Atom probe tomography

“Wheel, fire, microscope.” This statement was made at a nanotechnology-related conference,¹ at which the speaker was making a point on the importance of microscopes in the world of technology. What is the connection of this to atom probe tomography? The length scale at which human knowledge and endeavor operates is directly linked to the scale of our microscopies. We have long been able to conjecture about our reality at length scales beyond our ability to observe, but practical developments always require observation. For example, Greek philosophers were able to hypothesize about the existence of atoms,² but it took 25 centuries before atoms were first observed by humans with, in fact, a field ion microscope (FIM).³ Microscope technology is as important to human survival and advancement as the mechanical (wheel) and chemical (fire) technologies that we use. It took us longer to develop a workable microscope technology, but it is no less important.

At this early stage of the “century of nanotechnology,” we have microscopies that resolve $0.02nm$ on a surface [scanning tunneling microscope (STM)] or single atoms in a specimen [FIM and transmission electron microscope (TEM)]. Surely, that must be good enough. Well, we need even better. The ideal microscope would reveal the position and identity of all atoms in a specimen for as far as the user wishes to see. Is this realistic? It is not possible in any scanning probe instrument or electron microscope. To date, this objective has not been achieved but today’s atom probes approach this ideal. In this article, the state of atom probe tomography is reviewed with an emphasis on how the technique might be applied in current and future applications.

Atom probe technology is in the midst of a revolution which is at least as profound as the development of what is often referred to as the “three-dimensional atom probe”⁴ (3DAP) some $20years$ ago. In simple terms, there are three serial steps involved in obtaining information from a scientific instrument: specimen preparation, data collection, and data analysis. Any one of them can be a bottleneck of the technique. In atom probe work, data collection has been the principal bottleneck historically. Data collection in today’s atom probes takes about $1h$ or less. A few years ago it took days to accomplish the same task. The bottleneck has thus shifted to specimen preparation. New methods have been developed in specimen preparation also so that it, too, is quick and may even be automated. Thus, two key limitations to wider adoption of atom probe tomography have been addressed. What about the third step? It is not simple to categorize the speed of the data analysis step since it can range from trivial and instantaneous to difficult and ponderous. Once an analysis method has been developed, it may be automated. Therefore, this last step has the potential to be fast, and in the not-too-distant future, automated analyses could be used in applications such as repetitive process-monitoring tasks.

In the coming decade, it should be possible for the atom probe to make the transition from research laboratories to factories. Indeed, this is already happening. The technology has its limitations but the information it provides is unique and valuable and it is worth the effort to bring the richness of atom probe tomography to bear on the burgeoning activities of this century of nanotechnology.

The development of the atom probe technique traces its point of origin to in 1928 when, following Oppenheimer’s calculation of the time required to ionize an atom in vacuum,⁵ Fowler and Nordheim⁶ proposed a theory of electron emission from a solid by quantum mechanical tunneling. In 1935, Müller introduced a new type of microscope in which a needle-shaped specimen was placed in front of a phosphor screen.⁷ When a suitable negative voltage was applied, a field electron emission image from the specimen was produced on the phosphor screen. Due to the field enhancement, field strengths of the order of $1–10Vnm−1$ could be produced with the application of a few thousand volts on a needle with an end radius of about $1μm$⁠. The images produced with this field emission microscope had a spatial resolution of the order of $2nm$ and contained both crystallographic and electron work function information.

The FIM was a radical evolution of the field electron emission microscope.⁸ Four distinct changes were required for Müller’s new microscope to be successful in imaging atoms for the first time in October 1955.³ The polarity of the specimen was changed from negative to positive potential, an image gas was introduced into the system, the specimen was cooled to cryogenic temperatures, and finally, significantly sharper needle-shaped specimens were required. The field ion microscope is elegant in its simplicity of design and operation. The basic operation of a field ion microscope involves mounting a sharp needle on a cryogenically cooled stage in an ultrahigh vacuum system. A trace (typically $1×10−5mbar$⁠) of helium or neon image gas is introduced into the system. An increasing positive voltage is slowly applied to the specimen. As the voltage on the specimen is increased, the image gas atoms close to the apex of the specimen are polarized and attracted to the specimen where they are partially thermally accommodated to the cryogenic temperature. When the voltage on the specimens is increased to generate a sufficiently high field (typically $30–50Vnm−1$⁠), the image gas atoms on the surface are field ionized. This high field necessitates the use of needle-shaped specimens. The resulting ions are projected towards a phosphor screen positioned about $50mm$ in front of the specimen. The field ion image is the result of field ionization of gas atoms over individual atoms on the surface of the entire apex region of the specimen. Southon measured and explained the $I-V$ characteristic curve of a field ion microscope.⁹ By further increasing the field on the specimen, the surface atoms themselves can be field ionized and evaporated from the specimen. This field evaporation process thereby removes any surface films or damage produced during specimen preparation and also permits the interior of the specimen to be examined. Gomer’s seminal book on field emission and field ionization is a classic text in the field.¹⁰ 

For many years, these low intensity field ion images were produced directly on the phosphor screen. Nowadays, microchannel plate image intensifiers are positioned immediately in front of the phosphor screen not only to increase the gain by a factor of approximately 1000 but also to convert the incident ions into electrons which excite the phosphor screen more efficiently. Very few other improvements have been required in Müller’s original design apart from replacing the static reservoir of cryogenic liquid with a closed-cycle helium cryogenerator so that the temperature of the specimen can be varied continuously in the range of $15–110K$ to accommodate different materials.

In addition to the atom probe discussed below, a number of other related technologies also evolved from the field emission and field ion microscopes, for example, the field emission sources commonly used in electron microscopes, field emitter arrays,^11,12 and field ion sources for ion beams and propulsion sources.¹³ In addition, the topographiner developed by Young et al.¹⁴ and later developments of the scanning tunneling microscope and the atomic force microscope¹⁵ are directly related to the field ion micro-scope.

The next evolution of the technology was developed in order to identify the atoms that were imaged in the field ion microscope. Müller and Panitz,¹⁶ in 1967, and Müller et al.,¹⁷ in 1968, introduced the atom probe field ion microscope (APFIM). The atom probe is a combination of a field ion microscope and a mass spectrometer, as shown in Fig. 1. A small hole in the center of the microchannel plate and phosphor screen serves as the entrance aperture to the mass spectrometer and defines the extent of the region to be analyzed. An einsel lens is sometimes used between the phosphor screen and the detector to focus the divergent ion beam onto the detector. The specimen is mounted on a goniometer so that all points or atoms on the surface of the specimen can be selected for analysis. A circular counterelectrode positioned up to a few millimeters from the specimen is typically used in atom probes to define the field. In most types of atom probe, the ion travels in a field-free region after passing through the counterelectrode aperture. Although individual atoms can be analyzed, the atom probe is generally used to collect the atoms that originated in a small cylindrical volume where the diameter of the cylinder is defined by the effective size of the aperture. The evolution of the atom probe from the field ion microscope is shown in Fig. 1. Although an early atom probe with a magnetic sector mass spectrometer was constructed,¹⁸ the time-of-flight mass spectrometer has been universally adopted for the atom probe. Apart from the inherent simplicity of the design, the main advantages of the time-of-flight mass spectrometer are that it is equally sensitive to all elements (this is true when microchannel plates are used as the first stage of detection and the ions have at least $3keV$ energy) and no a priori knowledge of the type of ions or composition is required.

In the time-of-flight mass spectrometer, ions are field evaporated from the specimen by the application of either a high voltage positive-polarity pulse to the specimen or a negative-polarity pulse to a counterelectrode a few millimeters in front of the specimen. The amplitude of the pulse voltage is typically 15%–20% of the standing voltage on the specimen. The amplitude of the pulse, the standing voltage and the specimen temperature are all selected to ensure that all atoms have the same probability of being field evaporated on the application of the pulse and that effectively no atoms are field evaporated by the standing voltage. If these voltages are not selected appropriately, preferential evaporation may occur on the standing voltage, or preferential retention of the higher-evaporation-field elements may occur and lead to biased compositions. The number of atoms field evaporated in any pulse from the entire specimen apex region can readily be adjusted by adjusting the standing voltage. Typically, one or fewer atoms are evaporated per pulse to make sure that the detector system can correctly encode each atom. With position-sensitive detectors on 3DAPs, it is common to keep the average number of atoms per pulse to less than 0.1 to avoid multiple hits on the detector.

