SoLid: A short baseline reactor neutrino experiment

The SoLid experiment, short for Search for Oscillations with a Lithium 6 detector, is a new generation neutrino experiment which addresses the key challenges for high precision reactor neutrino measurements at very short distances and with little or no overburden. The primary goal of the SoLid experiment is to search for very short distance neutrino oscillations as a probe of eV-scale sterile neutrinos. This paper describes the SoLid detection principle, the mechanical design and the construction of the detector. It then reports on the installation and commissioning on site near the BR2 reactor, Belgium, and finally highlights its performance in terms of detector response and calibration.


Contents
SoLid, or Search for oscillations with a Lithium 6 detector, is a very short baseline neutrino oscillation experiment, located near the BR2 reactor of the SCK ⋅ CEN in Belgium. Its main purpose is to perform a precise measurement of the electron antineutrino energy spectrum and ux as a function of the distance travelled by antineutrinos between the reactor core and their interaction in the detector. These measurements will be primarily used to search for the existence of one or more sterile neutrinos corresponding to mass eigenstates of order ∆m 2 ∼ 1 eV 2 . Secondarily, the shape of the energy spectrum will serve as a reference measurement for electron antineutrinos originating from the ssion of 235 U. In order to achieve these goals, the SoLid experiment aims to detect electron antineutrinos with an average e ciency of 30%, reconstruct their energy with a resolution of 14%/ E(MeV), and obtain an overall Signal to Background ratio (S/B) of order unity.
Operating very close to the reactor core and at sea level, where large cosmic and reactor backgrounds are produced, combined with small installation spaces, represents new challenges in terms of background rejection capabilities. Compared to the contemporary very-short baseline neutrino experiments near reactors [1][2][3][4][5], the SoLid detector has some unique features, which are described extensively in [6]. It uses a nely 3D segmented plastic scintillator, combined with scintillation screens that contain 6 Li. The high segmentation and the pulse shape discrimination allow to actively identify and reduce all categories of background, limiting the need for large passive shielding. Moreover, the materials used, the robustness and compactness are also attractive for future reactor monitoring applications.
After demonstrating the applicability of the composite scintillator technology, a full-scale prototype module, SM1, with a ducial mass of 288 kg was operated near the BR2 reactor in 2015. Based on the performance of the prototype module [7], improvements were made to the original detector design before proceeding to the construction of a 1.6 ton detector in 2016-2017. The SoLid detector installation was completed in February 2018 and was successfully commissioned near the BR2 reactor in spring 2018. In this paper, we will rst give a complete description of the SoLid detector: its detection principle, its mechanical design, the construction phase and the quality assurance process. We will then describe the dedicated front-end electronics and the data acquisition system. In the third part, we will present the BR2 reactor core near which the SoLid experiment operates. Finally, we will present the data taking operation and describe how the detector response is simulated and how the experiment is calibrated in-situ.

Detector layout and design 2.1 Detection principle
The SoLid detector is designed to be a highly 3D segmented detector (8000 voxels/m 3 ) based on a dual scintillation technology. Electron antineutrinos will interact primarily in the active detector volume via inverse beta decay (IBD) on hydrogen nuclei, producing a positron and a neutron in the nal state:ν e + p → e + + n. Experimental approaches use the coincidence technique, which consists of detecting both the positron and the neutron, within a short time window, typically up to hundreds of microseconds [8]. The neutron generally thermalizes via elastic collisions in the detector, after which it can be captured by nuclei with a high neutron capture cross section. As such it typically induces a scintillation signal that is delayed in time with respect to the scintillation light caused by the positron and its corresponding annihilation gamma-ray photons. The time delay between the two signals can be tuned by the choice of neutron capture elements and their concentration and distribution in the detector. Neutrino oscillation experiments typically vary in their choice of scintillator, the neutron capture element, and the way these are incorporated in the scintillator.
SoLid opted for a combination of two scintillators. One is polyvinyl toluene (PVT), a relatively cheap plastic scintillator that is generally easy to machine in any desired shape or geometry, and the other is ZnS(Ag) used together with 6 LiF to capture thermal neutrons via the reaction: 6 3 Li + n → 3 1 H + α (4.78 MeV), (2.1) for which the decay products in turn induce scintillation in the ZnS(Ag) scintillator. The PVTbased scintillator is of the type EJ-200 produced by ELJEN Technology. It is a general purpose plastic scintillator that emits on average 10 000 photons per MeV of energy deposited by electrons of 1 MeV in the blue-violet wavelength band with a peak emission wavelength of 425 nm. The choice of PVT is mainly motivated by its good light yield and its linear energy response over a wide range of energies ranging from 100 keV to several MeV. It combines a long optical attenuation length with a scintillation pulse decay time of 2.1 ns. The 6 LiF:ZnS(Ag) scintillator for neutron detection is produced by SCINTACOR, in the form of thin screens. These so-called neutron detection screens, emit photons at a peak emission wavelength of 450 nm. The nature of the neutron capture reaction and the longer scintillation decay time of 10 microseconds for the 6 LiF:ZnS(Ag) scintillator allows for a pulse shape discrimination between signals induced in the neutron detection screens via nuclear interaction, hereafter denoted as NS, and signals induced via electromagnetic processes in the PVT, denoted as ES.
The detection technology, the materials and the geometrical arrangement of the main components of the SoLid detector are the same as for the SM1 prototype and are outlined in an earlier paper [7]. Based on the performance of this full-scale prototype and on studies with a dedicated test bench, described in [9], several design improvements were made to optimize the uniformity of the detector response and to maximise the light collection e ciency. These improvements are outlined below.

