Operational experience with the CMS pixel detector in LHC Run II

The CMS pixel detector was repaired successfully, calibrated and commissioned for the second run of Large Hadron Collider during the first long shutdown between 2013 and 2015. The replaced pixel modules were calibrated separately and show the expected behavior of an un-irradiated detector. In 2015, the system performed very well with an even improved spatial resolution compared to 2012. During this time, the operational team faced various challenges including the loss of a sector in one half shell which was only partially recovered. In 2016, the detector is expected to withstand instantaneous luminosities beyond the design limits and will need a combined effort of both online and offline teams in order to provide the high quality data that is required to reach the physics goals of CMS. We present the operational experience gained during the second run of the LHC and show the latest performance results of the CMS pixel detector.

A : The CMS pixel detector was repaired successfully, calibrated and commissioned for the second run of Large Hadron Collider during the first long shutdown between 2013 and 2015. The replaced pixel modules were calibrated separately and show the expected behavior of an un-irradiated detector. In 2015, the system performed very well with an even improved spatial resolution compared to 2012. During this time, the operational team faced various challenges including the loss of a sector in one half shell which was only partially recovered. In 2016, the detector is expected to withstand instantaneous luminosities beyond the design limits and will need a combined effort of both online and offline teams in order to provide the high quality data that is required to reach the physics goals of CMS. We present the operational experience gained during the second run of the LHC and show the latest performance results of the CMS pixel detector.

K
: Detector alignment and calibration methods (lasers, sources, particle-beams); Performance of High Energy Physics Detectors; Pixelated detectors and associated VLSI electronics in the front end and the analog address level calibration of pixels. In order to cope with the enormous data rate, the detector features analog zero-suppressed readout, which means that pixels with accumulated charges below a certain threshold are discarded. Thresholds are minimized in an iterative way to maintain high efficiency. The calibrations further include pulse height optimization and optical conversion adjustments. Then the timing of the readout is measured and tuned during the first collisions (figure 1) in a fine delay scan where the best setting is determined by simultaneously maximizing hit -and charge collection efficiency.

Good detector fraction
All modules were repaired during LS1, but in the commissioning eight modules showed faulty behavior ( figure 2). In addition, during the operation in nominal magnetic field of 3.8 T, a sector (a power and a read-out group) in a half-shell of the barrel signaled an under-voltage problem (figure 2b). There was no solution found to recover all modules but a stable configuration was found that allowed to keep layer 1 and 3 modules functional. The good detector fraction during stable operation is 98.3% for the barrel and 99.98% for the forward pixels where only a single readout chip (ROC) is not operational (figure 3).

High voltage bias scan
In order to achieve a better resolution, the high voltage bias of the barrel modules was increased from 150 V to 200 V. During first collisions, a high voltage scan [5] was carried out, and the cluster properties and hit efficiency were measured to validate the new setting (figure 4), which is expected to remain optimal throughout the year.

Lorentz angle
In CMS the magnetic field is parallel to the beam which causes the charge carriers to drift within the pixel sensors. This in turn creates larger clusters, which allow an improved position estimate and better resolution. For barrel layer modules the direction of the Lorentz drift is perpendicular to the beam. For them, the so-called grazing angle method [6] is used. In this procedure, the drift of electrons within the sensor is plotted against the production depth (figure 5a) and fitted with a linear function from which the tangent of the Lorentz angle is extracted. The forward pixel modules are slightly rotated about the radial axis of the blades in order to allow charge drift in these modules as well. There, the Lorentz angle is measured with the minimum cluster size method (figure 5b). The goal is to find the minimum of the V-like shape of the average cluster size vs. incidence angle. The cotangent of this angle equals the tangent of the Lorentz angle.
-3 - The measurement in the barrel is shown in figure 6a for various operational temperatures as a function of the applied bias voltage. These measurements are important calibration constants in the offline hit reconstruction. They are monitored since the beginning of the detector's operation (figure 6b). A monotonously increasing trend with accumulated radiation is observed.

