Practical analyses of the structures of ultrathin multilayers in
tunneling magneto resistance (TMR) and Magnetic Random Access Memory
(MRAM) devices have been a challenging task because layers are very thin,
just 1–2 nm thick. Particularly, the thinness (∼1 nm) and
chemical properties of the AlOx barrier layer are critical to its magnetic
tunneling property. We focused on evaluating the current TEM analytical
methods by measuring the thickness and composition of an AlOx layer using
several TEM instruments, that is, a round robin test, and cross-checked
the thickness results with an X-ray reflectometry (XRR) method. The
thickness measured by using HRTEM, HAADF-STEM, and zero-loss images was
1.1 nm, which agreed with the results from the XRR method. On the other
hand, TEM-EELS measurements showed 1.8 nm for an oxygen 2D-EELS image and
3.0 nm for an oxygen spatially resolved EELS image, whereas the STEM-EDS
line profile showed 2.5 nm in thickness. However, after improving the
TEM-EELS measurements by acquiring time-resolved images, the measured
thickness of the AlOx layer was improved from 1.8 nm to 1.4 nm for the
oxygen 2D-EELS image and from 3.0 nm to 2.0 nm for the spatially resolved
EELS image, respectively. Also the observed thickness from the EDS line
profile was improved to 1.4 nm after more careful optimization of the
experimental parameters. We found that EELS and EDS of one-dimensional
line scans or two-dimensional elemental mapping gave a larger AlOx
thickness even though much care was taken. The reasons for larger measured
values can be found from several factors such as sample drift, beam
damage, probe size, beam delocalization, and multiple scattering for the
EDS images, and chromatic aberration, diffraction limit due to the
aperture, delocalization, alignment between layered direction in samples,
and energy dispersion direction in the EELS instrument for EELS images. In
the case of STEM-EDS mapping with focused nanoprobes, it is always
necessary to reduce beam damage and sample drift while trying to maintain
the signal-to-noise (S/N) ratio as high as possible. Also we confirmed
that the time-resolved TEM-EELS acquisition technique improves S/N
ratios of elemental maps without blurring the images.