Proximity effect in a two-dimensional electron gas coupled to a thin superconducting layer

Introduction

Topological superconductors host zero-energy Majorana bound states at their edges that are highly sought for applications in topological quantum computing [1-3]. The two proposals to realize topological superconductivity that have received the most attention to date involve engineering Majorana bound states in either low-dimensional semiconducting systems [4-23] or in ferromagnetic atomic chains [24-32]. After the first signatures of topological superconductivity were observed [33-37], much of the experimental focus was placed on developing more suitable devices for realizing robust topological superconducting phases. One of the most significant experimental advances of the past few years was the successful epitaxial growth of thin layers of superconducting Al on InAs and InSb nanowires [38-42]. The intimate contact between the semiconductor and superconductor in these devices ensures a hard induced superconducting gap. Recently, this epitaxial growth technique has been applied also to InAs two-dimensional electron gases (2DEGs) [43-47].

The proximity effect has been theoretically studied recently in both strictly one-dimensional (1D) [48] and quasi-1D [49] wires coupled to thin superconducting layers. In both instances, a strong proximity coupling induces a large band shift on the semiconducting wire. This band shift is comparable to the level spacing in the superconductor, (which is δE_s ~ 400 meV for a superconductor thickness of d ~ 10 nm and a Fermi velocity of Al of v_F ~ 2 × 10⁶ m/s). In both cases, this large band shift makes it very challenging to realize a topological phase when utilizing thin superconducting layers.

In this paper, we extend the works of [48,49] to the 2D limit. We show that the large band shift that plagues the 1D case persists also in two dimensions. First, we show that the self-energy induced in an infinite 2DEG coupled to a superconductor of finite thickness is equivalent to that induced in an infinite wire coupled to a 2D superconductor of finite width (corresponding to the theoretical model of [48]), with the simple replacement of a 1D momentum by the magnitude of a 2D momentum. Analyzing the self-energy, we find that the induced gap in the presence of only Rashba spin–orbit coupling can be made comparable to the bulk gap of the superconductor only if the tunneling energy scale exceeds the large level spacing of the superconducting layer. As in the 1D case, the large tunneling energy scale induces a large band shift on the 2DEG and makes it very challenging to realize a topological phase. We also show that while the band shift can be significantly reduced by increasing the thickness of the superconducting layer, the topological phase is still difficult to realize if the 2DEG/superconductor interface remains very transparent.

Model of the Proximity Effect

The system we consider consists of a 2DEG with strong Rashba spin–orbit interaction (SOI) proximity-coupled to an s-wave superconductor of thickness d, as shown in Figure 1. The 2DEG-superconductor heterostructure is described by the action

The action of the 2DEG in Nambu space is given by

where ω is a Matsubara frequency, k = (k_x,k_y) is the momentum,

and

is a spinor of Heisenberg operators describing states in the 2DEG. The Hamiltonian density is

where ξ_k = k²/2m_2D − μ_2D (m_2D and μ_2D are the effective mass and chemical potential of the 2DEG, respectively, and ), α is the Rashba SOI constant, and σ_x,y,z (τ_x,y,z) are Pauli matrices acting in spin (Nambu) space. The superconductor is described by the BCS action,

where

is a spinor of Heisenberg operators describing states in the superconductor and the Hamiltonian density is

with m_s, μ_s, and Δ the effective mass, chemical potential, and pairing potential of the superconductor, respectively. Local tunneling at the interface between the two materials is assumed to conserve both spin and momentum,

where t is the tunneling amplitude. We must take the 2DEG to be located at some finite z_2D (0 < z_2D < d) due to the breakdown of the tunneling Hamiltonian approach for the case where the 2DEG is located at the boundary of the superconductor. The breakdown of the tunneling Hamiltonian results from our neglect of the thickness of the 2DEG (for related calculations in which the finite thickness is taken into account, see [50-53]). However, as shown in [48], choosing (where is the Fermi momentum of the superconductor) yields good agreement with numerical calculations in which there is no issue with placing the 2DEG strictly at the boundary.

