In 2017 JEOL introduced their new cryo-EM transmission electron microscope based on the ARM platform. The microscope has an in-column electron energy filter, or Ω-filter (“omega-filter”), in the intermediate optics between the objective lens and the projector lens. The omega filter passes a small range of electron energies. If the zero-loss peak is centred on an energy-selecting slit, inelastically scattered electrons, with their incoherent contribution to the image, are removed and the image contrast is improved. For biological materials, where, for every elastically scattered electron, about three or four electrons are inelastically scattered, the improvement in contrast can be quite pronounced.

Inserting an energy filter into the column is a bold undertaking, because it cannot be bypassed during microscope alignment. In-column filters are generally more complicated because the dispersed & filtered electron beam has to be steered back onto the microscope column axis. For the Ω-filter, the beam looks like the Greek letter Ω turned on its side, caused by beam deflection of four excited sector magnets (Figure 1). The electron beam emerging from the Ω-filter is a chromatic fan, with scattered electrons moving in different directions according to their energy. A narrow (retractable) slit at the exit of the Ω-filter removes inelastically scattered electrons and the energy-filtered beam continues along the optic axis to the next intermediate lens.

The physical design of the Ω-filter is optimized for one particular beam energy, which makes operation at other energies sub-optimal. The design has to factor in the sector magnet excitations, their radii of curvature, the inclination of the entrant & exit edges (for focusing and aberration minimization), and the sector magnet separations. The ray paths of the electrons create two achromatic images and stigmatic crossovers (diffraction patterns/images) within and outside the magnets and has no overall magnifying capability, i.e. the filter has unity magnification,

Figure 1. Energy-filtering according to the in-column (left) and the post-column (right) concepts. The dispersion of the unfiltered rays (thin, black lines) result in energy selection (coloured, thick lines) using dispersing elements (prisms).

In contrast, a post-column filter is simply bolted to the bottom of the microscope column and the electron beam is bent just once, by a single sector magnetic prism, to disperse the electron beam. At this point the beam is traveling parallel to the floor and the post-prism optics renders an image or spectrum onto a camera. A microscope with a post-column filter tends to look like the capital letter “L”. Further, because the filter entrance aperture sits below the viewing screen, the alignment of the microscope is totally separate. The filter is aligned (often automatically) after the microscope alignment. Further, because the post-column filter is relatively simple, it can be operated at different beam energies, once the requisite factory alignments have been made. So, for the materials science community who employ several different beam energies, e.g. for beam damage mitigation, a post column filter is a natural choice.

So why might JEOL bother with the complexity and rigidity of an Ω-filter for their cryo-EMs? From a technical point of view, there are at least three possible reasons:

First, routine cryo-electron microscopy is likely to be performed at a single beam energy for which the energy filter and microscope was designed. This has two advantages: first, the electron optics stays relatively constant, which improves lens stability (improves the incoherent damping envelope) and the magnification accuracy.is maintained. This is critical, because reconstructing single particles from aggregates of hundreds to thousands of images requires particle sizes to be consistent. JEOL instruments that I have calibrated are routinely better than 0.1% accurate.

The second reason may be that, by incorporating energy filtering as standard, it frees up the operator to choose a camera of her choice. At present, commercially available post-column filters tie the user to one brand of camera, which may not be in the user’s best interest, especially if using beam energies lower than 300 kV. However, CEOS have announced their intention to supply post-column filters in future with a choice of camera.

However, the third reason may be that the acceptance of an in-column filter, which is a measure of how much information it can pass (or transmit unhindered) is bigger than a post-column filter (Uhlemann & Rose, 1996). The significance of this is that, for a diffraction limited image, the maximum number of useful energy filtered pixels is determined by the electron wavelength and the energy filters acceptance, according to:

The acceptance, T, scales with energy-selecting slit size with a power determined by residual aberrations in the filter. Table 1 shows the maximum acceptances of various energy filter designs. The Fritz-Haber Institute Ω-filter (“FHI filter”) is fully 3^rd order corrected using sextupoles between the sector magnets. Most startling of all is the Mandoline filter acceptance. This filter was developed for the SESAM project (sub-eV sub-ångström microscope) at Stuttgart that was manufactured by Zeiss. The since-measured acceptance of the Mandoline filter is actually about a factor of 5 times smaller than the number stated (Essers et al, 2010), but it is extraordinarily good – an acceptance of 17400 nm² gives the total number of useful pixels of about 2.7 billion pixels or a chip with 55000 pixels square, more than ten times the size of current largest chips. Therefore, a microscope fitted with an in-column Mandoline filter would, in any given image, capture 100 times more particle images than a post-column filter. Perhaps this is the direction JEOL intend to go?

Table 1. Theoretical acceptances of various energy filters. Measured values exist for the Mandoline (*, T=17400) and Post-column filter (**, T=274) both at 1 eV slit width.

Note: measured acceptance for the post-column filter is taken from Kothleitner & Hofer (2003).

(or why you should always calibrate your data sets )

I have spent the good part of a day trying to reconcile three different data sets, each using a different signal, on two different microscopes, of the same area. This is not uncommon in Materials Science, where it is necessary to pass the electron probe more than once, over an area of interest, to collect and correlate various signals, e.g. EDX spectrum, diffraction pattern & high angle thermal diffuse scattering. The single-pass/multiple-signal microscope paradigm isn’t quite here yet.

The reconciliation of the data is hampered by the fact that the contrast in the various signals cannot be correlated using cross-correlation techniques. Identifying marks, e.g. holes and edges are often critical to the success of the enterprise. However, the task is hampered by an even more significant problem – the magnification, and by extension the probe stepping distances, are wrong. How wrong? Would you believe it if I said more than 25% wrong? What may surprise you is that this relative error is, unfortunately, very common.

