Efeito de eventos únicos em transistores MOS: classificação dos eventos via redes neurais profundas

Electronic devices are susceptible to defects caused by ionizing radiation, and the use of these devices is increasingly required in embedded applications that operate in aggressive environments (presence of radiation) such as space, nuclear reactors and particle accelerators. Among the most damaging defects are Single Event Effects (SEE). The effect is caused by a single ionized particle that, depending on several factors, can cause logical inversions in digital electronic devices, or even render the device inoperative. The study of these phenomena is of great importance in the creation of national technology, as they are basic requirements for generating resistant components radiation. Through unprecedented experiments in Brazil, involving the CITAR Project (Integrated Circuits Tolerant to Radiation), the appropriate environment was created to carry out these studies, since for the reproduction of these phenomena it is necessary to use a particle accelerator that is capable of generate heavy ion beams with low flux. In this work, the results obtained from the ionizing particle radiation experiment in a MOSFET p-type transistor are evaluated, including the creation of a simulated representation of the real device, using the SENTAURUS tool. Emissions of heavy ions in the component were simulated with the same characteristics as the ion beams used in the laboratory, with the expectation of obtaining the same response generated by the real device. Finally, through machine learning techniques, an algorithm was created capable of classifying the different events recorded during field experiments, as well as evaluating spurious signals that make up the data obtained. As a result of the simulations, we approximate the simulation of the 3N163 device to the electrical characteristics presented by the real devices, and through the training of a deep neural network using the data measured in the field capable of classifying 97% of accuracy