In the absence of tunneling, the spectrum of the 2DEG consists of two spin–orbit-split subbands described by

When the finite-size quantization scale of the superconductor greatly exceeds the gap, , the first few subbands of the superconductor follow a linearized form given by ()

where δE_s = πv_F/d is the level spacing in the superconductor (v_F = k_F/m_s is the Fermi velocity) and . When the thickness of the superconducting layer is much smaller than its coherence length, = πv_F/Δ, the level spacing of the layer greatly exceeds its gap, .

The spectra of the 2DEG and the superconductor are plotted in Figure 2. Provided that , the bands of the 2DEG and superconductor intersect at high energies . Since we impose momentum conservation (in addition to energy conservation) in Equation 6, the subbands are coupled only at the intersection points. Thus, a weak tunnel coupling induces anticrossings in the spectrum, as indicated in Figure 2, which leads to a shift in the subbands of the 2DEG. Additionally, the tunnel coupling opens a superconducting gap at the Fermi momenta of the 2DEG; however, due to the intersection points lying at very large energies, the gap opened in the 2DEG is very small. A large gap can only be induced if tunneling is strong enough to overcome the large energy mismatch similar to δE_s.

To determine the self-energy of the 2DEG induced by the superconductor, we integrate out the superconducting degrees of freedom. After integrating out, the 2DEG can be described by the effective action

with the self-energy given by

In Equation 10, is the Green’s function of the bare superconductor (in the absence of tunneling), which satisfies

Imposing a vanishing boundary condition at z = 0 and z = d, we find a solution to Equation 11 given by

where = 2m_s(μ_s ± iΩ) − k² and Ω² = Δ² + ω² [48,54]. The Green’s function of a bulk superconductor, expressed in real space, is

where, in evaluating the integral, we make a semiclassical expansion k_± = k_Fφ ± iΩ/(v_Fφ) ≡ ζ ± iχ (valid in the limit ) and define a quantity that parametrizes the trajectories of states in the superconductor. Substituting the Green’s function (Equation 12) into the self-energy (Equation 10), we find

where we define

with γ = t²/v_F, an energy scale determined by the tunneling strength. The quantity Γ_k,ω can be interpreted as an effective quasiparticle weight, as it takes values of 0 < Γ < 1, and is responsible for inducing superconductivity in the 2DEG, while δμ_k,ω corresponds to a tunneling-induced shift in the effective chemical potential of the 2DEG. Quite surprisingly, the self-energy in Equation 14 and Equation 15 coincides with that of a nanowire coupled to a two-dimensional superconductor with finite width as found in [48], with the simple replacement of a 1D momentum by the magnitude of a 2D momentum.

Results and Discussion

Induced gap and band shift

Using the self-energy derived in the previous section, we first calculate the size of the proximity-induced gap in the 2DEG. Once we find an expression for the gap, we estimate the tunneling strength needed in order for the gap in the 2DEG to be comparable to that in the superconductor. We then add a Zeeman term to the Hamiltonian of the 2DEG and estimate the Zeeman energy needed to reach the topological phase in such a setup.

It is convenient to work in the chiral basis in which the normal Green’s function of the 2DEG is diagonal. To this end, we introduce a unitary transformation

with , which can be used to convert between the spin (σ) and chiral (λ) bases, . The Green’s function in the spin basis is given by . Rotating to the chiral basis, we find a Green’s function given by

where = ω/Γ_k,ω, = ξ_k − δμ_k,ω ± αk and = Δ(1/Γ_k,ω − 1). The spin–singlet pairing induced by the superconductor appears as intraband chiral p-wave pairing (of the form p_x ± ip_y) when expressed in the chiral basis.