The problem arises because of the hysteretic nature of the objective lens magnetic field. The actual magnetization of the lens is a function of its excitation history. For a microscope that is used by other users, sometimes with different beam energies (& lens excitations), the magnetization accrues small changes that, even when the lens current is adjusted to a previously identical value, the actual magnetization is different. Quite simply put, the scale bar put onto your image by the microscope computer cannot be trusted!

To be fair, some microscopes are better than others. The JEOL 4000EX II I used to use at Cambridge Materials had a magnification error of less than 0.1%, i.e. the scale marker on the negative was accurate to better than 1 in 1000! However, this microscope was unusual – it had atop-entry holder and the the optimal lens current (of 7.14 amps) was held constant all the time. Other machines though, had significantly bigger errors, 1 part in 4 (25%) or even more – the greatest error was over 30% wrong.

So, what is the solution? FEI (now Thermo Fisher) microscopes have the option of choosing a button (one of the six hot-keys on the control pad L1-3 and R1-3) that allows all the lenses, or a subset of them, to undergo a lens cycling procedure (“Lens Normalization”). This reverses the currents in the lenses and brings them back again to undergo a cycle around the hysteresis loop. This certainly brings the errors down, but the actual error can still be big (<10% in my experience).

However, the only satisfactory method I have ever found is to calibrate the microscope immediately after the data acquisition. I use a calibration sample and acquire a data set (or two)in the same image mode, with the same objective lens setting and the same magnification. Focusing the calibration sample is done with just one control – the Z-height (vertical displacement of the sample relative to the probe focus). The calibration samples I use are the excellent Ted Pella waffle grating replica (product number 673 or 677) or the Agar cross-grating samples (product number S106), because the scan sizes I use cover the 100 to 1000 nm size range. For high-resolution imaging where the field of view is small, I have a tripod polished silicon <110> samples, which can be mounted in a double tilt holder and imaged at the zone axis. Until the single-pass/multiple-signal microscope becomes de rigeur, the calibration sample will be, and continue to be, probably one of the most important samples in the sample box.

Despite the previous nights conference meal and party at The Midland Hotel, the pace of the meeting did not let up. Todays topics of advanced instrumentation & nuclear materials proved to be really interesting and was very engaging.

Advanced instrumentation & Techniques

Gerald Kothleitner’s (Technical University of Graz, Austria) talk addressed the issue of elemental compositional accuracy and high spatial resolution, i.e. can we have both? (in short, yes, but up to a point). He looked at recent use of zeta-factors for EDX and how they’re needed for absolute quantification. The current multiple SDD detector geometry (a la Super-X, FEI) poses several challenges, in particular the variation of the zeta-factor across the SDD detector face that differs for each element. For <1% accuracy an accurate zeta-factor depends on many factors, including sample tilt angles, elemental composition (light elements particularly) and specific line (Thomas Walther’s talk looked at this too on Tuesday, day 1). Elemental estimation by EELS has a significant problem with channeling, i.e. the ‘capture’ of the electron beam down the atomic columns, which leads to skewing of the apparent ionisation rate. At zone axes this is a massive problem, inaccuracies in the absolute chemical composition by a factor of 2 or 3 are not uncommon and the only apparent way to mitigate this is to tilt the beam *away* from the zone (Yifeng Liao & Laurie Marks paper last year sprang to mind). Kothleitner proposes something similar – precessing a sub-angstrom probe at >2 degrees from a zone axis (for SrTiO3). Technically this will be very hard, i.e. very thin samples (<20nm) with massive aberration-free patches (>50mrad) in which only a fraction is used by the probe forming aperture. This is a problem that won’t go away quickly.

Lewys Jones talk (Oxford, UK) studied the use of quantitative STEM HAADF for determining the 3-dimensional structure of nano-particles to recover the surface structures; concave surfaces (positive curvature) seem to be particularly active and desirable. Oxford have done a fine job looking at the experimental conditions that determine the both the peak and integrated intensities on an atomic column, the latter measure being the more linear with the number of atoms and more forgiving in the number of parameters required to be known. While lateral atom location is easily measured with aberration corrected probes (plus adjustments for drift and scan distortion), the height of each atom is currently obtained using DFT (using ONE-TEP) by finding the minimum energy configuration along the beam (z-axis). For me, the amount of time needed with this approach is extremely expensive, unless the scanning confocal configuration, which Peter Nellist looked a little while ago can bring the z-location of atoms down to the angstrom level. The biggest challenge with the simple STEM-ADF + DFT approach will be to make this fully automated so that a microscope computer can be left to study the several thousand particles needed to get an accurate measure of their surface-characteristics and, hence, their reactivity characteristics. Can we automate ADF-STEM in the same way that cryo-EM is for biological EM? As a final caution, Lewys also pointed out that, relying on human perception to choose the ‘best’ particles to analyze, lead to significant bias in the way the population was selected studied, with larger particles being more attractive than small ones.

The next two talks addressed vortex beams in crystals, i.e. a beam carrying orbital angular momentum about the propagation axis with the two main theoretical frameworks, Bloch-wave theory and multislice, given by Budhika Mendis (Durham University, UK) and Scott Smith (University of Glasgow) respectively. Budhika’s talk studied the evolution of the orbital angular momentum expectation value <L_z> in the crystal, showing how it oscillates with varying amplitude, but fixed periodicity (along the propagation direction). He related this to the excitation of the (non-dispersive) 1s and the (dispersive) non-1s states in the crystal. The excitation of the 1s Bloch wave is dependent on the probe width and how many neighboring atomic columns are illuminated- the atom on the optic axis has zero excitation by symmetry (it is perfectly dark in Mike Berry’s words). The eigenenergy difference between the 1s and the non-1s states leads to rapid oscillation in <L_z>, and the slower, more complicated evolution determined by the spectrum of non-1s states. I would expect these to vary in a very complicated way with very small crystal mis-tilts and, if quantitative vortex-beam STEM is to become mainstream, the simple 1s channeling model (advocated by Dirk van Dyck) will need to be dropped quickly.
The Glasgow talk looked at the conditions (beam energy, illumination angle) for which left-hand and right-hand pairs of chiral crystals differ in their diffraction patterns, in this case quartz along the 3-fold screw axis. In short, low thicknesses with non-overlapping diffraction orders are needed to provide the most variance between patterns. The variation with beam energy appears to be small.What did spring to mind is what would happen when the winding pitch of an electron beam approaches that of the chiral pitch in the crystal (this would need a very low beam energy since the wavelength would have to be of the order of angstroms, i.e. eV not keV)? In this case the number of times the electron beam ‘crosses’ a plane will be inversely proportional to the difference and sum of the helical pitches for para- and contra- rotating beams, i.e. would a sympathetic helical beam encounter (diffuse) scattering less often than antisympathetic vortex states?

