betaFeedback

WorksPhysics and Astronomy

Understanding disorder in Silicon quantum computing platforms: Scattering mechanisms in Si/SiGe quantum wells

Abstract

Motivated by recent experiments on Si/SiGe quantum wells with a co-design of high electron mobility and large valley splitting [B. Paquelet Wuetz, et al., Nature Communications 14, 1385 (2023); D. D. Esposti, et al., arXiv:2309.02832], suitable for a Si-based spin qubit quantum computing platform, we examine the role of disorder by theoretically calculating mobility and quantum mobility from various scattering mechanisms and their dependence on the electron density. At low electron densities ne<4×1011n_e < 4 \times 10^{11} cm−2^{-2}, we find that mobility is limited by remote Coulomb impurities in the capping layer, whereas interface roughness becomes the significant limiting factor at higher densities. We also find that alloy disorder scattering is not a limiting mechanism in the reported high-mobility structures. We estimate the critical density of the disorder-driven low-density metal-insulator transition using the Anderson-Ioffe-Regel localization criterion and qualitatively explain the breakdown of the Boltzmann-Born theory at low densities. We also estimate the critical density by considering inhomogeneous density fluctuations induced by long-range Coulomb disorder in the system, and find a larger critical density compared to the one obtained from the Anderson-Ioffe-Regel criterion. For quantum mobility, our calculation suggests that remote and distant background impurities are likely the limiting scattering sources across all density. Future measurements of quantum mobility should provide more information on the distribution of background impurities inside the SiGe barriers. Moreover, we extend our theoretical analysis to the effect of quantum degeneracy on transport properties and predict the mobility and the critical density for the metal-insulator transition in spin-polarized high-mobility structures under an external parallel magnetic field.

Introduction

For spin qubits in gate-defined silicon quantum dots, it is important to minimize disorder potential fluctuations to better control charging energies and tunneling between quantum dots, which requires high mobility of the hosted two-dimensional electron gas (2DEG) [1, 2, 3]. Moreover, background charge noise, necessarily associated with unintentional random impurities in the environment, is the primary mechanism causing charge noise induced decoherence in semiconductor qubits, and therefore, a comprehensive understanding of the background impurity disorder is essential for good qubit performance [3, 4, 5, 6]. Additionally, a large valley splitting energy is important to limit leakage from the computational Hilbert space in the lowest conduction band, which is crucial for maintaining high-fidelity qubit operations [3, 7, 8, 9, 10, 11, 12]. Recent advancements in isotopically purified 28^{28}Si/SiGe quantum wells (QWs) at a low temperature of 1.7 K have shown promising improvements in channel static disorder, as indicated by an enhanced mobility 3×105 cm2/Vs3 \times10^5\ \mathrm{cm^2/Vs} and a smaller critical density of metal-insulator transition nc=7×1010 cm−2n_c = 7 \times10^{10}\ \mathrm{cm}^{-2}, while maintaining a relatively high mean valley splitting 0.2 meV [13, 14]. (Isotopic purification suppresses nuclear spin noise for semiconductor based spin qubits in the Si platform, providing long spin coherence times and enabling high-fidelity gate operations [15].) In comparison, previous experimental reports of large valley splitting $> 0.2\ \mathrm{meV}$ in Si/SiGe QWs showed 5 times lower mobility $< 6 \times 10^4\ \mathrm{cm^2/Vs}$ [16, 17, 18]. On the other hand, high mobility of $6.5 \times 10^5\ \mathrm{cm^2/Vs}$ was reported in conventional Si/SiGe heterostructures but with low valley splitting $35 - 70\ \mu\mathrm{eV}$ {citep}[19]. Therefore, the co-design of high electron mobility and large valley splitting calls for reinvestigating the scattering sources inside the Si/SiGe QWs reported in Refs. {citep}[13, 14]`. This is the

goal of our theoretical work, particularly in the context of the importance of the Si/Ge quantum computing platform.

In this study, we employ the Boltzmann-Born transport theory [20] to theoretically investigate both transport and quantum mobility (μ\mu and μq\mu_q). Our focus is on quantifying the effects of various scattering mechanisms, namely remote impurities at the semiconductor-oxide interface (RI), background impurities (BI), interface roughness (IR), and alloy disorder (AD). By analyzing experimental mobility data from 5 nm and 7 nm Si/SiGe quantum wells reported in Refs. [13, 14], we find that RI predominantly limits μ\mu at electron densities below 4×1011 cm−24 \times10^{11}\ \mathrm{cm}^{-2}, while IR becomes the limiting factor at higher densities ranging from 4−6×1011 cm−24 - 6 \times10^{11}\ \mathrm{cm}^{-2}. Despite limited experimental data on quantum mobility (μq=3.0±0.5×104 cm2/Vs\mu_q = 3.0 \pm0.5 \times10^4\ \mathrm{cm^2/Vs} at the highest density 6×1011 cm−26 \times10^{11}\ \mathrm{cm}^{-2} [21]), our theoretical models indicate that μq\mu_q is likely constrained by RI and distant BI scattering across all electron densities. Since μq\mu_q is much more sensitive to RI and distant BI scattering compared to μ\mu [22], we need more experimental data on μq\mu_q across a wider range of densities to better understand the distribution for BI in SiGe barriers. We also show that AD is quantitatively irrelevant in all experimental density ranges for these high-mobility structures. We address the limitations of the Boltzmann-Born theory at low electron densities ne≲2×1011 cm−2n_e \lesssim2 \times10^{11}\ \mathrm{cm}^{-2} with considerable disagreement between the theory and experimental data, although our theory explains the higher density data quantitatively. A possible explanation for this discrepancy at low densities is that the 2DEG is broken into inhomogeneous electron puddles separated by long-range Coulomb disorder potential barriers, so that both the perturbative Born approximation and linear screening are no longer applicable. The system eventually undergoes a metal-insulator transition (MIT) at a critical density ne=ncn_e = n_c, where transport in the low-density ne≳ncn_e \gtrsim n_c region should be described by percolation through charged puddles [23, 24, 25, 26, 27, 28, 29, 30]. This MIT is, in fact, a crossover (happening sharply over a narrow density range around ncn_c because of the rapid failure of screening of the long-range Coulomb disorder) and not a quantum phase transition [26, 27]. We estimate ncn_c in the experimental systems [13, 14] by using the Anderson-Ioffe-Regel (AIR) criterion [31, 32], although the failure of the Boltzmann-Born theory at low densities indicates that the critical density estimated from the AIR criterion should be smaller than the actual critical density (percolation threshold) observed in experiments and should be used as a lower bound estimate of the percolation threshold. We also estimate ncn_c by considering inhomogeneous density fluctuations induced by long-range Coulomb disorder in the system [33, 34], and find a larger ncn_c compared to the one obtained from the AIR criterion. Moreover, we extend our theoretical analysis to the effect of quantum degeneracy on transport properties and predict the mobility and the critical density for the metal-insulator transition in spin-polarized high-mobility structures, which can be tested in future experiments by applying an external parallel magnetic field.

The Si/SiGe heterostructures reported in Refs. [13, 14] consist of a strained Si QW of width ww surrounded by a Si0.69_{0.69}Ge0.31_{0.31} upper barrier of thickness d=30d = 30 nm and a thick strained relaxed Si0.69_{0.69}Ge0.31_{0.31} bottom barrier of several μ\mum. The upper barrier is passivated with a 1 nm Si capping layer. A metallic top gate is separated from the capping layer by a do=10d_o = 10 nm oxide dielectric spacer. The electron density nen_e is tuned by the gate up to 6×10116 \times10^{11} cm−2^{-2}. The bottom barrier is placed on top of a step-graded Si1−x_{1-x}Gex_x strain-relax buffer layer, which has a decreasing Ge concentration xx until it reaches the Si substrate with x=0x = 0. A schematic illustration of the Si/SiGe heterostructure is shown in Figure 1. In our calculations, we model the possible disorder sources as follows. There are uniform background charged impurities outside (inside) the QW with 3D concentration N1N_1 (N2N_2). The remote charged impurities at the semiconductor-oxide interface have a 2D concentration nrn_r. Because the top gate is relatively close to the semiconductor-oxide interface, we take into account the gate screening when we calculate the remote and background charged impurity scattering. The interface roughness between the Si QW and the Si0.69_{0.69}Ge0.31_{0.31} barrier is characterized by a typical height Δ\Delta and a lateral size Λ\Lambda. The conduction band offset from the twofold valley degenerate conduction band Δ2\Delta_2 in the strained Si to the sixfold valley degenerate conduction band Δ6\Delta_6 in the relaxed Si0.69_{0.69}Ge0.31_{0.31} is taken to be V0=180V_0 = 180 meV [3, 35]. The 2DEG wave function penetrates the barriers and scatters from the alloy disorder. Scattering mechanisms considered in our theory are the only low-temperature scattering mechanisms operational in high quality SiGe 2D structures, where the crystal quality is good enough to rule out short-range scattering by atomic point defects and vacancies [36], but random charged impurities are invariably present in the system along with interface roughness and alloy disorder. Since our interest is in understanding the low-temperature mobility, we neglect all phonon scattering and restrict our calculations to T=0T = 0. We believe that our comprehensive transport theory is quantitatively accurate at high carrier densities, and poorer at lower densities although it should remain qualitatively (and semiquantitatively) valid at lower densities above the critical density for the 2D MIT.

