Abstract : Nowadays, embedded systems and smart cards are part of everyday life. With the proliferation of these devices the need for security increases. In order to meet this demand, cryptographic algorithms are applied. However, even if the algorithms on mobile devices are secure from a cryptanalytical point of view, the secret they use can be revealed by attacking the cryptographic implementation. Indeed an adversary with physical access to the device can benefit from its characteristics or influence its behavior. Methods that observe the activity of a device are considered as passive attacks. In contrast, active attacks try to manipulate the computation and benefit from the erroneous results. These last methods are also called fault injection attacks. This thesis deals with fault attacks. It focuses on practical validations of theoretical attack on symmetric cryptographic algorithm using non-invasive attack. First a new global method to inject fault called setup time violation attack on both ASICs and FPGAs has been studied and carried out on different AES implementations. Then local and semi-invasive optical fault attacks by means of laser beam have been performed on a software implementation. Beside this analysis work, some countermeasures have been investigated. It has been notably shown that Differential with Precharge Logics, already good countermeasure against passive attacks, provide excellent resilience properties against fault attacks. We have demonstrated theoretically and shown practically that information hiding (such as WDDL) makes it difficult to mount fault attacks, since faulty outputs reveal no information about the keys.