The afternoon plenary speaker was Max Haider (CEOS GmbH, Germany). There are now over 500 aberration correctors around the world (just under 500 as of November 2014) and this explosion has been preceded by nearly 50 years of tackling the problem of correcting aberrations in electron lenses. The current limitation to deep, deep sub-angstrom imaging (<1 bohr radius) is electrodynamic noise in the liner tubes of the microscope. This can be understood as a form of Johnson noise in the electron beam as it passes the inductive (& capacitive) elements that make up the microscope column. The proximity of the electron trajectories and their overall path-length are contributory factors, so, either making the microscopes smaller or making the corrector elements work at lower powers is needed. The lowest image spread limit attainable is about 40pm and achieved by going to very high beam energies (MeV). This is the approach that the late that Akira Tonomura was advocating: Hitachi obtain a resolution of about 42pm in the million-volt energy range. On this pessimistic note, Max suggested that, perhaps, the main problem not resolution but contrast? For this reason the effort is now directed to improving the resolution a lower beam energies <50keV whilst maintaining sub-angstrom resolution. This will almost certainly require cold-field emission guns, not the Schottky sources used at present. CEOS are now working with Ute Kaiser (Ulm University, Germany) and FEI Ltd on the SALVE project that was initially partnered with Zeiss. Dr Haider hinted that operability of the aberration corrector is still an issue, although he tried to play that down. Parasitic aberrations are also a problem and, in my opinion, CEOS are chasing Nion in trying to correct out beyond 6th order (Nion now partially correct out to 7th order).

Microscopy of Nuclear Materials

The UK Government has committed to 80% reduction of CO2 gases by 2050 and nuclear power seems to be the only technology capable of reaching this target without the tides of wind tubines that the Daily Mail screams about on a regular basis. Increasing the fleet of nuclear power stations is becoming an extremely urgent issue. For young researchers looking for mid-term career advice, nuclear power is worth keeping an eye on.

Erwan Olivieri (CNRS, France) has been using high energy (keV) ion beams as a way of damaging ‘radiation hard’ cladding materials like SiC in situ. He’s been watching helium bubbles, i.e. in pre-irradiated material, shrink as the material is damaged by the high-energy ions. Sputtering on the back-surface seems to increase the rate of bubble disappearance. Further, he’s looked at CuNb superconducting wires designed to carry the currents needed for 70-100 tesla field necessary to confine tokamak plasmas in nuclear fusion reactors. Here the copper atoms displace first (with increasing dose) and then the Nb atoms. This is not surprising since the transfer of kinetic energy is highly dependent on the mass ratio of the ion and struck atom. This prompted the following question for me: How realistic is radiation damage by ions (where coulomb scattering is active) compared to high energy neutrons (spin-dependent). Perhaps the time has come to have in-situ TEMs connected to neutron sources?

The Glasgow group (Kirsty Annand) are lending their expertise to studying oxidation of zirconium alloys in nuclear fuel rod cladding tubes in conjunction with Mhairi Gass at Amec Foster & Wheeler. They’re using DualEELS to measure the near-edge structure and small (2-3eV) core-level shifts of very high ionization energies (Zr L2,3 at 2225 eV) to determine the oxidation state and electronic properties of the zirconium alloy, which appears to go through cycles of oxide thickening with an unusual 150-200 hour period. Their FIB lift-out samples seem to give quite thick samples and the interfaces between the different valencies of zirconium were hard to discern. This project is still in its infancy.

Standing in for Helen Freeman (Leeds University, UK) was Rik Brydson showing the effects of radiation on a neutron-moderator/ absorber material like graphite. Instead of ions, they’re using high (200keV) energy electron beams (above the damage energy threshold) to damage graphite before lowering to 80keV (sub-threshold) to study the sp2 bond content carbon, via the pi-star peak in the C-K edge. The pi-star seems to be a better measure of structural integrity than the bulk plasmon energy as a proxy for density. The plasmon energy shift is too small, probably because the contribution of pi-bonded electrons to the electron density is rather small.

Ian Griffiths, standing in for Alex Warren (both at Bristol University, UK) were studying precipitation in corrosion resistant 316 austenitic steels, i.e. Cr-rich phases. They’re using FIB lift-outs of chromium-rich areas identified in the SEM and studying the phases there in the Oxford-based JEOL ARM200CF using STEM EDX and EELS. Two phases in particular, chi and G phase appear in these areas. The identification of G-phase as being a gamma-prime type alloy was contentious and generated a lively debate afterwards.

Lastly, Simon Dumbill (National Nuclear Laboratory, UK) gave a nice talk on the way the nuclear industry is (re)embracing electron (and ion) microscopy. They’re using the FIB to make TEM-ready samples, which reduces the amount of hot, i.e. radioative, material by a factor of 100000 times. These are much safer to handle and allows non-nuclear labs to analyze important materials. However, the (radioactive) material sputtered away during the FIB milling is a major issue (most EM manufacturers won’t touch a radioactive microscopes, even for servicing) and NNL are now developing some sputter shield/blade that can present a large surface area along the line-of-sight of the sputtered material and be disposed of safely without contaminating the microscope. Finally, some of the segregation profiles seen 20 years ago, e.g. Nb & B along steel grain boundaries, might have been right all along. The 20-30 angstrom composition profiles seen then are now being seen (again) with smaller (aberration-free) probes carrying more current. The analysis times are now hundreds of times faster with these newer instrument s and NNL are eagerly waiting for their new JEOL ARM200CF at the end of 2015 for precisely this sort of work.

