from the remaining side using another half-panel consisting of honeycomb core material and a CFRP skin.
In order to provide light-tightness to the module, it is finished by closing the sides with CFRP sidewalls.
\Vref{fig:module-side-view} shows the schematic side view and cross section of a finished fibre module.
At this stage, the scintillating fibres are only exposed to the outside between the plastic end pieces of the fibre mats on both sides of the module.
They will later be covered by the SiPM arrays (see \cref{sec:sipm}).

\begin{figure}
  \centering
  \includegraphics[width=1.0\textwidth]{figures/scifi/Fibre_Module_Sideview}
  \caption{
    Cross section and side view of a fibre module of the \scifitracker{}.
    The dimensions shown in this image are given in \unit{mm}.
    Image taken from Ref.~\autocite{fibre-modules-design-review}.
  }
  \label{fig:module-side-view}
\end{figure}

Special modules need to be produced to account for the beam pipe as indicated in \vref{fig:T3}.
They feature a rectangular \qtyproduct{13 x 11.5}{\centi\meter} cutout on one side of the module.
To allow for this, the two fibre mats on this side of the module are produced with a reduced length.
The originally intended circular cutout was found to be difficult to realise due to the mirror foil that needs to be applied to the fibre mat ends~\autocite{revised-geometry}.

The material budget of the individual components of a \scifitracker{} module are listed in \vref{tab:module-material-budget}.
The aluminium end plugs are not considered because they lie outside the acceptance of the experiment.
The total radiation length for one module is $\qty{0.984}{\percent} X_0$ which translates to $\qty{3.936}{\percent} X_0$ for one tracking station consisting of four detection layers.
This value fulfills the required limit on the material budget as listed in \cref{sec:performance-requirements}.
While the Nomex honeycomb structure accounts for \qty{95}{\percent} of the module's thickness, it only contributes about \qty{31}{\percent} to the total radiation length.
Note that the total thickness given in \vref{tab:module-material-budget} is slightly larger than the nominal value of \qty{41.6}{\milli\meter} due to submersion of the glue into the honeycomb~\autocite{fibre-modules-design-review}.

In total, 128 fibre modules needed for the detector plus spares were produced at the two module assembly centres between 2016 and 2019.
Of these, about \qty{20}{\percent} are special modules with a rectangular cutout on one side to account for the beam pipe in the centre of each of the 12 detection layers.

\begin{table}
  \centering
  \caption{
    Material budget for a scintillating fibre module.
    Data taken from~\autocite{fibre-modules-design-review,pdg}.
  }
  \label{tab:module-material-budget}
  \begin{tabular}{@{}lrrr@{}} \toprule
    Material & Thickness $\Delta z$ [\unit{\micro\meter}] & Radiation length $X_0$ [\unit{\centi\meter}] & $\Delta z / X_0$ [\unit{\percent}] \\ \midrule
    Scintillating fibres & \num{1350} & \num{33.2} & \num{0.407} \\
    Kapton foil & \numproduct{2 x 25} & \num{28.6} & \num{0.017} \\
    Panel assembly glue & \numproduct{4 x 75} & \num{36.1} & \num{0.083} \\
    Nomex honeycomb & \numproduct{2 x 20000} & \num{1310.0} & \num{0.305} \\
    Carbon fibre skin & \numproduct{2 x 200} & \num{23.3} & \num{0.172} \\ \addlinespace
    \textbf{Total} & \num{42100} & & \num{0.984} \\ \bottomrule
  \end{tabular}
\end{table}


\section{Silicon Photomultipliers}
\label{sec:sipm}
Silicon photomultipliers (SiPMs) are solid state devices that allow for the detection of individual photons.
Combined with their small dimensions and high granularity they meet the needs for a high resolution tracker like the SciFi~\autocite{scifi-tdr}.

\subsubsection{Operating Principle}
The functionality of solid state photon detectors is based on a $p$-$n$ junction, which refers to the boundary between $n$- and $p$-doped regions inside a semiconductor material.
Close to the boundary, electrons from the $n$-doped region diffuse into the $p$-doped region and, vice versa, electron holes from the $p$-doped region diffuse into the $n$-doped region.
By recombination of electrons and holes in the area around the boundary, the so-called depletion zone, the amount of free charge carriers is reduced significantly resulting in a low electrical conductivity.
The diffusion continues until the electrical potential that is generated in the process counteracts the motion of the charge carriers.