Detection cell
The basic detection cell consists of a 5x5x5 cm 3 PVT cube, of which two faces are covered with neutron detection screens. Positrons with an energy of 10 MeV travel less than 48 mm in PVT, which implies that the majority of the IBD positrons will be stopped in the same cell as in which they are produced. In order to extract the scintillation photons produced in the PVT or in the neutron detection screens, 4 grooves with a 5×5 mm 2 square cross section are machined in four di erent faces of each cube. Each groove accommodates an optical bre with a square cross section of 3×3 mm 2 that guides the light to an optical sensor at the edge of the detector. All detection cells are optically isolated via a DuPont Tyvek wrapping of type 1082D, whose thickness has been increased from 205 to 270 µm to reduce the optical transparency. The neutron detection screens are cut into squares of 5×5 cm 2 and positioned, using no glues or optical gels, on two adjacent faces of the PVT cube. The two cube faces that are covered with neutron detection screens are the one that faces the reactor core, perpendicular to the Z-axis, and the one that is perpendicular to the X-axis, facing the electronic readout boxes that are mounted on one detector side. A schematic view of a detection cell together with the coordinate system and the position of the neutron detection screens is shown in Fig. 1. The scintillation light produced in the neutron detection screens is optically coupled to the PVT cube via the air trapped in between the two surfaces. The bulk of the neutron detection screens have a 225 µm thick MELINEX-339 re ective backing. The addition of this backing on the neutron detection screens with respect to the prototype module, combined with the overall improved light detection in the cells increases the amplitude of the NS signals and improves the NS-ES waveform discrimination. By doubling the amount of neutron detection screens per cell, the capture e ciency for thermal neutrons in the SoLid detector is increased by a factor 1.3 and the capture time is reduced from 102 to 65 microseconds, compared to the SM1 prototype [7].

Light collection
The scintillation photons produced in each detection cell are extracted and guided by 92 cm long double clad wavelength shifting bres, of type BCF-91A, produced by St.Gobain. One end of each optical bre is covered by a Mylar foil with a re ective aluminium coating, and the other end is coupled to a Hamamatsu type S12572-050P multi-pixel photon counter (MPPC), containing 3600 pixels, arranged in a 3×3 mm 2 matrix. The position of the MPPC and mirror alternates between the parallel bres to mitigate the attenuation of light in the bres and to ensure a more uniform light response throughout the detector (see Fig. 1). The detection cells are arranged into a detection plane of 16×16 cells, where each row and column of cells is read out by the same set of two optical bres, accounting for a total of 64 optical bres per detector plane, as shown in Fig. 2.

Plane & module design
The detection planes, with a cross sectional surface of 0.8×0.8 m 2 , are surrounded by a lining of white high-density polyethylene (HDPE) with a thickness of 46.0 and 46.8 mm, respectively in the vertical and horizontal directions (see Fig. 2). The HDPE bars act as re ectors for neutrons that would otherwise escape the detector. Each plane is structurally supported by a hollow frame of extruded aluminium that has been chrome coated to act as a Faraday cage for the MPPCs and their wirings. Each bre protrudes through the HDPE lining and the frame where it is capped o on each end with two di erent plastic 3D printed caps. One cap holds an MPPC sensor, while the other end holds the aluminized Mylar mirror (see Fig. 1). Optical contact with both the mirror and the MPPC is ensured with a drop of optical gel. The MPPC bias voltage and signal is carried on twisted pair ribbon cables that are routed through the hollow frame and are terminated on one of the frame sides in four insulation displacement connectors (IDCs) each grouping 16 MPPC channels. The front-end electronics, which is described in section 3, is self-contained in an aluminium encasing mounted on one side of each detection plane. Each detection plane is nally covered with two square Tyvek sheets on each of its light sensitive faces to further ensure optical isolation from its neighbouring planes. Frames and their attached readout electronics are grouped together by 10 units to form a detector module, mounted on a trolley (see Fig. 2). Each module can be operated as a standalone detector and has its own power supply and trigger electronics mounted on an overhead rail (see section 3). The SoLid detector currently includes a total of 5 detector modules, accounting for a total of 50 detector planes and corresponding to a ducial mass of 1.6 ton. The front and back planes of the detector are capped with a HDPE re ective shielding with a thickness of 9 cm. Under normal detector operations all modules are closely grouped together with an average spacing of 0.5 mm between two modules.

Cell production and assembly
The construction of the SoLid detector started in December 2016 and took roughly 14 months. The progress of the detection cells (wrapped cubes) production and plane assembly is shown in Fig. 3. The PVT cubes were extracted from 104×52×6.3 cm 3 PVT slabs and individually machined by an industrial partner in Flanders using CNC milling machines, with 0.2 mm tolerance on the cube and groove dimensions. After milling, all cubes were visually inspected for mechanical damage before being transported to the integration site at Universiteit Gent. There all cubes were washed with a light soap detergent to remove lubricant from the milling process and dried overnight. During frame production, two types of neutron detection screens were used. The cells contained in the bulk of the detector are all equipped with neutron detection screens that have a backing with a thickness of 225 µm, while all cells located at the outer edge of each frame received neutron detection screens without re ective backing material.  Each cube was weighed with a digital scale with a precision of 1 mg, before and after being equipped with neutron detection screens and wrapped with Tyvek. The two neutron detection screens for each detection cell were also individually weighted. Each detection cell was marked with a bar code sticker that allows for tracking of the production history in a dedicated SQL database. This database includes the bare and wrapped weights of each cell. During a period of 8 months a total of 13228 cubes were washed, inspected, wrapped and catalogued. Only 3% of all produced PVT cubes were rejected due to quality issues. The accuracy of the weights, combined with the tracking of the production batches revealed a small shift in cell mass during the production process, which falls well within the tolerances used in the cell quality control. The mass distributions of the PVT and neutron detection screens of the 50 detection planes are shown in Fig. 4. The mean weight of all PVT cubes equals 119.7 g with an RMS of 0.1 g, which allows to control at per mille level the proton content. The di erence in mass between the neutron detection screens with and without re ective backing can be observed in Fig. 4. Each of the 50 detection planes was assembled and equipped by hand in its aluminium frame. The position of each MPPC in the detector is stored in the construction database, together with its breakdown voltage.