Final thoughts
The EMAG & MMC 2015 meeting was really excellent. It was well organized (by the Royal Microscopical Society), well attended and the themes were really interesting. Finally, the conference week went extremely well with the use of my Brompton bike. I got around Manchester quickly and easily and I would not go to another conference without it.

Travelling companions – Brompton bike and luggage on the train home.

The great thing about EMAG is that it brings together microscopists of all flavors and today I had some time with the scanning probe microscopy (SPM) community. Perhaps the most impressive talk today was Philip Moriarty‘s “Pauli’s principle in dynamic force microscopy: Do we really see chemical bonds?”. Unlike most of the talks, Philip addressed an important issue that many tend to ignore – over-interpretation of images. He put it succinctly thus:

“AFM still has the problem that, if it swims like a duck, waddles like a duck and quacks like a duck, it can still be a goose.”

The problem is with images like these:

Images of molecules taken with tip-functionalized AFM (P. Moriarty, MMC+EMAG 2015)

These images use an AFM tip which is terminated with a molecule of known structure. Leo Gross used a simple carbon monoxide to get the first beautiful images of pentacene. The (dynamic force) image rendered is that of the tip-sample  interaction (specifically the potential energy landscape). These can be modeled with Lennard-Jones (or Morse) potentials between the tip and sample molecules (simple) but tends to be interpreted as bond-charge density distribution of the sample molecule only, often with the help of full density-functional electronic modeling (complicated). Basically, the SPM community has been seduced into overly complicated interpretation and it is time for an Occam shave. The answer to this problem of over-interpretation has been the use of social media sites like PubPeer:

PubPeer slide showing that 1014 views this particular paper has attracted (P. Moriarty, MMC&EMAG 2015).

A shows of hands showed that only about 3 or 4 people in an audience of about 30+ knew about this site (I was not aware one of them). PubPeer gives people a platform to criticize and discuss the science behind these groundbreaking papers and root out overly complicated interpretation. For him, these tools are crucial in the fight for clear ideas to come to the fore. This is not only a scientific issue, but a moral issue for how, often tax-payers money, is used for research.

Other talks I saw focused on in-situ experiments where the sample is manipulated either thermodynamically (temperature, pressure, chemical potential) or mechanically (force and displacement).

The Ze Zhang (Zhejian University, China) talk (given in absentia by Jun Yuan) of the mechanical properties of nanowires was extremely interesting. Sub-micrometer nanowires are being pulled or bent in the microscope and the evolving nanostructure monitored. The mode of plastic deformation (perfect dislocations versus partials + twinning) in copper nanowires changes at about the 200nm thickness size (perfect dislocations above this). Further, nano-particles of silver between a gold and tungsten tips shows behavior of both a liquid (morphology) and solid (crystal fringes).

Eric Prestat (Manchester) showed some lovely work on high pressure (atmospheric, 100kPa) and 900C using Protochips holders. He’s been watching the distribution of palladium-silver change during annealing and iron-oxide (ilmenite) reduce. Chemical mapping under these conditions with x-rays (EDX) seems to be easier than EELS because of the X-ray transparency of the confining windows of the sample holder.

Ed Boyes (York) showed some of the STEM images of platinum catalysts taken operated at 0.1Pa at 500C showing that it is possible to maintain angstrom resolution under these conditions. He ended with the sobering point that, industrial catalysts operate at 170 bar (17 MPa, i.e. 170 million times higher) than that operated in the microscope. One wondered if a benchmark pressure of 1MPa (10 bar) is possible with todays holder/microscope technology?

Electron microscopy of nano-particles in liquids (Teresa Roncal-Herrero, York) was even more impressive. She was using gold chloride (AuCl3) in ethanol and watching gold precipitate onto nano-rods. The electron beam has a massive effect – electron doses seem to have a profound effect on the rate of precipitation and extremely low-dose imaging STEM techniques are needed to prevent incorrect reaction kinetics being measured.

Finally, two talks on the magnetic properties of nano-magnetism using off-axis holography were presented. David Reyes (CEMES CNRS, France) gave a nice talk on copper-cobalt multilayers in nanowires grown by electrodeposition and how the magnetization changes the single-domain/vortex magnetization states. Trevor Almeida (Imperial College London), in collaboration with DTU (Denmark) and ER-C (Julich, Germany) gave a lovely talk using the magnetic field of the microscope lenses to identify the magnetic state of paleomagnetic materials. Both of these talks were able to isolate the magnetic contribution to the recovered electron phase, i.e. remove the mean inner potential contribution, and provide quantitative magnetizations for their sample. The influence of Rafal Dunin-Borkowskis work was evident in the way the data was taken and presented in these talks.

So, after a full day of talks these are a few thoughts on some of the talks presented thus far…

Julie Gonnissen (with Sandra van Aert) from the EMAT group at Antwerp gave possibly one of the most convincing talks on the TEM vs STEM debate. They’re using a Bayesian approach to calculate the posterior probability density function (pdf) to calculate the likelihood for a certain number of atoms in a column (given the intensity in an image). This approach allows them to calculate the error of mis-assignment (“getting the number of atoms wrong”) directly and she showed that, at the two electron doses considered (ten thousand and ten million electrons per square Angstrom), the STEM annular dark field has a low error rate (<20%) compared to the negative Cs imaging technique (NCSI) developed at the Ernst Ruska center at Julich. It seems STEM is still less error prone than TEM.

