† Corresponding author. E-mail:
Electron magnetic circular dichroism opens a new door to explore magnetic properties by transmitted electrons in the transmission electron microscope. However, obtaining quantitative magnetic parameters, such as spin and orbital magnetic moment with element-specificity, goes a long way along with the development and improvement of this technique both in theoretical and experimental aspects. In this review, we will give a detailed description of the quantitative electron magnetic circular dichroism (EMCD) technique to measure magnetic parameters with spin-specificity, element-specificity, site-specificity, and orbital-spin-specificity. The discussion completely contains the procedures from raw experimental data acquisition to final magnetic parameters, together with the related custom code we have developed.
The newly developed techniques in the modern advanced transmission electron microscope (TEM) aim to reach the ultimate limitation of spatial resolution and energy resolution to resolve the subtle details and fine changes of atomic structure, chemical composition, and electronic structure. However, magnetic information is usually rarely explored to keep pace with these rapid developments because of the limited magnetic characterization techniques throughout the history of this field in the TEM. Overall, several magnetic imaging techniques are involved, namely Lorentz microscopy, off-axis electron holography, and differential phase contrast.[1,2] There is no denying that they have provided plenty of magnetic information in different ways. Still, the pursuit of new techniques with much higher spatial resolution and more detailed magnetic information has never stopped though it is developing slowly.
In 2006, Schattschneider et al.[3] for the first time demonstrated that an electron magnetic circular dichroism (EMCD) signal can be detected in the TEM by transmitted electrons after his proposal of using electron energy-loss spectra (EELS) based on specific diffraction geometry in 2003,[4] which is reported as the EMCD technique. As a counterpart of x-ray magnetic circular dichroism (XMCD),[5,6] based on the transmitted electrons instead of the expensive synchrotron radiation source, EMCD provides the same magnetic parameters with element specificity. Paired with the complementary analytical methods for crystallographic, electronic, and chemical information at atomic scale in modern advanced TEM, EMCD has become a new powerful technique for the comprehensive study of magnetic materials at nanometer scale in recent years.
EMCD differs from previous magnetic characterization methods in the TEM as it is presented as a spectroscopy technique. It provides a new insight of magnetic measurement in the following three aspects. Firstly, it is the same as XMCD with both element specificity and spin specificity, which is an advantage comparing with these phase contrast based on magnetic imaging techniques that are bulk sensitive. The element with different spin orientation has definitely resolved a deeper understanding of magnetic behaviors. Secondly, EMCD provides intrinsic magnetic parameters, spin and orbital magnetic moment, as it can directly probe the spin characters of electrons. Note that phase contrast methods reply on the phase shift caused by magnetic potential, which is an assemble of spin electrons with interaction. Finally, the new record of spatial resolution for EMCD has just been broken from ∼1 nm[7–9] to atomic-plane resolution by introducing the chromatic-aberration-corrected TEM[10] or using the beam-shift EMCD technique.[11] This record for the former one should be the best one among reported results for all magnetic characterization techniques in the TEM.[10] In addition, EMCD is still being developed now and new achievements have already been made, such as EMCD for polycrystalline,[12] in-plane EMCD technique,[13] and EMCD in Lorentz mode.[14,15] Along with these characteristics and advantages, one of the most challenging and attractive issues for EMCD is to obtain reliable quantitative magnetic parameters at nanometer scale, that is the quantitative EMCD technique and the main topic of the present review.
However, EMCD is actually a complicated technique because of its electron-based detection source, additional strict requirements according to its basic principle, etc. Even so, lots of efforts have been devoted to promote the development of the EMCD technique both in theoretical and experimental aspects in past years, such as the improvements of spatial resolution and signal-noise-ratio (SNR),[7–9,16,17] the fundamental understanding of its theory[18–22] and quantitative measurements of spin and orbital magnetic moments with element-specificity and site-specificity.[23–27] Concerning the topic of this review, we summarize these issues for quantitative EMCD magnetic measurement as follows. First of all is the dynamical diffraction effects. In the periodic crystal structure, the incoming electron wave function becomes a superposition of Bloch waves, and dynamical diffraction effects become quite remarkable to affect the intensity and distribution of EMCD signal in the diffraction plane.[21,28,29] In addition, it leads to additional coefficients in the sum rules, complicating the quantitative EMCD technique.[20] As the crystal serves as a beam splitter to create the chiral EELS signal, monocrystalline of the probed area, and dynamical diffraction effects are unavoidable. The second one is the strict diffraction geometry, which is critical to provide the conjugate positions for signal acquisition. The diffraction geometry is limited to the systematical reflection condition, such as two-beam and three-beam cases, and extended to zone axis now.[30] Besides, the asymmetry of different diffraction geometries and its influence on the quantitative magnetic measurement has already been discussed, which refers to the accuracy of quantification.[23,25,31–35] The third one is the notoriously low intensity and SNR of EMCD signals. A high-quality signal leads to highly reliable quantitative measurements. Optimizing experimental conditions and improving SNR will be of significant benefit to quantitative EMCD technique. The last but not the least should be the complex and tedious data processing and signal extraction. There are some pitfalls here and much attention should be paid to minimize the errors.
