August_EDFA_Digital

ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 21 NO. 3 30

RECENT ADVANCES IN VLSI CHARACTERIZATION USING THE TEM (continued from page 27)

spacing in the respective direction. Using PED, strain—or the lack of it—in the channel region can be determined even in the presence of the gate, and compared to predictions by process simulations such as technology computer-aided design (TCAD). The secondmainstreamapplication of PED is orienta- tionmapping of Cu interconnects or other polycrystalline materials. In orientation mapping, the electron beam is rastered over the sample and a diffraction pattern is recordedevery fewnanometers. The recordedpatterns are then compared to a database for the material-at-hand in order to determine the sample orientation at each point. This is illustrated in Fig. 4, which depicts an orientation map of a Cu interconnect structure. Maps such as these can help an FA engineer understand which textures are more prone to electromigration failure than others. SIGNAL PROCESSING FOR DATA ANALYTICS In electron microscopy, the radiation damage that can occur when illuminating the sample often limits the electron dose that can be applied. This in turn limits the S/N ratio of the data and often leads to noisy data maps. This is especially true for radiation sensitivematerials such as ultra-low K (ULK) dielectrics. Inorder toobtaindatawith a lownoisemarginwithout an additional electron dose, smart signal processing can be employed. A dramatic increase in the signal-to-noise

ratio can be achieved by using amathematical procedure called principle component analysis (PCA). [4] In PCA, the observed variables are reduced to a smaller number of principle components that account for most of the vari- ance found, in this case, signal variance. This technique significantly reduces noisewithout compromising spatial or energetic resolution. The effect of PCA on a Mn map of a Cu interconnect structure is depicted in Fig. 5. In this Cu line, a Mn-doped Cu seed layer was deposited before the metal line was filledwith electroplated Cu. As Fig. 5 shows, a typical EELS measurement results in a noisy Mn map. The electron exposure had to be limited in order not to destroy the sur- roundingULKdielectric. Applying PCA for noise reduction, however, clearly depicts theMn-doped seed layer without any increase in electron dose. Advanced signal processing analytics like PCA canhelp the FA engineer to “squeeze out” information from data, whichmay appear hopelessly noisy to the naked eye. It is also used regularly to obtain concentrationmaps andpro- files of radiation sensitivematerials like ULK dielectrics. [5] ELEMENTAL ELECTRON TOMOGRAPHY The introduction of fast elemental mapping has sig- nificantly propelled TEM failure analysis forward. The holy grail, however, would be full 3D mapping of all the elements comprising a device or defect. This goal can be achieved using chemical electron tomography in an analytical TEM.

Electron tomography in theTEMhasbeen used in the semiconductor industry for a couple of years. A sample is imaged over an extended range of tilt angles using a special specimen holder. The multitude of images or projections then undergoes amathemati- cal back-projection process to yield the 3D structure of the sample. Similarly, chemical sensitive tomography is realized by replacing the scanning TEM images with chemically dispersive maps, either EDX [6] or EELS maps. [7] From these maps, 3D data can be extracted element by element. An EDX tomogram is recorded as follows: The sample is tilted around a fixed axis from - 70° to + 70° in steps of 5°, and elemental maps (similar to Fig. 2) are recorded at each step. Any deviation in imaging conditions

Fig. 4 Typical orientation map of a Cu interconnect structure obtained with PED. The grain orientation in each direction is color coded according to the stereographic projection below. The mean grain size for this interconnect segment is 39 nm.

edfas.org