Advance to Controlling one to a Few Hundred Atoms at Microsecond Timescales Using AI Control of Electron Beams

The work should lead to control one to a few hundred atoms at microsecond timescales using AI control of electron beams. The computational/analytical framework developed in this work are general and can further help develop techniques for controlling single-atom dynamics in 3D materials, and ultimately, upscaling manipulations of multiple atoms to assemble 1 to 1000 atoms with high speed and efficacy.

Scientists at MIT, the University of Vienna, and several other institutions have taken a step toward developing a method that can reposition atoms with a highly focused electron beam and control their exact location and bonding orientation. The finding could ultimately lead to new ways of making quantum computing devices or sensors, and usher in a new age of “atomic engineering,” they say.

This could help make quantum sensors and computers.

“We’re using a lot of the tools of nanotechnology,” explains Li, who holds a joint appointment in materials science and engineering. But in the new research, those tools are being used to control processes that are yet an order of magnitude smaller. “The goal is to control one to a few hundred atoms, to control their positions, control their charge state, and control their electronic and nuclear spin states,” he says.

While others have previously manipulated the positions of individual atoms, even creating a neat circle of atoms on a surface, that process involved picking up individual atoms on the needle-like tip of a scanning tunneling microscope and then dropping them in position, a relatively slow mechanical process. The new process manipulates atoms using a relativistic electron beam in a scanning transmission electron microscope (STEM), so it can be fully electronically controlled by magnetic lenses and requires no mechanical moving parts. That makes the process potentially much faster, and thus could lead to practical applications.

Using electronic controls and artificial intelligence, “we think we can eventually manipulate atoms at microsecond timescales,” Li says. “That’s many orders of magnitude faster than we can manipulate them now with mechanical probes. Also, it should be possible to have many electron beams working simultaneously on the same piece of material.”

Atomic soccer

The power of the very narrowly focused electron beam, about as wide as an atom, knocks an atom out of its position, and by selecting the exact angle of the beam, the researchers can determine where it is most likely to end up. “We want to use the beam to knock out atoms and essentially to play atomic soccer,” dribbling the atoms across the graphene field to their intended “goal” position, he says.

“Like soccer, it’s not deterministic, but you can control the probabilities,” he says. “Like soccer, you’re always trying to move toward the goal.”

In the team’s experiments, they primarily used phosphorus atoms, a commonly used dopant, in a sheet of graphene, a two-dimensional sheet of carbon atoms arranged in a honeycomb pattern. The phosphorus atoms end up substituting for carbon atoms in parts of that pattern, thus altering the material’s electronic, optical, and other properties in ways that can be predicted if the positions of those atoms are known.

Ultimately, the goal is to move multiple atoms in complex ways. “We hope to use the electron beam to basically move these dopants, so we could make a pyramid, or some defect complex, where we can state precisely where each atom sits,” Li says.

This is the first time electronically distinct dopant atoms have been manipulated in graphene. “Although we’ve worked with silicon impurities before, phosphorus is both potentially more interesting for its electrical and magnetic properties, but as we’ve now discovered, also behaves in surprisingly different ways. Each element may hold new surprises and possibilities,” Susi adds.

The system requires precise control of the beam angle and energy. “Sometimes we have unwanted outcomes if we’re not careful,” he says. For example, sometimes a carbon atom that was intended to stay in position “just leaves,” and sometimes the phosphorus atom gets locked into position in the lattice, and “then no matter how we change the beam angle, we cannot affect its position. We have to find another ball.”

Science Advances – “Engineering single-atom dynamics with electron irradiation” (2019).


Atomic engineering is envisioned to involve selectively inducing the desired dynamics of single atoms and combining these steps for larger-scale assemblies. Here, we focus on the first part by surveying the single-step dynamics of graphene dopants, primarily phosphorus, caused by electron irradiation both in experiment and simulation, and develop a theory for describing the probabilities of competing configurational outcomes depending on the postcollision momentum vector of the primary knock-on atom. The predicted branching ratio of configurational transformations agrees well with our atomically resolved experiments. This suggests a way for biasing the dynamics toward desired outcomes, paving the road for designing and further upscaling atomic engineering using electron irradiation.

Controlling the exact atomic structure of materials is an ultimate form of engineering. Atomic manipulation and atom-by-atom assembly can create functional structures that are hard to synthesize chemically. e.g., exactly positioning atomic dopants to modify the properties of carbon nanotubes and graphene. Nitrogen (N) or phosphorus (P) dopants might be useful in quantum informatics due to nonzero nuclear spin.

Successful atomic engineering requires understanding of two parts:
(i) how the desirable local configurational changes can be induced to increase the speed and success rate of control and
(ii) how to scale up basic unit processes into feasible structural assemblies of 1 to 1000 atoms to produce the desired functionality.

Historically, scanning tunneling microscopies have demonstrated good stepwise control of single atoms, leading to physicochemical insights and technological advances. However, their scalability and throughput are severely limited by the mechanical probe movements. Recently, aberration-corrected scanning transmission electron microscopy (STEM) has emerged as a versatile tool for characterizing the precise atomic structure of materials. Despite its very early stage of development, STEM also shows great promise as a tool for atomic manipulation: In two-dimensional (2D) graphene, Si dopants are found to be stepwise controllable, and iterating these basic steps enables long-range movement with a high throughput, whereas in a 3D silicon crystal, the projected location of Bi dopants was also manipulated. However, imprecise understanding of the dynamics of the basic steps, which involves relativistic electron-nucleus collisions, electronic excitation and relaxation, dynamic ion trajectories, momenta dephasing, and heat conduction, add uncertainties to this technique. While the traditional theory of radiation damage provides a basis for understanding, instead of trying to minimize beam effects, atomic engineering seeks to utilize them to achieve desired configurational changes. Concepts like the displacement threshold energy Ed, which is, in most cases, approximated as scalar, turn out to be too crude to guide the design of the precise cross-sections of different configurational outcomes.

Now researchers use STEM to both drive and identify the atomic motion of individual phosphorus (P) dopants in graphene and construct a theoretical scheme for evaluating their relative probabilities with respect to electron energy and momentum direction.

They have categorized the dynamics into four types:
(A) direct exchange and
(B) Stone-Wales (SW) transition, which conserve the atoms, and
(C) knockout of a carbon neighbor and
(D) replacement of the dopant atom by C, which do not conserve the local composition.

4 thoughts on “Advance to Controlling one to a Few Hundred Atoms at Microsecond Timescales Using AI Control of Electron Beams”

  1. BTW what is the distinction between ‘computer controlled’ & ‘AI controlled’ ?
    Has ‘AI’ just become a buzzword?

Comments are closed.