Quality assurance
Before being integrated in a detection module, each detection plane was tested on the so-called Calipso test bench, shown in Fig. 5 and described in detail in [10]. This test bench consists of a robot that can position a calibration source in front of a SoLid plane with millimetre accuracy. A polyethylene (PE) neutron collimator is added when performing NS calibration in order to increase the neutron capture rate. In addition, a dedicated 22 Na self-triggering calibration head was designed for the ES calibration. The Calipso test bench served primarily as an automated quality control system. As such it provided an early detection of typical construction quality issues such as missing neutron detection screens, bad bre connections, malfunctioning MPPCs and wrong cabling which were all resolved before integration in a detector module. It also allowed to perform an initial test of the electronics and DAQ system before mass production. As a result, for a nominal bias of 1.5 V above each MPPC breakdown voltage, an average gain of about 22 Analogue-to-Digital Conversion units (ADC) per pixel avalanche (PA) was determined with an RMS of 3%. This was further re ned with in-situ equalizations during detector commissioning at the reactor site to achieve a gain equalized to 1.4% across the whole detector (see section 5). The quality assurance campaign with Calipso allowed to have a preliminary calibration of all the detection cells. Calipso measured the light yield by using a 22 Na gamma source in coincidence with an external trigger to remove background. The measured Compton edges caused by the interaction of the 1270 keV gamma rays are used to extract the light yield using two consistent methods based on an analytical t and a template method described in [10]. The average light yield was observed to be larger than 70 PA/MeV/cell corresponding to a stochastic energy resolution of 12% which is consistent with the SoLid physics requirements [9]. The response of the detector to neutrons was also evaluated using a 252 Cf source emitting neutrons with a mean energy of 2 MeV in order to determine the relative di erence in neutron response across the detector and to validate the neutron trigger settings as described in [11]. Because of the dependence on moderation and detector geometry, the absolute neutron capture and reconstruction e ciency is determined insitu, as will be detailed in the section 7. Combining the capture and reconstruction e ciency, the total relative dispersion of this e ciency across the detection cells is 5%.

Container integration
The detector and its electronics are installed in a cooled cargo container with dimensions of 2.4×2.6×3.8 m 3 as shown in Fig. 6 and 7. The container is further customized for thermal insulation and feed through of cooling lines. A dedicated patch panel, located on the side of the container, bundles all the connectors needed for the electronics (power supply, readout), the container instrumentation and the ethernet communication. The 5 detector modules are positioned o -center in the container in order to allow for access and service space (see Fig. 6, 7 and 12). They are mounted on a rail system, that allows for an accurate and robust positioning and alignment (see CROSS calibration system in section 2.6). The electronics are cooled by a chiller system which is described later in section 3.1. Due to the dimensioning of the chiller system and its radiators it is possible to cool down and control the ambient air temperature in the container to a precision of 0.2 degrees Celsius. Under normal data taking circumstances, the ambient temperature of the SoLid detector is kept at a xed value of 11 degrees Celsius. In order to keep the relative humidity of the air inside the detector at acceptable levels the container is permanently ushed with dry air that enters the container at a low ow rate of 5 m 3 /hour. Environmental observables such as pressure, temperature and humidity in the container are constantly monitored by means of a custom sensor network that is controlled and readout by a Raspberry-Pi device. This speci c readout is interfaced with the data acquisition of the experiment. During nominal data taking, the gamma background is monitored by a NaI scintillator (PMT coupled) located inside the container and the airborne radon concentration is monitored by a radon detector.

CROSS calibration system
In order to perform in-situ calibrations of the electromagnetic energy response and of the neutron capture e ciency, a calibration robot, CROSS, is mounted on top of the SoLid detector inside the container, as shown in Fig. 7. First, each of the modules is mounted on a trolley, which is itself mechanically connected by a pivot link to a linear actuator (SKF -CAHB10). This actuator allows to move the module carriage on the rails by a few centimeters, which is needed to insert small radioactive sources between modules during calibration. These displacements are monitored to an accuracy of better than 5 mm by mechanical position sensors mounted on the ground rail of the detector. As such a total of six calibration air gaps of 30 ± 5 mm can be created sequentially on both sides of each module.
The calibration robot that straddles the whole detector along its longitudinal axis is equipped with a holder for radioactive calibration sources as well as four capacitive sensors (BCS M18BBH1-PSC15H-EP02). Each module contains aluminium reference pins and stainless steel screws located on its top. Three capacitive sensors allow to monitor the longitudinal position of the robot by detecting the module reference pins. The fourth capacitive sensor ensures that the air gap is sufciently large by measuring the distance between the stainless steel screws. Once the calibration robot is positioned between two modules, the source holder can further be moved along the Xand Y-axes. As such it can scan an area of 6 cells on the left and right sides of the plane center and 6 and 4 cells respectively above and below the plane center, covering nearly half of the detection plane's surface (see Fig. 7). The radioactive source is installed manually on the calibration arm from the outside of the container and the shielding.