In the field evaporation process, the potential energy of an atom on the surface due to the voltages on the specimen, $neV$⁠, is converted into kinetic energy $12mν2$⁠, when the ion leaves the specimen. Here, $n$ is the number of electrons removed from the atom during field evaporation, $e$ is the charge on an electron, $V$ is the total applied voltage on the specimen, $m$ is the mass of the atom, and $ν=d∕t$ is the speed of the atom which, to a good approximation, is taken to be constant because the ions reach a terminal speed within a few tip radii, $d$ is the distance between the specimen and the detector and $t$ is the time of flight. Therefore, the mass-to-charge-state ratio $m∕n$ of the ion is given by

where $α$ is a factor less than 1, $Vdc$ is the standing voltage on the tip, and $Vpulse$ is the pulse voltage amplitude. The time of flight is determined with a time-to-digital converter (TDC) from the time between the application of the field evaporation pulse and the impact of the ion on the single ion detector at the end of the mass spectrometer. In order to reduce the background noise level, flight times are measured only over a timing window when ions are expected to strike the single atom detector. The duration of this window depends on the flight distance from the specimen to the single ion detector and is equal to the time interval for the entire periodic table from hydrogen to the highest mass-to-charge state ratio expected. Therefore, the length of this window and the time required to process the data impose a limit on the maximum possible pulse repetition rate. The mass resolution of the classical APFIM can be better than $1∕100$⁠. The mass resolution depends on the stability of the power supply for $Vdc$⁠, as shown by Wagner et al.¹⁹ In this article it was shown that it is possible to observe the four major isotopes of tungsten.

It was quickly recognized in these early voltage-pulsed atom probes that the full potential energy was not transferred to the ions as they were field evaporated from the surface of the specimen [hence, the use of $α$ in Eq. (1).] For electrical pulsing, these “energy deficits” are not necessarily small. The magnitude of the energy deficits depends on the pulse fraction employed and the value of $Vdc$⁠. Müller and Krishnaswamy devised methods to compensate for energy deficits through the addition of electrostatic lenses in the mass spectrometer.²⁰ These electrostatic lenses were based on sections of a toroid proposed by Poschenrieder.²¹ These lenses typically increase the mass resolving power or mass resolution²² of the instrument from $Δm∕m$ of less than $1∕100$ to better than $1∕1000$⁠. A commercial APFIM with a Poschenrieder lens, the APFIM 100, was made in the 1980s by Vacuum Generators of England. These lenses have the additional benefit of reducing the background noise as only those ions with the correct energy are detected. Therefore, ions that are created by the standing voltage, such as residual image gas atoms and preferentially evaporated specimen atoms, are eliminated. Values of $Δm∕m$ of up to 6000 have been achieved for ion beams with very small acceptance angles.

An alternative method that is appropriate to larger acceptance angles is the reflectron lens or energy mirror.²³ In the reflectron lens, ions are retarded and reflected back to the detector by an electric field. There is a nominally planar surface at which ions are time focused, with location independent of the mass-to-charge-state ratio. The detector is located at this surface. The achievable mass resolution of a reflectron system is ultimately limited by the magnitude of the energy deficit. Multistage reflectrons have also been proposed to improve the mass resolution of the atom probe even further.²⁴ A commercial reflectron-compensated APFIM was made in the 1980s by Applied Microscopy of England.

Historically, most atom probes have used a voltage pulse to field evaporate atoms from the specimen. However, the specimen material must have electrical conductivity greater than about $102S∕cm$ (heavily doped semiconductor) for the pulse to be effective. This has limited the application of the voltage-pulsed atom probe primarily to metals. Work on photodissociation of adsorbed molecules on field ion specimen surfaces with laser (photon) pulses was pursued by several authors in the late 1970s.^25–29 This work led to the concept of using short-duration laser pulses focused onto the apex of the specimen to induce field evaporation at the standing voltage on the specimen. Tsong²⁶ described experiments with pulsed field evaporation of tungsten in the field ion image mode with laser pulsing and identified many of the advantages of pulsed-laser-field evaporation for atom probe operation. Kellogg and Tsong³⁰ reported the first time-of-flight data from a pulsed-laser atom probe (PLAP) and concluded that the mechanism was due to thermally assisted field evaporation. They also showed that pure silicon can be pulse evaporated in the PLAP. The pulsed-laser atom probe is usually applied to the analysis of higher resistivity materials such as semiconductors. Laser pulsing has the added advantage that the energy deficits found in voltage pulsing are not present. This makes it possible to achieve high mass resolution without the need for energy-compensating devices. The extremes of this approach can be seen in the work of Liu et al.³¹ who used an $8m$ meter long flight path with $300ps$ laser pulses to achieve mass resolving power of 4000 in a one-dimensional atom probe. They were able to resolve the energy distributions of the evaporated ions to determine binding energies and correlate them with known cohesive energies of several elements.

Despite the advantages of laser pulsing, the technology essentially disappeared from the active atom probe developments in the 1990s. The reasons had to do with the extra expense of the lasers, the difficulty of achieving and maintaining alignment of the laser, and questions about the range of operating parameters for obtaining the correct composition of the specimen. Since about 2002, there has been a resurgence of activity as the technical advantages of laser pulsing, advances in laser technology, and need for analysis of semiconducting and insulating materials have combined to make this route more attractive. In particular, laser technology that produces ultrashort pulses $(100fsto10ps)$ with high optical quality in reliable systems has become available in the past half decade. Efforts to explore the benefits of very high intensity pulses (⁠$<1ps$⁠, $>1012W∕m2$⁠) have been conducted by Bunton et al.,³² Gault and co-workers,^33–35 and Cerezo et al.,³⁶ since about 2003 with particular emphasis on the mechanism of pulsing of field evaporation.

The early work on this topic³⁰ was performed with laser pulses of many nanosecond duration. All evidence from that early work and much of the more recent work has consistently indicated a thermal pulsing mechanism. Direct photoionization and pulsing by the electric field of the photon pulse have been considered in addition to thermal effects (see discussion in Ref. 37). Recently, following experiments with subpicosecond $(120–350fs)$ laser pulses, Gault et al.³³ have proposed that the field evaporation may be pulsed by the electric field developed at the specimen apex in a process later described as optical rectification by Vella et al.³⁸ Whether this latter pulsing mechanism may prevail and under what conditions is not yet established. The attraction of alternative pulsing mechanisms is the prospect that the pulse duration may be directly linked to the duration of the laser pulse $(<1ps)$ as opposed to the duration of the thermal pulse $(<1ns)$⁠, which includes the long time for the specimen to cool and depends on the specimen geometry and thermal diffusivity. Very high mass resolution might then be possible. The attractiveness of thermal pulsing is that it is general, and with fine focusing of the laser, even poor thermal conductors produce quality results.

Laser pulsing in 2006 is much more precise than in most of the early work. Pulse duration, irradiated volume magnitude, and absorbed energy are orders of magnitude smaller and much more tightly controlled and understood. Laser pulsing of atom probe specimens is rapidly becoming the preferred method of operation for atom probe tomography, at least for nonmetal specimens. As shown below, important data for semiconductor applications are now being obtained with these instruments. The first commercial pulsed-laser atom probe was shipped in 2006 by Imago Scientific Instruments Corporation.³⁹ Cameca SA, (Ref. 40) and Oxford nanoScience, Ltd.,⁴¹ now a division of Imago, are also developing and will soon be shipping atom probes with ultrashort-pulse laser pulsing. All of the work on laser pulsing is directly applicable to the three-dimensional atom probe.

The atom probe field ion microscope described in the previous section is very inefficient in the amount of specimen analyzed due to the small acceptance angle of the mass spectrometer. Therefore, some new variants of atom probe were developed to overcome that limitation. The $10cm$ atom probe, now known as the imaging atom probe (IAP), was developed by Panitz.^42,43 The imaging atom probe used a pair of spherically curved channel plates and phosphor screen positioned $10–15cm$ in front of the specimen so that the flight distances from the specimen to all points on the detector were identical. This single ion detector could be momentarily energized to record only the ions of a small range in time of flight, i.e., a single mass-to-charge-state ratio. Panitz used a gating procedure to detect different species. This procedure is viable as long as one knows a priori what species will be detected. Therefore, two-dimensional maps of the solute distribution of different elements could be obtained. This instrument has not served a significant role for metallurgical or semiconductor applications due to difficulties in simultaneously quantifying the concentrations of the different elements. However, this instrument enabled significant contributions to be made in the study of charge-state distributions and trajectory aberrations and also in imaging the shape of biological molecules.⁴⁴ 

In the 1980s a series of new atom probes was developed and introduced a new era in the history of the atom probe. As these instruments produce three-dimensional images of the internal structures of specimens from many slices, each containing a few atoms, this technique has been termed atom probe tomography (APT). Essentially, as atoms on the surface of the specimen field evaporate one at a time and fly toward a two-dimensional position-sensitive detector, their hit position in $x$ and $y$ is recorded. Gradually, each atom on the surface evaporates and exposes the underlying layer(s). The sequence of atom hits on the detector can be used to track both the serial evaporation of atoms in a given layer and the serial evaporation of the layers. The three-dimensional image is thus reconstructed from this combination of two-dimensional hit positions and field evaporation sequence.