\paragraph{Forward Bias}
By applying a positive external voltage to the $p$-side with respect to the $n$-side (forward bias) the diffusion process is reversed.
When the bias voltage is large enough, the depletion zone becomes very thin and the $p$-$n$ junction becomes conductive allowing for a (forward) current flow.

\paragraph{Reverse Bias}
Applying a negative external voltage to the $p$-side with respect to the $n$-side (reverse bias) has the opposite effect:
The depletion zone expands and the electrical conductivity stays low.
Only a small (reverse) current can flow. \\

\noindent
A photodiode makes use of operating a $p$-$n$ junction with a reverse bias voltage applied.
Photons with sufficient energy can create electron-hole pairs in the depletion zone via the photoelectric effect.
A free electron that is generated by this process is called (primary) photoelectron (pe).
Since the resulting current flow only consists of individual charge carriers, photodiodes are not suitable for low intensity applications like the \scifitracker{}.

Avalanche photodiodes (APD) overcome this issue by operating at higher voltages.
The thereby increased electrical field at the $p$-$n$ junction accelerates the primary photoelectrons up to energies at which they can create electron-hole pairs themselves.
This process can repeat itself several times causing an avalanche of charge carriers that is measurable as an electrical current.

When the applied voltage is increased even further, beyond the breakdown voltage \Vbd{}, a single primary photoelectron can cause a self sustaining avalanche (Geiger mode).
The released charge is independent of the number of primary photoelectrons and results in a permanent current if unquenched.
In practice, the avalanche is typically interrupted by inserting a resistor in series to the photodiode.
As the current flow increases during the emerging avalanche, the applied voltage is lowered below the breakdown voltage by the so-called quench resistor $R_\text{Q}$ such that no more charge carriers can be created.

SiPMs are composed of an array of APD pixels that are operated in Geiger mode (G-APD).
They are connected in parallel such that the resulting signal is proportional to the number of pixels with a triggered avalanche.
If the probability for multiple photons hitting the same pixel is low, SiPMs allow for the counting of individual photons.
This is given when the number of pixels is much greater than the number of incident photons.
The amplitude of a signal is typically stated in units of generated primary photoelectrons ($X\,\text{pe}$), which is equivalent to the number of involved pixels in low light conditions.

\subsubsection{Correlated and uncorrelated Noise}
SiPMs are subject to different types of noise that affect the operation.
One distinguishes between correlated and uncorrelated noise.

\paragraph{Correlated Noise}
The emergence of correlated noise is, as the name suggests, correlated to the occurrence of an initial signal.
A large contribution comes from crosstalk.
Crosstalk originates from infrared photons that are emitted in a pixel during an avalanche.
These photons can reach neighbouring pixels causing another discharge.
Depending on the region where the photon is absorbed, the resulting pulse either happens instantaneously (\textbf{direct crosstalk}, within a few \qty{100}{\pico\second}) or with a significant delay up to \qty{100}{\nano\second} (\textbf{delayed crosstalk}) with respect to the initial signal~\autocite{sipm-for-scifi}.

Delayed crosstalk must not be confused with \textbf{afterpulses}, which have a different formation mechanism.
Afterpulses occur due to defects in the silicon lattice that can act as charge traps.
A charge carrier created during an avalanche can get caught in these traps.
When it is released at a later time, it can trigger another discharge in the same pixel causing a second, delayed pulse distributed over a few \qty{100}{\nano\second}~\autocite{sipm-for-scifi}.

\paragraph{Uncorrelated Noise}
Instead of being excited by an incident photon, an electron-hole pair in the depletion zone can also be generated by thermal excitation.
The resulting signal is indistinguishable from a photon induced avalanche.
This type of noise can occur randomly and is independent of any initial signal.
Because of that, it is also referred to as \textbf{dark counts}.
Due to the thermal origin, the dark count rate (DCR) strongly depends on temperature.
In addition to that, the DCR increases after irradiation. \\

\noindent
Various combinations of the different types of noise can occur.
For example, it is not uncommon that a dark count arises in conjunction with a crosstalk event.
The amplitude of the resulting signal is thereby increased making it more difficult to be identified as noise.





