Tải bản đầy đủ - 0 (trang)
Plucking, positioning, and perturbing atoms at silicon surfaces

Plucking, positioning, and perturbing atoms at silicon surfaces

Tải bản đầy đủ - 0trang

Fig. 3 STM image of a dangling bond wire with accompanying schematic illustration of the

H:Si(100) surface and the wire. Taken from Ref. 21.

A significant number of papers have been published on this depassivation

technique, including a well-written review by Walsh and Hersam in 200921

(from which Fig. 3 is taken). I shall therefore not focus on the physics and

chemistry of H-depassivation in the following sections, but, rather, discuss

how hydrogen-resist patterning has recently been used to fabricate atomicscale devices – in effect, to complement the Walsh-Hersam review with a

discussion of key results over the years since its publication. In particular,

the generation of arrays and ‘wires’ of silicon dangling bonds via the

removal of hydrogen atoms has been explored in some depth theoretically

as the basis of a rather novel type of atomic logic circuitry, as discussed in

Section 3.3.

Atom manipulation is also possible on the bare (i.e. unpassivated) Si(100)

surface and in Section 3.4 I describe a series of recent results from our group

at the University of Nottingham which involve the switching of bistable

silicon atoms via chemical force alone. This is a very different approach to

atomic-level control as compared to the H-depassivation technique: atoms

are switched between two states via direct modification of a chemical bond

(a toggling of bond angle), rather than the injection of tunnelling electrons.

3.1 Ohm’s law: how low can we go?

In January 2012 Weber et al.22 reported that they had succeeded in fabricating and characterising a wire formed from heavily-doped silicon which

was a single atomic layer ‘high’, only four atoms wide, and exhibited a

resistivity of 0.3 mO cm. The wire was fabricated, contacted, and embedded

in a silicon substrate using a combination of STM lithography of H:Si(100),

Nanoscience, 2013, 1, 116–144 | 121

Fig. 4 Ohmic conduction in atomic wires. (A) A four-atom-wide wire fabricated on a

H:Si(100) surface via removal of hydrogen using an STM tip. (B), (C) Higher resolution images

showing the atomic structure of the wire before and after phosphine dosing, respectively. (D)

Theoretical modelling of the electron density within, and in the vicinity of, the wire. (E) Fourprobe resistance plotted against wire length. Note linearity. Taken from Ref. 22.

doping via exposure to PH3 (resulting in an effective doping density of

1021 cm À3 which, as the authors pointed out, is three orders of magnitude

beyond the level required to cross the Mott insulator-metal transition), and

subsequent capping of the structure using molecular beam epitaxy overgrowth of silicon (Fig. 4).

This major advance in nanoelectronic device fabrication was accompanied by key new insights into fundamental electronic behaviour at the

atomic limit. Remarkably, the heavily doped silicon wires were found to

have Ohmic conductance (i.e. their resistivity was independent of wire

diameter or length) due to the very small separation between donors

(B1 nm, i.e. less than the Bohr radius). The abrupt doping profile – ranging

from B1015 cmÀ3 outside the wire to 1021 inside – yields very effective

charge confinement. Moreover, and perhaps surprisingly, the atomic wires

tolerate extremely high current densities (5 Â 105 A cmÀ2), comparable to

those in state-of-the-art copper interconnects.

3.2 A single atom transistor

Following hot on the heels of their report of scaling of Ohmic conductance

down to the atomic limit, Simmons and collaborators produced a single

atom transistor.23 This was also fabricated using the H-depassivation

technique and represents a remarkable achievement in silicon processing

and associated device physics. An image of the device prior to its encapsulation in a silicon overlayer is shown in Fig. 5 where a silicon atom

122 | Nanoscience, 2013, 1, 116–144

Fig. 5 A single atom transistor. (a) Device geometry fabricated using H-depassivation on

a H:Si(100) surface; (b) High resolution image of active region showing ejected silicon

atom which is a signature of phosphorous incorporation; (c) Stability diagram for device.

From Ref. 23.

‘ejected’ from the underlying substrate between the source and drain electrodes is clearly observed. This ejection of silicon arises from the dosing of

the surface with phosphine in order to produce not only the highly doped

drain, source, and gate contacts visible in the STM image of Fig. 5(a) but

also to incorporate a single phosphorous atom between the contacts.