Readout system design
The readout system is custom-made and based on a combination of analogue/digital front-end electronics and Field-Programmable Gate Array chips (FPGA). It brings together compactness, low power consumption (< 1 kW), exibility and high reliability for unattended operation on restricted access. All MPPC signals are equalized, synchronized (< 1 ns) and continuously digitized at 40 Msample/s. The use of zero suppression techniques (ZS), combined with pulse shape trigger algorithms, results in a data reduction factor of around 10 k, down to 20 Mb/s, with negligible dead time (see Tab. 1).
The readout system operates on three levels: plane, module and full detector. Each of the 50 single detection planes has its own readout system, mounted directly on its side within a dedicated aluminium enclosure (see Fig. 8). It contains all the front-end electronics to run in autonomous mode, as described below. Each detector module is equipped with a heat exchanger and a services box that contains a DC-DC voltage converter to power the module, clock and synchronization distribution board, network patch panel and Minnow JTAG programming system. The module clock board (master/slave mode) provides a common clock fan-out to synchronise the ten associated digital boards. A master clock-board allows to run the ve detector modules synchronously. MPPCs are connected via an interface place using twisted-pair ribbon cables that terminate into insulation displacement connectors. (Right) Diagram of a ten planes detector module with its services box and its heat exchanger, placed below to take o heat generated by the electronics [11].
The front-end electronics of a single detection plane consists of two 32-channel analogue boards, a 64-channel digital board, together with a power distribution system and an Inter-Integrated Circuit module that reads out four environmental sensors mounted inside the hollow frame. These environmental sensors monitor temperature and humidity levels throughout the detector. The two analogue boards are connected to the cathodes of the 64 MPPCs of the plane. They provide a common 70 V power supply, as well as individual trim bias voltages (0-4 V) used to equalize the amplitude response of each MPPC individually (see section 5.1). Before being sent to the digital boards and in order to perform more accurate time stamp and amplitude measurements, the fast MPPC pulses (a few ns) are read out in di erential AC coupled mode, ampli ed, band-pass ltered and shaped by a charge integrating operational ampli er to stretch the signal over several digital samples of 25 ns each.
The two analogue boards are connected to a 64-channel digital board for digitisation and trigger. Each digital board has eight 8-channel ADCs, operating at a rate of 40 MHz with 14 bit resolution. Digital boards are controlled and read out over a 1 Gbit/s optical Ethernet connection. A Phase-Locked Loop is included, which allows the digital boards to operate in standalone mode using an internally generated clock, or run synchronised to an external clock signal. Triggers and readout logic are implemented in a Xilinx Artix-7 (XC7A200) based FPGA device. JTAG con-nectors are included for remote rmware programming. Trigger signals from each digital board are propagated to all other detector planes by using two duplex 2.5 Gbit links (copper cables). A complete description of the detector electronics is given in [11].
The entire readout electronics is coupled very close to the detector, within aluminium enclosures, inside the chilled container. Both act as a Faraday cage, providing shielding from outside electronics noise. The top and bottom sides of these enclosures have openings to allow air ow cooling. The electronics are cooled by six fans mounted between the services box and the plane electronics enclosures, pushing air downwards towards a heat exchanger which is capable of removing the 200 W of heat generated by each module (see Fig. 8). The radiator unit is based on circulating water containing 18% propylene glycol, connected to a chiller that operates nominally at a temperature of 5 degrees Celsius. It also acts as an overall cooling source to lower the ambient temperature inside the insulated detector container. As the environment temperature inside the container is maintained to 11 degrees Celsius, MPPC responses are stabilized at 1.4% level and the MPPC dark count rate is reduced by a factor of three compared to operation at room temperature.

Online triggers and data reduction
Multiple triggers and data reduction techniques have been implemented at the FPGA level [12]. The trigger strategy for neutrinos relies solely on triggering on a NS signal. As the NS scintillation process is characterized by a set of sporadic pulses emitted over several microseconds (see section 2), the NS trigger algorithm involves tracking the time density of peaks in the waveform [11]. All algorithm parameters have been optimized during deployment: the amplitude threshold on waveform local maxima to be counted as a peak is set to 0.5 PA, the size of the rolling time window is xed at 256 waveform samples (6.4 µs) and the number of peaks, required in the window, is set to 17 (see Fig. 9). These default values correspond to a trigger e ciency of 75% and a purity of 20% during nominal reactor ON periods. The 80% non-neutron triggers are mostly muon signals, which can be distinguished using an o ine identi cation (see section 5.2 and 7). For each NS trigger, a large space-time region is read out in order to encapsulate all signals from the IBD interaction. Three planes are read out on either side of the triggered plane, with a large time window of 500 µs before the trigger and 200 µs after the trigger. The NS trigger rate, which does not change signi cantly depending on reactor operation, uctuates around 80 Hz corresponding to a data-rate of 15 MB/s (see Tab. 1).
Two additional triggers are also implemented to measure background and to survey the detector stability. A threshold trigger has been implemented to record high amplitude ES signals, such as muons. The default physics mode threshold is 2 MeV with a X-Y coincidence imposed. This gives a trigger rate of about 2.1 kHz and data-rate of 2 MB/s during nominal reactor ON periods. It decreases by around 10% during reactor OFF periods (see Tab.1 and Fig. 18). A periodic trigger has also been implemented in order to monitor continuously the stability of the MPPCs, as well as any noise contributions. The entire detector is read out for a time window of 512 samples without zero suppression, with a default trigger rate of 1.2 Hz, giving a data rate of 3.9 MB/s (see Tab. 1). The three triggers include storing MPPC waveforms for o ine analysis. A zero suppression value at 1.5 PA, respectively 0.5 PA in NS mode, allows to remove the pedestal contribution, whilst retaining all MPPC signals. It results in a waveform compression factor of around 50 (resp. 500) [11]. Table 1 summarizes the di erent trigger parameters and their associated data rates.   Table 1: Summary of trigger settings and associated data rates during reactor ON physics data [11].
The readout software runs on a disk server, located very close to the detector. It provides 50 TB of local storage, that is split into two data partitions, which are periodically swapped and cleared. All the data are rst transferred to the Brussels HEP Tier 2 data centre, then subsequently backed up at CC-IN2P3 in France [13] and at Imperial College in the UK using GRID tools, which are used for o ine processing and simulation production. 4 The BR2 reactor at SCK⋅CEN

