The Mu3e Experiment searches for the lepton-flavour violating decay of a positive muon to two positrons and an electron. A detailled description of the experiment can be found on the homepage at the Paul Scherrer Institute (PSI).
In order to reconstruct the decay electrons as precisely as possible, we are building a silicon detector with 300 million pixels. To reduce multiple scattering in the detector material, the sensors will be thinned to 50 microns, which is made possible by the novel HV-MAPS technology, which we also use in P2.
The experiment is designed to reach a sensitivity of one in 1016 muon decays, which requires observing two billion decays every second. This in turn leads to a data rate of about 1 Tbit/s. The task of our group is to read out these data and to reconstruct the electron tracks with the best possible precision. In order to achieve this with the necessary speed and within a limited space and power budget, we use programmable logic (FPGAs), optical links and computers with powerful graphics processing units. Currently we are active in the following areas:
Using a detailed simulation, we optimize the detector design. This requires precise modelling of muon decays as well as the geometry and material properties of the detector. In close collaboration with our partners in the Mu3e collaboration, we work to find solutions that are both performant in terms of the physics reach and technically feasible.
In order to verify the simulation, we regularly perform test beam measurements at MAMI in our institute and at external sites such as DESY or PSI.
The simulation also allows us to predict data rates, which in turn shapes the design of the read-out electronics. The electronics design is then fed back into the simulation in order to estimate the effects of dead time, buffer overflows etc.
One of the main challenges for the Mu3e experiment is the reconstruction of up to two billion particle tracks every second. This with limited computing power, but still aiming for ultimate precision.
One read-out frame will contain up to one hundred electron tracks. Reconstructing these tracks from hundreds of hits in the pixel detector without assigning hits to the wrong track is a challenging problem of pattern recognition. We attempt to tackle this challenge using fast and efficient algorithms tailored for FPGAs and graphics processing units.
The next step is then determining the electron momenta through a track fit. As the tracks are strongly distorted by multiple scattering in the detector material, also this step poses a challenge.
The final Mu3e detector will produce about a Tbit of data per second. Reading out these data is one of the tasks of our group. We use FPGAs directly on the detector to synchronize the data and perform zero-suppression. Optical links then forward the data to read-out boards, where another set of FPGAs serves as switches, assuring that every reconstruction PC sees data from the complete detector. We prepare the firmware for all these FPGAs, test the fast links and are involved in the design of the required printed circuit boards. Within the PC we develop firm- and software for fast data transfers from the optical receiver to the memory of the graphics card via direct memory access.
Between the raw data of the experiment and the final measurement lies a careful data analysis. Our group is preparing for this by analysing simulated data and test-beam measurements. Thus we can hone our skills and optimize our algorithms in order to obtain the most sensitive measurement of the branching fraction mu -> eee.