The deterministic placement of a single dopant atom within a device is a

major and impressive advance in semiconductor processing. Fuechsle

et al.23 went on to investigate the charge transport properties of the single P

atom between the contacts, finding that below the voltage regime required

for transistor operation, it acted as a quantum dot. Three energy levels/

charge states corresponding to an ionized, neutral, and negatively charged

dot (D ỵ , D0, and D respectively) are observed in plots of the source-drain

voltage vs. the applied voltage on the gate (a so-called stability diagram),

Fig. 5(b). The ability to retain the quantum states and the charging energy

of the P dopant, despite its close proximity to the highly doped contacts/

gates (and their associated electrostatic potential) is particularly important

in terms of the ultimate limits of Moore’s law. It would seem that scaling

working devices down to the single atom limit is indeed possible.

Nanoscience, 2013, 1, 116–144 | 123

3.3 Towards dangling bond logic gates

Christian Joachim (CEMES-CNRS, Toulouse) and his collaborators24–27

have theoretically, and comprehensively, explored the design rules for logic

gates based around the type of H-depassivation mechanisms so elegantly

exploited by Simmons and co-workers. The structure of a NAND gate,

comprising two atomic scale inverters, is shown in Fig. 6. Switching of the

input states of the device is achieved by the manipulation of H atoms (and,

thus, the control of dangling bonds (DBs)) – (de)hydrogenation of dimers

makes a dramatic difference in the transmission spectrum of the conducting

channels between the gold pads. Design principles for a variety of gates

and switches have been put forward based on this strategy.26 Joachim and

co-workers have also couched intramolecular charge transport processes in

terms of logic operations – an approach which has been termed quantum

Hamiltonian computing and which is detailed at length in an important

paper published at the start of 2012.25 It is perhaps worth noting that charge

transport via a pattern of interacting dangling bonds on the H:Si(100) surface, assuming it is decoupled from the bulk substrate, has many parallels

with intramolecular conduction channels. Schofield and co-workers28 have

also very recently shown that dangling bonds on the H:Si(100) surface can be

formed such that there is wavefunction overlap without bond formation.

Fig. 6 Design for a NAND gate formed via placement of dangling bonds on an otherwise

hydrogen-passivated Si(100) surface. Data are input via the depassivation/repassivation of

dangling bonds. Taken from Ref. 26.

124 | Nanoscience, 2013, 1, 116–144

Although the fabrication of devices similar to that shown in Fig. 6 is

beyond current capability, important insights into the operation of the DB

logic gates can be derived from tunnelling spectroscopy with an STM. This

is a natural measurement to undertake given that the fabrication method is

based around the injection of electrons using an STM. There are two key

issues to address, however, and both relate to the state of the STM tip. First,

for direct comparison between experiment and theory, the state of the tip

including, in particular, the electronic states involved in tunnelling to/from

the sample, needs to be known. (The traditional assumption that the tip has

a flat density of states in the energy range of interest is in many cases

unwarranted). Second, the tip state also fundamentally underpins the ability

to (de)hydrogenate in the first place.

Given that NC-AFM has the demonstrated potential to provide significantly higher resolution imaging capability than possible with STM, an

interesting question arises: is it possible to (de)hydrogenate using chemical

force alone, i.e. with an atomic force microscope? This precise problem,

albeit in a somewhat different context, was studied by K. Eric Drexler back

in the early nineties.29 Drexler put forward a concept known as molecular

manufacturing, which has attracted a great deal of controversy. Putting

aside the more controversial aspects for now,30 at the core of molecular

manufacturing lies a demonstrably valid idea: chemical reactions can be

driven on an atom-by-atom basis purely by (chemo)mechanical force. A

significant number of density functional theory calculations have explored

the viability of this process, termed mechanosynthesis by Drexler, for diamond surfaces.31–35 In a pioneering experiment in 2008,36 Custance and

co-workers demonstrated just how high a degree of control was possible for

mechanosynthetic reactions, using the NC-AFM technique to drive

exchanges between atoms at a silicon surface and at the apex of a scanning


Before attempting to manipulate atoms at the H:Si(100) surface using

(NC-)AFM, it is obviously important first to establish the conditions

required for atomic resolution imaging. Although a number of studies of

H:Si(100) using NC-AFM were reported a decade ago,37,38 and important

theoretical work addressed the critical contribution of the tip state to image

contrast39 (which is a perennial issue in SPM (see ‘The Trouble with Tips’