The several variants of this instrument type primarily use different types of position-sensitive single ion detectors, as shown in Fig. 1, and include the following. The optical atom probe (OAP) uses a phosphor screen recorded with a charge coupled device (CCD) camera.^45–47 The position-sensitive atom probe^48,49 (PoSAP) measures the relative charge on the three anodes of a wedge-and-strip or backgammon detector.⁵⁰ The tomographic atom probe (TAP) measures the charge on several adjacent anodes of a $10×10$ square array of anodes.^51,52 The optical position-sensitive atom probe features a primary detector with a phosphor screen and a pair of secondary image-intensified detectors consisting of an $8×10$ array of anodes, a phosphor screen, and a CCD camera.⁵³ The optical tomographic atom probe (OTAP) uses a combination of a linearly segmented anode, a phosphor screen, and a CCD camera.⁵⁴ These instruments are collectively referred to as three-dimensional atom probes (3DAPs).⁴ The OAP was the first of these concepts that recorded each atom’s position and time of flight independently. However, the first true 3D data was obtained with the PoSAP. These events marked a seminal development in the history of the atom probe.

Position-sensitive detectors for the 3DAP must meet some of the most demanding performance specifications of any application. They should ideally be able to correctly record two or more like atoms that have evaporated on the same field evaporation pulse next to each other. This means that the events are barely separated, if at all, in space or time on the detector. If a detector cannot correctly encode more than one simultaneous hit, then the probability of an evaporation event during any given voltage pulse must be kept low $(<10%)$ so that the probability of multiple atom evaporation events is kept very small $(⪡1%)$⁠. Thus, the data collection rate of an atom probe is greatly impacted by the ability of the position-sensitive detector to record multiple hits, and orders of magnitude in data collection rate are at stake. The original PoSAP had a serial detector that could not accept multiple hits. Subsequently, detectors have addressed this need in various ways. The TAP had a large degree of parallel detection with its array of 96 detector plates. The optical PoSAP and OTAP used complementary optical signals to gain some multihit capability. In all of these embodiments, the cycle time of the detector is a consideration that can limit the repetition frequency of the atom probe.

Delay line detectors^55,56 have an inherent capacity to encode near-simultaneous hits in space or time. Their use on atom probes was first suggested by Andrén and Stiller,⁵⁷ and they have been successfully applied in all of the most recent commercial instruments. The crossed delay line (CDL) detector features a pair of delay lines placed perpendicular to each other in place of the phosphor screen or anodes used in the other variants of 3DAP. When the electrons from the microchannel plate strike the delay line at some location, they travel to each end of the line. The position of the ion’s impact is therefore determined from the arrival times measured at each end of the delay line. These times are converted into true $x$ and $y$ distances on the detector. Today’s crossed delay line detectors and high speed digital timing systems provide $1000×1000pixels$ at rates up to $500000ions∕s$⁠.³⁹ The electronics are designed to discriminate between hits separated in space or time by more than about $5ns$ (time resolution on a delay line detector is equivalent to spatial resolution across the detector.) Da Costa et al.⁵⁸ have used novel digitization and deconvolution of the timing signals in a delay line detector to improve this multihit resolution to about $1.5ns$⁠. The computational burden on the digital electronics system needed to process this information can limit the cycle time of the instrument, but that concern is greatly diminished with fast processors. Cameca offers this capability on their instruments.⁴⁰ Multihit capability is also enhanced by the addition of a third detector axis^59–61 that is used to resolve the ambiguity between two simultaneous hits (⁠$x1$⁠, $x2$⁠, $y1$⁠, $y2$⁠) and to determine the correct $xy$ pairs. This detector feature can improve multihit capability significantly and is now used on commercial detectors from Oxford nanoScience⁴¹ and Imago.³⁹ 

The data from the position-sensitive detector, i.e., the pixel coordinates from the CCD camera or the charges from the anodes, are converted into true $x$ and $y$ distances on the detector and then scaled by the magnification to obtain the $x$ and $y$ coordinates of the atom in the specimen. Small trajectory aberrations due to interactions with the neighboring atoms as the atom leaves the surface limit the spatial resolution in this $x-y$ plane to $∼0.2nm$⁠. The third or $z$ coordinate, which is determined from the position in the evaporation sequence, must take into account the detection efficiency of the mass spectrometer, typically $∼60%$⁠, and the atomic density of the material under analysis. The identity of the ion is determined from the mass-to-charge-state ratio. This assignment process is straightforward as there are only one or two typical charge states for most elements, or a limited number of molecular ions, e.g., $M2$⁠, $M3$⁠, and $MX$ where $M$ is a metal atom and $X=H$⁠, C, N, and O atoms. In cases where there are overlaps, the natural abundances of the other isotopes can generally be used to deconvolute the overlap for the analyzed volume.⁶² The composition is determined from the number of atoms of each type in the analyzed volume. For overlapping peaks that can be deconvoluted, the identity of any individual atom in the overlapping peaks is not known independently but can be displayed in images as a statistical combination of the overlapping isotopes.

Some of these three-dimensional atom probes are equipped with reflectron lenses to improve the mass resolution. However, the field of view is necessarily restricted to $<0.05sr$ by the reflectron lens. In energy-compensated variants of 3DAP, the mass resolution is generally of the order of $Δm∕m=∼1∕500$ full width at half maximum (FWHM) at $m∕n=30$⁠. In contrast, the mass resolution of an instrument without energy compensation is low (⁠$Δm∕m=∼1∕100$ FWHM at $m∕n=30$⁠). However, with this latter, unencumbered configuration, it is possible to collect data from a large field of view (⁠$>1.5sr$⁠). Two commercial companies have offered reflectron-compensated 3DAPs: the ECOPoSAP from Oxford nanoScience, Ltd.⁴¹ and the ECOTAP from Cameca SA.⁴⁰ 

These three-dimensional atom probes have been used to study a wide range of metallurgical materials.⁶³ However, there are several types of samples, such as devices or thin films fabricated on semiconducting wafers, where the needle-shaped specimens are difficult to fabricate. These specimen preparation methods are described below. About the same time, Nishikawa and co-workers^64–66 were also interested in the analysis of planar specimens and they devised the concept of a positionable counterelectrode that is close enough to the specimen to permit only one protrusion to be selected for analysis. Nishikawa and co-workers called this instrument the scanning atom probe (SAP) as the counterelectrode can be scanned along the surface until the field evaporated ions from a protrusion are detected. Perhaps the most important result of this new approach to atom probe design was the fact that it radically altered the shape of the specimen and made “microtip” geometries possible. Microtips lend themselves to preparation from planar structures much more so than do needles. Furthermore, when coupled with a “local electrode” (see below), they may be arrayed on a single specimen holder which greatly amplifies the throughput of an instrument and increases the prospects for iterative and combinatorial analyses.

Kelly et al.^67,68 recognized the importance of the proximity or “local” effects of the counterelectrode with the specimen and associated the name, local electrode atom probe, or LEAP^® with this configuration. In the LEAP, the cryogenically cooled planar or needle-shaped specimen is mounted on a nanopositioning stage and pointed towards a funnel-shaped local electrode, as shown in Fig. 1. The specimen is generally roughly aligned to the aperture in the local electrode with the aid of a pair of long range optical microscopes. The detected field evaporated ions are used to finalize the alignment.

For a small (⁠$20–50μm$ diameter) local electrode positioned close to the specimen (i.e., less than one aperture diameter away), the field at the apex of a specimen is greater at a given extraction potential than for a traditional remote (⁠$5–10mm$ diameter, $4–10mm$ from the specimen) counterelectrode. Because the voltages are lower, it is economically possible to build suitable pulsers (⁠$2ns$ FWHM pulse with amplitudes of up to $2000V$⁠) that can operate at pulse repetition rates up to $250kHz$⁠. This pulsing rate is two orders of magnitude faster than previous instruments. In order to correctly determine the position of ions at these higher rates, a delay line detector is needed for its high data rate capability.