Figure 1. A schematic illustration of the high-mobility Si/SiGe heterostructure reported in Refs. [13, 14], where ww is the thickness of the strained Si quantum well. The oxide and the top Si0.69_{0.69}Ge0.31_{0.31} barrier is separated by a Si capping layer of thickness 1 nm.

The rest of this paper is structured as follows. § 2 discusses the general formalism of the Boltzmann transport theory to calculate the mobility and quantum mobility. We evaluate the BI and RI scattering rates using the delta-layer 2DEG approximation in § 3. The infinite potential well approximation together with the local field correction, applied to RI, BI, and IR scattering, is discussed in § 4, while § 5 employs a finite potential well approximation to calculate the AD scattering. Gate-screened charged impurity scattering is explored in Section VI. In § 7, we estimate the critical densities of MIT using the Anderson-Ioffe-Regel (AIR) criterion and qualitatively explain the breakdown of the Boltzmann-Born theory at low densities. We mention that the AIR criterion is often referred to as the Mott-Ioffe-Regel (MIR or IRM) or just the Ioffe-Regel criterion, but we believe that AIR is a more appropriate terminology here since our criterion is specific to disorder induced incoherent scattering associated with destructive interference as envisioned by Anderson rather than the Mott criterion involving the transport mean free path being equal to lattice spacing, which plays no role in doped semiconductors of interest to us. We also estimate the critical density by considering inhomogeneous density fluctuations induced by long-range Coulomb disorder in the system and compare the result to the AIR critical density. Section VIII examines the effect of spin/valley degeneracy on the mobility and the AIR critical density. Finally, we conclude and summarize in § 9.

Mobility and quantum mobility

In this section, we apply the Boltzmann transport theory to calculate the mobility and quantum mobility at T=0T = 0 [20, 37, 38, 39, 40, 41, 42]. The transport mobility is given by μ=eτ/m⋆\mu= e\tau/m^{\star}, and the conductivity is eneμen_{\mathrm{e}}\mu, where nen_{\mathrm{e}} is the 2DEG concentration, m⋆=0.19m0m^{\star} = 0.19m_{0} is the transverse effective mass of an electron in Si, m0m_{0} is the free electron mass,

1τ=4m⋆πℏ3∫02kFdq4kF2−q2(q2kF)2⟨∣U(q)∣2⟩.(1)\frac{1}{\tau} = \frac{4m^{\star}}{\pi\hbar^{3}}\int_{0}^{2k_{\mathrm{F}}}\frac{dq}{\sqrt{4k_{\mathrm{F}}^{2}-q^{2}}}\left(\frac{q}{2k_{\mathrm{F}}}\right)^{2}\left\langle\lvert U(q)\rvert^{2}\right\rangle. \tag*{(1)}

is the transport scattering rate, and U(q)U(q) is the screened potential of a given scattering source. kF=4πne/gk_{\mathrm{F}} = \sqrt{4\pi n_{\mathrm{e}}/g} is the Fermi wave vector. g=gsgvg = g_{\mathrm{s}}g_{\mathrm{v}} is the total degeneracy of electrons, where gsg_{\mathrm{s}} is the spin degeneracy and gvg_{\mathrm{v}} is the valley degeneracy. The quantum mobility is given by μq=eτq/m⋆\mu_{\mathrm{q}} = e\tau_{\mathrm{q}}/m^{\star}, where the quantum (single particle) scattering rate reads

1τq=2m⋆πℏ3∫02kFdq4kF2−q2⟨∣U(q)∣2⟩.(2)\frac{1}{\tau_{\mathrm{q}}} = \frac{2m^{\star}}{\pi\hbar^{3}}\int_{0}^{2k_{\mathrm{F}}}\frac{dq}{\sqrt{4k_{\mathrm{F}}^{2}-q^{2}}}\left\langle\lvert U(q)\rvert^{2}\right\rangle. \tag*{(2)}

The quantum scattering rate can be indirectly measured through the Dingle temperature TD=ℏ/2πτqT_{\mathrm{D}} = \hbar/2\pi\tau_{\mathrm{q}} which characterizes the damping of Shubnikov-de Haas (SdH) oscillations. In general, τq<τ\tau_{\mathrm{q}} < \tau since the transport mobility ignores all forward scattering. (For strictly zero range s-wave scattering the two scattering times are trivially equal, but for long-range disorder scattering the two could vary by orders of magnitude.) For charged impurity scattering, the potential correlator averaged over impurity positions is given by

⟨∣U(q)∣2⟩=∫−∞+∞dz N(z)U12(q,z),(3)\left\langle\lvert U(q)\rvert^{2}\right\rangle= \int_{-\infty}^{+\infty} dz\,N(z)U_{1}^{2}(q,z), \tag*{(3)}

where N(z)N(z) is the 3D concentration of impurities at a distance zz from the center of the 2DEG, and U1(q,z)U_{1}(q,z) is the screened Coulomb potential for an impurity located at zz:

U1(q,z)=2πe2κqϵ(q)∫−∞+∞dz′ ∣ψ(z′)∣2e−q∣z−z′∣,(4)U_{1}(q,z) = \frac{2\pi e^{2}}{\kappa q\epsilon(q)}\int_{-\infty}^{+\infty} dz^{\prime}\,\lvert\psi(z^{\prime})\rvert^{2}e^{-q\lvert z-z^{\prime}\rvert}, \tag*{(4)}

where ψ(z)\psi(z) is the confinement wave function along zz direction and the dielectric function is given by

ϵ(q)=1+(qTF/q)Fc(qw)[1−G(q)].(5)\epsilon(q) = 1 + (q_{\mathrm{TF}}/q)F_{\mathrm{c}}(qw)[1-G(q)] . \tag*{(5)}

Here, qTF=g/aBq_{\mathrm{TF}} = g/a_{\mathrm{B}} is the Thomas-Fermi screening wave vector, aB=κℏ2/m⋆e2a_{\mathrm{B}} = \kappa\hbar^{2}/m^{\star}e^{2} is the effective Bohr radius, κ=11.9\kappa= 11.9 is the dielectric constant of Si, and ww is the QW width. G(q)=q/(gq2+kF2)G(q) = q/(g\sqrt{q^{2}+k_{\mathrm{F}}^{2}}) is the local field correction using the Hubbard approximation [43, 44] which reflects the suppressed screening at q≳kFq \gtrsim k_{\mathrm{F}} (with G=0G = 0 being the RPA, which overestimates short wavelength screening). The form factor Fc(qw)F_{\mathrm{c}}(qw) is given by

Fc(qw)=∫−∞+∞dz∫−∞+∞dz′∣ψ(z)∣2∣ψ(z′)∣2exp⁡(−q∣z−z′∣).(6)F_{\mathrm{c}}(qw) = \int_{-\infty}^{+\infty} dz\int_{-\infty}^{+\infty} dz^{\prime}\lvert\psi(z)\rvert^{2}\lvert\psi(z^{\prime})\rvert^{2}\exp(-q\lvert z-z^{\prime}\rvert). \tag*{(6)}

In Eq. (4) we ignore the screening from the top gate. To include the gate screening, we should add the Coulomb potential of the image charges and replace e−q∣z−z′∣e^{-q\lvert z-z^{\prime}\rvert} by (e−q∣z−z′∣−e−q∣2dg−z−z′∣)\left(e^{-q\lvert z-z^{\prime}\rvert}-e^{-q\lvert2d_{\mathrm{g}}-z-z^{\prime}\rvert}\right), where dg=d+do+w/2d_{\mathrm{g}} = d+d_{\mathrm{o}}+w/2 is the distance from the gate to the center of the 2DEG.