below), quantitative experimental verification of theoretical predictions was

lacking. In common with other experimentalists, our group has found that

NC-AFM of the H:Si(100)-(2 Â 1) surface very frequently results in inverted

contrast, where the H atoms appear as depressions in a constant frequency

shift scan, rather than as maxima. (The same phenomenon has been

observed for the hydrogen-passivated Ge(100) surface40). As described by

Sharp et al.,41 the contrast inversion effect arises due to passivation of

dangling bonds at the tip with hydrogen (Fig. 7). A simple model of a Hpassivated tip interacting with a H:Si(100) surface provided good agreement

with experimental force curves.

Of course, if the precise tip termination affects the NC-AFM imaging

process to this extent, then any type of manipulation event – such as

hydrogen extraction or deposition – will be similarly heavily influenced by

the state of the probe. We examined this for the H:Si(100) surface using

Nanoscience, 2013, 1, 116–144 | 125

Fig. 7 Comparison of experimental and theoretical force-distance curves for (a) a danglingbond-terminated tip, and (b) a hydrogen-passivated tip. The latter provides better agreement

with the experimental data. From Ref. 41.

DFT calculations41 and, as might be expected, only certain classes of tip

apex allow for the possibility of hydrogen transfer to a surface (which is of

particular interest with regard to ‘error correction’ in the fabrication and

actuation of dangling bond logic gates). For some tip types, hydrogen

deposition is precluded. We return to the vexed question of tip control in

Section 6 below.

3.4 Flipping bistable atoms

In the absence of hydrogen, the clean Si(100) surface forms a c(4 Â 2)

reconstruction (Fig. 8) whose fundamental structural unit, as for H:Si(100), is

a dimer – a pair of silicon atoms. Unlike the hydrogen-terminated surface,

however, bare silicon dimers are not planar; instead, they buckle so that one

atom moves out of the surface plane and its partner moves towards the

substrate (as compared to the unbuckled unit). Accompanying this structural

modification, which may be thought of as a Jahn-Teller effect, is a movement

of electronic charge across the pi-like orbital arising from the interaction

of the dangling bond on the dimer atoms. This coupled structuralelectronic distortion shifts surface states out of the band gap and makes the

(defect-free) Si(100)-c(4 Â 2) reconstruction semiconducting.

There has been a great deal of controversy, however, regarding the

ground state of the Si(100) surface. As discussed by Yoshida et al.,42 the

source of this controversy was a lack of appreciation for the extreme sensitivity of the Si(100) surface dimers to external perturbations due to, for

example, a scanning probe and/or the scattering of low energy electrons. An

NC-AFM study43 in 2006 conclusively demonstrated that the tip-sample

interaction could strongly perturb the c(4 Â 2) surface such that the ‘down’

atom of each dimer would be ‘pulled up’ by the probe, giving rise to an

apparent (2 Â 1) periodicity in the images.

Given this ability to perturb relatively large surface areas, the perhaps

obvious question to ask is whether a NC-AFM tip could flip a single isolated dimer between its two buckled configurations. We have addressed this

question in a series of studies over the last couple of years,18,44–46 focussing

in particular on the influence of the potential energy landscape of the surrounding surface on the propensity for dimer flipping. Figure 8(a) is a

126 | Nanoscience, 2013, 1, 116–144

Fig. 8 Silicon dimer flipping using (chemo)mechanical force. (A) Ball-and-stick models of the

‘native’ c(4 Â 2) reconstruction of the pristine Si(100) surface, comprising rows of dimers in

alternate buckling configurations, and two defect structures where the ‘zig-zag’ buckling of a

row is interrupted. Also included is a schematic of the dimer flipping process and its relationship to binary encoding. (B) An example of dimer flipping showing frequency shift vs. tipsample separation curves (and corresponding before and after NC-AFM images) for the

creation (upper) and removal (lower) of a two-phason state. Images in A courtesy of Adam

Sweetman and Sam Jarvis. (B) taken from Ref. 18.

simple schematic of the general scheme we adopted, inspired by previous

theoretical papers47–49 which mused on the possibility of carrying out

information storage and, indeed, logic operations via modification of the

orientation of silicon dimers. Some researchers have couched the dimer

orientation problem in terms of ‘mechanical’ spin states, in analogy with the

Ising model description of ferromagnetism. A number of fascinating questions then arise: Do the mechanical spins interact? If so, is it even possible to

‘flip’ a single dimer? If dimer ‘spins’ are coupled, is logic possible? What is

the origin of the coupling?