Hao Yang’s talk (Nellist Group, Oxford) was a nice chance to see some of the data that did not make it into their recent Nature Communications paper. This included Eshelby-twist displacement plots (versus defocus) for two mixed dislocations (opposing screw components) and an edge dislocation where the (Eshelby) displacements follow different trajectories. This was a nice reminder that the state of the art instruments can be used to solve genuinely difficult materials problems like dislocation core structures.

Andy Stewart’s talk (University of Limerick) on electron diffraction tomography was a witty take on an tricky experiment. They’re taking 2-3000 diffraction patterns over a large tilt range (with precession) to estimate the kinematic diffraction pattern of a variety of material including dose-sensitive materials like insulin. They’re taking diffraction tilt series with a total electron dose up to 4 electrons-per square Angstrom. Fortunately those few electrons are concentrated into a number of peaks that make the job manageable. Diffraction is certainly the way to go when so few electrons can be used.

Peter van Aken from Stuttgart gave a beautiful talk on the excitation of surface plasmons from tapered gold rods used for scanning near-field optical microscopes. This was a detailed look at the surface plasmon eigenmode evolution (in time and space) as the electron probe approaches (in an aloof way [pdf]) the tip of the cone. The dynamical variation in the electric field as the electron approaches was beautiful to behold and is an area rich in relativistic electrodynamic wave theory. This was a very though-provoking talk.

Finally, Quentin Ramasse (SuperSTEM & University of Liverpool) showed some of the images of doped graphene (with silicon, boron and nitrogen) and how, even with low beam energies, atoms can be triggered to move between atomic sites. The near-edge structure of the (dopant atom) ionization edge is providing to be the killer technique for telling if the dopant atom is displaced along the beam axis by using DFT models. The cross-sections for these processes can be measured with 3 significant figure accuracy.

One thing that I took away from this is that 2D materials (planes/sheets) are proving to be a very fruitful materials system to study with very short depth-of-focus probes in aberration corrected microscopes. The microscope are still just structure projectors within this materials paradigm and the use of such instruments for thicker (3-dimensional) materials has yet to start in earnest. People like David Muller and Peter Nellist are two of a very small number or people taking this seriously.

Ondrej Krivanek (right) and  former Director of the SuperSTEM facility Andrew Bleloch (left), inspect some modifications adapted for the Nion pole-pieces. Photograph taken at the Old Cavendish Laboratory (3rd March 2010).

In August last year a remarkable paper appeared in the journal Nature with the title “Vibrational spectroscopy in the electron microscope” authored by Ondrej Krivanek, Tracy Lovejoy, Niklas Dellby, Toshihiro Aoki, Ray Carpenter, Peter Rez, Emmanula Soignard, Jiangtao  Zhu, Phil Batson, Maureen Lagos, Ray Egerton and Peter Crozier [1]. It was the culmination of a huge amount of work to make the electron microscope a tool for studying phonons (or atomic vibrations) at the nanoscale for the first time. It was an extraordinary moment.

For Ondrej Krivanek, this was the “closing of the circle”, a return to the problem of getting very high energy resolution energy loss information about a sample using a carefully structured electron probe. His first significant electron-optical achievement was the Gatan imaging filter in 1992 [2]. This remarkable spectrometer could be attached to any microscope and had the ability to form images with a range of energy selected. With its CCD camera the GIF quickly became a tool for mapping the chemistry, measuring imaging and diffraction information accurately and created a whole new paradigm of routine quantitative microscopy. The GIF was a huge commercial success and hundreds of GIFs now sit at the bottom of many microscopes.

The second significant advance was the quadrupole-octupole aberration corrector for a STEM for which he and Niklas Dellby formed a new company, Nion (a concatenation of the first 2 letters of their first names) in 1997. Their aberration corrector was designed to perform imaging and analysis with a sub-angstrom probe with a decent current (>100 pA). Because the aberration corrector had to sit between the electron source and the objective lens (the lens that sits next to the object under study) it required an intimate knowledge of the electron-optical system it sat in. Ondrej’s experience with the VG STEMs at Cambridge allowed them to design and build a system that fitted into the VG HB501 run by Phil Batson at IBM [3]. Their Nature paper of 2002 showed that their proof of principle unit worked [4].

However, integrating an aberration corrector into an existing (and inherently flawed) VG STEM column was a humbling experience – the liner-tube on the VG was flimsy, the lenses drifted with changes of excitation and the top-entry sample holders gave no height adjustability. Nion decided that they had to build a whole new electron-optical platform from scratch if they were to go into the deep sub-angstrom realm. One of the wisest decisions they made was to stick with a cold field emission gun (cFEG) which, with better UHV seals, allowed them to use the brightest electron source available. The first Nion STEMs were an extraordinary advance in small-probe formation, but they were not the first to achieve commercial success with an aberration corrected microscope [5]. CEOS GmbH headed by Max Haider, Joachim Zach and heavily influenced by Harald Rose had designed a working aberration corrector for SEM in 1995 [6] and for TEM in 1998 [7].

However, Ondrej and Niklas’s technological advance last year was a coup de grace. Although their electron monochromator was first described in 2009 with the design aim of 10meV resolution or better with about 30pA of current [8]. The first results of this system were described in Microscopy in 2012 [9]. In 2013 Ondrej presented their latest results “12meV” at EMAG 2013 (I was there and the excitement elicited by this result was palpable).  His closing line in that talk suggested a physical limit of about 3meV resolution. To get appreciable current in such a narrow energy spread would require increasing the brightness of the gun, perhaps through aberration correction in the gun. Everyone left the auditorium with a stupid grin on their face – this was like a bunch of kids meeting Willy Wonka for the first time!

Figure 1 shows just what they’ve done to the energy resolution [1]. Their monochromatic zero-loss peak width is 25 times better than the cFEG width and about 50-75 times better than the Schottky thermoinic emitter emitter used by other manufacturers (500 – 750 meV ZLP width).