Delta layer 2DEG approximation

For simplicity, assuming that the gate is far away dg≫dd_{\mathrm{g}}\gg d and the wavefunction perpendicular to the quantum well has a delta function profile ∣ψ(z)∣2=δ(z)\lvert\psi(z)\rvert^{2}=\delta(z), the form factors reduce to identities and U1(q,z)=2πe2κqϵ(q)e−q∣z∣U_{1}(q,z)=\frac{2\pi e^{2}}{\kappa q\epsilon(q)}e^{-q\lvert z\rvert}. In this section, we disregard the local field correction, thus making the dielectric function the RPA form, ϵ(q)=1+qTF/q\epsilon(q)=1+q_{\mathrm{TF}}/q. We will address the effect of form factors and local field correction in Section IV, and the effect of gate screening in Section VI. We use a 2-impurity model to calculate the scattering rate following Refs. [41, 45]. That is, we assume that there are two delta layers of impurities. One layer is located at a distance dd from the 2DEG, which represents the remote charged impurities in the capping layer. The other layer is at z=0z=0, which represents oxygen-related background charges in the QW [46] or an effective description of other short-range scattering sources such as the interface roughness [47, 48]. The impurity distribution is given by

N(z)=nrδ(z−d)+n2δ(z).(7)N(z) = n_{\mathrm{r}}\delta(z-d)+n_{2}\delta(z). \tag*{(7)}

The corresponding transport scattering rate τ−1=τRI−1+τS−1\tau^{-1}=\tau_{\mathrm{RI}}^{-1}+\tau_{\mathrm{S}}^{-1} reads [40, 41, 45]

1τRI=nr2πℏm⋆(2g)2f(2kFd,qTF2kF),(8)\frac{1}{\tau_{\mathrm{RI}}} = n_{\mathrm{r}}\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^{2}f\left(2k_{\mathrm{F}}d,\frac{q_{\mathrm{TF}}}{2k_{\mathrm{F}}}\right), \tag*{(8)}
1τS=n22πℏm⋆(2g)2f(0,qTF2kF).(9)\frac{1}{\tau_{\mathrm{S}}} = n_{2}\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^{2}f\left(0,\frac{q_{\mathrm{TF}}}{2k_{\mathrm{F}}}\right). \tag*{(9)}

The quantum scattering rate τq−1=τqRI−1+τqS−1\tau_{\mathrm{q}}^{-1}=\tau_{\mathrm{qRI}}^{-1}+\tau_{\mathrm{qS}}^{-1} reads

1τqRI=nr2πℏm⋆(2g)2fq(2kFd,qTF2kF),(10)\frac{1}{\tau_{\mathrm{qRI}}} = n_{\mathrm{r}}\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^{2}f_{\mathrm{q}}\left(2k_{\mathrm{F}}d,\frac{q_{\mathrm{TF}}}{2k_{\mathrm{F}}}\right), \tag*{(10)}
1τqS=n22πℏm⋆(2g)2fq(0,qTF2kF),(11)\frac{1}{\tau_{\mathrm{qS}}} = n_{2}\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^{2}f_{\mathrm{q}}\left(0,\frac{q_{\mathrm{TF}}}{2k_{\mathrm{F}}}\right), \tag*{(11)}

and the dimensionless functions f(a,s)f(a,s) and fq(a,s)f_{\mathrm{q}}(a,s) are defined as

f(a,s)=∫01s2e−2ax1−x2(x+s)2 dx,(12)f(a,s) = \int_{0}^{1}\frac{s^{2}e^{-2ax}}{\sqrt{1-x^{2}(x+s)^{2}}}\,dx , \tag*{(12)}
fq(a,s)=∫01s2e−2ax1−x2(x+s)2 dx.(13)f_{\mathrm{q}}(a,s) = \int_{0}^{1}\frac{s^{2}e^{-2ax}}{\sqrt{1-x^{2}(x+s)^{2}}}\,dx . \tag*{(13)}

In the density range

g16πd2≪ne≪g316πaB2,(14)\frac{g}{16\pi d^{2}} \ll n_e \ll\frac{g^{3}}{16\pi a_{B}^{2}}, \tag*{(14)}

the limits 2kFd≫12k_F d \gg1 and qTF/2kF≫1q_{\mathrm{TF}}/2k_F \gg1 are satisfied and we have f(a,s)=(2a3)−1f(a,s)=(2a^{3})^{-1} and fq(a,s)=(2a)−1f_q(a,s)=(2a)^{-1} [49] which lead to the well-known results for RI scattering [50, 51, 52, 53]

1τRI=nr(2g)2πℏ8m⋆(kFd)3,(15)\frac{1}{\tau_{\mathrm{RI}}}=n_r\left(\frac{2}{g}\right)^2\frac{\pi\hbar}{8m^{\star}(k_Fd)^3}, \tag*{(15)}
1τqRI=nr(2g)2πℏ2m⋆kFd.(16)\frac{1}{\tau_{q\mathrm{RI}}}=n_r\left(\frac{2}{g}\right)^2\frac{\pi\hbar}{2m^{\star}k_Fd}. \tag*{(16)}

and the corresponding ratio is given by

(ττq)RI=(2kFd)2≫1.(17)\left(\frac{\tau}{\tau_q}\right)_{\mathrm{RI}}=(2k_Fd)^2\gg1. \tag*{(17)}

The large difference between τ\tau and τq\tau_q reflects the diffusive character of the electron dynamics on the Fermi surface scattered from remote impurities [22]. The physical meaning of τ/τq\tau/\tau_q is the typical number of small-momentum forward scattering events needed to change the direction of momentum by an angle of order π\pi [53]. Next, we calculate the short-range scattering rate τS−1\tau_S^{-1} and τqS−1\tau_{qS}^{-1} contributed from a delta layer of impurities at z=0z=0. The related integrals f(0,s)f(0,s) and fq(0,s)f_q(0,s) at s>1s>1 can be calculated analytically:

f(0,s)=s2π+2s31−s2+2s3(2−s2)sec⁡−1(s)(s2−1)3/2,(18)f(0,s)=s^2\pi+\frac{2s^3}{1-s^2}+\frac{2s^3(2-s^2)\sec^{-1}(s)}{(s^2-1)^{3/2}}, \tag*{(18)}
fq(0,s)=s1−s2+s3sec⁡−1(s)(s2−1)3/2.(19)f_q(0,s)=\frac{s}{1-s^2}+\frac{s^3\sec^{-1}(s)}{(s^2-1)^{3/2}}. \tag*{(19)}

In the density range ne≪g3/16πaB2n_e\ll g^3/16\pi a_B^2, the limit s=qTF/2kF≫1s=q_{\mathrm{TF}}/2k_F\gg1 is satisfied and f(0,s)=fq(0,s)=π/2f(0,s)=f_q(0,s)=\pi/2, so that τS−1\tau_S^{-1} and τqS−1\tau_{qS}^{-1} are given by [40, 45]

1τS=1τqS=n2π2ℏm⋆(2g)2,(20)\frac{1}{\tau_S}=\frac{1}{\tau_{qS}}=n_2\frac{\pi^2\hbar}{m^{\star}}\left(\frac{2}{g}\right)^2, \tag*{(20)}

which is independent of the electron concentration. The ratio τ/τq=1\tau/\tau_q=1 reflects that short-range impurities are efficient in backscattering the electrons – this is the situation in regular 3D metals where τ\tau and τq\tau_q are basically the same.