In addition, the ions that are field evaporated from the specimen reach the local electrode in a significantly shorter time (tens of picoseconds) and thereafter are shielded from time-varying fields. This reduces the major source of the energy deficits over the entire field of view. Therefore, the local electrode atom probe can achieve high mass resolution (better than $Δm∕m$ of $1∕600$ FWHM at $Al+$⁠) over a $100nm$ field of view. This mass resolution is sufficient to separate the individual isotopes of all the elements. The energy spread of the ions is not completely eliminated and the peaks in mass spectra from a LEAP^® have a greater width at their base than spectra from reflectron-compensated instruments. The principal impact of this is in detecting elements in very low concentration whose peak location resides on the longer-flight-time side of a large peak. The large field of view permits up to $∼250000$ atoms per atomic layer to be collected. Data sets with more than $108$ atoms have been obtained at collection rates of greater than 20 000 correctly positioned ions per second.

The benefits of the LEAP^® design concept, large field of view with high mass resolution, high data collection rate, and planar microtip capability, are mutually inclusive. It is difficult to have one without the other: a 50 times larger field of view would require 50 times longer acquisition without a concomitant increase in data collection rate. These features are incorporated into a commercial instrument³⁹ that achieves the high performance gains originally envisioned for this geometry.

The LEAP^® achieves major performance gains that represent more than just speed for speed’s sake. Practical applications require that sufficiently large volumes are investigated to sample the features of interest. The largest accessible volume that the technique can characterize dictates what materials problems can be tackled. The high data collection rate and large field of view of the LEAP^® microscope extends this range by a couple orders of magnitude, as shown in Fig. 2. The larger accessible volumes have been found to be critical to resolving questions in many cases.

Laser pulsing of the LEAP^® geometry has also been developed. It should be noted that some of the performance gains of the voltage-pulsed LEAP^® are enabled by the advent of modern laser pulsing. For example, high mass resolution can be achieved in a large field of view with a laser pulsed instrument, and high repetition rates are also possible with a laser. The commercial laser-pulsed instruments, the LEAP^® 3000X from Imago,³⁹ the LA WA TAP from Cameca,⁴⁰ and the 3DAP X from Oxford nanoScience⁴¹ are examples of such systems. It should also be noted that larger area detector systems are being developed by all of the commercial suppliers and that advances in this area are expected in the next few years. Fields of view in excess of $200nm$ have been achieved on aluminum specimens already.⁶⁹ The laser pulsed LEAP^® retains the advantage that it works on microtip specimen geometries.

Another major development in atom probe instrumentation occurred when the large angle reflectron compensator (LARC™) was recently developed.⁷⁰ This component effectively overcomes the major limitation of reflectrons as used in 3DAP: the spatial position of ions is distorted according to the ion energy, i.e., chromatic aberration. With the LARC, the surface of time focusing of the ions is close to coincident with the surface of spatial focusing. The flight path length will be relatively long in a LARC which has two main ramifications: (1) the mass resolution in pulsed-laser mode can be very high (better than $1∕2000$ FWHM at $m∕n=30$⁠) due to the long time of flight, and (2) the maximum pulse repetition rate may have to be limited $(<500kHz)$ due to the long time of flight. With a LARC, it is thus possible to build 3DAPs with a large field of view $(>70nm)$ and high mass resolution in both pulsed-voltage mode and pulsed-laser mode. Oxford nanoScience⁴¹ currently offers such an instrument. The LARC technology is fully compatible with LEAP^® as well.

As with many experimental techniques, specimen preparation for atom probe tomography is the key to a successful experiment. Fortunately, there has been an explosion of new ideas and methods for atom probe specimen preparation similar to that which occurred in transmission electron microscopy. In fact, most of the methods developed for TEM can be directly applied to atom probe specimen preparation. Specimen preparation is often the last part of the analysis process that is automated. It can dictate whether the technology is adopted and practical or relegated to specialized laboratories. With the high speed of data analysis available in today’s atom probes, specimen preparation is now the main bottleneck in the process.

The various techniques can be viewed as a collection of tools for different types of materials including bulk specimens, thin films, surface regions, implanted regions, and microtips. Not all tools will be appropriate for any given job, and sometimes it takes several tools to finish the job. The more tools available, the more likely we will be to find a way to quickly and effectively make the best specimen. The right tool for any given job depends not just on the specimen but, perhaps more importantly, on what information the user seeks. In the following sections, every technique from the earliest to the most recent methods is described.

The essential geometry of an atom probe specimen is a sharp needle that tapers with a modest angle (⁠$<10°$ semiangle) to an apex between 50 and $150nm$ radius. Since the radius of the apex of the specimen increases during the experiment due to the continual removal of atoms from the surface, the aim of specimen preparation is to produce a needle with a starting end radius of less than $50nm$ and a high quality surface finish with no protrusions. This shape is dictated by the high field required for the field ionization and field evaporation processes. Atom probe specimens also require that they be self-supporting and, for voltage-pulsed atom probes, have an electrical resistance to the apex of about $102Ω$ or less. For the typical geometry above, the specimen resistance is dominated by the last $10μm$ or so. If the specimen material is uniform, this indicates a specimen resistivity of less than $∼0.05Ωcm$⁠. Historically, a small taper angle has been maintained along most of the shank of the needle. However, a cusp shape (i.e., the cross-section diameter increases faster than linearly with distance from the apex) provides some significant advantages as this shape has more mechanical stability and also reduces the electrical resistance of the shank due to the larger cross section.

Electropolishing has been the main method used to fabricate these needles from metal wires, whiskers, or blanks (typically $0.25×0.25×10mm3$⁠) cut from bulk material. Several reviews of these methods have been published,^71–75 so only a brief description of these methods will be presented.

The most common electropolishing method is to suspend the wire or blank in a thin $(6–8mm)$ layer of electrolyte that is floated on top of a dense inert liquid such as a polyfluourinated polyether. As electropolishing only occurs within the electrolyte, a necked region is formed in the center of the wire. If electropolishing is continued, the weight of the lower portion of the wire will eventually be too much to be supported by the necked region and the lower half will separate, forming two atom probe specimens. Generally, the specimen is transferred to a more dilute and controllable electrolyte just before separation so that a good surface finish and a smaller end radius is obtained. Occasionally, chemical polishing is used. An example of a quality electropolished metal atom probe specimen is shown in Fig. 3(a).

Fractured or blunt specimens may be resharpened with a micropolishing method.^73–76 In this method, a drop of electrolyte is suspended in a small diameter $(3mm)$ platinum wire loop. The apex of the needle is inserted into this drop to complete the cell, and a voltage is applied. The time and position of the specimen in the electrolyte may be adjusted to sculpt the end form into a desired shape and taper angle. This method may also be used to remove oxide films and any damaged region formed in other methods.

Pulse polishing may be used to remove extremely small amounts of material from the apex of the specimen to bring a microstructural feature, such as a grain boundary into the analyzable portion of the needle.^77–81 In this method, after some calibration experiments, a number of short-duration voltage pulses are applied to the specimen immersed in an electrolyte. This method is usually performed in conjunction with examination in a transmission electron microscope to monitor the progress. Ion milling in a low energy argon ion beam may also be used for the same purpose.

As in electron microscopy, ion beam methods of specimen preparation have been finding their way into the toolbox. Broad ion beam (BIB) methods were first used by Liddle et al.⁸² to make microtips on a surface of a wafer. They used $1μm$ polystyrene spheres as masking particles to protect the substrate from an ion beam (⁠$∼5keV$ $Ar+$ ions) impinging at normal incidence. As material is sputtered away by the ion beam, the masking particle shields the materials below it until it, too, is sputtered away. Liddle et al. noted that the polystyrene sputtered quickly relative to the semiconductors they were studying and suggested using particles with lower sputtering rates. Larson et al.⁸³ employed diamond particles to effectively create microtips suitable for analysis in a local electrode atom probe [Fig. 3(b)]. The microtips form into the shape of cones (at least for axially symmetric mask particles) with a semiangle equal to the elastic scattering angle of the incoming ions. The cones can be as sharp as $5nm$ radius at the apex. Camus et al.⁸⁴ considered what the optimal shape of the masking particle would be and found through modeling of the sputtering process that a shape similar to a typical electron-beam-induced deposit on a surface is best. With energetic ion beam sputtering there will always be implantation of the ions into the specimen. The low energy ions used in this method penetrate only a few nanometers which can be removed readily, or the implanted region may be ignored in the analysis of the data.