Figure 8(b) shows the results of a typical dimer flipping experiment. The

tip is positioned above a ‘down’ atom of a dimer at the Si(100)-c(4 Â 2)

surface and a Df(z) spectrum is acquired. (Although we can’t ‘see’ the down

atom of the dimer in the NC-AFM image, we know where it is by symmetry). At a certain critical tip-sample separation – in this case, about 1.6 A˚

closer to the surface than the tip position used for imaging – there is a sharp

discontinuity in the frequency shift vs. z curve due to the sudden jump of the

lower dimer atom towards the tip. On retraction of the tip, the dimer

remains in its flipped conformation, as is clear from Fig. 8(b).

What is also very clear from Fig. 8 (b) is that two dimers, rather than one,

have been flipped. (This arrangement of buckling orientations can be

Nanoscience, 2013, 1, 116–144 | 127

described as a ‘two phason’ configuration). This is a general result: we have

to date never succeeded in flipping a single dimer on the Si(100)-c(4 Â 2)

surface, correlated flips of pairs (or larger numbers) of dimers always

occurs. This correlated flipping mechanism is intriguing because it hints at

the possibility of carrying out logic operations via the ‘communication’ of

silicon dimers48 (albeit a type of logic which necessitates UHV conditions

and operates at 5 K with a sub-Hz bandwidth). But the devil is in the detail –

dimer flipping is extremely sensitive to the potential energy landscape of the

surface which, in turn, is modified dramatically by the presence of defects.

This is demonstrated in Fig. 8(c) where a row of dimers is pinned in the

buckled state by a boron-related surface defect. (Our p-type Si(100) samples

are heavily boron doped).

Hence, attempting to engineer specific logic gates ((N)AND, (N)OR

etc. . .) will be plagued by problems due to the influence of the surrounding

(imperfect) surface. Nonetheless, the potential of mechanical logic using

silicon dimers remains a topic worth exploring, with perhaps the possibility

to scale down at least some elements of Babbage’s fascinating difference

engine approach50 to the atomic level.

The dimer flipping experiment also raises very interesting questions

regarding the dynamics of the flip events and, in particular, the ability of

(density functional) theory to model the process. We found that DFT did

not reproduce our experimental inability to flip single dimers; density

functional calculations using the SIESTA code invariably predicted that

three neighbouring dimers with the same buckling configuration in a single

row is a stable state.18,44 This prompted us to carry out a detailed

exploration of the potential energy landscape for the process (both for the

‘pristine’ surface and in the presence of defects) using the nudged elastic

band method, as shown in Fig. 9. Although this landscape explains many of

our experimental results, a key issue remains to be resolved: how does the

Fig. 9 Calculated variation in the potential energy surface as a function of different defect

types for a tip-induced transition from the c(4 Â 2) structure shown in (i) to the two-phason

state shown in (iii), via the ‘three-in-a-row’ configuration shown in (ii). From Ref. 44.

128 | Nanoscience, 2013, 1, 116–144

system transit to the two phason configuration seen in Fig. 8(b), given that

the energy barrier is much greater than the available thermal energy?

Defects clearly modify the landscape (Fig. 9) but to date we have not

managed to model a defect structure which collapses the energy barrier to

be crossed so as to adopt the two phason arrangement. There are two

possibilities: DFT and associated theoretical methods are simply not up to

the task and are missing a key piece of the underlying physics/chemistry, or

the barrier is not a numerical artefact and a novel crossing mechanism is at

work. One pathway not accounted for by the NEB calculation is a quantum

mechanical tunnelling from one dimer state to another. Although this might

seem quite a remote possibility given the total mass of the Si-Si dimer unit, it

is perhaps worth noting that there is strong evidence that objects as large as

CO51 and Co atoms52 can tunnel through diffusion barriers comparable (in

energy and width) to those present in the dimer flipping process.