Pixel thresholds
A general goal is to minimize pixel thresholds in order to maximize the cluster size. For this reason, the average pixel thresholds and noise are important quantities that are monitored throughout the year. The measurement is done by injecting varying amounts of charge to each pixel using an -4 - internal circuit and determining the point where the fitted error function reaches 50% efficiency. In 2015, the threshold of new modules rapidly increased with irradiation (figure 7a). A similar trend was observed also in Run I [5]. The noise (figure 7b) quickly reached similar values as that of the old modules, which no longer experience such large changes due to irradiation.

Hit resolution
Two algorithms are used in CMS for the estimation of the hit positions. One, called the generic algorithm, estimates the hit position based on the charge of the first and last pixel in a cluster. It gives a very fast and relatively accurate hit position estimate (this is used in the high level trigger). The other, called the template algorithm, uses predefined cluster shape templates to find a very accurate hit position. It is based on a detailed simulation (PIXELAV [7]) of the expected pixel -6 - cluster shapes in x and y projections for different track incidence angles. PIXELAV also describes radiation damage induced changes. The hit resolution was measured on layer 2 and disk 1 with tracks that have hits on layer 1, and layer 3 or disk 2. The tracks were then re-fitted without the hit in the middle, and the residual between the original and the interpolated hit positions were measured. The residual distribution is then fitted with a student-t function. In order to subtract the effect of the two other layer measurements, the width of the fit is divided by √ 3/2 to get the intrinsic pixel resolution. The template algorithm (which is used in the final fit of the reconstructed tracks) showed superior performance in all cases. A larger improvement was seen in the barrel pixel measurements (figure 11) and a minor one for the forward pixels (figure 12). The resolution mesurement is seen to depend on the running conditions. In the end-of-year reconstruction, a better alignment was achieved which improved the first seven points significantly.

Hit efficiency
The hit efficiency (defined in [8]) is also monitored throughout the year. As seen in previous measurements in Run I [9], there is an efficiency loss due to a design limitation of the read-out chips in the current pixel detector. For high instantaneous luminosities, the internal buffer of the readout chips tend to overflow with data, which causes dynamic data losses. The losses are measured in different LHC running conditions. The detector was designed for luminosities up to 10 34 cm −2 s −1 , which was reached and surpassed later in 2016. The measurements are shown in figure 13. The innermost layer reached inefficiencies of multiple percents, while the other layers and disks showed better performance with efficiencies above 99% in general. The efficiency of layer 1 is shown in figure 14 for various number of colliding bunches in the LHC fills. In order to account for these inefficiencies, double column losses were incorporated in the CMS Simulation [10].

Single Event Upsets
Single Event Upsets (SEUs) are changes in the state of control registers of the detector caused by ionizing particles. They can degrade or interupt data taking. The typical effect is the disappearance of single pixels (which have a negligible effect) or all clusters in readout chips or larger detector fractions. These intermittent problems are called soft errors, which can be cured by reprogramming the detector. In Run I, several recovery mechanisms were introduced to deal with SEUs [5].
The average rate at which read-out chips turn fully inefficient was recently estimated to be one chip in every 2 pb −1 , using measurements in data. A previously introduced recovery mechanism successfully recovered them in cases where a rapid series of errors were detected on a specific channel. Recently an extra automatic recovery mechanism was introduced that fires after reaching a threshold of 50 pb −1 since the last occasion.  Hit efficiency of the innermost pixel layer as a function of LHC bunch crossing number for different number of colliding bunches in the ring. The average number of inelastic proton-proton collisions (pile-up) was between 12 and 15. The first colliding bunch is preceded by the abort gap (which is comprized of always empty bunches between 3300-3600). After the abort gap, there is enough time for the internal buffers of the readout chip to clear and hence recover full efficiency. For lower number of colliding bunches, there are also smaller gaps between the bunch trains where the efficiency recovers partially.

Conclusions
The CMS Pixel detector was successfully commissioned and calibrated after the Long Shutdown 1. During Run II, it showed an overall good performance that matched design expectations. The detector maintained an excellent resolution that initially even surpassed the Run I measurements due to the lowered operational temperature and increased barrel bias voltages. The hit efficiency, was above 99% except for the innermost layer which saw multiple percent losses. In 2017, when the LHC luminosities are expected to increase further, the upcoming upgrade of the detector is -9 -

JINST 11 C12057
expected to cure the inefficiency of the current detector. This will enable to maintain and surpass the physics goals of the CMS collaboration.