In the other 2-impurity model, we replace the delta-layer remote impurities by uniform background charged impurities outside the well with a 3D concentration N1N_1. The impurity distribution is given by

N(z)=N1Θ(∣z∣−w/2)+n2δ(z),(21)N(z)=N_1\Theta(|z|-w/2)+n_2\delta(z), \tag*{(21)}

The corresponding scattering rates due to uniform background charged impurities are given by

1τBI=N12πℏm⋆(2g)2∫∣z∣>w/2f(2kFz,qTF2kF) dz,(22)\frac{1}{\tau_{\mathrm{BI}}}=N_1\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^2\int_{|z|>w/2}f\left(2k_Fz,\frac{q_{\mathrm{TF}}}{2k_F}\right)\,dz, \tag*{(22)}
1τqBI=N12πℏm⋆(2g)2∫∣z∣>w/2fq(2kFz,qTF2kF) dz.(23)\frac{1}{\tau_{q\mathrm{BI}}}=N_1\frac{2\pi\hbar}{m^{\star}}\left(\frac{2}{g}\right)^2\int_{|z|>w/2}f_q\left(2k_Fz,\frac{q_{\mathrm{TF}}}{2k_F}\right)\,dz. \tag*{(23)}

In the electron density range

g16πw2≪ne≪g316πaB2,(24)\frac{g}{16\pi w^2}\ll n_e\ll\frac{g^3}{16\pi a_B^2}, \tag*{(24)}

the limits 2kFw≫12k_Fw\gg1 and qTF/2kF≫1q_{\mathrm{TF}}/2k_F\gg1 are satisfied and we get the analytical expressions of the BI scattering rates

1τBI=N1w(2g)2πℏ2m⋆(kFw)3,(25)\frac{1}{\tau_{\mathrm{BI}}}=N_1w\left(\frac{2}{g}\right)^2\frac{\pi\hbar}{2m^{\star}(k_Fw)^3}, \tag*{(25)}
1τqBI=N1w(2g)2πℏln⁡(2ds/w)m⋆kFw,(26)\frac{1}{\tau_{q\mathrm{BI}}}=N_1w\left(\frac{2}{g}\right)^2\frac{\pi\hbar\ln(2d_s/w)}{m^{\star}k_Fw}, \tag*{(26)}

where dsd_s is the cutoff distance of the BI distribution so that impurities at ∣z∣>ds|z|>d_s can be ignored. Otherwise, τqBI−1\tau_{q\mathrm{BI}}^{-1} diverges logarithmically due to the contribution of small-momentum scattering from the long distance. We choose ds=dg/2d_s=d_g/2 as half the distance from the gate to the center of the 2DEG, considering that charged impurities close to the gate should be effectively screened [54]. We will discuss this choice of dsd_s and the gate screening effect in more detail in Section VI.

In the density range described by Eqs. (14) and (24), we see from Eqs. (15) and (25) that τRI−1\tau_{\mathrm{RI}}^{-1} and τBI−1\tau_{\mathrm{BI}}^{-1} are proportional to kF−3k_F^{-3}, which leads to a mobility proportional to ne3/2n_e^{3/2}. On the other hand, τS−1\tau_S^{-1} in Eq. (20) does not depend on kFk_F, so it results in a constant mobility that does not depend on nen_e. As a result, the total mobility described by the 2-impurity model τ−1=τRI−1+τS−1\tau^{-1}=\tau_{\mathrm{RI}}^{-1}+\tau_S^{-1} or τ−1=τBI−1+τS−1\tau^{-1}=\tau_{\mathrm{BI}}^{-1}+\tau_S^{-1} should increase as ne3/2n_e^{3/2} at low density and eventually saturate at high density. We can compare the density range ne=1−6×1011 cm−2n_e=1-6\times10^{11}\ \mathrm{cm}^{-2} in the high mobility structures reported in Refs. [13, 14] with the density ranges written in Eqs. (14) and (24). We use the spin/valley degeneracy gs=2g_s=2 and gv=2g_v=2 for the high mobility structures reported in Refs. [13, 14]. In the density range ne=1−6×1011 cm−2n_e=1-6\times10^{11}\ \mathrm{cm}^{-2} the Fermi energy calculated using g=4g=4 is EF=0.6−3.8 meVE_F=0.6-3.8\ \mathrm{meV} which is much higher than the valley splitting 0.2 meV0.2\ \mathrm{meV} reported in Refs. [13, 14], so our choice of g=4g=4 is correct. In Section VIII, we will discuss the effects of quantum degeneracy on the transport properties in more detail. Using g=4g=4, we find that the experimental density range is well within the range described in Eq. (14) where the upper bound is g3/16πaB2≈1013 cm−2g^3/16\pi a_B^2\approx10^{13}\ \mathrm{cm}^{-2} and the lower bound is g/16πd2≈1010 cm−2g/16\pi d^2\approx10^{10}\ \mathrm{cm}^{-2}. Therefore, the asymptotic expressions for RI rates, Eqs. (15) and (16), should be a good approximation. On the other hand, the lower bound in Eq. (24) is g/16πw2≈2.4×1011 cm−2g/16\pi w^2\approx2.4\times10^{11}\ \mathrm{cm}^{-2}, which means that the asymptotic expressions for BI scattering rates Eqs. (25) and (26) are good only when ne≫2.4×1011 cm−2n_e\gg2.4\times10^{11}\ \mathrm{cm}^{-2}. At low densities ne≪2.4×1011 cm−2n_e\ll2.4\times10^{11}\ \mathrm{cm}^{-2}, we have

1τBI=1τqBI=N1(2ds−w)π2ℏm⋆(2g)2,(27)\frac{1}{\tau_{\mathrm{BI}}}=\frac{1}{\tau_{q\mathrm{BI}}}=N_1(2d_s-w)\frac{\pi^2\hbar}{m^{\star}}\left(\frac{2}{g}\right)^2, \tag*{(27)}

and the background impurity scattering behaves similarly to short-range impurity scattering with an effective 2D impurity concentration N1(2ds−w)N_1(2d_s-w). Hence, as the density is reduced, τBI−1\tau_{\mathrm{BI}}^{-1} and τqBI−1\tau_{q\mathrm{BI}}^{-1} transition from a long-range behavior, as expressed in Eq. (27). The crossover electron density is at ne=2.4×1011 cm−2n_e = 2.4 \times10^{11}\ \mathrm{cm}^{-2}.

Figure 2 shows the results of the best-fit mobility curves (solid black) and the corresponding prediction of quantum mobility (dashed black) by performing the numerical integrals of Eqs. (8), (9) and (22). The experimental data from Refs. [13, 14], represented by blue and red curves with shaded areas indicating error bars, are also included for comparison. The results of the combination of remote and short-range impurity scattering are shown in Figure 2 (a) and (Figure 2b), while the results of the combination of background and short-range impurity scattering are shown in Figure 2 (c) and (Figure 2d). The fitting impurity densities are nr≈2×1012 cm−2n_r \approx2 \times10^{12}\ \mathrm{cm}^{-2}, n2≈1.4n_2 \approx1.4–2.8×109 cm−22.8 \times10^{9}\ \mathrm{cm}^{-2} and N1≈10×1015 cm−3N_1 \approx10 \times10^{15}\ \mathrm{cm}^{-3}, in reasonable agreement with the numbers reported in Refs. [35, 40, 46, 55, 56, 57, 58]. In both 2-impurity models, short-range impurity scattering is the limiting factor for transport mobility at high densities ne=4n_e = 4–6×1011 cm−26 \times10^{11}\ \mathrm{cm}^{-2}. The dominant scattering mechanism at lower densities is RI or BI, depending on the choice of the impurity model. We see that the first 2-impurity model (τ−1=τRI−1+τS−1\tau^{-1} = \tau_{\mathrm{RI}}^{-1} + \tau_{\mathrm{S}}^{-1}) fits the mobility data at intermediate to high electron density ne=2.5n_e = 2.5–6×1011 cm−26 \times10^{11}\ \mathrm{cm}^{-2} within the experimental uncertainty. At the highest density ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2} for both quantum wells, the predicted quantum mobility is around 1.0×104 cm2/Vs1.0 \times10^{4}\ \mathrm{cm}^{2}/\mathrm{Vs}, while the measurements of SdH oscillations give a larger value (3.0±0.5)×104 cm2/Vs(3.0 \pm0.5) \times10^{4}\ \mathrm{cm}^{2}/\mathrm{Vs} [21].