In this review, we will give a detailed description of procedures on quantitative EMCD magnetic measurement, along with the custom code we have developed to simulate dynamical coefficients, process the raw data, extract intrinsic magnetic chiral dichroism signals, and calculate magnetic parameters and errors. The flowchart is displayed in Fig.
The experimental EMCD signal suffers from dynamical diffraction effects, thus the quantitative relationship between EMCD and dynamical diffraction theory should be established to calculate the dynamical coefficients for the extraction of intrinsic MCD signals. In this part, the basic theory will be introduced first. Then, the input/output parameters of the custom code for calculating dynamical coefficients will be discussed, taking the example of YIG under different experimental diffraction conditions.
The EMCD signal contains two EELS spectra from conjugate positons. The theory of dynamical diffraction effects and the EMCD technique has been established over the past few years. Two parts are involved: dynamical diffraction effects and mixed dynamic form factors (MDFFs). The details can be found in Refs. [18] and [19]. Based on their theory, the final double differential scattering cross-section (DDSCS) at opposite chiral positions of ‘+’ and ‘–’ in the diffraction plane are given by our group as follows:[24,25]
Here we denote
Based on the theoretical frame described above, several kinds of software have been developed, such as the Bloch wave (BW) by Loffler et al.,[36] the BW + Matlab code by our group,[24,25] and also the Modified Automatic Term Selection (MATS) algorithm by Rusz et al.[37,38] In particular, the algorithm for MATS is optimized to avoid summation of negligible terms and to improve the calculation speed. Moreover, MATS is able to deal with the case of convergent beam, e.g., in the scanning TEM (STEM) mode,[38] rather than the assumption of plane-wave illumination in the case of BW software. Here, we will focus on the case of plane-wave illumination with our own developed Matlab code together with the BW software.
The detailed expressions of dynamical coefficients are presented in Eqs. (
After that, our custom Matlab code is used to calculate dynamical coefficients based on Eqs. (
To present the characters of site specificity and spin specificity for the quantitative EMCD technique, it is better to choose the magnetic materials with the same element sited at different crystallographic positions. Here, we take YIG as an example. In addition, YIG has a very complex crystallographic and magnetic structure, and it is suitable to demonstrate the generality of the quantitative EMCD technique. YIG has a ferromagnetic and garnet structure. Yttrium ions are located at the dodecahedral sites, and iron ions are located at octahedral (oct) and tetrahedral (tet) sites, respectively. The magnetic moments of octahedral and tetrahedral Fe ions are antiparallel in alternate {111} planes. The magnetic structure can be written as
Here, we choose three kinds of diffraction conditions to meet the requirement of quantitative measurement for the enhancement of Fe at octahedral and tetrahedral sties, that is, (4-44) three-beam and two-beam cases,
In the EMCD experiment, we take two EELSs at particular positions in the diffraction plane and the EMCD signal is the difference of these two EELS spectra. The simulated distribution of EMCD signals in the diffraction plane under certain diffraction conditions will provide a practical guide to acquire an EMCD signal with high intensity and SNR. In this part, we will talk about some key points in the EMCD experiments and data processing procedures to get EMCD signals.
EMCD is based on critical diffraction geometry, thus the accurate diffraction condition is the determinant precondition. The Kikuchi lines are firstly used to tune the orientation close to the expected orientation. For fine alignment, beam tilt is often much more effective. The diffraction pattern should be checked before and after the signal acquisition to make sure that the diffraction conditions are not changed during the experiments, which might be caused by sample drift or damage. The sample thickness plays a decisive role and the larger thickness generally will lead to a less obvious or even disappearing EMCD signal. The optimized thickness can be deduced from simulations as mentioned in Section
The EMCD signals are acquired at two conjugate positions, so the detector positions are very strict. The objective aperture (OA) or entrance aperture (EA) of the gatan image filter (GIF) system can be used to select the chiral positions as shown in Fig.
