6533b825fe1ef96bd1282ed3

RESEARCH PRODUCT

Contribution to the analysis of signals obtained by dynamic photon emission for the purpose of studying very large scale integration circuits

subject

description

Scaling progresses has the benefit of making chips always more powerful. On the other hand, when there is a failure, the analysis of such advanced devices has became more sensitive. The defect localization step of this process is the critical one. Indeed, the aim is to find transistors which dimensions range in several nanometers on a device which surface is several square centimeters.Optical techniques like dynamical photon emission, also named Time Resolved Imaging (TRI), have proved to fit in such context. The later is based on the acquisition and exploitation of photons emitted by a switching CMOS structure. Due to its physical bacground, this tool has a limited invasive effect and can be used to analyze defect generating faults during a dynamical stimulation of the device. The complexity of the chips manufactured in advanced technologies has brought out some physical and technical limitations which can jeopardize analysis performed with this tool. To be more specific, signal over noise ratio can be quite low, so as the spatial resolution compared to the studied structures. In addition, complex circuits require long test sequences, generating huge quantities of photons to analyse. As a conclusion, all of these phenomenon forbid a simple manual procedure if ones expect to extract the emission signature of the defect in such data.The work reported in this thesis aims to develop new approaches of processing at the post-acquisition level, in order to solve or workaround the various aforementioned issues. It will enable the analyst to formulate an even better and more precise diagnosis.The task consists in extracting and synthesizing the information available in large amount of noisy signals. With that superpose in mind, two main approaches have been studied and developed. The first one establish a mapping of one parameter the electrical signals varying through time and space inside the acquisition area. It is based on a mixture of signal processing tools for 2D 1D signals. The second approaches uses data mining. More precisely, it combines clustering to statistical analyses of the resulting classes in order to find an emission event which is unexpected or having unusual properties, suggesting a candidate for failure. These two processes are complementary as they bring different information to the analyst.