Figure 2. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the delta-layer approximation for the 2DEG without local field correction. Subfigures (a) and (b) incorporate remote and short-range impurity scattering [Eqs. (8), (9), and (10)], whereas (c) and (d) consider background and short-range impurity scattering [Eqs. (9), (22), and (23)]. Fitting parameters: (a) nr=1.9×1012 cm−2n_r = 1.9 \times10^{12}\ \mathrm{cm}^{-2} and n2=2.8×109 cm−2n_2 = 2.8 \times10^{9}\ \mathrm{cm}^{-2}; (b) nr=1.7×1012 cm−2n_r = 1.7 \times10^{12}\ \mathrm{cm}^{-2} and n2=1.8×109 cm−2n_2 = 1.8 \times10^{9}\ \mathrm{cm}^{-2}; (c) n2=2.1×109 cm−2n_2 = 2.1 \times10^{9}\ \mathrm{cm}^{-2} and N1=11×1015 cm−3N_1 = 11 \times10^{15}\ \mathrm{cm}^{-3}; (d) n2=1.4×109 cm−2n_2 = 1.4 \times10^{9}\ \mathrm{cm}^{-2} and N1=12×1015 cm−3N_1 = 12 \times10^{15}\ \mathrm{cm}^{-3}.

On the other hand, the other 2-impurity model (τ−1=τBI−1+τS−1\tau^{-1} = \tau_{\mathrm{BI}}^{-1} + \tau_{\mathrm{S}}^{-1}) fits the experimental mobility data in a narrower density range ne=3n_e = 3–6×1011 cm−26 \times10^{11}\ \mathrm{cm}^{-2}, with the predicted quantum mobility around 10×104 cm2/Vs10 \times10^{4}\ \mathrm{cm}^{2}/\mathrm{Vs} at the highest density ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2}, which is a factor of 3 larger than the experimental value [21]. Since the typical momentum transfer of RI scattering is much smaller than that of BI scattering, the ratio (μ/μq)RI≫(μ/μq)BI(\mu/\mu_q)_{\mathrm{RI}} \gg(\mu/\mu_q)_{\mathrm{BI}} [22], which explains a factor of 10 difference in the predicted quantum mobility shown in Figure 2. In Section VI, we will explore how including gate screening effects in RI scattering calculations can improve the agreement with experimental quantum mobility data. However, it is important to note that the experimental accuracy for quantum mobility data is generally less reliable than for transport mobility, largely due to the indirect method of extracting quantum mobility through the Dingle temperature [22]. Given these considerations, a semi-quantitative agreement between our theoretical predictions and experimental data should be regarded as satisfactory. At low densities ne<2×1011 cm−2n_e < 2 \times10^{11}\ \mathrm{cm}^{-2}, the theoretical mobility curves deviate significantly higher than the experimental data. We should emphasize that this deviation at low densities does not arise from any new scattering mechanism ignored in the theory. The dominant scattering mechanism at low densities ne<4×1011 cm−2n_e < 4 \times10^{11}\,\mathrm{cm}^{-2} is RI if we use the 2-impurity model Eq. (7) [or BI if we use the 2-impurity model Eq. (21)]. Rather, this deviation indicates the systematic failure of the Boltzmann-Born theory at low densities, a topic we will address in more detail in § 7, particularly in the context of the metal-insulator transition.

Infinite Potential Well

In this section, we use an infinite potential well approximation in order to address the quantum well confinement better, where the 2DEG wave function is given by ψ(z)=2/wcos⁡(πz/w)Θ(w/2−∣z∣)\psi(z) = \sqrt{2/w}\cos(\pi z/w)\Theta(w/2 - |z|) [45, 59]. The justification of the infinite potential well approximation will be discussed at the end of this section. Substituting this wavefunction into Eq. (4), we obtain the single-impurity Coulomb potential

U1(q,z)=2πe2κqϵ(q){F0(qw)e−q∣z∣,∣z∣>w/2,G0(q,z),∣z∣<w/2.(28)U_1(q,z) = \frac{2\pi e^2}{\kappa q \epsilon(q)} \begin{cases} F_0(qw)e^{-q|z|}, & |z| > w/2, \\ G_0(q,z), & |z| < w/2. \end{cases} \tag*{(28)}

where

F0(x)=8π2x(x2+4π2)sinh⁡(x/2),(29)F_0(x) = \frac{8\pi^2}{x(x^2+4\pi^2)}\sinh(x/2), \tag*{(29)}

and

G0(q,z)=8π2(1−e−qw/2cosh⁡qz)qw(4π2+q2w2)+2qw[1+cos⁡(2πz/w)]4π2+q2w2.(30)G_0(q,z) = \frac{8\pi^2(1-e^{-qw/2}\cosh qz)}{qw(4\pi^2+q^2w^2)} +\frac{2qw[1+\cos(2\pi z/w)]}{4\pi^2+q^2w^2}. \tag*{(30)}

Both F0(qw)F_0(qw) and G0(q,z)G_0(q,z) monotonically decrease as a function of qq and weaken the scattering potential at large momentum qq. The screening form factor Fc(qw)F_c(qw) in Eq. (6) reads

Fc(x)=20π2x3+3x5−32π4(1−e−x−x)x2(4π2+x2)2.(31)F_c(x) = \frac{20\pi^2x^3+3x^5-32\pi^4(1-e^{-x}-x)}{x^2(4\pi^2+x^2)^2}. \tag*{(31)}

This form factor enters the dielectric function ϵ(q)\epsilon(q) and weakens the screening at large momentum qq.

First, we calculate the charged impurity scattering rates separately and see how well they can interpret the data individually. The 3D impurity concentration is given by

N(z)=nrδ(z−d−w/2)+N1Θ(∣z∣−w/2)+N2Θ(w/2−∣z∣),(32)N(z) = n_r\delta(z-d-w/2)+N_1\Theta(|z|-w/2)+N_2\Theta(w/2-|z|), \tag*{(32)}

where nrn_r is the RI concentration in the capping layer, N1N_1 (N2N_2) is the BI concentration outside (inside) the quantum well. The theoretical results of mobility and quantum mobility are shown in Figure 3, based on the assumption that only one term in Eq. (32) is nonzero. Figure 3 (a) and (Figure 3b) show the results of nr≠0n_r \ne0 and N1=N2=0N_1=N_2=0. This RI-dominant mobility can fit the data at intermediate densities ne=2−3×1011 cm−2n_e=2-3\times10^{11}\,\mathrm{cm}^{-2} within the margin of experimental uncertainty. The predicted quantum mobility is around 1.0×104 cm2/Vs1.0\times10^4\,\mathrm{cm}^2/\mathrm{Vs} at the highest density ne=6×1011 cm−2n_e=6\times10^{11}\,\mathrm{cm}^{-2}.

Figure 3. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the infinite potential well approximation without the local field correction. (a) and (b) assume only RI near the capping layer nr≠0n_r \ne0. (c) and (d) assume only BI outside the quantum well N1≠0N_1 \ne0 and no BI inside N2=0N_2 = 0, while (e) and (f) assume N1=0N_1 = 0 and N2≠0N_2 \ne0. Impurity concentrations and roughness parameters for fitting: (a) nr=3.1×1012 cm−2n_r = 3.1 \times10^{12}\ \mathrm{cm}^{-2}; (b) nr=2.6×1012 cm−2n_r = 2.6 \times10^{12}\ \mathrm{cm}^{-2}; (c) N1=14×1015 cm−3N_1 = 14 \times10^{15}\ \mathrm{cm}^{-3} and N2=0N_2 = 0; (d) N1=12.6×1015 cm−3N_1 = 12.6 \times10^{15}\ \mathrm{cm}^{-3} and N2=0N_2 = 0; (e) N1=0N_1 = 0 and N2=10.3×1015 cm−3N_2 = 10.3 \times10^{15}\ \mathrm{cm}^{-3}; (f) N1=0N_1 = 0 and N2=5.5×1015 cm−3N_2 = 5.5 \times10^{15}\ \mathrm{cm}^{-3}.