One of the principal advantages of the masked BIB approach is that large numbers (hundreds) of microtips may be made in parallel. It also relies on inexpensive low energy ion guns. The masking particles can be as simple as diamond polishing grit. With randomly located masking particles, the technique is not location specific. Many other possibilities exist for depositing masking particles. Photon and electron lithography and electron-, ion-, or photon-beam-induced decomposition of a gas are just some examples. In principle, it is possible to make masking particles at specific locations with any of these methods.

Focused ion beam (FIB) instruments have become a mainstay of specimen preparation in electron microscopy. Over the past decade, similar methods have been used with increasing regularity for the preparation of atom probe specimens. The first efforts using a focused ion beam were reported by Waugh et al.,⁸⁵ but it was not until efforts by Larson et al.,^86–90 Schleiwies,⁹¹ and Thompson et al.⁹² to make specimens from multilayer films that the technique became widely adopted. The use of coupon extraction from bulk or wafer materials was pioneered by Kelly et al.⁹³ and Thompson et al.⁹⁴ to make semiconductor specimens and by Miller et al.⁹⁵ for metallic specimens.

FIB-based specimen preparation methods have made it possible to fabricate atom probe specimens from a significantly wider range of materials and alternative specimen geometries than traditional electropolishing methods. The wide range of materials includes multilayer and surface films,^92,96 semiconductor devices,⁹⁴ carbon filaments and whiskers, ceramics, and oxides including magnetite.^97,98 FIB-based methods have been developed to fabricate specimens from powders, ribbons, and thin sheets.^95,99,100 Lift-out or coupon extraction methods⁹⁵ have been used for site-specific specimen preparation of features such as grain boundaries,⁹⁹ interphase interfaces, low volume fraction or inhomogeneously distributed phases, coarse precipitates,¹⁰¹ as well as ion-implanted and subsurface regions.¹⁰² In addition, the FIB may be used as a final step for electropolished specimens to correct electropolishing artifacts and ensure a circular cross section with a smooth taper angle.¹⁰³ 

Generally, the FIB-based methods require a two-stage process. In the first stage, a post or a blank is fabricated or extracted from the region of interest. The blank is mounted on a support post with Pt or W deposition. The post or the blank is then sharpened into the atom probe needle-shaped specimen by an annular milling technique in which the outer and inner diameters of a circular mask and the milling current are decreased progressively.

A typical example of flat-topped posts⁹⁰ that were prepared in silicon by deep reactive ion etching (the Bosch process) is shown in Fig. 4(a). Because the posts contain a flat silicon surface on which standard thin film preparation and deposition may occur, this type of specimen is expected to be representative of actual structures. Wafers with regions of etched posts and unetched flat space may be made so that physical property measurements may be made and correlated with the APT characterization. Nishikawa and Kimoto⁶⁵ and Kuhlman and Wishard¹⁰⁴ noted that similar structures can be made with a diamond dicing saw, as shown in Fig. 4(b). Posts as small as $7μm2$ and $100μm$ tall have been made.¹⁰⁴ This mechanical approach enables existing structures such as thin films and bulk metals to be prepared. After the films are deposited, these posts can be removed from the wafer and can be attached to metal needles with an electrically conducting silver epoxy. Alternatively, if the posts are sufficiently far apart and the wafer is less than $1cm2$⁠, they may be analyzed in a local electrode atom probe without removal from the wafer. In either case, the posts must be sharpened with the annular milling method.

Prefabricated arrays of posts, both flat topped and presharpened, have been made¹⁰⁵ in the form of small coupons that are etched into silicon wafers (Fig. 5). Each coupon can greatly facilitate the preparation of specimens from a particular deposition and can be considered as witness coupons for process monitoring. The presharpened arrays or microtips can be used with no additional specimen preparation. In some cases, the curvature of the substrate of the presharpened array may not matter [see, for example, the extensive work by the groups in Göttingen and Münster ^106,107] and may more closely approximate spherical catalyst particles. Microtip arrays permit the rapid characterization of many identical tips for improved statistical analysis and potentially bypass one of the three bottlenecks in atom probe tomography.

One of the major concerns of the FIB-based technique is the damage caused to the specimen by the ion beam. Commercial instruments use gallium liquid-metal ion sources with accelerating voltages between 2 and $30keV$⁠. Both gallium implantation and ion mixing produce damage and artifacts in the specimen. The atom probe can be used to measure the level and depth of gallium implantation into the specimen. Dual beam instruments are beneficial as the scanning electron beam may be used to image the specimen rather than the gallium ion beam, thereby reducing damage. In order to minimize this damage, a thin protective Pt or W cap is generally deposited on the surface with either the electron or ion beam before milling. This cap may also be used to monitor the precise position of the surface. The use of low accelerating voltages $(2–5keV)$ for the final stage of annular milling has been found to be beneficial in minimizing the gallium implantation depth and hence the damaged region.^88,94

Comprehensive texts on field ion microscopy, atom probe, and atom probe tomography studies have been published elsewhere.^63,108 The following examples illustrate some of the types of studies that may be performed on metals, alloys, and other types of materials.

The pressure vessels used in nuclear reactors have been studied extensively with atom probe tomography due to their technical importance.¹⁰⁹ These materials become embrittled in service due to the interaction of the solutes in the steel vessel with the vacancies and other products created by the incident neutrons. Atom probe tomography has revealed that there are several contributions to this embrittlement including ultrafine copper-enriched precipitates, phosphorus clusters, and segregation to dislocations and grain boundaries. An example of a pair of atom maps from a model pressure vessel steel (⁠$Fe–0.1wt%$ Cu, 1.6% Ni, 1.6% Mn, trace P) is shown in Fig. 6. In this representation, a colored sphere is plotted at the position of each atom. Typically, only subsets of all the atoms, such as some of the different solute atoms, are used in this type of representation so as not to obscure the features present. In this local electrode atom probe example containing a total of over $7×106$ atoms, only the copper (green) and phosphorus (orange) atoms are shown. A high number density $(∼1.5×1023m−3)$ of small copper-enriched precipitates is evident in this atom map. These precipitates were also found to be enriched in manganese and nickel. In addition, a phosphorus atmosphere around a dislocation is visible. The radius of gyration $(∼1.6nm)$ or size of each copper-enriched precipitate may be estimated from the positions of the copper, manganese, and nickel atoms in the precipitate. The number density is estimated from the number of precipitates in the analyzed volume. Their composition can also be estimated from the numbers of different types of atoms within an envelope defining the extent of the precipitate.¹¹⁰ 

Many different classes of nickel-based superalloys have also been characterized by atom probe tomography.^111–113 These studies have primarily focused on the partitioning of the alloying elements between the different phases, solute concentration profiles across interphase interfaces and grain boundaries, and the site occupancy of solute additions in the ordered phases. For example, atom probe tomography has been used to identify the precipitates in the most common nickel-based superalloy Alloy 718.¹¹⁴ There had been significant debate in the literature whether some of the precipitates in this niobium-containing superalloy were the $L12$-ordered $Ni3Al$ $γ′$ phase or the $DO22$-ordered $Ni3Nb$ $γ″$ phase as both of these phases can contain significant levels of Al, Ti, and Nb. An atom map of an Alloy 718 sample that was isothermally annealed to produce the undefined precipitates is shown in Fig. 7. The iron- and chromium-enriched $γ$ matrix surrounding the small $γ′$ precipitates is clearly evident. A close examination of the individual precipitates reveals that they contain aluminum- and titanium-enriched regions characteristic of the $γ′$ phase and niobium- and titanium-enriched regions characteristic of the $γ″$ phase. Some precipitates were found to consist of four such regions.

The morphology and distribution of phases may be characterized. Isoconcentration surfaces are used to visualize the interface between regions of different compositions such as the $γ′$ precipitates and the surrounding $γ$ matrix. An isoconcentration surface representation from a CMSX4 superalloy is shown in Fig. 8. In this $32×106$ atom example, the edges of two large $γ′$ primary precipitates are evident. A distribution of finer, roughly spherical $γ′$ precipitates is also seen in the $γ$ channels between these larger $γ′$ precipitates. A precipitate-free zone close to the $γ′$ primary phase is also evident. The large field of view and the significantly larger number of atoms collected in the local electrode atom probe is extremely beneficial in the characterization of this type of coarse microstructure.