The BR2 reactor
The BR2 reactor (Belgian Reactor 2) is a materials testing reactor operated by the nuclear research center SCK ⋅ CEN in Mol (Belgium). Since its start-up in 1963, it is one of the most powerful research reactors in the world and thus plays an important role in nuclear material and fuel R&D. It is also widely used for production of medical isotopes and neutron transmutation doped silicon [14]. The BR2 reactor is a pressurized "tank-in-pool" type reactor, cooled with water and moderated by its beryllium structure and water (see Fig. 10). It has a unique twisted design with inclined channels to obtain a compact core (50 cm e ective diameter, 90 cm height). The BR2 reactor uses highly enriched uranium fuel (HEU: 93.5% 235 U) at powers varying between 40 and 100 MW th . It thus produces a very high neutron ux, up to 10 15 n/cm 2 /s, and provides an intense source of antineutrinos up to about 2⋅10 19ν e s. At the end of the SM1 prototype physics run, the BR2 reactor was shut down for a period of one year and a half, and has undergone a thorough overhaul. The BR2 operation was restarted in July 2016. In practice, the reactor operates at a nominal power of about 65 MW th , for 160 to 210 days per year, during cycles of about three to four weeks (ON period). There are on average 6 cycles of reactor ON periods per year, that alternate with interim maintenance periods of the same duration (OFF period). The Solid experiment takes advantage of the OFF periods to perform calibration campaigns and background measurements.

Detector integration on site
The SoLid detector is located at level 3 of the BR2 containment building in direct line-of-sight of the nominal reactor core center. This is the third detector installed at this location by the collaboration, after the two prototypes, NEMENIX [6] and SM1 [7]. The 50 detector planes are oriented perpendicularly to the detector-reactor axis, and as close as possible to the reactor core. As such, the sensitive volume of the SoLid detector covers a baseline of [6300 mm − 8938 mm] away from the nominal center of the BR2 reactor core (see Fig. 11 and Fig. 12). As the aluminium reactor vessel is totally immersed in water, its radiation is properly shielded. Moreover, at this oor of the containment building, no other experiments surround the detector and all neighbouring beam ports have been shielded with 20 cm thickness of lead. It thus ensures stable and low reactor induced background conditions. Nevertheless, in order to mitigate the atmospheric and cosmic backgrounds, which were determined experimentally with SM1 and compared with a full-chain G 4-based Monte-Carlo simulation [15], a passive shielding surrounds the detector (see Fig. 12 and Fig. 13). The top of the detector is shielded with a 50 cm PE layer made of 2.5 cm thick PE slabs that are staggered to avoid gaps. The PE slabs are supported by a steel sca olding straddling the container and surrounded by a 50 cm thick water wall on the four sides of the container. In order to capture remaining thermalized neutrons, thin cadmium sheets with a thickness of 2 mm are sandwiched between the passive shielding and the container. ). The SoLid position system is based on three Cartesian coordinates along perpendicular axes in a right-handed system. The Z-axis is perpendicular to the detector planes and its direction points away from the nominal center of the BR2 reactor core. The Y-axis points upward towards the zenith, and the X-axis points to the right side of the detector, when facing the reactor.
The environment of the BR2 containment building is continuously monitored and registered by the BR2 Integrated Data Acquisition System for Survey and Experiments (BIDASSE). During SoLid operation, environmental parameters, such as temperature, humidity and pressure, outside and inside the containment building, are constantly monitored. Also the background radiation is monitored using gamma and beta detectors placed in the vicinity of the SoLid container. So far, these variables are used as a cross check of the data coming from the container instrumentation, i.e. environmental sensors, NaI scintillator and airborne radon detector mentionnd in section 2.2. Finally, the thermal power is continuously monitored by two methods (thermal balance measurement and ionization of nitrogen in the cooling water), which allow its determination with 5% absolute uncertainty. It shows the reactor core (red) submerged in water (blue) and the detector geometry including the detector module placement (yellow and blue rectangles) inside the cargo container, the rail system (dark grey rectangle), container insulation (purple) and passive water shielding (dark green).

Neutrino ux modeling
For each cycle, i.e. for a given fuel loading map and operation history, detailed simulations of the BR2 reactor core are performed to calculate the emitted antineutrino spectrum. In addition, the computation of the spatial ssion distribution, combined with a dedicated tracking algorithm, allows to obtain the detector acceptance, de ned as the fraction of emitted antineutrinos that pass through the detector. The geometrical acceptance, which is about 0.11%, depends slightly on the fuel loading map. The emitted antineutrino spectrum is computed using the conversion and summation methods [16]. The conversion method is based on the prediction of the ssion rates as a function of time using a MCNPX (or MCNP6) 3D model of the reactor core interfaced with the evolution code CINDER90 and combined with the converted β − spectra, measured at ILL reactor in Grenoble, France (see Fig. 14). In addition, the MURE code [17] allows to compute the o -equilibrium e ects to adapt the converted spectra to the irradiation time of the antineutrino experiment. The summation method uses the same MCNPX/CINDER90 software combined with the amount of in-core β − emitters and consists in summing all the individual beta branches composing the total spectrum weighted by the beta decay activities [18] (see Fig. 14). Systematic e ects coming from the thermal power uncertainty, modeling uncertainty as well as nuclear data, will also be estimated. The current calculations indicate that at typical power settings of BR2, the SoLid experiment is subjected to IBD interaction rates between 11.5 and 14.5 mHz (1000-1250 events per day).