For the detector positions, the asymmetry has already been emphasized before both in the signal acquisition and quantitative measurement.[25] To meet the requirement of symmetry, the dynamical coefficients for nonmagnetic signals should be equal for the ‘+’ and ‘–’ positons to make sure that the nonmagnetic signals are canceled out for pure EMCD signals. Then the magnitude of asymmetry is defined as the relative error of dynamical coefficients for nonmagnetic components,
After the EMCD experiments, we have only got the raw data of paired EELS. First of all, energy calibration should be done for all the spectra. If the dual EELS GIF system is available, it is easy to do the calibration with respect to the low loss peak. Otherwise, the oxygen pre-edge or the onset of L3 edge of Fe is used. After that, the following whole processing procedures are displayed in Fig.
For each group of data, the plus and minus EELS should be firstly processed with pre-edge background subtraction. The backgrounds for the original EEL spectra at ‘+’ and ‘–’ EEL spectra are fitted in the pre-edge region following the power-law function of
The plural scattering will also influence the measurement of magnetic parameters.[40] The removal of plural scattering for ‘+’ and ‘–’ background-subtracted edges is conducted in the Digital Micrograph software or our custom Matlab code using the Fourier-ratio deconvolution with a zero-loss modifier function[41] as shown in Fig.
Although the ‘+’ and ‘–’ EELS spectra are acquired at the same conditions, such as acquisition time and collection angle, a slight variation of intensity exists and should be removed. At last, the ‘+’ and ‘–’ deconvoluted edges are normalized by the integration of the intensity in a post-edge window between 50 eV and 100 eV after the onset energy of the L3 edge to against the effects of asymmetry, based on an assumption that in the post-edge region the magnetic signal is negligible and only non-magnetic spectral components remain. By subtracting the two spectra, the EMCD signal is obtained in Fig.
The experimental EMCD signals still contain the dynamical coefficients and only the ratio of orbital and spin magnetic moment can be obtained based on the EMCD sum rules. To extract the intrinsic MCD signals for spin and orbital magnetic moment separately, the further processing is necessary. In addition, due to the low SNR of EMCD signals, plenty of datasets should be taken to improve the accuracy of quantitative magnetic measurement. The lowest SNR and standard of data volume are not evaluated here. This is because the experiments here are conducted in the TEM mode. The line or area scanning of EMCD signals in the STEM mode is required to discuss these issues, in which case the signals with different SNR can be extracted from the datasets. This part should be quantitatively discussed in the future. Therefore, normalization of all datasets and statistical methods are introduced in this part.
A series of experimental EMCD spectra will be acquired with different intensities at different positions. It is obvious that signals with different intensities will lead to an inconsistent SNR. Therefore, it is better to keep the total intensity at the same level. To put them together the normalization is needed. Note that the normalization here is different from the post edge normalization for EMCD signals above. Based on Eq. (
The matrix form of Eq. (
Now the intrinsic MCD signals free from dynamical diffraction effects are obtained, similar to the XMCD signals. Therefore, it is ready to apply the sum rules for magnetic parameters.
Here, we will talk about the sum rules first. The XMCD technique has its own sum rules to get quantitative magnetic parameters based on the experimental spectra,[43,44] while it is not entirely suitable to EMCD, for which the dynamical diffraction effects should be taken. In 2007, the analytical form of sum rules for EMCD is derived and the quantitative magnetic parameters, the ratio of spin, and the orbital moment for the magnetic element are obtained after applying it to experimental results.[20,21] The formula of EMCD sum rules is similar to XMCD sum rules. However, it should be noted that the orbital and spin magnetic moments are expressed explicitly in terms of both experimental spectra and dynamical diffraction coefficients, which is not the same case in XMCD. The sum rules used here are the formula of XMCD (Eqs. (
Applying the sum rules to the intrinsic MCD signals, the magnetic parameters of YIG are obtained as listed in Table
The error analysis is necessary to evaluate the reliability of quantitative measurement and the error bars from different sources are given to help understand and improve the accuracy of the quantitative EMCD technique. The estimation of these errors involves the deviations from data pre-processing, the noise in EMCD spectra, the measurement of sample thickness, and the calculations of theoretically obtained dynamical coefficients. The total error is obtained by adding them together. We will give a respective discussion in the following.