Figure 1. The energy distribution of electrons traveling with and without monochromation (units are milli-electron volts, meV). The “0” in the figure corresponds to a zero energy loss (i.e. total kinetic energy of 60 keV). Electrons to the right of “0” are slower. The 25-fold reduction in peak width is the result of the monochormator. Taken from [1].

So, what is special about this latest advance? Figure 2 shows what all the fuss is about. The bump in figure 2 is a phonon-loss peak from hexagonal boron nitride – the excitation of a lattice vibration (longitudinal optical phonon) as the solid is struck by electrons. Think of this as the characteristic ‘ping’ as electrons bounce of the atoms in the crystal. While this was not the first time phonons were detected with an electron beam (that belonged to Boersch, Geiger and Stickel [10]), this was the first time they can be detected with a spatial resolution of about a nanometer.

Figure 2. The zero loss peak (ZLP) is at 0 meV. The hexagonal boron nitride LO phonon loss peak is at 173 meV. Taken from [1].

The implications of this work are enormous. The tone of a vibrating material changes when defects or interfaces are introduced (think of a damaged drum-skin or a frayed guitar string). Characteristic modes are created in multilayered semiconductors that may (or may not) scatter electrons and excitons in optoelectronic devices. Nanostructured materials like carbon nanotubes and graphene all have distinct tone depending on their size, which are subtly altered with defects and impurities. A lastly, questions like, for example, do dislocations hum? (do dislocations have a broad-band phonon spectrum?) can be answered, simply by mapping the energy loss spectra around these defects. We are now, finally, into a realm that traditional optical spectroscopy has beaten electron microscopy hands down.

In my humble opinion, the technological ramifications of this instrument are extremely interesting:

1. The gun, monochromator and spectrometer can never be regarded as separate systems again. There is too high a penalty paid when each of these operates without the same reference voltages and current supplies.

2. Mapping chemistry using the GIF for energy-filtered TEM (EFTEM) is now rather crude. EFTEm has been shown to be too dose-inefficient (you waste a lot of electrons) and spectrometers designed specifically for STEM-EELS are the way to go. The one saving grace for the GIF is the direct-detection cameras for biological TEM imaging. Here the ability to remove inelastically scattered electrons improves the reconstruction of small proteins and ribosomes.

3. Schottky sources are too dim. Although they tolerate poor vacuums, the low brightness of the Schottky gun (standard or extreme) is simply not high enough to get more than a pico-amp of current at similar probe specifications. The latest cFEG guns developed by Hitachi for very high voltages uses (MeV) operates at pressures 1000 times lower (in terms of vacuum pressure) and, when the tungsten <310> cFEG tip is kept clean, you get a 3-5 fold improvement in brightness [11,12].

The UK is now in the envious positions of having one of these remarkable microscopes operating at the SuperSTEM facility. I will be watching what comes out of there with baited breath.

[1] O.L. Krivanek, T.C. Lovejoy, N. Dellby T. Aoki, R.W. Carpenter, P. Rez, E. Soignard, J. Tzu, P.E. Batson, M.J. Lagos, R.F. Egerton & P. A. Crozier (2014) Nature 514, 209-212
[2] O.L. Krivanek, A.J. Gubbens, N. Dellby & C.E. Meyer (1992) Microsc. Microanal. Microstruct. 3, 187-199
[3] The Cambridge physics group (mIcrostructural physics) were well-known for tinkering with microscope hardware – the last VG-STEM to be housed there had numerous veroboard circuits attached to it that controlled various home-built systems designed to stabilise the energy loss spectra or switch the direction of the beam in the McMullen spectrometer.
[4] P.E. Batson, N. Dellby & O.J. Krivanek (2002) Nature Vol. 418, 617-620
[5] N. Dellby, N.J. Bacon, P. Hrncirik, M.F. Murfitt, G.S. Skone, Z.S. Szilagyi and O.L. Krivanek (2011) The European Physical Journal Applied Physics 54, 33505
[6] J. Zach & M. Haider (1995) Optik 98, 112-118
[7]  M. Haider, S. Uhlemann, E. Schwann, H. Rose, B. Kabius & K. Urban (1998) Nature 392, 768-769
[8] O. Krivanek, J. P. Ursin, N.J. Bacon, G. Corbin, N. Dellby, P. Hrncirik, M. Murfitt, C.S. Own and Z.S. Szilagyi (2009) Phil. Trans. R. Soc. A367, 3683-3697
[9] O. Krivanek, T.C. Lovejoy, N. Dellby & R.W. Carpenter (2012) Microscopy 62 (1), 3-21
[10] H. Boersch, J. Geiger, & W. Stickel (1966) Phys. Rev. Lett. 17, 379-381
[11] B. Cho, K. Shigeru  C. Oshima (2013) Rev. Sci. Instm. 84, 013305
[12] K. Kasuya, T. Kawasaki, N. Moriya, M. Arai & T. Furutsu (2014) J. Vac. Sci. Techno. B 32 (3) 031802

I am indebted to Dr Arthur Blackburn for bringing to my attention articles 11 and 12.

Having just read the news on the Institute of Physics news web-page, I was intrigued to see this item on simulating neutron damage in materials using ion beams instead, by Gary Was at the University of Michigan. By virtue of their lack of charge, neutrons are very difficult to control either in terms of their energy or their degree of collimation, especially in small (university) laboratories. So, the use of energy-specific ion beams instead is a real breakthrough in being able to synthesise conditions corresponding to actual nuclear power plants.