Before continuing, let us simplify the parameters Γ_k,ω and δμ_k,ω. We will focus on the limit where the thickness of the superconducting layer is much smaller than its coherence length, (equivalently, ), and where the normal layer is located close to the edge of the superconductor, . Because of the large Fermi surface mismatch between the 2DEG and superconductor, we must have (or, equivalently, φ ≈ 1); in the following, we neglect the momentum dependence by setting φ = 1 (which is justified as long as we only consider momenta ). In the limit , the parameters simplify to

where we drop the subscript (k,ω) because both Γ and δμ are now independent of frequency and momentum. In expanding Equation 15 to arrive at Equation 18, we assumed that (therefore, these expressions break down when k_Fd/π→n, with ).

The spectrum of the proximitized 2DEG is determined by the poles of the retarded Green’s function. After analytic continuation iω→E + i0⁺, we find two branches of the spectrum from Equation 17 given by

where μ_eff = μ_2D + δμ is an effective chemical potential of the 2DEG. The spectrum describes an s-wave superconductor with Rashba-split bands and an excitation gap

We see that the size of the excitation gap is determined by the parameter Γ. When , the full bulk gap of the superconductor is induced in the 2DEG, while for , a very small gap is induced. In order to have an induced gap comparable (but not equal) to the bulk gap, we require that neither nor is satisfied. However, to realize this situation requires a tunneling strength

where we have assumed that (k_Fz_2D)²/sin²(k_Fd) ~ 1. If the tunneling strength is chosen as in Equation 21, the band shift measured at k = 0, E_±(0), is

Therefore, the scale of the band shift is also set by the level spacing in the thin superconducting layer. We note that while the quantity δμ is bounded only by the chemical potential of the superconductor μ_s (as the tunneling Hamiltonian approach itself should break down for γ ~ μ_s), the band shift saturates to E_±(0) ~ δE_s in the limit (where ).

We plot the spectrum of the 2DEG (see Equation 19) in Figure 3. In the weak-coupling limit (Figure 3a), there is a rather small band shift but a negligible superconducting gap is opened in the 2DEG. In the strong-coupling limit (Figure 3b), we show that while a larger gap is induced, the band shift is very large.

Topological transition

We now add a Zeeman splitting Δ_Z to the Hamiltonian of the 2DEG such that

Such a Zeeman splitting can arise due to the application of an out-of-plane magnetic field [4,5] (though orbital effects are not incorporated here) or due to the proximity of a magnetic insulator [8]. Also, it is possible to apply an in-plane magnetic field (to avoid unwanted orbital effects) to reach the topological phase if the 2DEG has a finite Dresselhaus SOI, as shown in [9]. An in-plane magnetic field in the presence of only Rashba SOI is not sufficient to reach the topological phase because it does not open a gap in the Rashba spectrum. The spectrum in the presence of the Zeeman splitting, which again is determined by poles in the retarded Green’s function , is given by

where we have used E_g = Δ(1 − Γ) as in Equation 20. Therefore, we find a gap-closing topological transition at k = 0 for the critical Zeeman splitting

In the case of a very large band shift, and , the topological transition is given by [note that Γ drops out of Equation 25 in this limit].

We now provide an estimate of the Zeeman splitting at which we expect the k = 0 gap-closing transition to occur experimentally in an Al/InAs 2DEG heterostructure. Given the thickness of the superconducting Al layer of d = 10 nm [44], we estimate a level spacing of δE_s = = 413 meV (taking v_F = 2 × 10⁶ m/s). Therefore, if a sizable gap is induced in the 2DEG, as observed experimentally, typical values for the band shift are of the same order of magnitude as the level spacing, Γδμ ~ 400 meV. Then, provided that the chemical potential cannot be controlled over such a large scale by external gates, the critical Zeeman splitting needed to reach the topological phase is = δμ ~ 400 meV. Such a large Zeeman splitting cannot be achieved in the 2DEG without destroying superconductivity in the thin layer. We also note the possibility that, by coincidence, the band shift vanishes (or becomes small); from Equation 18, we see that δμ = 0 if k_Fd = cot⁻¹(1/k_Fz_2D) + nπ (for ). In this special case, which requires the thickness of the superconducting layer to be finely tuned on the scale of its Fermi wavelength, there is no band shift to prevent one from tuning into a topological phase. However, for most devices, the large band shift makes it very challenging to realize a topological phase.