Visualising (intra)molecular force-fields and submolecular structure

A sub-field of scanning probe microscopy which has expanded rapidly over

the last couple of years is the use of NC-AFM to provide extremely high

resolution images of submolecular structure. As for so many step changes in

SPM capability, it was the IBM research labs - in this case, the IBM Zurich

team of Leo Gross, Fabian Mohn, Nikolaj Moll and Gerhard Meyer (in

collaboration with Peter Liljeroth of Utrecht University) – who made key

advances, publishing remarkable images of pentacene that showed the

intramolecular bonding ‘framework’ in detail10 (Fig. 10(a)). Gross et al.

argued that to enable the type of high resolution imaging seen in Fig. 10(a),

two aspects of the experimental set-up were key. First, the tip was passivated with a CO molecule (otherwise the strong interaction with the scanning probe perturbed or moved the pentacene molecule) and, second,

imaging was carried out in the Pauli repulsion regime of the tip-sample

interaction potential.

Fig. 10 (A) Ball-and-stick model of pentacene and a NC-AFM image taken with a COfunctionalised tip in the Pauli exclusion regime of the tip-sample potential (Ref. 10); (B) Model

of the PTCDA molecule and an STM image taken using the scanning tunnelling hydrogen

microscopy (STHM) protocol introduced by Temirov et al. (Ref. 53).

Nanoscience, 2013, 1, 116–144 | 129

Since Gross et al.’s striking data were published, both the IBM group and

a number of other research teams have produced high resolution NC-AFM

images (using the qPlus variant of the technique, for the reasons discussed in

Section 2). An important review written by Leo Gross was published in

Nature Chemistry in March 2011.9 I will therefore largely forego a discussion of the results presented in that review and instead focus in this

section on developments since its publication.

Before moving on to cover recent examples of intramolecular resolution

using NC-AFM, however, it is essential that I highlight a novel approach to

high resolution STM imaging whose discovery pre-dates the publication of

the IBM pentacene work. This is the scanning tunnelling hydrogen microscopy (STHM) technique introduced by Temirov et al.53–56 Included in Fig.

10(b) is an image of PTCDA (3,4,9,10-perylenetetracarboxylic-dianhydride)

on a Au(111) surface, acquired using STHM and where it is clear that the

image resolution is comparable to that attained in the NC-AFM data.

This is largely because the two techniques share the same key elements – a

passivated tip (H-passivated in the case of Termirov et al.’s results) and

operation within the Pauli repulsion regime – although the contrast

mechanisms are of course rather different. For STHM, the passivated tip

translates variations in tip-sample force into a modulation of the junction

conductance. As can be seen from Fig. 10(b) the image resolution far

exceeds that observed in conventional STM because the latter is sensitive

only to variations in electron density in a relatively narrow energy window

close to the Fermi level. In STHM, as pointed out by Temirov et al.,55 it is

variations in the total electron density – and the information on chemical

structure embedded within it – which are probed.

On the basis of a series of combined Green’s function-local orbital density

functional theory calculations, Martı´ nez et al.57 have very recently put

forward a slightly different argument regarding the attainment of ultrahigh

resolution in STHM. They propose that hydrogen molecules are dissociated

at the tip and that the H atoms dramatically modify the density of states

(DOS) at the Fermi level. It is this modification of DOS at EF which they

claim gives rise to the enhanced resolution. It should be noted, however,

that Temirov and colleagues53 previously listed a number of experimental

observations to support their claim that it is molecular, rather than atomic,

hydrogen (or deuterium) that underpins the STHM imaging mechanism.

If it is indeed H atoms, rather than H2 molecules, at the tip which are

responsible for the intramolecular contrast in STHM then there is the

exciting potential to combine STHM and NC-AFM measurements of

molecules adsorbed on the H:Si(100) (or H:Ge(100)) surface by exploiting

the passivated state of the tip which results simply from scanning the surface, as described in Section 3.3 above. This in turn could provide important

new insights into the relationship between molecular conformation and

function for molecular logic gates on semiconductor substrates.

The precise atomic and electronic structure of the tip apex of course

underpins all SPM and over the last couple of years there has been rapidly

growing interest in characterising and controlling the probe state to a much

greater extent than ever before (which I return to in Section 6). Before

leaving STM to focus on recent NC-AFM work, I would like to briefly

130 | Nanoscience, 2013, 1, 116–144

Tài liệu bạn tìm kiếm đã sẵn sàng tải về

Plucking, positioning, and perturbing atoms at silicon surfaces

Tải bản đầy đủ ngay(0 tr)