Advanced Algorithms for Silicon Tracking Detectors

Nowadays particle physics experiments rely on position-sensitive silicon sensors for vertex reconstruction, particle tracking and increasingly even for momentum measurement. The Belle II experiment at KEK (Tsukuba, Japan) will investigate CP violation, and applies two different types of silicon sensors, which measure decay vertices of B mesons delivered by the collider SuperKEKB. The innermost Pixel Detector PXD features two layers of novel DEPFET (Depleted P-channel Field Effect Transistor) sensors, which provide precise, unambiguous 2D position measurements with a coarse timing resolution. The next four layers of the Silicon Vertex Detector SVD use double-sided silicon microstrip sensors, which offer 2D position measurements with high timing accuracy, but sometimes can introduce ambiguities. Both sensor technologies will have to team up symbiotically to unleash their full power: precise, unambiguous vertex reconstruction with accurate timing. Primary Research Goal Correction of Position Measurement Bias: Pulse-height readout of the silicon sensors allows for position measurements far more precise than the distance between individual sensing elements (i.e. pixels for the PXD, strips for the SVD). This is done by interpolation, using the signal amplitudes of the involved sensing elements. However, the relation between amplitudes and position often is highly nonlinear, leading to a bias of the measured position. Note that all tasks related to track reconstruction depend on precise position measurements: detector alignment, track finding, track fitting and vertex reconstruction. I plan to join the Belle II tracking group to implement a correction of this effect in the Belle II software framework, and to study its effect on the quality of track fitting and vertex reconstruction. It will be the first time ever that a correction of this effect is applied in a large-scale collider experiment. Stretch goal Combination of PXD and SVD Online Data: During PXD`s long readout time it accumulates a lot of uninteresting background hits, and it is not feasible to store all of them. Using the Fast Hough Transform algorithm one can find and fit tracks in the SVD, and define so-called regions of interest around these tracks` extrapolations to the PXD layers. Only PXD hits inside these regions are accepted to be sent to the data acquisition. This happens at trigger level implemented in fast FPGA logic, and makes the readout of the PXD possible in the first place. It will be the first time that particle tracking is carried out online, before a trigger decision. If time allows, I want to join the activities of the PXD group and contribute my knowledge about silicon sensors in general and the SVD in particular to the project. Both research goals need detailed knowledge about the PXD, including the sensors, the readout chain, and the software framework. Therefore I have to present in Munich, where the DEPFET sensor was invented and is produced, and where the heart of the PXD community resides.

Modern particle physics is a huge endeavor, bringing together numerous people from many different fields of expertise. There are the people running the accelerators, like the LHC (Large Hadron Collider) at CERN (Geneva, Switzerland), others are responsible for the large detectors, which try to make sense of the mess that is a particle collision. For such a large undertaking, it is vital to establish proper communication between the different fields. Only then can you make the most out of the vast trove of information which is generated by such experiments. There are five main fields of expertise needed to run a particle detector: 1. Sensor physics: Sensor hardware and electronic readout. 2. Reconstruction algorithms: Particle identification and track reconstruction. 3. Triggering: Decisions which collision events to store and which to ignore. 4. Data analysis: Searching for e.g. new particles in the stored data. 5. Theory: Predictions of what we should detect. Particle physics experiments nowadays heavily rely on silicon sensors to measure the positions of particles created in the collisions. Lately, silicon sensors are also used to measure the particles energies. While building and operating these sensors is a hardware- centric task, the data processing of the sensors signals is done in software, most often by people who dont know the intricacies of the involved sensors. This research project now sits right at the interface between sensor physics and reconstruction algorithms. My self-set task was to improve the information the sili- con sensors hand over to the reconstruction algorithms. Think of it as a translation process, in which I take into account the special properties of the sensor, and trans- late them into the higher language of the subsequent software. This software expects the sensor to report a particle position and a statement about the uncertainty of said position. It is relatively easy to calculate a position from the sensors response. In contrast, giving a well-founded estimate of the positions accuracy is another matter, because it depends on a number of variables. There are well-established algorithms for calculating the particles position and accuracy, that take into account the non- linearities of the sensors response. Surprisingly to me, they have not been used in large particle physics experiments up to now. So I took one of these algorithms, and implemented it in the reconstruction software of the Belle II experiment in Japan. Furthermore, I took the first steps towards using a neural network for a similar task. This should allow the people working on reconstruction algorithms to have stronger faith in the data they get from the silicon sensors.