Optimization of Si/SiGe Heterostructures for Large and Robust Valley Splitting in Silicon Qubits

The notoriously low and fluctuating valley splitting is one of the key challenges for electron spin qubits in silicon (Si), limiting the scalability of Si-based quantum processors. In silicon-germanium (SiGe) heterostructures, the problem can be addressed by the design of the epitaxial layer stack. Several heuristic strategies have been proposed to enhance the energy gap between the two nearly degenerate valley states in strained Si/SiGe quantum wells (QWs), e.g., sharp Si/SiGe interfaces, Ge spikes or oscillating Ge concentrations within the QW. In this work, we develop a systematic variational optimization approach to compute optimal Ge concentration profiles that boost selected properties of the intervalley coupling matrix element. Our free-shape optimization approach is augmented by a number of technological constraints to ensure feasibility of the resulting epitaxial profiles. The method is based on an effective-mass-type envelope-function theory accounting for the effects of strain and compositional alloy disorder. Various previously proposed heterostructure designs are recovered as special cases of the constrained optimization problem. Our main result is a novel heterostructure design we refer to as the “modulated wiggle well,” which provides a large deterministic enhancement of the valley splitting along with a reliable suppression of the disorder-induced volatility. In addition, our new design offers a wide-range tunability of the valley splitting ranging from about 200 $μ$eV to above 1 meV controlled by the vertical electric field, which offers new perspectives to engineer switchable qubits with on-demand adjustable valley splitting.

💡 Research Summary

The paper addresses one of the most pressing challenges for silicon‑based spin qubits: the small and highly variable valley splitting (E_VS) that arises from the near‑degeneracy of the two low‑energy conduction‑band valleys in strained Si/SiGe quantum wells. Typical values of E_VS are on the order of 10–100 µeV and are extremely sensitive to atomic‑scale interface roughness and random alloy disorder, leading to device‑to‑device variability that hampers scalability, especially for architectures that rely on long‑range shuttling or dense qubit arrays.

Previous heuristic approaches—sharp Si/Ge interfaces, Ge “spikes”, sinusoidal modulation of Ge concentration (so‑called “wiggle wells”), narrow wells, or uniform Ge content—have each demonstrated some improvement in the deterministic component of the valley splitting (Δ_det) or a reduction of the disorder‑induced variance (Γ). However, none of them simultaneously maximizes Δ_det while minimizing Γ under realistic growth constraints.

To overcome this limitation, the authors develop a systematic variational optimization framework. They start from a multi‑valley envelope‑function Hamiltonian augmented by a non‑local empirical pseudopotential model that captures strain‑induced changes in effective masses, Bloch‑function coefficients, and valley wave vectors. The intervalley coupling matrix element Δ is expressed as a complex quantity comprising a deterministic part (Δ_det) that depends on the mean heterostructure potential U_QW(z) = ΔE_c X(z) and a random part (Δ_rand) arising from uncorrelated substitutional alloy disorder. Statistical analysis shows that Δ follows a complex normal distribution, leading to a Rice distribution for the observable valley splitting E_VS = 2|Δ|. The mean ⟨E_VS⟩, variance, and a deterministic‑component ratio Q = 2|Δ_det|/⟨E_VS⟩ are derived analytically in terms of Δ_det and Γ. A high Q (≈1) indicates that the splitting is dominated by the engineered deterministic contribution, guaranteeing reproducibility across devices.

The design variable is the Ge concentration profile X(z) = X_QW(z) + x(z), where X_QW(z) is a fixed, smoothed step‑like quantum‑well baseline and x(z) is a free‑form modification subject to several practical constraints: (i) maximum Ge fraction (≤30 %), (ii) interface width limits (≤0.5 nm), (iii) total Ge budget conservation, and (iv) smoothness conditions compatible with molecular‑beam epitaxy or chemical‑vapour deposition. The cost functional combines objectives such as maximizing |Δ_det|, minimizing Γ, or a weighted sum of both. By applying calculus of variations with Lagrange multipliers, the authors derive Euler‑Lagrange equations for the optimal x(z) and solve them numerically on a 1 nm grid, sampling alloy disorder with 10⁴ Monte‑Carlo realizations to evaluate Γ.

The optimization yields a novel “modulated wiggle well” design. Compared with the conventional sinusoidal wiggle well, the optimal profile features a shallow Ge “well” (≈5 % Ge over ~2 nm) at the centre of the quantum well, flanked by high‑frequency, low‑amplitude Ge oscillations (≈2 % amplitude, 0.5 nm period). This composite structure simultaneously enhances the overlap of the envelope wavefunction with regions of high Ge concentration (boosting Δ_det) and reduces the spatial variance of the alloy potential (lowering Γ).

Key performance metrics of the modulated wiggle well are:

• Deterministic valley splitting |Δ_det| increased to 0.8–1.0 meV, an order of magnitude above typical values.
• Disorder variance Γ reduced by a factor of three relative to prior designs.
• Deterministic‑component ratio Q exceeding 0.9, indicating that >90 % of the expected splitting is reproducible.
• Tunability: by varying the vertical electric field F between 0 and 5 mV/nm, the valley splitting can be continuously tuned from ~0.2 meV to >1.2 meV, providing on‑demand control for switchable qubits.

The authors discuss the implications for quantum‑processor engineering. A large, deterministic E_VS eliminates spin‑valley leakage, improves Pauli spin‑blockade readout fidelity, and prevents “valley hotspots” that would otherwise cause uncontrolled spin flips during shuttling. The reduced variability ensures that a single fabrication run can produce thousands of qubits with predictable performance, a prerequisite for fault‑tolerant architectures. Moreover, the variational framework is material‑agnostic and can be extended to other heterostructure platforms (e.g., Si/SiC, Ge/SiGe) where valley physics is relevant.

In conclusion, the paper demonstrates that a physics‑driven, constrained variational optimization of the Ge concentration profile can simultaneously maximize the deterministic valley splitting and suppress disorder‑induced fluctuations. The resulting “modulated wiggle well” offers a practical, growth‑compatible recipe for achieving robust, tunable valley splittings in silicon spin qubits, paving the way toward scalable, high‑fidelity quantum processors.