The high spatial resolution of the technique enables the early stages of phase separation and their evolution of the phases to be characterized. Classic examples of this type of study are the phase separation that occurs within the miscibility gap in the iron-chromium system^115–117 and the formation and evolution of Guinier-Preston (GP) zones in aluminum alloys.¹¹⁸ In these examples, the size, composition, and morphology of the phases and solute segregation at interphase interfaces may be estimated. An example of the morphological information that may be obtained is illustrated in the phosphorus isoconcentration surface of a $Pd40Ni40P20$ bulk metallic glass that was isothermally annealed below the glass transition temperature shown in Fig. 9. An isotropic interconnected microstructure of two amorphous phases was found to precede primary crystallization.

One of the most difficult microstructural features to characterize is solute segregation to dislocations due to their small extent and the low dislocation densities in most materials.^119–121 However, several APT characterizations of dislocations have been performed. In order to improve the dislocation density, these studies have been mainly applied to cold-worked or mechanically alloyed materials or neutron-irradiated materials. An exceptional example of boron segregation to dislocations in B2-ordered FeAl was obtained by Blavette et al.¹²² Other examples include phosphorus and carbon segregation to dislocations in neutron-irradiated pressure vessel steels as shown in Fig. 10 ^{120,123–125} and segregation to dislocations in mechanically-alloyed oxide-dispersion-strengthened ferrite steel¹²⁶ and low carbon steels.^127,128

Segregation of solute species to two-dimensional interfaces such as grain boundaries and interphase interfaces can have a major impact on a wide range of materials properties in real applications. This phenomenon has been widely studied with the atom probe since its inception. However, the advent of three-dimensional imaging in APT has made this a far more powerful endeavor. Indeed, three-dimensional images are necessary for a general evaluation (no special orientation or conditions) of two-dimensional buried interfaces. This type of application clearly demonstrates the power of APT.

Marquis et al. studied Mg segregation to the interface of $8nm$ diameter $Al3Sc$ precipitates in an $Al–2.2at.%$ $Mg–0.12at.%$ Sc alloy with both APT and first principles computations.¹²⁹ An image of a single precipitate and a proximity histogram¹³⁰ composition profile derived from the same heat treatment conditions are shown in Fig. 11. In this case, the proximity histogram of several images provides quantitative evaluation of interfacial segregation on a subnanometer scale. These data were compared quantitatively with first principles calculations to arrive at the understanding that the Mg segregation is in equilibrium thermodynamically and is due to attractive second and fourth nearest neighbor interactions between the Mg and Sc atoms. The fact that Mg additions in this alloy lead to reduced faceting of the precipitates was explained in the modeling as due to less crystallographic anisotropy in the interface energy between the precipitate and the Al alloy matrix. The good agreement between experiment and modeling of the interfacial excess of Mg allowed the authors to estimate with confidence the reduction in the interfacial energy between the precipitate and matrix upon addition of Mg, which is also consistent with reduced precipitate coarsening rates of $Al3Sc$ in an AlMgSc alloy when compared to coarsening of $Al3Sc$ in an Al–Sc alloy.

Multilayer thin film structures are used in a variety of ways in information storage technologies. They may be used for storage media, read/write components in hard drives, magnetic memory structures, and spintronics. These materials have received a large amount of attention from practitioners of atom probe tomography.⁹⁶ In addition, research has been performed on thin films deposited onto pre-formed (usually metal) needles.¹⁰⁶ Other researchers have performed fundamental studies of phase stability and interdiffusion.^107,131,132

There is an excellent match between the strengths of the atom probe technique and the needs of this application area. The ability to characterize buried interfaces compositionally in a three-dimensional image is the key. The properties of devices built from these thin film structures depend on both chemical intermixing or diffuseness and morphological roughness at buried interfaces and also on such parameters as segregation to the interfaces and even layer thickness. Atom probe tomography provides information on all of these parameters.

The flat-topped-post specimen preparation method for thin metal films has been used extensively in the past five years. Many basic questions concerning thin metal films have been answered by this time by atom probe investigation. These include origin of an asymmetry in interface diffuseness in $Cu∕CoFe$ layers,¹³³ the reasons for smoothing of rough interfaces during layer growth,¹³⁴ and the role of oxygen in sputter gases in giant magnetoresistance (GMR) performance.¹³⁵ 

As APT has emerged as a powerful force in materials characterization, it has become apparent that molecular dynamics (MD) simulations of materials is a synergistic partner in much of these studies for several reasons. Firstly, APT provides information on a length scale that is well matched to MD simulation. Indeed, APT is one of the few experimental techniques that lends itself to direct comparisons with MD simulations. Secondly, molecular dynamics simulation makes it possible to add temporal evolution of structures to the static information obtained from APT. In the following examples, MD simulation has been utilized to enhance significantly the outcome of a study of multilayer metal structures. More generally, it has been applied to a variety of APT studies, and its use is likely to increase significantly in this field.

In a typical multilayer film device, there may be tens of layers with five or more layer materials. Since the interfaces between these layers usually play a critical role in determining their properties, it is essential that the interfaces be examined and understood. Any possible improvement in magnetic or electrical properties can lead to significant increases in storage speed and density. If one considers all possible types of interfaces, there can be tens of permutations in a single device. In practice, subsets of these layered structures are usually prepared for a detailed analysis. The examples below illustrate the use of APT (and MD simulation) for basic studies of multilayer structures. Eventually, it is expected that materials characterization of the types shown below may even be extended to routine evaluation of production devices.

One frequently utilized layer combination in the information storage industry is the paramagnetic Cu spacer layer sandwiched between layers of ferromagnetic $Co–10at.%Fe$⁠. These structures are designed using theoretical principles to determine optimal layer thicknesses and compositions. The best properties are obtained in general when the interfaces are perfectly abrupt (low diffuseness) and the layers are flat (low roughness). In projection imaging such as in a transmission electron microscope (TEM), it is not possible to distinguish between diffuseness and roughness unless the scale of the roughness is larger than the specimen thickness (typically $40nm$ or more.)

An experimental evaluation of a $CoFe∕Cu∕CoFe$ multilayer sandwich is shown in Figure 12(a), which is a subset of a much larger data set. These layers were grown upward in the figure so that the CoFe layer was deposited onto NiFe and so on. Visually, in Fig. 12(a), the CoFe-on-Cu interface appears more diffuse than the Cu-on-CuFe interface. A composition profile along the growth direction upward was obtained from this subvolume and is shown in Fig. 12(b). There is a clear suggestion from this profile that the CoFe-on-Cu interface is more diffuse, as evidenced by the more gradual transition in the Cu composition at around $8nm$ on the profile. This observation has been made multiple times on different depositions with different instruments and has always been repeatable. It was not expected when it was first found. It was also not clear initially whether this effect was due to the fact that Cu has a lower homologous temperature at the deposition conditions and is therefore able to mix with the CoFe more than vice versa or whether the surface energies of Cu and CoFe were different enough to account for differences in mixing. A MD simulation that was already part of an independent, ongoing study was compared with these data Fig. 12(c),¹³³ and an excellent match between experiment and model was found. The composition profile derived from the model [Fig. 12(d)] reproduces the experimental data almost exactly. Interface widths of 1.08 $(±0.18)nm$ and 0.4 $(±0.14)nm$ for the CoFe-on-Cu and Cu-on-CoFe interfaces, respectively, were measured in the experimental data and compare well with $∼1.44$ and $∼0.33nm$ from the model.

Similarly, the smoothing that occurs with each deposition of Cu layers was observed and quantified with APT (Ref. 134 and 136) and was shown to be due to the low surface energy of Cu {111} which are normal to the growth direction.¹³⁷ This example illustrates well the power of combining experimental characterization and modeling at the atomic scale. The authors know of no other experimental technique that can provide the information needed to make this comparison. Indeed, the good agreement between the simulated and experimental data demonstrates, perhaps for the first time, that APT analysis can be used to check atomistic modeling and that atomistic modeling can then be used to extend our understanding of the material.

Even though there is a need for low electrical resistivity of the specimen $(<0.05Ωcm)$ in voltage-pulsed atom probes, thin dielectric layers may be analyzed so long as they are less than about $4nm$ thick. Kuduz et al.¹³⁸ have performed APT studies of thin $(2.5nm)$ aluminium oxide layers sandwiched between nickel-iron layers (Fig. 13). These stacks are candidates for use in data storage devices which utilize the tunneling magnetoresistance (TMR) effect. Despite the fact that alumina is one of the highest resistivity materials known, these thin dielectric layers were successfully run in a voltage-pulsed atom probe.