\subsubsection{Basic electrical Model}
Following Ref.~\autocite{sipm-technical-guide}, a basic electrical model of a G-APD can be developed to derive the key characteristics of the signal pulse.
It consists of the breakdown voltage \Vbd{}, capacity $C_\text{J}$ and ohmic resistance $R_\text{S}$ of the G-APD, as well as the quench resistor $R_\text{Q}$ and bias voltage \Vbias{}.
A conceptual switch $S$ is responsible for initiating ($S$ is closed) and terminating ($S$ is open) the avalanche.

\begin{figure}
  \centering
  \includegraphics[width=1.0\textwidth]{figures/scifi/SiPM_Model}
  \caption{
    Electrical circuit of a basic equivalent model for an externally biased G-APD (left).
    The plot on the right shows the signal pulse after closing (at $t_0$) and re-opening (at $t_\text{max}$) the switch $S$.
    Image inspired from Ref.~\autocite{sipm-technical-guide}.
  }
  \label{fig:sipm-model}
\end{figure}

\Vref{fig:sipm-model} (left) shows the electrical circuit of the model.
In the absence of incident photons and ignoring dark noise, the switch $S$ is open.
The capacitor $C_\text{J}$ is fully charged by the applied bias voltage \Vbias{}.
No current is flowing.
Then, representative for an incident photon triggering an avalanche, the switch $S$ is closed at time $t_0$.
The capacitor $C_\text{J}$ begins discharging through the series resistor $R_\text{S}$ as a flowing current
\begin{equation}
  I_\text{S}(t) \sim 1 - \text{exp}\left(-\frac{t}{R_\text{S}C_\text{J}}\right) \equiv 1 - \text{exp}\left(-\frac{t}{\tau_\text{S}}\right)
\end{equation}
with a time constant $\tau_\text{S}$\,.
At $t_\text{max}$, which is in the order of \qty{1}{ns} after $t_0$~\autocite{sipm-technical-guide}, the maximum current
\begin{equation}
  I_\text{max} = \frac{V_\text{BIAS}-V_\text{BD}}{R_\text{Q}+R_\text{S}} \equiv \frac{\OV}{R_\text{Q}+R_\text{S}}
\end{equation}
is reached.
The introduced variable \OV{} is called overvoltage.
At this time, the applied voltage to the G-APD drops down to approximately \Vbd{}, which is not sufficient to sustain the avalanche.
The switch $S$ opens and the capacitor $C_\text{J}$ is recharged by a current
\begin{equation}
  I_\text{Q}(t) \sim \text{exp}\left(-\frac{t}{R_\text{Q} C_\text{J}}\right) \equiv \text{exp}\left(-\frac{t}{\tau_\text{Q}}\right)
\end{equation}
with a time constant $\tau_\text{Q}$\,, which is also referred to as the recovery time.
The shape of the resulting signal is shown in \vref{fig:sipm-model} (right).
For $R_\text{Q} \gg R_\text{S}$ and thereby $\tau_\text{Q} \gg \tau_\text{S}$, the total charge $Q$, i.e. area under the signal shape, can be derived as
\begin{equation}
  Q = I_\text{max} \cdot \tau_\text{Q} = \frac{\OV}{R_\text{Q}+R_\text{S}} \cdot R_\text{Q} C_\text{J} = \OV \cdot C_\text{J}\,.
\end{equation}
Since the avalanche is triggered by a single primary electron with elementary charge $e$, the charge amplification factor -- or gain $G$ -- is given by
\begin{equation}
  \label{eq:sipm-gain}
  G = \frac{Q}{e} = \frac{\OV \cdot C_\text{J}}{e}\,.
\end{equation}
Typical operation values for the gain of SiPMs lie between $10^5$ and $10^7$~\autocite{sipm-technical-guide}.






\subsubsection{Application at the \scifitracker{}}

\begin{figure}
  \centering
  \includegraphics[width=0.8\textwidth]{figures/scifi/SiPM_with_zoom}
  \caption{
    Hamamatsu H2017 128-channel SiPM array with flex cable and connector as used for the \scifitracker{} (left).
    A zoomed in view on the active area with the gap between the two dies is shown on the right, as well as an illustration of the position of one SiPM channel with respect to the fibre mat.
    Image adapted from Ref.~\autocite{axel-diss}.
  }
  \label{fig:sipm-zoom}
\end{figure}