Figure 3 (c) and (Figure 3d) show the results of N1≠0N_1\ne0 and N2=nr=0N_2=n_r=0, which fit the data at intermediate densities ne=2.5−5×1011 cm−2n_e=2.5-5\times10^{11}\,\mathrm{cm}^{-2}. The predicted quantum mobility is around 10×104 cm2/Vs10\times10^4\,\mathrm{cm}^2/\mathrm{Vs} at the highest density ne=6×1011 cm−2n_e=6\times10^{11}\,\mathrm{cm}^{-2}. Figure 3 (e) and (Figure 3f) show the results of N2≠0N_2\ne0 and N1=nr=0N_1=n_r=0, which fit the data at high densities ne=4−6×1011 cm−2n_e=4-6\times10^{11}\,\mathrm{cm}^{-2}. The corresponding ratio τ/τq∼1\tau/\tau_q\sim1 is expected, as the background impurities inside the QW should behave similarly to the short-range impurities because of strong screening by the electrons. In general, switching the dominant impurity distribution from RI (nr≠0n_r\ne0) to BI (N1≠0N_1\ne0) and finally to a different BI distribution (N2≠0N_2\ne0) results in a progressive flattening of the mobility μ(ne)\mu(n_e) and a decrease in the ratio μ/μq\mu/\mu_q towards unity.

In order to better fit the mobility data, we use interface roughness as an additional scattering mechanism that limits mobility at high density [27, 48, 51, 59, 60, 61, 62]. Using the correlator of local well width variations ⟨Δ(r)Δ(r′)⟩=Δ2exp⁡(−∣r−r′∣2/Λ2)\langle\Delta(r)\Delta(r')\rangle=\Delta^2\exp(-|r-r'|^2/\Lambda^2), where Δ\Delta is the typical roughness height and Λ\Lambda is the roughness lateral size, the scattering potential due to interface roughness is given by

⟨∣UIR(q)∣2⟩=πϵ2(q)Δ2Λ2(∂E∂w)2e−q2Λ2/4,(33)\left\langle|U_{\mathrm{IR}}(q)|^2\right\rangle =\frac{\pi}{\epsilon^2(q)}\Delta^2\Lambda^2 \left(\frac{\partial E}{\partial w}\right)^2e^{-q^2\Lambda^2/4}, \tag*{(33)}

where EE is the ground state energy of the potential well along zz direction. For an infinite potential well, the ground state energy E=E0=ℏ2π2/2mz∗w2E=E_0=\hbar^2\pi^2/2m_z^*w^2, and ∂E/∂w=2E0/w\partial E/\partial w=2E_0/w. Here mz∗=0.92m0m_z^*=0.92m_0 is the longitudinal effective mass of an electron in Si. Using Eqs. (33), (1), and (2), we obtain the transport relaxation rate τIR−1\tau_{\mathrm{IR}}^{-1} and quantum rate τqIR−1\tau_{q\mathrm{IR}}^{-1} due to interface roughness scattering. At low electron densities ne≪g/4πΛ2n_e\ll g/4\pi\Lambda^2 where kFΛ≪1k_F\Lambda\ll1 and qTF/2kF≫1q_{TF}/2k_F\gg1 are satisfied, we obtain

τIR−1=1τ03π8(2kFqTF)2,(34)\tau_{\mathrm{IR}}^{-1}=\frac{1}{\tau_0}\frac{3\pi}{8}\left(\frac{2k_F}{q_{TF}}\right)^2, \tag*{(34)}
τqIR−1=1τ0π4(2kFqTF)2,(35)\tau_{q\mathrm{IR}}^{-1}=\frac{1}{\tau_0}\frac{\pi}{4}\left(\frac{2k_F}{q_{TF}}\right)^2, \tag*{(35)}

where τ0−1\tau_0^{-1} is given by

τ0−1=2π4ℏm∗(m∗mz∗)2Δ2Λ2w6.(36)\tau_0^{-1}=\frac{2\pi^4\hbar}{m^*}\left(\frac{m^*}{m_z^*}\right)^2\frac{\Delta^2\Lambda^2}{w^6}. \tag*{(36)}

We find that both μIR\mu_{\mathrm{IR}} and μqIR\mu_{q\mathrm{IR}} are proportional to ne−1n_e^{-1} at low electron densities ne≪g/4πΛ2n_e\ll g/4\pi\Lambda^2. In the high-density range g/4πΛ2≪ne≪g3/16πaB2g/4\pi\Lambda^2\ll n_e\ll g^3/16\pi a_B^2, we obtain

τIR−1=1τ03π4(2kFqTF)2(kFΛ)−5,(37)\tau_{\mathrm{IR}}^{-1}=\frac{1}{\tau_0}\frac{3\sqrt{\pi}}{4}\left(\frac{2k_F}{q_{TF}}\right)^2(k_F\Lambda)^{-5}, \tag*{(37)}
τqIR−1=1τ0π4(2kFqTF)2(kFΛ)−3,(38)\tau_{q\mathrm{IR}}^{-1}=\frac{1}{\tau_0}\frac{\sqrt{\pi}}{4}\left(\frac{2k_F}{q_{TF}}\right)^2(k_F\Lambda)^{-3}, \tag*{(38)}

where we find μIR∝ne3/2\mu_{\mathrm{IR}}\propto n_e^{3/2} and μqIR∝ne1/2\mu_{q\mathrm{IR}}\propto n_e^{1/2} [63], similar to the long-range remote impurity scattering results [cf. Eqs. (15) and (16). The crossover of the μIR∝ne−1\mu_{\mathrm{IR}} \propto n_e^{-1} to the μIR∝ne3/2\mu_{\mathrm{IR}} \propto n_e^{3/2} behavior is at ne=g/4πΛ2n_e = g/4\pi\Lambda^2, where μIR\mu_{\mathrm{IR}} reaches its minimum

μIR,min≈ε0m∗qTF2Λ24=e8πℏ(mz∗m∗)2qTF2w6Δ2.(39)\mu_{\mathrm{IR,min}} \approx\frac{\varepsilon_0}{m^*}\frac{q_{\mathrm{TF}}^2\Lambda^2}{4} = \frac{e}{8\pi\hbar}\left(\frac{m_z^*}{m^*}\right)^2\frac{q_{\mathrm{TF}}^2 w^6}{\Delta^2}. \tag*{(39)}

For the high mobility structures reported in Refs. [14] with w=7w = 7 nm, we equate the saturation mobility value ≈30×104 cm2/Vs\approx30 \times10^4\ \mathrm{cm}^2/\mathrm{Vs} to μIR,min\mu_{\mathrm{IR,min}}, and obtain a rough estimate of Δ≈5\Delta\approx5 Å. This value is in reasonable agreement with the roughness height measured from scanning transmission electron microscopy (STEM) images reported in Refs. [13, 14].

Figure 4 (a) and (Figure 4b) are calculated using the combination of RI and IR scattering, while (c) and (d) are calculated using the combination of BI and IR scattering. When we fit the experimental data, we restrict the RI (BI) concentration to be the same for both the 5 nm and 7 nm quantum wells. This restriction is based on the fact that the experimental uncertainty of mobility is relatively small ∼10%\sim10\% for different devices with the same structure design [13, 14], indicating that the sources of disorder should be similar from device to device. As a result, the best-fit charge impurity concentrations are nr=2.5×1012 cm−2n_r = 2.5 \times10^{12}\ \mathrm{cm}^{-2} and N1=12×1015 cm−3N_1 = 12 \times10^{15}\ \mathrm{cm}^{-3}. To explain the difference in the peak mobility between the 5 nm and 7 nm quantum wells, the roughness parameters are chosen differently within the range Δ=4.5\Delta= 4.5–8.08.0 Å and Λ=16\Lambda= 16–2828 Å. The roughness parameters that we choose are of the same order of magnitude as those used in Refs. [27, 47, 48, 62].

Figure 4. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the infinite potential well approximation without the local field correction. Subfigures (a) and (b) incorporate remote impurity and interface roughness scattering, whereas (c) and (d) consider background impurity and interface roughness scattering. Impurity concentrations and roughness parameters for fitting: (a) nr=2.5×1012 cm−2n_r = 2.5 \times10^{12}\ \mathrm{cm}^{-2}, Δ=4.5\Delta= 4.5 Å, and Λ=27\Lambda= 27 Å; (b) nr=2.5×1012 cm−2n_r = 2.5 \times10^{12}\ \mathrm{cm}^{-2}, Δ=8\Delta= 8 Å, and Λ=28\Lambda= 28 Å; (c) N1=12×1015 cm−3N_1 = 12 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=5\Delta= 5 Å, and Λ=16\Lambda= 16 Å; (d) N1=12×1015 cm−3N_1 = 12 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=7\Delta= 7 Å, and Λ=20\Lambda= 20 Å.