The formation of the aluminium oxide tunnel barriers in spin tunnel junction structures¹³⁹ is usually carried out by depositing metallic Al on the underlying ferromagnetic layer (such as a Ni, Fe, or Co alloy) and then by oxidizing. The stoichiometry of the oxide layer was found to be correct (at. % O/at. % $Al=1.50$⁠) when sufficient oxygen was available in the atmosphere during sputter deposition of the oxide, and it was deficient in oxygen when a native oxide was allowed to form on a thin aluminum metal film. Some nickel and iron were also found in the sputter-deposited alumina layer. This latter finding was unexpected. Both of these findings illustrate the power of APT for atomic-scale evaluation of processes and materials that could not have been obtained by any other means.

Because of the difficulty with pulse evaporating higher resistivity materials in voltage-pulsed atom probes, little work has been performed on ceramics. Early work on laser pulsing showed that pulsed evaporation in high resistivity materials such as alumina,¹⁴⁰ silicon,¹⁴¹ and silicon carbide¹⁴² was possible. With the recent resurgence of PLAP, more high resistivity materials have been examined.^102,143 However, much more work needs to be done to show that the correct compositions and quality 3D projections can be obtained. Progress is being made, and there are certainly many deserving materials characterization challenges in ceramics.

Because of its technological importance, several efforts have been made to analyze silicon in an atom probe. Work with PLAP has already been cited above. In addition, Melmed et al.¹⁴⁴ were able to analyze high resistivity silicon in a voltage-pulsed atom probe by using long-duration voltage pulses. They found that pulse lengths of $17ns$ FWHM were sufficient to analyze $70Ωcm$ Si, $80ns$ FWHM for $8600Ωcm$ Si, and $320ns$ FWHM for $2.5×104Ωcm$ Si. Melmed et al. pointed out that the resistance of the specimen, which is dominated by the final tens of micron length near the apex, determines the time constant of the electrical circuit that must transmit the pulse. The $RC$ time constant must be comparable to the pulse duration. In this work the three isotopes of silicon and their ionization states were determined but no observation of the dopants in the silicon were reported.

Observation of dopants in Si has been of great interest in the atom probe community for many years. Because of their generally low concentration (⁠$10appm$ or less), it is necessary to study highly doped Si or collect at least $106$ atoms in an analysis. The first three-dimensional image of boron in Si came from work on a voltage-pulsed energy-compensated 3DAP. The specimen was made by FIB sharpening of Si posts as in the work on multilayers above (see Ref. 136 for example). This silicon work was not published except as a note in a conference proceedings.¹⁴⁵ 

With the advent of commercial ultrafast pulsed lasers, laser pulsing of semiconductor specimens has become the method of choice for atom probe tomography. Pulsed-laser analysis provides superior mass resolution and sensitivity and, in addition, is capable of analyzing undoped Si structures of high resistivity. Pulsed-laser APT has been used to map one-, two-, and three-dimensional distributions of boron, arsenic, and phosphorus with high spatial and compositional resolution. Some examples include the observation of arsenic implant profiles in silicon,¹⁰² boron segregated to grain boundaries in polysilicon,¹⁰² and the lateral diffusion profile of boron in a transistor from a source/drain into the gate region.¹⁴⁶ Figure 14 (Ref. 147) shows an image of a high energy implant of the $boron10$ isotope into silicon. This material is a Standard Reference Material of the National Institute for Standards and Technology.¹⁴⁸ Secondary ion mass spectrometry (SIMS) has been used extensively to characterize the boron depth distribution in this material as shown in the profile in Fig. 14. The profile obtained by APT analysis on a pulsed-laser LEAP^® shows an excellent match with SIMS, the accepted metrology standard for this type of measurement. Note that analytical sensitivity better than $1018cm−3$ is demonstrated in this data set.

Dielectrics for the gate in transistors for logic and processor microelectronics are a critical component that is receiving a lot of development attention currently. In particular, high-$k$ dielectrics based on rare earth oxides such as hafnia are the focus of much of the activity. Because they are deposited on high resistivity silicon, it is necessary to use laser pulsing to analyze these materials. Figure 15 shows a section of an image that contains a hafnia-based dielectric prepared by IMEC.¹⁴⁹ The distribution of hafnium, silicon, and oxygen through the layer is of particular interest. Note the significant increase in hafnium near the silicon substrate. This is an early confirmation of major industry fears that silicon can getter oxygen from the hafnium oxide layer which alters the dielectric behavior of the layer in significant detrimental ways. Note also the very high spatial resolution achieved in this $1.5nm$ (FWHM of Si distribution) layer. The application of this capability to transistor development and analysis is likely to be especially important to the semiconductor industry.

Camus et al.¹⁵⁰ performed atom probe compositional analysis of a rutile $(TiO2)$ specimen derived from xenolith from Jagersfontein in a wide-field-of-view atom probe. The specimen resistivity was very high (about $6×105Ωcm$⁠). Atom probe spectra taken at low tip temperature $(80K)$ showed almost exclusively atomic species, that is, Ti, O, and $O2$⁠. However, as the probing temperature was increased towards and to room temperature, they saw more and more molecular species, i.e., mostly TiO and some $TiO2$⁠. They further analyzed the relative amounts of the $O+$ and $O++$ in the mass spectrum and suggested that the detection of molecular ions may correlate with the ionization state of the oxygen in the solid. This latter proposal has not been fully explored or verified.

Kuhlman et al.^97,98 have studied naturally occurring metamorphic magnetite (predominately $Fe3O4$⁠) that contains disk-shaped Al–Mn–Fe spinel precipitates in the structure. Field ion microscopy was used to show that the precipitates are similar to that found by other techniques.⁹⁷ Atom probe analysis was performed on both a reflectron-compensated 3DAP and a LEAP^®. These materials have a relatively low resistivity of about $0.05Ωcm$⁠. LEAP^® analysis was obtained from a matrix region between the precipitates. The image is compositionally uniform and all peaks in the mass spectrum may be readily identified (Fig. 16). Several metal-oxide molecular ions are formed as evident in the mass spectrum. The composition of the magnetite (Table I), especially with respect to the total oxygen content, is the principal result. For stoichiometric magnetite $(Fe3O4)$⁠, oxygen is $57.1at.%$ of the material. In this work, the oxygen content was found to be $51.3at.%$⁠. The discrepancy may be either real, in which case the oxygen content is off stoichiometry and low, or it may be instrumental, in which case there is oxygen loss during the analysis (e.g., evaporation between pulses.) A high pulse fraction of 25% was used in this work to ensure that no evaporation occurred between pulses. A possible explanation for a low oxygen count involves the peak at $m∕n=16$⁠. This peak is usually $O+16$ in most materials. However, if 77% of this peak is the molecular species $O2++16$⁠, then the oxygen stoichiometry is correct. A similar explanation was found to apply to carbon atoms in carbides in steels.¹⁵¹ Despite this question and the difficulty of running this material in a voltage-pulsed atom probe, the results are encouraging for the prospects of analyzing ceramic materials with APT.

The vast majority of specimens previously analyzed using atom probes have been metals due to the need for low electrical resistivity during voltage pulsing. Consequently, there is only a modest history of atom probe analysis of biological or organic specimens. In 1975, Machlin et al. reported two-dimensional $1–2nm$ resolution field ion microscope (FIM) images of freeze-dried tRNA dimers.¹⁵² A major problem discussed in some detail by Machlin et al. is that the high field strengths for FIM imaging $(∼5–15V∕nm)$ appeared to cause desorption and fragmentation of adsorbed biomolecules.¹⁵² In the 1980s, Panitz reported FIM and field desorption imaging of biological materials including ferritin^153,154 and unstained DNA.¹⁵⁵ Other than a few additional publications by these research groups, this is essentially the entire history of biological atom probe and FIM imaging.

There has also been some research using atom probes to examine nonbiological organic materials. Nishikawa et al. examined diamondlike carbon films with a nonimaging (one dimensional) scanning atom probe to obtain compositional depth profiles with sufficient mass resolution to discriminate among carbon clusters as large as $C16$ with differing numbers of hydrogen atoms.^156,157 Other investigators used FIM and atom probes to examine graphite fibers,¹⁵⁸ the intrinsically conductive polymer (ICP) polypyrrole,¹⁵⁹ and phthalocyanine.¹⁶⁰ These efforts demonstrate that field ionization in an atom probe can be used to determine compositional structure in covalently linked carbon-based structures.