Backgrounds
The SoLid detector is also subjected to various background processes that contaminate the IBD samples for nal analysis. Because the primary physics trigger is set to detect thermal neutrons interacting in the neutron detection screens, most backgrounds are related to either the production of neutrons via processes other than IBD interactions, or processes that excite the ZnS(Ag) scintillator embedded in the neutron detection screens. Some background processes exhibit a clear time structure between the triggered NS time and preceding ES signals, while others have a random time structure and are called accidental. Reactor independent backgrounds dominate our data sample and can be extracted from data collected during reactor OFF periods validated by dedicated simulation models. Reactor dependent backgrounds are very scarce and are monitored using a dedicated NaI gamma ray detector and with dedicated control samples that are depleted of IBD events, such as samples of triggered NS signals prior to detected ES signals. In all cases we try to validate the background composition and the in uence of selection criteria by using dedicated Monte Carlo simulations, wherever they are available. A detailed description and treatment of these models falls beyond the scope of this paper. Instead we summarize the main background processes and their origin below.
A rst source of neutrons to which the detector is constantly exposed is of atmospheric origin. These neutrons are produced by cosmic ray spallation when high energy primaries collide with atmospheric nuclei. Neutrons can penetrate much further into our atmosphere than the electromagnetic component and are shown to produce a complex energy spectrum [19] ranging from sub-eV to multi-GeV. The ux of atmospheric neutrons is simulated using the Gordon model as described in [19], scaled to the BR2 reactor site elevation and latitude, and cross-checked with the more general purpose CRY generator [20]. The ux contains slow and fast neutrons that induce a di erent response in the SoLid detector. Slow neutrons that enter our detector can, in combination with an accidental coincidence of an ES signal such as those induced by gamma rays, produce signals similar to IBD events. The detector timing and spatial segmentation with corresponding topological selections can largely suppress this background. The passive water shield of 50 cm surrounding the experiment, combined with the Cd sheets placed on the outer walls of the container help to thermalize and capture some of the epithermal neutrons. The fast neutron component is able to penetrate the detector and can induce highly energetic proton recoils resulting in ES signals. If the neutron further thermalizes inside the detector it can be captured and induce a NS trigger. As such it introduces a time correlated background that dominates the selected IBD events samples for ES signals with energy above 5 MeV. This background is mainly suppressed by timing and ES signal multiplicity requirements.
Cosmic ray muons are also known to induce spallation reactions in materials near or inside the SoLid detector that produce neutrons or radioisotopes. The rate of neutron production increases with muon energy and with material density. The rate and spectrum is modelled using the CRY generator [20] by simulating cosmic ray showers on a surface that lies 30 m above the BR2 building and by tracking all shower components through the building and detector geometry. Roughly one third of the spallation neutrons are produced inside the detector, while the rest is created in surrounding structures. The techniques to mitigate the corresponding accidental and time correlated background are similar to those to reduce the atmospheric neutron background. Cosmic muons themselves are used as a calibration tool, as they generally leave a reconstructed track in the detector. In some cases, however, muons can clip the detector edges, leaving an isolated energy deposit that can contribute to the accidental backgrounds in the detector. Muons can also decay in the detector, resulting in the detection of the Michel electron or positron with a characteristic delay corresponding to the muon life time. The rate and spectrum of cosmic ray muons are modelled using CRY, but are cross-checked by other models by Guan [21] and Reyna [22].
Intrinsic radioactivity of detector materials or airborne isotopes are another source of backgrounds. The airborne isotope of 222 Rn can produce several alpha and beta particles along its decay chain. Its presence inside the detector container is therefore monitored by a dedicated Rn detector based on the RADONLITE and RADONPIX technology [23], developed at CERN. Another source of intrinsic radioactivity are trace fractions of Bi isotopes contained in detector materials, in particular the neutron detection screens. The 214 Bi isotope is the most troublesome and is part of the long 238 U decay chain. It decays to 214 Po via β − emission with a half-life of roughly 20 minutes and a Q β of 3 MeV. The resulting Po isotope has a half life of 164 µs and emits an energetic alpha particle that can cause a scintillation of the ZnS(Ag) scintillator of the neutron detection screens. The half life of 214 Po is very similar to the thermalization and capture time of fast neutrons in the SoLid detector. This background, referred to as BiPo, dominates at prompt energies below 3 MeV and is di cult to mitigate. This BiPo background is modelled by generating random decay vertices in the neutron detection screens throughout the detector, followed by the subsequent decays with corresponding half lifes and energies. The use of cube and bre topology information allows to localize the spatial origin of the alpha particle, while timing and energy can be used to tag the ES signal. In addition, also the integrated energy of the NS signal can be used to discriminate neutrons and alphas from the 214 Po decay.
Other backgrounds can be broadly categorized as accidentals and consist of random coincidences of ES signals that are typically induced by gamma rays and thermal neutrons in the surroundings of the detector. The accidental distribution can vary with reactor power, but can be easily extracted from data itself, using negative time di erences between the ES and the NS signals. Accidentals contribute only marginally to the selected IBD events sample.

Channel characterization and equalization
During nominal operations, the gain of the MPPCs is set to around 31 ADC counts per PA, which corresponds to a mean over-voltage of 1.8 V above the avalanche breakdown value of each sensor. This over-voltage setting was optimized for neutron e ciency during the commissioning of the detector at BR2. It is a compromise between photon detection e ciency, pixel cross talk and thermal dark count rate. The amplitude response of the sensors is equalized by an automatic procedure that rst consist of nding the individual break down voltage of each MPPC, which is spread with a standard deviation of around 2 V over all the sensors. For a given channel, the linear relationship between gain and voltage is determined by performing a voltage scan. This procedure allows to equalize the gain of all the channels with a spread around 1.4%, where the dominant uncertainty is the precision of the gain-nder itself (see Fig. 15). MPPC sensors typically have a high dark count rate, which is the main reason why the detector is cooled inside an insulated container. The rate also strongly depends on the over-voltages applied. Under nominal running conditions, i.e. at a mean over-voltage of 1.8 V and at a temperature of 11 ○ C, the mean dark count rate is 110 kHz per channel, which is uniform across the detector. The MPPC pixel cross talk, which corresponds to the probability that a pixel avalanche triggers an avalanche in a neighbouring pixel, also depends on the bias voltage and amounts to 20% for an over-voltage of 1.8 V [11]. Long term trends of the MPPC response are highly stable, as shown on Fig. 16.

