Processamento multicore para reconstrução online de energia por meio de redes neurais

Energy reconstruction in Calorimeters is a process executed in pulses generated in its electrodes to estimate the energy of subatomic particles that go through its material. Non-Gaussian characteristics of the noise in calorimeters operating at high frame rates demonstrate that non-linear estimation methods are more appropriate. Nevertheless, such methods tend to have a high computational cost, which makes their online implementation an inconvenience. Is presented with this work, the implementation of an embedded neural network in a multicore process in FPGA, capable of operating at an acquisition rate above 40 MHz, as well as an implementation in the Calorimeter of the ATLAS Experiment.