For the data pre-processing, we have used different pre-edge windows, low-loss spectra and normalization regions in each step as discussed in Section
For the noise in EMCD spectra, the standard errors for the magnetic parameters are calculated from a series of independent pairs of EMCD spectra as shown in Table
The measurement error of sample thickness is about ±1.7 nm determined by convergent beam electron diffraction (CBED) and low-loss EELS. The dynamical coefficients corresponding to the thickness varying from 45.0 nm to 48.4 nm (t = 46.7 nm) are calculated to give the error. However, its influence on magnetic parameters is negligible, for the change of dynamical coefficients in this range of thickness is very flat as shown in Fig.
The error of dynamical coefficients is difficult to be evaluated. The possible uncertainties in the imprecise knowledge of experimental dynamical diffraction conditions will undoubtedly contribute to the error of magnetic parameters. These factors, such as the orientation of sample, detector positions, convergence and collection angle, tilt angle from zone axis, partial coherence of the incident beam, and the defects of the sample, will contribute to the uncertainty of dynamical coefficients, while only some of them can be determined currently. For example, the estimation for the positons of detector and tilt angle is done here. The center of the detector is moved to right and left with 0.1G4-44 along the systematic row with a diameter of 0.5G4-44 under the two-beam case during the calculations, which is close to the position of the collection aperture in the experiments. The change of dynamical diffraction coefficients is less than about 6.5%. For the tilt angle from the [1-10] zone axis, we take the angles of 8.4°, 9.4°, and 10.4° respectively to calculate the dynamical coefficients and the variation is less than 3%. Therefore, the actual error should be larger than 9.5%. However, the exact value is actually unknown. As they all have an effect on the dynamical coefficients during the quantitative measurement, we finally use the 20% random error of dynamical coefficients to estimate it (errors in Tables
That is to say, the quantitative EMCD technique is heavily dependent on the simulation of dynamical diffraction effects. It is obvious that the calculation of dynamical coefficients is very complicated. Then it is impossible to obtain accurate dynamical coefficients due to the uncertainty discussed above. Therefore, advanced statistical methods with specific physical significance and boundary conditions, such as the multivariate curve resolution (MCR)[35,42,46] that has already been tested, should be developed or combined together to solve these problems for further development of quantitative EMCD technique in the future.
In this review, the detailed procedures for the quantitative EMCD technique are presented taking the example of YIG. The whole process can be summarized as follows: acquiring experimental chiral EELS signals and pre-processing for EMCD signals, simulating dynamical coefficients, extracting intrinsic MCD signals, calculating magnetic parameters and error analysis. The related custom code (QEMCD, quantitative EMCD)[47] can be divided into two parts: simulating dynamical diffraction effects by BW software and custom Matlab code, and extracting intrinsic signals and magnetic parameters by custom Matlab code, which can be required from the corresponding author Dong-Sheng Song. The magnetic parameters, orbital and spin magnetic moment with spin specificity, element specificity, site specificity, and orbital spin specificity can finally be measured for magnetic materials with different crystallographic structures.
However, there are still some issues of the quantitative EMCD technique that should be addressed. The first one is the low intensity and SNR of EMCD signals though it has already been optimized by simulating the dynamical diffraction effects. With the XFEG electron gun and K2 camera with direct electron detection installed on the EELS spectrometer,[47,48] the SNR can be further greatly improved with extremely increased source brightness and detection efficiency. Under the current experimental conditions, the smallest total magnetic moment per atom that has been quantitatively measured is
[1] | |
[2] | |
[3] | |
[4] | |
[5] | |
[6] | |
[7] | |
[8] | |
[9] | |
[10] | |
[11] | |
[12] | |
[13] | |
[14] | |
[15] | |
[16] | |
[17] | |
[18] | |
[19] | |
[20] | |
[21] | |
[22] | |
[23] | |
[24] | |
[25] | |
[26] | |
[27] | |
[28] | |
[29] | |
[30] | |
[31] | |
[32] | |
[33] | |
[34] | |
[35] | |
[36] | |
[37] | |
[38] | |
[39] | |
[40] | |
[41] | |
[42] | |
[43] | |
[44] | |
[45] | |
[46] | |
[47] | |
[48] |