The \scifitracker{} utilises a total of \num{524288} SiPM channels that are grouped in 128-channel arrays.
The multichannel arrays referred to as H2017 are custom-made and produced by Hamamatsu\footnote{Hamamatsu Photonics K.K., 325-6, Sunayama-cho, Naka-ku, Hamamatsu City, Shizuoka Pref., 430-8587, Japan}.
One array is \qty{32.54}{\milli\meter} wide and consists of two dies with 64 channels each.
The silicon dies are protected with a thin \qty{105}{\micro\meter} epoxy layer~\autocite{axel-diss}.
After mounting in the detector, the insensitive area between two adjacent SiPM arrays amounts to \qty{480}{\micro\meter}, while the specifications state the gap between the two dies to be \qty{220 \pm 50}{\micro\meter}~\autocite{sipm-specifications}.
The geometry was chosen such that the width of multiple arrays match the fibre mat and module dimensions.
One mat is instrumented with four SiPM arrays, which translates to 16 arrays for a fibre module on each end.
\Vref{fig:sipm-zoom} shows a photo of a H2017 package with a zoomed in view on the individual SiPM channels.
The package is mounted on a Kapton flex PCB (Printed Circuit Board) that is about \qty{12}{\centi\meter} long.
Two 80-pins, \qty{0.5}{\milli\meter} pitch connectors\footnote{DF12(3.0)-80DP-0.5V by Hirose Electric Co., Ltd.} are placed on the other end of the flex PCB to provide the SiPM connections to the front-end electronics (see \cref{sec:fee}).

\begin{figure}
  \centering
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/SiPM_PDE}
    \caption{
      Measured photon detection efficiencies (PDE) based on a pulse counting approach for an H2017 SiPM array for different wavelengths $\lambda$ and overvoltages $\Delta V$.
      Image taken from Ref.~\autocite{sipm-for-scifi}.
    }
    \label{fig:sipm-pde}
  \end{minipage}
  \hfill
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/SiPM_Noise}
    \caption{
      Probabilities for different types of correlated noise as determined during the quality assurance of \num{5000} H2017 SiPMs.
      Variations in the order of $\pm \qty{1}{\percent}$ were observed between the different devices.
      Image taken from Ref.~\autocite{olivier-diss}.
    }
    \label{fig:sipm-noise}
  \end{minipage}
\end{figure}

The SiPM channel pitch is \qty{250}{\micro\meter} thus matching the diameter of the scintillating fibres.
The height is with \qty{1625}{\micro\meter} about \qty{20}{\percent} larger than the fibre mat to allow for slight misalignments during the assembly.
One channel is composed of \numproduct{4 x 26} pixels each covering an area of \qtyproduct{57.5 x 62.5}{\micro\meter}.
As shown in \vref{fig:sipm-pde}, the large pixel size allows for high photon detection efficiencies up to \qty{44}{\percent} at the nominal overvoltage $\Delta V = \qty{3.5}{\volt}$~\autocite{sipm-for-scifi} for the relevant wavelengths from \qtyrange{400}{600}{\nano\meter} (compare with \vref{fig:emission-spectrum}).
Additionally, it enables the application of isolating trenches to suppress crosstalk.
\Vref{fig:sipm-noise} shows the crosstalk probability for different overvoltages $\Delta V$, as well as the probabilities for other types of correlated noise.
At the nominal overvoltage $\OV = \qty{3.5}{\volt}$, the total probability for correlated noise amounts to about \qty{7}{\percent}.
It is split equally between the two types of crosstalk: direct and delayed.
Afterpulses practically do not play a role for H2017 arrays.

The typical gain for an H2017 SiPM array is $G = \num{4e6}$ at the nominal overvoltage.
\Vref{fig:sipm-gain} shows the dependence of the gain on the overvoltage for one channel.
As derived in \vref{eq:sipm-gain}, it follows a linear relation.
The recovery time is found to be $\tau_\text{Q} = \qty{84.6\pm0.2}{\nano\second}$.
The breakdown voltage \Vbd{} ranges from \qtyrange{51.0}{52.5}{\volt} as determined among 20 devices from the production batch.
Within one array, it typically varies by $\qty{\pm300}{\milli\volt}$~\autocite{axel-diss}.
When operated at the same voltage around the nominal overvoltage $\OV = \qty{3.5}{\volt}$, this results in a gain uniformity of better than \qty{10}{\percent} as required in the TDR~\autocite{scifi-tdr}.