From Figure 4 we see that the 2-impurity model combining RI and IR scattering fits the mobility data in a wider density range ne=1.8n_e = 1.8–6×1012 cm−26 \times10^{12}\ \mathrm{cm}^{-2}, while the other 2-impurity model combining BI and IR fits the data in a narrower density range ne=3n_e = 3–6×1012 cm−26 \times10^{12}\ \mathrm{cm}^{-2}. Therefore, RI scattering is more likely the dominant scattering mechanism for transport mobility at low densities, whereas at high densities the transport mobility is limited by IR scattering. This behavior is similar to that in Si-SiO2_2 MOSFET 2DEG systems, where the IR scattering is much larger leading consequently to lower maximum mobilities.

In the above calculations, we ignore the local field correction G(q)=q/(gq2+kF2)G(q) = q/(g\sqrt{q^2+k_F^2}) in the dielectric function Eq. (5). The results that include the local field correction are presented in Figure 5 and Figure 6. Incorporating this correction leads to a reduction in the estimated impurity concentrations and the ratio of τ/τq\tau/\tau_q. This can be understood because the local field correction weakens the screening at a large momentum q≳kFq \gtrsim k_F, so the disorder is more effective in scattering the electrons. Given that τ\tau is more sensitive to large-momentum scattering compared to τq\tau_q, the local field correction consequently reduces τ\tau to a greater extent than τq\tau_q for the same disorder configuration, leading to a smaller τ/τq\tau/\tau_q. To analytically incorporate the local field correction G(q)G(q) into the scattering rate calculations, we evaluate the qq integral in Eqs. (1) and (2) in an approximation where the square-root singularity at q=2kFq = 2k_F is integrated, but the form factors and G(q)G(q) are taken at q=2kFq = 2k_F. As a result, the expressions of charged impurity scattering rates Eqs. (15), (16), (20), (25), (26), and (27) should be multiplied by an additional factor F0(2kFw)[(2kF/qTF)ϵ(2kF)]−2F_0(2k_F w)[(2k_F/q_{TF})\epsilon(2k_F)]^{-2}, where the local field correction G(q)G(q) enters the dielectric function ϵ(q)\epsilon(q) defined in Eq. (5). At low densities kFΛ≪1k_F\Lambda\ll1, the expressions of IR scattering rates Eqs. (34) and (35) should be multiplied by [(2kF/qTF)ϵ(2kF)]−2[(2k_F/q_{TF})\epsilon(2k_F)]^{-2}. While at high densities kFΛ≫1k_F\Lambda\gg1, the typical momentum transfer is Λ−1\Lambda^{-1} so Eqs. (37) and (38) should be multiplied by [(2/qTFΛ)ϵ(2/Λ)]−2[(2/q_{TF}\Lambda)\epsilon(2/\Lambda)]^{-2}.

Figure 5. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the infinite potential well approximation with the local field correction. (a) and (b) assume only RI near the capping layer nr≠0n_r \ne0. (c) and (d) assume only BI outside the quantum well N1≠0N_1 \ne0 and no BI inside N2=0N_2 = 0, while (e) and (f) assume N1=0N_1 = 0 and N2≠0N_2 \ne0. To fit the data, the impurity concentrations and roughness parameters are chosen as (a) nr=2.5×1012 cm−2n_r = 2.5 \times10^{12}\ \mathrm{cm}^{-2}; (b) nr=2.1×1012 cm−2n_r = 2.1 \times10^{12}\ \mathrm{cm}^{-2}; (c) N1=9.7×1015 cm−3N_1 = 9.7 \times10^{15}\ \mathrm{cm}^{-3} and N2=0N_2 = 0; (d) N1=8.7×1015 cm−3N_1 = 8.7 \times10^{15}\ \mathrm{cm}^{-3} and N2=0N_2 = 0; (e) N1=0N_1 = 0 and N2=7.2×1015 cm−3N_2 = 7.2 \times10^{15}\ \mathrm{cm}^{-3}; (f) N1=0N_1 = 0 and N2=3.8×1015 cm−3N_2 = 3.8 \times10^{15}\ \mathrm{cm}^{-3}.

Figure 6. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the infinite potential well approximation with the local field correction. Subfigures (a) and (b) incorporate remote impurity and interface roughness scattering, whereas (c) and (d) consider background impurity and interface roughness scattering. Fitting parameters: (a) nr=2.0×1012 cm−2n_r = 2.0 \times10^{12}\ \mathrm{cm}^{-2}, Δ=3.5 A˚\Delta= 3.5\ \text{\AA}, and Λ=30 A˚\Lambda= 30\ \text{\AA}; (b) nr=2.0×1012 cm−2n_r = 2.0 \times10^{12}\ \mathrm{cm}^{-2}, Δ=6.5 A˚\Delta= 6.5\ \text{\AA}, and Λ=30 A˚\Lambda= 30\ \text{\AA}; (c) N1=8.1×1015 cm−3N_1 = 8.1 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=3 A˚\Delta= 3\ \text{\AA}, and Λ=22 A˚\Lambda= 22\ \text{\AA}; (d) N1=8.1×1015 cm−3N_1 = 8.1 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=6 A˚\Delta= 6\ \text{\AA}, and Λ=22 A˚\Lambda= 22\ \text{\AA}.

Now we address the justification of the infinite potential well approximation employed in our calculations. This approximation is justified under two conditions. The first condition is that the subband gap of the infinite potential well, Esg=3ℏ2π2/2m⋆w2E_{sg} = 3\hbar^2\pi^2/2m^{\star}w^2, should satisfy the inequality EF≪Esg≪V0E_F \ll E_{sg} \ll V_0, where EFE_F is the Fermi energy of the 2DEG and V0V_0 is the conduction band offset of the quantum well. In the case of the 7 nm quantum well reported in Ref. [14], Esg≈25E_{sg} \approx25 meV, EF<4E_F < 4 meV, and V0≈180V_0 \approx180 meV [3, 35], so the first condition EF≪Esg≪V0E_F \ll E_{sg} \ll V_0 is met. The second condition is that the asymmetric electric field from the top gate should not significantly change the effective well width. If the electric field is strong enough to tilt the bottom of the conduction band toward a triangular potential well, then the effective well width becomes smaller. Using the Fang-Howard wavefunction [64], we can roughly estimate the effective well width at the highest density where the effect of the electric field Egate=4πene/κE_{\mathrm{gate}} = 4\pi e n_e/\kappa is the strongest.

Howard wavefunction [64], we can roughly estimate the effective well width at the highest density where the effect of the electric field Egate=4πene/κE_{\mathrm{gate}} = 4\pi e n_e/\kappa is the strongest. At the highest density ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2}, the estimated smallest effective well width is wFH=6/b≈5.5w_{FH} = 6/b \approx5.5 nm, where b=(48πmz⋆e2ne/κℏ2)1/3b = (48\pi m_z^{\star}e^2n_e/\kappa\hbar^2)^{1/3}. Given that the estimated smallest effective well width from the Fang-Howard wavefunction, wFH≈5.5w_{FH} \approx5.5 nm, does not significantly differ from the original well width w=5−7w = 5 - 7 nm, it suggests that the bending of the conduction band bottom due to the electric field from the top gate should not substantially alter the results. This reasoning justifies the use of the infinite potential well approximation in our calculations. One shortcoming of both the strict 2D and the infinite square well potential is that they cannot incorporate the short-range alloy disorder scattering arising from the SiGe alloy on the barrier outside the well. We address this shortcoming in the next section by using a finite well confinement model where the carrier wavefunction is allowed to tunnel into the barrier regime.