Detector operation and data quality monitoring
Run operations are controlled via a dedicated Python-driven web application, the "SoLid Data Quality Monitor" (SDQM). It automatically processes a small fraction of each run ( rst GB) using the SoLid reconstruction and analysis software. Output measurements and distributions of the detector as well as in-situ environmental sensors are read out periodically, as show in Fig. 17 and stored in an online database and is continuously inspected via a web application. The rates obtained from the monitoring database are shown in Fig. 18. The NS trigger rate stays stable irrespective to the reactor operation. Once the muons contamination has been removed, the NS rate is around 18 Hz and is strongly correlated to the airborne radon concentration which is monitored by a Rn detector (see section 2). The transition between the reactor ON and OFF periods can only be seen by the relatively small change in the threshold trigger rate, which is strongly correlated to the gamma rate measured by the NaI detector. The SoLid detector segmentation provides a powerful tool for identifying cosmic muons crossing the detector. Muons deposit their energy in a large number of cells along their path. Their o ine reconstruction thus relies on a spatial clustering that groups all signals from neighbouring bres, an energy requirement to reject low energetic secondary signals, and nally, a requirement on the bre multiplicity. An example of a reconstructed muon track inside the detector is displayed in Fig. 19.
The reconstructed muon rate, which is about 250 Hz, can be used as a standard observable, providing uniformity maps of the detector response and an e ective tool to control the stability over time. As expected, we observe a linear relationship between the muon rate and the atmospheric pressure (see Fig. 20). The tracking algorithm also computes the muon path length in each cell by tting the dE dx distributions. It is then possible to continuously monitor the stability of the detector response during physics mode. As shown in Fig. 21, the variation of the energy scale is below 2% over a data taking period of two months. Lastly, the muon reconstruction also allows to verify the time synchronisation of the detector channels. Indeed, the time in which a muon crosses the detector is negligible compared to the DAQ sampling time and the deposited energy in each cell has to be detected simultaneously. As shown on Fig. 20, the channels are synchronized within 5.908 ± 0.002 ns.

Simulation
The simulation of the SoLid detector is divided in two parts: one part models the energy loss and scattering of particles, including neutrons, in the SoLid detector and the reactor hall, while the second stage models the optical system of the detector, including the scintillator response, the optical transport, the photon collection by the MPPCs and the electronics response.

G 4 model
The rst part, SoLidSim, is implemented using the G 4 simulation library [24]. In order to accurately model the scattering of fast neutrons, the propagation of cosmic showers through the detector and the creation of spallation products in high-Z materials surrounding the detector, a detailed geometry model of the detector surroundings is made. This model, as graphically shown in Fig. 11, is based extensively on detailed blueprints of the reactor building and survey measurements performed prior to detector installation and includes as main features the majority of the concrete and steel structures of the BR2 containment building, including the cylindrical containment building inner and outer walls and dome cap, the concrete oors of level 3, where the detector is located, level 2 below the detector and levels 4 to 7 situated above the detector. Speci c features such as staircases, elevator shafts, crane passways, and access holes are included as well. Special care is taken to model in detail the reactor fuel tank, the water pool and its concrete walls with beam ports including concrete and steel plugs, the 20 cm thick lead shielding wall in between the SoLid detector and the radial beam port facing the reactor core. The inclusion of  these structures can be switched o in the tracking of particles through the detector to save time and computing power for simulations of IBD events or background processes occurring inside the detector. The geometry of the detector includes besides the sensitive volume of the detector, all HDPE neutron re ectors, all metal structures surrounding the sensitive volume, including the electronics housing, the CROSS system, all mounting rails, the container insulation and steel walls, the passive water and PE shielding surrounding the detector and its support sca olding.

Readout simulation
After modelling the energy deposits or the creation of secondary particles in the detector and its surroundings with SoLidSim, the energies deposited in the sensitive volume of the detector are translated in detected pixel avalanches in the MPPCs connected to bres surrounding the energy deposit. This is modelled in a standalone readout software library, ROSim, which is speci c to SoLid. For the optical part, it includes the modelling of the scintillation photon production in the PVT and in the ZnS scintillator of the neutron detection screens, in particular the non-linearity corrections to the energy response using Birks' law, the loss of scintillation photons due to scattering and absorption in the PVT cubes, the neutron detection screens and the wrapping material, the attenuation of the wavelength shifted photons in the optical bres, the re ectivity and absorption losses in the mirrored bre ends. The readout simulation also takes into account the measured dark-count rate, which is generated uniformly across the detector at a rate of 110 kHz per channel. The probability of cross-talk in a neighbouring pixel, of about 20%, is also considered. After an avalanche is triggered in an MPPC, the pixel is insensitive for incoming photons for a short time period. This pixel recovery is modelled by an exponential recovery of the pixel bias voltage with a time constant of 24 ns.
The last stage of the simulation takes into account the shaping and ampli cation of the MPPC signals, and adds a small amount of white noise. This noise is modelled as a random walk in the ADC sample amplitudes around the nominal baseline. The RMS of the noise corresponds to the values measured in data and equals 2 ADC, compared to the amplitude of a single pixel avalanche which is 32 ADC for an MPPC bias of 1.8 V above its breakdown voltage. The pulses are nally sampled with the same frequency and resolution as the SoLid ADCs and the data is stored in the same format as real data for processing by the reconstruction software. An example of an IBD event generated with the SoLidSim software and processed by the readout simulation is shown in Fig. 22 and compared to an observed IBD candidate.