Increasing thickness of superconducting layer

The self-energy appearing most frequently in the literature to describe proximitized nanowires and 2DEGs [55-60], which also has been used often in interpreting experimental results [40,42], is that induced by a bulk superconductor,

Equation 26 can be obtained by setting z_2D = d/2 and taking the limit d→∞ in Equation 15 (or, as it is usually done, by substituting the bulk Green’s function in Equation 13 when evaluating the self-energy in Equation 10). Hence, this self-energy describes a 2DEG embedded within a bulk superconductor, as shown in Figure 4a. To describe the case where a 2DEG is placed at the surface of a bulk superconductor (as shown in Figure 4b), the limit d→∞ should be taken in Equation 15 while keeping z_2D finite (or, equivalently, substituting the Green’s function of a semi-infinite (SI) superconductor when evaluating the self-energy in Equation 10). For this case, we obtain

The most notable difference is the presence of a nonzero band shift in the semi-infinite case. However, this band shift is significantly reduced compared to the case of a thin superconducting layer, as it saturates to E_±(0) ~ Γ_SIδμ_SI ~ Δ in the limit .

Although it may seem that a topological phase can be much more easily realized by simply increasing the thickness of the superconducting layer in order to reduce the band shift induced on the 2DEG, this is not the case. Crucially, both the bulk and the semi-infinite self-energies give the ratio γ/Δ as the relevant parameter determining whether the system is in the weak-coupling [, or equivalently ] or strong-coupling [(1 − Γ) ~ 1, or equivalently E_g ~ Δ] limit. This is in stark contrast to the limit of a thin superconducting layer, where a tunneling energy is required to open a gap E_g ~ Δ in the 2DEG. Therefore, because the tunneling energy γ is a property of the interface and should not be expected to change as the thickness of the superconducting layer is increased, this energy is fixed to γ ~ δE_s provided that the interface is transparent enough to induce a gap in the thin-layer limit (as seen in the experiments). If the thickness of the superconductor is increased, such that , the system will be deep within the strong-coupling limit; from Equation 26 and Equation 27, we find . The critical Zeeman splitting needed to induce a topological phase (see Equation 25) is therefore given by ~ Δ/Γ ~ γ ~ 400 meV. We note that in the case of a thin superconducting layer, the topological transition is pushed to large Zeeman splitting by very large δμ, which could possibly be compensated for if the chemical potential μ_2D has a large range of tunability. In the case of a bulk system, the topological transition is pushed to large Zeeman splitting by very small Γ, which cannot be affected by tuning μ_2D. Hence, even if the thickness d of the superconducting layer is made infinite, the topological phase transition is determined by the interfacial tunneling energy. In order to induce a topological phase more reliably, a much weaker coupling between a 2DEG and a bulk superconductor (such that γ ≤ Δ) should be sought. We also note that this result applies to the 1D model considered in [48] as well.)

Conclusion

We have studied the proximity effect in a two-dimensional electron gas (2DEG) strongly coupled to a thin superconducting layer, showing that the detrimental band shift shown in [48,49] to dominate the proximity effect in wires is also crucial in 2DEGs. In order to induce a sizable gap in the 2DEG, the tunneling energy scale must overcome the large level spacing within the superconductor. However, introducing such a large energy scale to the semiconductor induces a large band shift that makes it challenging to realize a topological phase. This challenge cannot be alleviated by simply increasing the thickness of the superconducting layer but requires a significant weakening of the proximity coupling afforded by the epitaxial interface.