\begin{figure}
  \centering
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/SiPM_Gain}
    \caption{
      Dependence of the gain on the overvoltage for one H2017 SiPM channel.
      The red line shows the result of a linear fit to the data.
      Image taken from Ref.~\autocite{sipm-for-scifi}.
    }
    \label{fig:sipm-gain}
  \end{minipage}
  \hfill
  \begin{minipage}[t]{0.48\textwidth}
    \centering
    \includegraphics[width=1.0\linewidth]{figures/scifi/SiPM_DCR}
    \caption{
      Dark count rate (DCR) as a function of the temperature for irradiated H2017 SiPMs at $\OV = \qty{3.5}{\volt}$.
      The coefficient $T_{1/2}$ that describes the temperature difference at which the DCR is reduced by $1/2$ is determined with an exponential fit in the range [$-40,-10$]$\,\unit{\celsius}$ and indicated by the solid lines.
      Image adapted from Ref.~\autocite{olivier-diss}.
    }
    \label{fig:sipm-dcr}
  \end{minipage}
\end{figure}

A key parameter for the operation of the \scifitracker{} is the rate at which dark counts occur (DCR), as it is main source of noise for the detector.
As described previously, the DCR strongly depends on both temperature and irradiation dose, specifically the \qty{1}{MeV} neutron-equivalent fluence (\unit{n_{eq}\per\centi\square\meter}).
Towards the end of the lifetime of the upgraded detector, the expected fluence that the SiPMs will have received is \qty{6e11}{n_{eq}\per\centi\square\meter}~\autocite{scifi-tdr}.
At that time, the DCR will reach values of several hundred \unit{MHz} at room temperature, while the DCR of an unirradiated SiPM is in the order of \unit{kHz}.
This drastic increase resulting from lattice deformations is mitigated by cooling of the SiPM arrays.
As shown in \vref{fig:sipm-dcr}, the DCR can be reduced by \qty{50}{\percent} with every decrease in temperature by about \qty{10}{\celsius}.
Thereby, at the end of LHC Run 4, manageable DCRs in the order of \qty{10}{MHz} can be reached at \qty{-40}{\celsius}.
Reaching low DCRs of irradiated SiPMs at \qty{-40}{\celsius} was a main criteria for the selection of the best suited technology during the R\&D phase of the \scifitracker{}.

The temperature of each array can be monitored with the help of a Pt1000 temperature sensor\footnote{Pt1000 SMD 0603 Class B by Heraeus Nexensos GmbH} that is mounted on the backside of the SiPM array.
The specific sensor was chosen to cover the temperature range of interest between \qtyrange[retain-explicit-plus,range-phrase={ and }]{-40}{+30}{\celsius} with a precision of \qty{\pm1}{\celsius}~\autocite{axel-diss}.


\section{Front-End Electronics}
\label{sec:fee}

The front-end electronics of the \scifitracker{} are located just outside of the geometrical acceptance on both ends of the fibre modules.
They provide the interface between the SiPMs and the control and data-acquisition (DAQ) system.
The connection to the SiPMs is established via flex cables as shown in \vref{fig:sipm-zoom}, while the communication with the back-end electronics happens over optical fibres.
The optical connections are divided into unidirectional links for the transmission of the data, and bidirectional control links.

\begin{figure}
  \centering
  \includegraphics[width=1.0\textwidth]{figures/scifi/ROB_Photo}
  \caption{
    Photograph of a SciFi Readout Box (ROB) with the top cover removed.
    The connection to the SiPM arrays is made via 16 connectors located at the bottom.
  }
  \label{fig:rob-photo}
\end{figure}

The front-end electronics are grouped into so-called Readout Boxes (ROBs), which are the physical units that are mounted to the fibre modules.
In total, the front-end electronics of the \scifitracker{} consists of 256 ROBs with more than half a million channels.
A photograph of one ROB with the top cover removed is shown in \vref{fig:rob-photo}.
It consists of two identical halves, which are referred to as HalfROBs.
Besides from sharing a common housing and cooling block, the HalfROBs are fully independent from each other in terms of power and control.

The front-end electronics follow a modular design.
Each HalfROB consists of one Master Board, four Cluster Boards, and four PACIFIC Boards.
The latter are the direct interfaces to the SiPMs.
Each PACIFIC Board houses four custom 64-channel ASICs (Application Specific Integrated Circuits) that perform the analogue processing and digitisation of the SiPM signals.
In the next step, the Cluster Boards perform a hit reconstruction and noise suppression on the digital pattern provided by the PACIFIC Boards with the help of two on-board FPGAs (Field Programmable Gate Arrays).
Lastly, the clustered data is encoded and shipped to the DAQ servers via optical transmitters located on the Master Board.
In addition, the Master Board is responsible for distributing power, clocks and control commands among all boards within one HalfROB.

