Electronic crystals and quasicrystals in semiconductor quantum wells: an AI-powered discovery

The homogeneous electron gas is a cornerstone of quantum condensed matter physics, providing the foundation for developing density functional theory and understanding electronic phases in semiconductors. However, theoretical understanding of strongly-correlated electrons in realistic semiconductor systems remains limited. In this work, we develop a neural network based variational approach to study quantum wells in three dimensional geometry for a variety of electron densities and well thicknesses. Starting from first principles, our unbiased AI-powered method reveals metallic and crystalline phases with both monolayer and bilayer charge distributions. In the emergent bilayer, we discover a new quantum phase of matter: the electronic quasicrystal.

💡 Research Summary

In this paper the authors present a first‑principles study of strongly correlated electrons confined in semiconductor quantum wells using a neural‑network‑based variational Monte‑Carlo (NN‑VMC) approach. The motivation stems from the well‑known limitation of conventional density‑functional theory (DFT) and Hartree‑Fock methods, which treat the homogeneous electron gas (HEG) as a uniform, three‑dimensional background and therefore miss essential physics when the electron system is confined to a finite‑thickness well. In a realistic quantum‑well geometry, three energy scales compete: the kinetic energy of the in‑plane motion, the Coulomb repulsion between electrons, and the subband spacing associated with quantization in the growth direction (z‑axis). The interplay of these scales gives rise to a rich phase diagram that cannot be captured by mean‑field approximations.

The authors model spin‑polarized electrons in a quantum well of thickness d, expressed in effective atomic units. They explore a broad parameter space defined by the Wigner‑Seitz radius rs (which controls the average electron density) and the well thickness d. To treat the many‑body wavefunction, they employ an attention‑based deep neural network that encodes pairwise and higher‑order correlations with millions of variational parameters. The network is trained by minimizing the variational energy using stochastic sampling within the VMC framework; automatic differentiation and efficient importance sampling allow rapid convergence even for large systems.

The resulting phase diagram reveals four distinct regimes:

1. Metallic phase – At high electron density (small rs) and thin wells, the system behaves as a conventional two‑dimensional electron gas with a delocalized Fermi‑liquid‑like wavefunction. The kinetic energy dominates and the electrons remain itinerant.

2. Monolayer Wigner crystal – When rs is increased or the well is made slightly thicker, the Coulomb repulsion overcomes the kinetic energy, and electrons crystallize into a triangular lattice confined to a single plane. This phase exhibits a dramatic drop in conductivity and long‑range positional order.

3. Bilayer electron crystal – For sufficiently thick wells, the electrons split into two parallel layers while retaining finite inter‑layer tunneling. Each layer forms a regular lattice, but the overall structure is a coupled bilayer. The inter‑layer spacing is set by the subband gap, and the system can be tuned continuously between the monolayer crystal and the bilayer crystal by varying d.

4. Electronic quasicrystal (new quantum phase) – The most striking discovery appears in the bilayer regime at low density and large d. Instead of a periodic lattice, the electrons self‑organize into a quasiperiodic pattern with five‑fold rotational symmetry. Classical energy calculations (Coulomb plus kinetic) would deem this configuration unstable compared with a regular bilayer crystal. However, the authors show that quantum zero‑point fluctuations (phonon‑like collective modes) lower the total energy of the quasiperiodic arrangement enough to make it the true ground state. They term this state an “electronic quasicrystal,” emphasizing that its stability is intrinsically quantum mechanical and has no classical analogue.

To substantiate the quasicrystal claim, the authors compute two contributions to the total energy: (i) the static Coulomb energy of the charge configuration, and (ii) the zero‑point energy obtained from the spectrum of vibrational modes derived from the NN‑VMC wavefunction. By scanning rs and d, they map out where the quasicrystal’s total energy falls below that of the competing bilayer crystal. The region of stability is roughly rs ≥ 30 (electron densities ≲ 10¹⁰ cm⁻²) and d ≥ 30 nm for GaAs/AlGaAs parameters.

The paper also discusses experimental implications. High‑mobility GaAs quantum wells with precisely controlled well width and carrier density are identified as realistic platforms. The electronic quasicrystal should manifest in diffraction experiments (electron microscopy or X‑ray) as characteristic Bragg peaks with five‑fold symmetry, and in transport as a sudden increase in resistivity. Moreover, external knobs such as gate‑induced electric fields, hydrostatic pressure, or asymmetric doping could be used to tune the inter‑layer tunneling and drive transitions between metallic, crystalline, and quasicrystalline phases, offering a versatile testbed for quantum phase control.

Methodologically, the work demonstrates that NN‑VMC can handle continuous three‑dimensional electron systems with thousands of particles, surpassing the size limitations of exact diagonalization and the correlation shortcomings of DFT. The attention‑based architecture efficiently captures long‑range entanglement, while the variational Monte‑Carlo optimizer provides unbiased energy estimates. This combination opens the door to first‑principles investigations of other low‑dimensional strongly correlated platforms, such as transition‑metal dichalcogenide heterostructures, moiré superlattices, and twisted bilayer systems.

In summary, the authors (i) develop an AI‑powered variational method capable of accurately describing the ground state of electrons in realistic quantum wells, (ii) map out a comprehensive phase diagram that includes metallic, monolayer crystal, bilayer crystal, and a newly predicted electronic quasicrystal, and (iii) provide concrete experimental signatures and pathways for verification. The discovery of a quantum‑stabilized quasicrystalline electronic order represents a landmark in condensed‑matter physics, illustrating how machine‑learning‑enhanced many‑body techniques can uncover entirely new phases of matter that were previously inaccessible to theory.