The intriguing thing for me was that this is the sort of thing that Argonne National Laboratories (ANL) has been up to with their IVEM-Tandem facility albeit with a much lower energy ion beam (0.5MeV instead of 5MeV in the IoP news article) and with a variety of specimen temperatures. People such as Mark Kirk (ANL) and Mike Jenkins (Oxford) have devoted huge amounts of time and effort trying to understand how defects arise in ion-irradiated materials, to good effect. They did this by making movies of their samples while they are being irradiated and watch for changes in the diffraction contrast for tell-tale signs that defects are being created. These defects tend to be dislocation loops which form by the condensation of vacancies left behind after the swift ion has displaced atoms and, being only a few nanometers across, these things are very hard to find.

However, with aberration-corrected electron microscopes it may be possible to see and watch these defects being created and evolving, especially with the latest state-of-the-art heating holders and fast cameras. So, together with a 5MeV ion accelerator, may we see a new era of ion-beam transmission electron microscopes some time in the future, especially with all the talk of nuclear power making a comeback?

1. You get what you want, where you want (specificity)

About 10 years ago I went to a seminar given by a member of our group, Dr Stephen Lloyd, in which he made a remarkable claim:

“I believe that the invention of the FIB was more important than the invention of the TEM” 

Being a TEM-geek, I guffawed at this and listened nonetheless. However, this one statement has stuck with me for a long time and the more I think and reflect on it, the more unsettled I became with my own view on this. In short, Stephen was correct, the FIB is more important than the TEM.

It is worth explaining why the utterance of this remark from Stephens lips was so shocking to me. First, Stephen is a no-nonsense guy who was one of the best microscopists I had ever met, so seeing a Master of His Art extolling the virtues of a different instrument was genuinely troubling. Second, the FIB was a relatively new tool and, being naturally conservative, Stephen wasn’t prone to the distraction of new toys just for the sake of it. Third, Stephen is a quietly contemplative person and listens carefully to anything he hears (this could be quite unsettling when he looked at you intently with those dark eyes). In short, Stephen had a good nose for bullshit, so, when he spoke, people listened carefully to what he had to say.

Ten years on from his seminar I have come to the conclusion that Stephen was right all along and this post summarizes why I have changed my mind.

Like people, it is the defects in materials that make them so interesting. Defects also tend to participate in the eventual failure of a material and tend to emerge, or be associated with, some sort of protrusion or localized change at the surface. Careful study in an optical microscope, or SEM, will reveal some feature and indicate that something interesting has happened down below. The FIB allows you to cut out that small volume of material around this potentially pathological defect and, if you know the crystallography, do so with a certain orientation in mind. The picture below exemplifies this quite nicely (it is one of the images Stephen used in his seminar). It shows, beautifully, how the material has deformed along deformation twins (running approximately diagonally) which seems to ‘squeeze’ material above and below into small wedges. The layers act as good markers of the material showing where the material moves when the material was stressed (compressed horizontally in this case).

Figure 1. InGaAs-GaAs multilayers (on indium phosphide) that have undergone bending creates deformation twins that intersect to form rucks. Taken from Stephen Lloyd et al (2005) Philosophical Magazine, Vol 85, issue 22, pages 2469-2490.

Defect analysis before FIB was torturous – it usually took days, if not weeks of carefully mechanically removing material, layer-by-layer, to ensure the defect of choice was not inadvertently removed. If the defect was lost, then the whole exercise was wasted. Being a successful defect analyst used to require incredible technical abilities in both mental and physical dexterity, which meant that there weren’t that many. Today, with the FIB, the job is a lot more accessible.

2. The volume of material is manageable and scientifically interesting

Down at the micrometer level (one thousand of a millimeter) the mechanical behavior of materials changes. They’re usually stronger and deform in new ways especially at lower temperatures below the brittle-to-ductile transition temperature. Creating micrometer-sized structures that can then be mechanically tested and then analyzed post-deformation allows us to see the changes in the structure. This technique was invented by Michael Uchic under the supervision of Bill Nix at Stanford. The FIB simply removes a large quantity of material, in an annular fashion, to leave a thin pillar of material secured at the base by the native material. Using a small punch, the pillars are compressed and the hardness measured.

Figure 2. A bright-field image of an aluminium-cobalt micro-pillar that has been compressed under a diamond punch. The material deforms with a combination of slip along the primary slip plane (diagonal bands), defect loops (DL) and linear defects (L).

3. FIB makes TEM analysis easy

Ten years ago less than 5% of the samples that went into the TEM were FIB-prepared. Now, that figure is about 50% FIB, 20% traditional mechanical polishing and ion-milling and 30% nano-particles. Earlier artefacts in FIB-prepared samples, e.g. curtaining and thickness variations, can now be avoided and the success rate of TEM sample preparation has shot up to about 90-95%. Further, with a successful FIB milling strategy, it is possible to make several samples per day. The thin lamellae are then mounted them onto special metal washers like that below.

Figure 3. A low magnification image of the sample mounting washer (right hand side) and the FIB-prepared sample sits on the bottom edge of pillar marked “C”. The mesh grid contains gold-on-carbon for tuning the Titan electron microscope. Taken by JonB (2013).

The FIB also allows you to mill to a specimen thickness of about 50-100nm, which avoids the problems of thick samples with weak contrast, but is thick enough to be mechanically robust to handling. Problems such a surface damage are still with us, rendering the surface layers (typically 2 to 5 nm) rather poor quality (amorphous).

4. TEM analysis is more popular than ever

The sheer numbers of people doing TEM has increased markedly over the past ten years. What was once an esoteric and “hard technique” is now practiced by many of the students and post-doctoral staff that I work with. With widespread usage it has been easier to source newer, better microscopes and try stuff we once thought was impossible.

Next week sees the start of Microscience Microscopy Congress 2014 in Manchester, which is the new formed by merging the younger meeting, Microscience, with the older Electron Microscopy and Analysis Group meeting, or EMAG as it is generally known.