A detailed description of the various components of the SciFi front-end electronics is given in \cref{ch:fee}.


\section{Infrastructure}
The operation of the \scifitracker{} is only enabled with the help of a reliable infrastructure.
It includes the provision with electrical power and cooling for the various components.
In addition, a mechanical support structure is required that holds the detector modules in place and routes the supply lines to the desired locations.
The different parts constituting the infrastructure are described in the following.


\subsection{Electrical Power}
Two different types of power supplies are used for the \scifitracker{}.
They can be distinguished by the voltage levels at which they are operated.

\subsubsection{Low Voltage}
\label{sec:lv}
The front-end electronics are powered by MARATON power supplies manufactured by W-IE-NE-R\footnote{W-IE-NE-R Power Electronics GmbH, Burscheid, Germany}.
These devices are specifically developed for the operation at the LHC in a magnetic and radiation environment.
This is also reflected by the acronym MARATON that stands for \textbf{\underline{Ma}}gnetism \textbf{\underline{Ra}}diation \textbf{\underline{To}}lerant \textbf{\underline{N}}ew power supply system.
Each unit offers 12 independent low voltage/high current channels that provide up to \qty{300}{\watt} each~\autocite{maraton}.

The MARATON system consists of three main components: the power box, a primary rectifier, and a remote controller.
The primary rectifier converts the \qty{230}{\volt} AC (alternating current) mains voltage to a regulated \qty{385}{\volt} DC (direct current) voltage that is fed to the power box.
In the MARATON power box, DC-DC converters are used to provide up to 12 low voltage floating output voltages.
With the help of the remote controller, each channel can be switched off or on independently.
In addition, it allows for the monitoring of the output voltages and currents.
The adjustment of the output voltage for each channel is possible via a potentiometer that is accessible with a screwdriver at the front of the power box and that can be set up to \qty{8}{\volt}.
In the same way, the overvoltage protection (OVP) and current limits can be modified.

At the \scifitracker{}, each MARATON channel delivers power for up to two ROBs.
To limit the voltage drop in the cables due to high currents, the power box is located near the detector in a harsh environment with high radiation and stray magnetic fields.
The connection to the primary rectifier is established with an Amphenol ECTA 133 circular connector, while two 37-pin D-SUB connectors are used for the communication with the remote controller.
The system is designed such that the primary rectifier and remote controller are operated in a safe environment under standard industrial conditions up to \qty{120}{\meter} apart from the power box.
The remote controller is integrated into the LHCb network by a TCP/IP connection.
Utilising the Open Platform Communications (OPC) standard, this allows for the control and monitoring of the system within the aforementioned limits from outside the LHCb cavern.


\subsubsection{High Voltage}
\label{sec:hv}
The SiPMs are biased by power supplies manufactured by CAEN\footnote{CAEN S.p.A., Viareggio, Italy}.
The modules of type A1539BP~\autocite{caen-module} feature \num{32} channels that are accessible through two DB25 connectors.
The output voltage of each channel can be set between \qtyrange[range-phrase={ and }]{0}{100}{\volt} and deliver up to \qty{20}{\milli\ampere}.
Even though this voltage range is not usually referred to as high voltage (HV), the term will be used throughout this thesis as counterpart to the \qty{8}{V} LV as provided by the MARATON power supplies.
The accuracy of the set voltage is stated as $\qty{\pm 0.05}{\percent}\qty{\pm 50}{\milli\volt}$.
The ramp-up and ramp-down rates can be individually adjusted for each channel.

The boards are equipped with both current and voltage protections.
The maximum output voltage can be set on a hardware level through a potentiometer that is accessible on the front panel.
It applies to all 32 channels.
Further restrictions on current and voltage limitations are adjustable in software on a per-channel basis.
In addition, it allows for the modification of the trip time after which a channel is turned off when the current limit is exceeded.