Figure 17 shows a mass spectrum obtained by Maruyama et al. from polypyrrole.¹⁵⁹ There are several important differences between this spectrum and those obtained from inorganics. Whereas in inorganics, the majority of field evaporated species is single ions, organics mostly form molecular ions. The molecular ions may be either clusters of like atoms (e.g., $C2$⁠, $C4,…$⁠) or different atoms (e.g., $CH2$⁠, CO,…). There is a large number of peaks that correspond to molecular ions from this polymer. Each of these molecular ions is slightly different in mass, sometimes by as little as one mass unit. As a result, it is difficult to unambiguously identify peaks in the spectrum because individual elements and their signature isotopic groupings are not so readily used as “fingerprints” in the spectrum. It is very important in this type of work to have a reliable calibration of time of flight. Even though single atoms are not extracted, the small size of the molecular ions means that high quality three-dimensional reconstruction at the subnanometer scale is still possible, at least to the precision of the molecular unit size. Although this history is limited, these reports suggest real potential to use atom probes to analyze conductive carbon-based materials, but the analysis will likely be more difficult than inorganics owing to a higher propensity for complexity and ambiguity.

There are three features of APT, which are coupled, that make it stand out among analytical instruments: three-dimensional compositional imaging at the nanoscale analysis of buried features and high analytical sensitivity. The image resolution is excellent, but electron microscopy and scanning probe instruments can exceed it for pure image resolution. It is the spatial resolution of APT as an analytical technique that is unmatched. Many scientific instruments offer a compositional imaging mode where characteristic signals are resolved spatially on a point-by-point basis. The spatial resolution of such images is the resolution of the technique when operated in a focused-probe mode. The highest spatial resolution technique of these is the scanning transmission electron microscopy (STEM) which can provide $1nm$ resolution laterally from x-ray emissions and $0.2nm$ resolution laterally from electron energy loss events. However, these images are integrated through the specimen thickness which is about $10nm$⁠. APT achieves better than $0.3nm$ in all three directions of a three-dimensional image simultaneously.

Because it is a three-dimensional technique, APT excels at an analysis of subsurface or buried features in specimens. These can include one-dimensional features such as dislocations, two-dimensional features such as layers and interfaces, and three-dimensional features such as precipitates. With a wide field of view, about 250 000 atoms per atomic layer can be recorded. For interfacial segregation the sensitivity is therefore about $40appm$/atomic layer and is limited principally by the number of atoms detected per layer. For bulk analysis, the sensitivity can be $10appm$ or better. It is typically limited by the background signal in the mass spectrum as long as enough atoms are collected.¹⁶¹ For an analytical imaging technique, this is a very high sensitivity. Thus APT, especially as practiced by the high speed, large field of view instruments, offers an unparalleled combination of compositional imaging at the atomic scale of multidimensional objects with high analytical sensitivity.

Given all of its excellent characteristics, it is tempting to think of APT as approaching the ideal microscope for nanoscale work. However, all analytical techniques have their limitations and it is important that the limitations of APT be kept in perspective. The following is a list of the most important limitations in order of potential significance for users.

This field is in its infancy. Many innovative methods have been developed for needle-shaped specimens and they will continue to play a role. But the future of specimen preparation for atom probe tomography is microtips. Methods and tools for making these specimens en masse with rapidity and automatically are just being developed. The main advantages of microtip preparation are small material volume required, parallel processing of large numbers, uniformity of shape and position (in regular arrays), easy access to thin films on substrates, and low invasiveness on the specimen. These advantages can be very important when attempting to analyze complex materials structures for the first time. There is sometimes a learning curve to be overcome for complex materials, and the more uniform and numerous the tips the easier it is to obtain quality data.

It should be realized that the mechanical stresses in the specimen near the apex approach the cohesive strength of the material.¹⁶² Specimen fracture is common in the atom probe. When there is a weak link, such as a weak transverse interface or layer or a surface notch, it may not be possible to obtain data. Methods that make it possible to run at lower fields in the atom probe can reduce this effect. The stress in a specimen increases with the square of the applied field. Wide field-of-view atom probes employ a lower average field at the specimen apex and in principle have a lesser probability of causing specimen fracture. On the basis of very limited data so far, this effect does appear to hold. Laser pulsing also is more gentle on the specimen since the standing field is kept some 10%–20% lower (stress is 20%–40% lower). Observations to date also tend to support the hypothesis that laser pulsing leads to fewer specimen fractures. This effect will be closely monitored as greater experience is gained.

The ideal nanoscale microscope detects all atoms. Alas, today’s atom probes utilize microchannel plates which detect 60% of all energetic ions hitting them.¹⁶³ The electronics convert as high as 99.9% of the amplified hits. The amplification process is nonselective, so there is no bias in the detection process on the quantification of composition. There are some applications for which 100% detection of ions would be needed but, for the most part, it is not a major concern. Higher detection efficiency is always better and improving the detection efficiency is an area for future research. It is possible to increase the detection efficiency to over 72% by placing a biased mesh in front of the microchannel plates to drive secondary electrons into the plate.¹⁶⁴ This approach has the disadvantage, however, that a mesh shadow is superimposed on the data and overall background noise is increased.

Though images are orders of magnitude larger now than a decade ago, the finite analyzed volume will still limit the application of the atom probe. The largest field of view in spatial units is about $200nm$ today. This characteristic length scale can readily be increased through incremental improvements by a factor of 2 or so, but it will not reach a micron easily. Common image data sets today are $100nm$ in diameter by $100nm$ in depth $(106nm3)$ with $107–108$ atoms. With the right specimen, image sizes of $109$ atoms are within reach. At today’s highest data collection rate of $106atoms∕min$⁠, it will take $17h$ of data collection. 100 ppb (parts per billion) sensitivity might be reached in such a specimen. These data collection rates depend on a material’s ability to withstand electrical pulsing or laser pulsing at $200kHz$ or greater. Enough practical experience has now been gained to show that this is not a problem.

If nanotechnologies are to flourish, the tools of the trade, both the tools for making nanostructures and the microscopes for seeing them, must be available. Microscopes of all types are needed, but atom probes are uniquely positioned to fulfill a crucial need at the finest scale. In order to realize this promise, it must be straightforward to gain useful knowledge from the atom probe.

We believe the instrument will continue its evolution out of the research laboratories toward widespread adoption. In order for this to happen, the instrument must become simple enough to use that less-than-expert users can expect to obtain reliable information in hours. There must first be major advances that simplify, facilitate, and automate specimen preparation. These types of advances have occurred in electron microscopy specimen preparation over the past two decades. Much of the technology developed there can be directly applied to atom probe specimen preparation. Tools like the FIB for sharpening tips and coupon extraction and automated etchers and coaters can be directly applied. Indeed, in the past three years the application of FIB to atom probe specimen preparation has increased specimen yield by two orders of magnitude and decreased the time per specimen by one order of magnitude. Entirely new methods and hardware will be needed as well. At the root of all this activity is the basis of a revolution in specimen preparation for atom probe tomography: the microtip specimen geometries first proposed by Nishikawa and co-workers.^64,65

There will need to be advances in the basic technology of atom probes to diminish the learning curve for operation and analysis. There is a logical progression that the technology can take if it delivers value at each stage from university research and government laboratories to research and development laboratories for commercial concerns, and finally to process monitoring and control for commercial concerns. The need for speed and reproducibility increases along this progression, but so too does the specificity of purpose. This progression requires largely market-driven development efforts, and the extent to which the end point is approached will be a good measure of how successful we have been at building a practical microscope for nanotechnology.

The bases of atom probe tomography, including field ion microscopy, have been in practice for over $50years$ but the technology is just beginning to gain wide acceptance. In the past decade major developments in specimen geometries and imaging capabilities of the atom probe have been made. A confluence of critical need in major established and new industries with these technological developments should propel the technology rapidly into many new applications. Continued innovation in specimen preparation and analysis of high resistivity specimens is needed and is likely to occur in the next few years.

One of the authors (T.F.K.) would like to thank David Seidman for suggesting that this article be written and the many individuals at Imago Scientific Instruments who have contributed to this article, especially Keith Thompson and David Larson. Research at the Oak Ridge National Laboratory SHaRE User Facility was sponsored by the Office of Basic Energy Sciences, U.S. Department of Energy under Contract No. DE-AC05-00OR22725 with UT-Battelle, LLC.