There was some disquiet at EMAG 2013, hosted by the University of York, when this was announced last year. I have always regarded EMAG as a gentle and friendly conference where space was given equally to PhD students, post-docs and tenured academics to speak. The program usually gives everyone a chance to present, regardless of their position in the academic food chain, and the organizing committee did a good job to keep everyone happy. The only weak spot with EMAG was the commercial exhibition. In a couple of EMAG meetings that I went to, the exhibition was held too far away from the talks and angry rumors were rolling around that various firms felt that they were getting poor value from the conference fee. The future of EMAG was at stake if the commercial sponsors weren’t happy. So it has come to pass.

Microscience on the other hand has been more of a commercial success because it was always considered to be an exhibition first and a conference second. To do this, Microscience has to be big. When conferences get big, they lose their informality and the serendipitous chance encounters tend to fall off, especially when there are numerous drinking establishments in the vicinity (the ‘beer compass’ at EMAG tended to point to the few, decent establishments). Further, with MMC being the size it is, it is likely that there are fewer places to hold it, limiting the chances that it will move around the country. My current understanding is that MMC is to be held at Manchester or London for the indefinite future in contrast to EMAGs wanderings around the United Kingdom in the past.

I hope that what appears to be a big, slightly daunting meeting will pass with every success and my colleagues will return with numerous stories of great talks, exciting new toys and new relationships and collaborations. EMAGs informal and quieter atmosphere may be a relic of the past, but I hope that its spirit will pass to MMC with a better, friendlier and more exciting future ahead of it.

Yesterday I watched the Gatan webinar “Joint EELS-EDS Spectrum Imaging for Fast-Mapping and Atomic Scale Analysis” presented by Paolo Longo, formerly a research student with the excellent Alan Craven at Glasgow University. This was way over-subscribed and a colleagues connection dropped out as a consequence, so this webinar addressed a topic a lot of people wanted to know about.

The presentation started with the basic physics, which seemed to be correct before launching into a series of examples, using a number of beautiful systems around the world, including the FEI Titan at Graz. This showed just how good electron energy loss spectroscopy (EELS) is relative to energy dispersive X-ray spectroscopy (EDS), especially in terms of the signal-to-noise ratio for the various elements. This was, without doubt, a microscopists wet-dream, with some of the most exquisite chemical maps I have ever seen. I was also genuinely surprised by the ability to map, with EELS, some extremely high energy losses, e.g. the Titanium K edge at 4.9 keV (Gatans spectrometers are doing things now I never thought I would see).

The over-riding impression I was left with was “why bother doing EDS at all?”, which is not surprising, because Gatan do not sell EDS spectrometers. The laws of physics seem to concur with Gatan’s view too: X-rays are difficult to deflect, collect and measure; light elements have poor X-ray yields (only 1-in-a-thousand ionized carbon atoms give out an X-ray, on average) and high-energy X-rays have a habit of lighting up other stuff you don’t necessarily want to see in the spectrum.

So, if EDS is so rubbish, why bother? Well, EDS has some strengths that EELS simply cannot match. They are:

1. EDS works for thick samples

Above about 100 nanometres the visibility of the EELS ionization threshold starts to drop. By 500 nanometres the signal is largely lost amongst the background signal, making the background removal difficult. EDS doesn’t suffer from this problem and, although the spatial resolution gets worse (see point 2 below) the EDS peaks just get bigger. Figures 1 and 2 capture the essence of the problem for silicon for four different thicknesses 30nm, 60nm, 120nm and 240nm.

Figure 1. The silicon K line at 1.74keV from thin, 30nm thick silicon (green line) to 240nm thick silicon (blue line).

Figure 2. Electron energy loss spectra of the silicon K edge (1.84keV) from exactly the same points as in figure 1. The thickness double between each spectrum and so the quantity of signal should rise in proportion.

 

2. Not everyone wants atomic resolution

Atomic resolution chemical maps require some extraordinary sample preparation, and many of my colleagues don’t have the time. Often they need to examine several samples within a few hours. Further, not every problem requires atomic resolution. I would say that the 1-to-100nm length scale is more important than the 0.1-to-1nm scale because it is at these are the sorts of distances in which elemental diffusion occurs and secondary precipitation and pathological inhomogeneities take place.

Figure 3. An EDS map of a precipitate in an alloy.

 

3. EDS is simple to set up

EDS is easier than an Amazon one-click(TM) transaction. The sample is giving out X-rays anyway and you simply ask the computer to collect the signal. EELS still requires faffing around to get the appropriate camera length and aligning the EELS spectrometer to get the beam and the appropriate range of signal you want. Incorporate the zero-loss peak and you are screwed – the CCD camera will saturate as it tries to handle several orders of magnitude of signal (six or seven orders of magnitude are not uncommon). The development of Dual EELS with the Glasgow group mitigates this to a large extent, but it still needs setting up by an experienced user (typically <20% of users at present).

4. EDS quantification is better

It is not difficult to get 0.1 atomic percent (1 at.% = one-in-a-hundred atom accuracy) with EDS. It is hard to get this with light elements, e.g. carbon, oxygen and nitrogen, which is important for materials like, e.g. steels, nickel and aluminium alloys, so EELS is worth pursuing for those elements alone. However, EELS will get down to 0.01at.% for elements like C, N and O, but for heavier elements, e.g. silver, indium, tungsten etc, the minimum detectable mas fraction can be several percent because either the edges have a delayed maximum (due to large angular momentum) or the ionization cross sections (a measure of the ease with which an atom ionizes) are small or overlap significantly with other edges. Further, a quick look through Egerton shows that the parameterized inelastic form factors (used to calculate the ionization cross section) have errors of 5% (K edges), 15-20% (L edges) and 30-40% (M & N edges) and are needed for absolute quantification. These are not small uncertainties. It was interesting to note that the none of the maps in the Gatan webinar were converted to compositions, either percentage or absolute concentrations (atoms per square nanometre).

From what I see in the laboratory it will be a long, long time before EDS is usurped completely by EELS. However, Gatan have surprised me on many occassions, so I won’t be completely surprised if I have to admit being wrong again in the future. Plus ca change.