At the \scifitracker{}, each HV channel biases four SiPM arrays corresponding to one fibre mat.
Intermediary switch panels allow to disable individual SiPM dies to not interfere with neighbouring dies in case of defects.
The HV modules are operated in CAEN crates of type SY4527~\autocite{caen-crate}.
Besides from housing up to 16 boards, each crate contains a fan tray section with six fans, and a \qty{600}{\watt} primary power supply unit (PSU).
Depending on the load, up to three additional booster PSUs can be installed.
The crates used for the \scifitracker{} will be expanded with one additional \qty{1200}{\watt} booster PSU of type A4533, allowing for the operation of ten A1539BP modules at maximum load.
For controlling the HV remotely, each crate includes a CPU and Gigabit Ethernet interface that allows for the integration into the LHCb network.



\subsection{Cooling}
Two different types of cooling circuits are implemented in the \scifitracker{}.

\subsubsection{Water Cooling}
During operation, the front-end electronics heat up significantly.
To limit the temperature to a maximum of \qty{50}{\celsius}, a water cooling scheme inspired by the OT is implemented~\autocite{scifi-tdr}.
To allow for an efficient heat transfer, the ROBs are tightly screwed on aluminium cooling blocks.
The cooling water flows through a straight copper pipe that traverses the cooling blocks of 5-6 neighbouring ROBs depending on the tracking station.
A \qty{180}{\degree} bend behind the last ROB guides the water to the return pipe that traverses the cooling blocks in reverse order, thereby also contributing to the cooling process.

Based on a heat dissipation close to \qty{100}{\watt} per ROB, it was estimated that a flow rate of \qty{2.5}{\liter/\minute} per row of 5-6 ROBs is required.
At this rate, the water temperature is expected to increase by \qty{3}{\celsius}.
It is foreseen to use chilled and demineralised water with a temperature of \qty{16}{\celsius} at the inlet~\autocite{water-cooling-requirements}.


\subsubsection{Novec Cooling}
\label{sec:novec}
As outlined in \cref{sec:sipm}, the SiPMs need to be cooled to \qty{-40}{\celsius} to mitigate the drastic increase of the dark count rate due to radiation damages.
A complex infrastructure is needed to achieve the required cooling performance.
The coolant used for that purpose is Novec 649 distributed by 3M\footnote{3M, Saint Paul, Minnesota, USA}.
The fluid has similar thermo-physical properties as Perfluorohexane $\text{C}_{6}\text{F}_{14}$ that has been widely used in previous low-temperature cooling applications at the LHC.
Comparable with $\text{C}_{6}\text{F}_{14}$, it is non-combustible, volatile, dielectric and practically non-toxic~\autocite{scifi-tdr}.
However, due to its much lower global warming potential (GWP) around 1, Novec has been chosen as the superior alternative.

\begin{figure}
  \centering
  \includegraphics[width=0.8\textwidth]{figures/scifi/Coldbox}
  \caption{
    Cross-sectional view of a SciFi cold box with its components (left), as well as a thermal simulation (right).
    The SiPM signals are transmitted through the flex PCB to the front-end electronics located above.
    Images adapted from~\autocite{scifi-tdr}.
  }
  \label{fig:coldbox}
\end{figure}

The coolant is circulated through a titanium cooling bar to which the SiPMs are glued.
The structure is surrounded by an insulating box, which is referred to as cold box.
A cross-sectional view of a cold box is shown in \vref{fig:coldbox}.
Each box contains sixteen SiPMs corresponding to one fibre module and ROB.
A pair of inlet and outlet bellows allow for a flexible connection of the individual cold boxes to the cooling circuit.
The Novec supply lines as well as the bellows are vacuum insulated.

To avoid condensation and ice formation inside the cold box due to the low temperatures, a system is implemented that constantly flushes the boxes with a dry gas (either dried air or nitrogen).
During the commissioning of the first detector parts (see \cref{ch:commissioning}), it was found that condensation can also occur towards the outside.
Therefore, a heating system has been developed that covers the surface of the cold boxes, as well as the Novec bellows and dry gas outlets.
It consists of Kapton isolated heating wires that are twisted around the mentioned parts.
Same as for the front-end electronics, the required heating power is provided by MARATON power supplies.