Data taking and calibration
The SoLid detector was commissioned between February and June in 2018, after which it entered stable physics operations. Since then the experiment has been in continuous operation during all subsequent BR2 reactor cycles and refuelling periods. In between reactor cycles, ample time was reserved for in-situ detector calibrations using the CROSS system, described in section 2, with several neutron and gamma sources, as described below. The periods during which the SoLid detector collected physics quality data during reactor on periods is summarized in Tab. 2, and the periods during which calibration data were taken are shown in Tab. 3. The integrated amount of data taking time under various conditions, together with the integrated BR2 reactor power at which the SoLid detector collected physics data, is shown in Fig. 23 over the course of one year of operation.

Neutron calibration
The neutron detection e ciency drives directly the IBD detection e ciency. In SoLid we aim to determine at percent level the relative and absolute neutron detection e ciency of each of   the detection cells. The dedicated NS trigger was optimized during the detector commissioning to ensure the largest possible neutron trigger e ciency, while keeping the data rate sustainable (see section 3). However, this come with a relatively low NS event purity of about 20%. The rst step consists in removing the muons contribution (see section 5.2). In order to reject remaining background, the second requirement is based on an o ine pulse shape discrimination using the integral over amplitude ratio. The results are displayed in Fig. 24. For the standard data taking in physics mode at BR2, the NS signals, whose rate does not depend on the reactor operation, can be well separated from the tagged ES events, with a purity above 99% after the selection requirements are applied.
In order to determine precisely the neutron reconstruction e ciency, two radioactive neutron sources are used: AmBe and 252 Cf for which both activities have been calibrated at the 2% precision level at the National Physical Laboratory (UK). In addition, dedicated G 4 Monte-Carlo simulations are performed to compute the neutron capture e ciencies for neutrons coming from both sources. In turn, the neutron reconstruction e ciency is evaluated cell per cell, with a mean total uncertainty (stat. + syst.) of about 2% (see Fig. 25). The SoLid detector has an average neutron reconstruction e ciency of 73.9 +4.0 −3.3 %. The absolute systematic uncertainties, which are below 5%, are estimated taking into account the di erence in e ciency for the two neutron sources and the uncertainty in the activity of those sources. Thus, we obtain an absolute detec- The red dashed line shows the selection requirement used for the particle identi cation. The right panel presents the projection on the integral over amplitude axis for selected and rejected events [10]. tion e ciency for IBD neutrons greater than 52%, with a relative uncertainty between detector module which are below 2%.

Energy scale
To be sensitive to aν e oscillation, the ES energy reconstruction needs to be measured accurately. In particular, the energy scale and its dependence upon the actual deposited energy should be known at the 2% level. To that end, the SoLid detector response is calibrated using γ sources at various energies.
During standard calibration runs, the energy scale in each detection cell is determined using a 37 kBq 22 Na gamma source. This source is placed at nine di erent positions in each of the 6 detector gaps, using the CROSS system. To reconstruct the total amount of light produced in a given cell, the total number of detected photons originating from the 22 Na decay spectrum per cell must be computed. To perform this operation, coincidences are searched for between the two vertical and the two horizontal sensors coupled to the four bres going through each cell. Finally the four amplitudes are summed taking into account the gain of each MPPC. Gammas from the 22 Na source (511 keV and 1270 keV) interact in the PVT mostly through Compton scattering. In addition, given the granularity of the detector planes, only a fraction of the total gamma energy is deposited within each PVT cube. Consequently, a broad visible energy spectrum needs to be reconstructed and tted for one or more Compton edges.
During the quality assurance process, two methods to tackle the latter issue were developed [10]. The rst method consists of tting the Compton edge pro le of the spectrum by an analytical function based on the Klein-Nishina cross section and the result is compared to the predicted value (see Fig. 26). The second method employs a Kolmogorov-Smirnov test and compares the measured energy spectrum to a G 4 simulated sample varying the energy scale and energy resolution (see Fig. 26). Although the two methods rely on di erent assumptions, the obtained results are consistent within 2% and meet the required energy scale precision [10]. For the standard data taking in physics mode at BR2, an average of 94 PA/MeV/cell was measured without MPPC cross-talk subtraction, which is estimated to be around 20%. The light yield is uniform across the whole detector, as in shown in Fig. 27, and is stable over time, as can be seen in Fig. 28. The variation of the mean value of the light yield and the RMS of its distribution are within 2% over a period of one year. For linearity studies, 207 Bi and AmBe radioactive sources are also used in two detector gaps, in addition to the 22 Na source. The light yield ratio measured with two di erence sources is consistent with what is expected for linear behavior as can be seen in Fig. 28.

Conclusion
The Solid collaboration constructed a 1.6 ton highly segmented neutrino detector based on an a ordable dual scintillator technology in the years 2016-2017. Its design parameters were improved after a measurement campaign with a single module prototype in 2015. As of the spring of 2018 the full size SoLid detector is in continuous operation at the BR2 research reactor of the SCK ⋅ CEN in Belgium. The BR2 reactor is operated with highly enriched 235 U fuel arranged in a very compact geometry, which reduces the uncertainties in the calculation of the incoming electron antineutrino ux and its energy spectrum. The detector has proven to run very stably over long periods of time and can be routinely calibrated with dedicated gamma and neutron sources with an in-situ system. The statistical energy resolution, the energy scale precision and the level of inter-channel response calibration all adhere to or surpass the SoLid design speci cations. A detailed geometry description and detector response simulation have been developed, allowing for a future validation and understanding of the physical and instrumental backgrounds and an optimisation of the neutrino detection and oscillation measurements.