\subsection{Mechanical Support Structure}
\label{sec:c-frames}
The individual detector modules are carried by C-shaped mounting frames.
The complete detector consists of a total of 12 of these so-called C-Frames that are arranged on both sides of the beam pipe.
They are mounted on guide rails, which facilitates the installation and also allows for quick access in case of required maintenances.

\begin{figure}
  \centering
  \includegraphics[width=0.8\textwidth]{figures/scifi/C-Frame}
  \caption{
    Schematic illustration of a C-Frame of type SX that carries the detector modules, front-end electronics, and the various infrastructure components.
    The \scifitracker{} consists of 12 of these frames.
    Image adapted from Ref.~\autocite{c-frames}.
  }
  \label{fig:c-frame}
\end{figure}

One C-Frame carries two layers with 5-6 fibre modules each depending on the associated tracking station.
While the modules in one layer are oriented vertically (i.e. in $y$-direction within the LHCb coordinate system), the modules in the second layer are tilted by \qty{5}{\degree} -- the so-called \textbf{s}tereo layer.
Having two different types of C-Frames is sufficient to form the typical \textit{x-u-v-x} geometry (see \cref{sec:upgrade-motivation}) in each station on both sides of the beam pipe.
They are denoted as C-Frames of type XS and SX, depending on the order of the two layers when having the vertical bar of the frame on the right-hand side.

In addition to the modules, the C-Frames also provide the framework for the various components of the detector infrastructure as described before.
They guide the low and high voltage cables to the ROBs, as well as the pipes for the water and Novec cooling.
Moreover, cable trays are mounted on the frames to allow for the installation of the optical cables for data transmission and control of the front-end electronics.
A schematic representation of a \cframe{} with its components is shown in \vref{fig:c-frame}.

\section{Tracking Performance}
\label{sec:testbeam}
During the preparation of this thesis and throughout the development phase of the \scifitracker{}, multiple test beam campaigns have been conducted in order to evaluate its performance, in particular the tracking capabilities.
The most recent one took place during two weeks in July 2018 at the CERN Super Proton Synchrotron (SPS) North Area~\autocite{sps-north-area}.
The beam in the H8 beamline was set to a width of about \qty{1}{\centi\meter} and provided pions and muons with momenta around \qty{180}{GeV/s}.

The experimental setup as shown in \vref{fig:testbeam-setup} consisted of two full-width half-length fibre modules with the mirror on one end and \num{16} SiPM arrays on the other.
The SiPM signals were read out by two ROBs that transferred the data via optical links to a mini version of the upgraded LHCb DAQ system~\autocite{minidaq}.
Reference tracks used for the determination of the tracking performance parameters were provided with an uncertainty of approximately \qty{8}{\micro\meter} by a TimePix3 telescope located downstream of the modules~\autocite{timepix}.

\begin{figure}
  \begin{subfigure}{0.49\linewidth}
    \centering
    \includegraphics[width=1.0\textwidth]{figures/scifi/Testbeam}
    \caption{Experimental setup}
    \label{fig:testbeam-setup}
  \end{subfigure}
  \hfill
  \begin{subfigure}{0.49\linewidth}
    \centering
    \includegraphics[width=1.0\textwidth]{figures/scifi/Resolution}
    \caption{Resolution}
    \label{fig:testbeam-resolution}
  \end{subfigure}
  \caption{
    Test beam setup consisting of two fibre modules along with the corresponding front-end electronics (a) and measured hit resolution in the bending direction ($x$-axis) within the LHCb coordinate system (b).
    The particle beam provided in the SPS North Area at CERN is indicated by the orange arrow.
  }
  \label{fig:testbeam}
\end{figure}

The campaign was a great success as it could be shown that a single hit efficiency beyond \qty{99}{\percent} can be reached.
In addition, as shown in \vref{fig:testbeam-resolution}, the resolution in the bending direction ($x$-axis) within the LHCb coordinate system was measured to be about \qty{65}{\micro\meter}.
Thus it could be verified that the corresponding performance requirements as listed in \cref{sec:performance-requirements} are fulfilled.

This test beam was the first one during which the full chain of front-end electronics along with the new DAQ system was tested.
In addition, all SciFi related components in use were (near-)production versions.
Using this setup, it was possible for the first time to read out the 32 optical links at the LHC clock frequency of \qty{40}{\mega\hertz} resulting in a total data rate of about \qty{150}{\giga\bit/s}.
Getting the two systems to work with each other was not a trivial task and required for extensive preparations, which have been a significant part in the course of this thesis.
The software that had to be developed for it also formed the basis for the commissioning of the front-end electronics.
More details on this follow in \cref{ch:commissioning}.