Finite Potential Well

In this section, to calculate the short-range alloy disorder scattering rate, we employ the finite potential well approximation to determine the wavefunction tail extending outside the quantum well. Furthermore, we recalculate the RI, BI and IR scattering rates using the finite potential well wavefunction, which enters the expressions of the form factors Eqs. (5), (6) and the lowest subband energy in Eq. (33). The comprehensive results combining the scattering mechanisms of RI (BI), IR and AD are shown in Figure 7 and Figure 8. The finite potential well Schrödinger equation reads

Figure 7. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the finite potential well approximation without the local field correction. Subfigures (a) and (b) incorporate RI, IR, and AD scattering, whereas (c) and (d) consider BI, IR, and AD scattering. Fitting parameters: (a) nr=2.4×1012 cm−2n_r = 2.4 \times10^{12}\ \mathrm{cm}^{-2}, Δ=4.5 A˚\Delta= 4.5\ \mathring{\mathrm{A}}, and Λ˚=50 A˚\mathring{\Lambda} = 50\ \mathring{\mathrm{A}}; (b) nr=2.4×1012 cm−2n_r = 2.4 \times10^{12}\ \mathrm{cm}^{-2}, Δ=9 A˚\Delta= 9\ \mathring{\mathrm{A}}, and Λ˚=37 A˚\mathring{\Lambda} = 37\ \mathring{\mathrm{A}}; (c) N1=11×1015 cm−3N_1 = 11 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=5 A˚\Delta= 5\ \mathring{\mathrm{A}}, and Λ˚=26 A˚\mathring{\Lambda} = 26\ \mathring{\mathrm{A}}; (d) N1=11×1015 cm−3N_1 = 11 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=7 A˚\Delta= 7\ \mathring{\mathrm{A}}, and Λ˚=30 A˚\mathring{\Lambda} = 30\ \mathring{\mathrm{A}}.

Mobility versus electron density for quantum wells with widths of 5 nm and 7 nm

Figure 8. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the finite potential well approximation with the local field correction. Subfigures (a) and (b) incorporate RI, IR, and AD scattering, whereas (c) and (d) consider BI, IR, and AD scattering. Fitting parameters: (a) nr=1.9×1012 cm−2n_r = 1.9 \times10^{12}\ \mathrm{cm}^{-2}, Δ=4.5 A˚\Delta= 4.5\ \text{\AA}, and Λ=22 A˚\Lambda= 22\ \text{\AA}; (b) nr=1.9×1012 cm−2n_r = 1.9 \times10^{12}\ \mathrm{cm}^{-2}, Δ=8 A˚\Delta= 8\ \text{\AA}, and Λ=35 A˚\Lambda= 35\ \text{\AA}; (c) N1=7.5×1015 cm−3N_1 = 7.5 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=5 A˚\Delta= 5\ \text{\AA}, and Λ=20 A˚\Lambda= 20\ \text{\AA}; (d) N1=7.5×1015 cm−3N_1 = 7.5 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=7.5 A˚\Delta= 7.5\ \text{\AA}, and Λ=24 A˚\Lambda= 24\ \text{\AA}.

−ℏ22m(z)d2ψ(z)dz2+V(z)ψ(z)=Eψ(z),(40)-\frac{\hbar^2}{2m(z)}\frac{d^2\psi(z)}{dz^2}+V(z)\psi(z)=E\psi(z), \tag*{(40)}

where m(z)=mz⋆=0.92m0m(z)=m_z^\star=0.92m_0 is the longitudinal effective mass for silicon at ∣z∣<w/2|z|<w/2 and m(z)=mB≃mz⋆m(z)=m_B\simeq m_z^\star at ∣z∣>w/2|z|>w/2 inside the Si1−xGex\mathrm{Si}_{1-x}\mathrm{Ge}_x barrier, since at x<0.85x<0.85, Si1−xGex\mathrm{Si}_{1-x}\mathrm{Ge}_x alloys are considered as “Si-like” material with almost the same effective masses [65, 66]. Note that Eq. (40) is an approximation for the situation with a variable effective mass, which applies very well in our system since m(z)m(z) varies little with zz. The effective finite potential well is described by

V(z)={0,∣z∣<w/2,V0−EF,∣z∣>w/2,(41)V(z)=\begin{cases} 0, & |z|<w/2,\\ V_0-E_F, & |z|>w/2, \end{cases} \tag*{(41)}

where V0=180 meVV_0=180\ \mathrm{meV} [3, 35] and EF=ℏ2kF22m⋆E_F=\frac{\hbar^2 k_F^2}{2m^\star} is the Fermi energy of the 2DEG. In the following we use VV to represent V0−EFV_0-E_F for simplicity. The bound state solutions (E<VE<V) for the lowest subband are

ψ(z)={Ce−η∣z∣,∣z∣>w/2,Dcos⁡(kz),∣z∣<w/2,(42)\psi(z)=\begin{cases} Ce^{-\eta|z|}, & |z|>w/2,\\ D\cos(kz), & |z|<w/2, \end{cases} \tag*{(42)}

where k=2mz⋆E/ℏ2k=\sqrt{2m_z^\star E/\hbar^2} and η=2mB(V−E)/ℏ2\eta=\sqrt{2m_B(V-E)/\hbar^2}. The boundary condition reads

Ce−ηw/2=Dcos⁡(kw/2),(43)Ce^{-\eta w/2}=D\cos(kw/2), \tag*{(43)}
CmBηe−ηw/2=Dmz⋆ksin⁡(kw/2).(44)\frac{C}{m_B}\eta e^{-\eta w/2}=\frac{D}{m_z^\star}k\sin(kw/2). \tag*{(44)}

The normalization of ψ(z)\psi(z) gives

C2∫w/2∞e−2ηz dz+D2∫0w/2cos⁡2(kz) dz=12.(45)C^2\int_{w/2}^{\infty}e^{-2\eta z}\,dz+D^2\int_0^{w/2}\cos^2(kz)\,dz=\frac{1}{2}. \tag*{(45)}

Combining with Eq. (43), one has

C=[1ηe−ηw+kw+sin⁡(kw)2kcos⁡2(kw/2)e−ηw]−1/2.(46)C=\left[\frac{1}{\eta}e^{-\eta w}+\frac{kw+\sin(kw)}{2k\cos^2(kw/2)}e^{-\eta w}\right]^{-1/2}. \tag*{(46)}

Dividing Eq. (44) by Eq. (43), one obtains

tan⁡(kw/2)=mz⋆mBηk.(47)\tan(kw/2)=\frac{m_z^\star}{m_B}\frac{\eta}{k}. \tag*{(47)}

Introducing dimensionless quantities E~=E/E0\tilde{E} = E/E_0, V~=V/E0\tilde{V} = V/E_0, where E0=ℏ2π2/2mz⋆w2E_0 = \hbar^2\pi^2/2m_z^\star w^2 is the ground state energy for infinite well, we arrive at [67]

E~+mBmz⋆E~tan⁡2(E~π/2)=V~.(48)\tilde{E} + \frac{m_B}{m_z^\star}\tilde{E}\tan^2\left(\sqrt{\tilde{E}}\pi/2\right) = \tilde{V}. \tag*{(48)}

The local energy fluctuation due to interface roughness reads

δE(r)=∂E∂wΔ(r),(49)\delta E(\mathbf{r}) = \frac{\partial E}{\partial w}\Delta(\mathbf{r}), \tag*{(49)}

where Δ(r)\Delta(\mathbf{r}) is the local variation of the well width at a position r\mathbf{r}. Taking the derivative with respect to ww for both sides in Eq. (48), we obtain the expression for ∂E/∂w\partial E/\partial w

∂E∂w=−2Ew{1+g(mBmz⋆,EV)V~−1/2}−1,(50)\frac{\partial E}{\partial w} = -\frac{2E}{w}\left\{1+g\left(\frac{m_B}{m_z^\star},\frac{E}{V}\right)\tilde{V}^{-1/2}\right\}^{-1}, \tag*{(50)}

where

g(x,y)=2π[x−1/2+y(x1/2−x−1/2)]−1(1−y)−1/2.(51)g(x,y)=\frac{2}{\pi}\left[x^{-1/2}+y\left(x^{1/2}-x^{-1/2}\right)\right]^{-1}(1-y)^{-1/2}. \tag*{(51)}

Substituting Eq. (50) into Eq. (33), we obtain the scattering potential for the interface roughness for a finite potential well.

We now estimate the correction introduced by the finite potential well model relative to the infinite potential well model. According to Eq. (48), we see that if V~→∞\tilde{V} \to\infty and mB=mz⋆m_B = m_z^\star, then E~=1\tilde{E} = 1. For finite values of V~\tilde{V}, we have 0<E~<10 < \tilde{E} < 1, implying that the effective well width is larger than ww due to wavefunction penetration into the barriers. For example, at electron density ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2}, we have V~≈10\tilde{V} \approx10 for a 5 nm QW and V~≈20\tilde{V} \approx20 for a 7 nm QW. In the limit V~≫1\tilde{V} \gg1 one can solve Eq. (48) perturbatively and obtain

E~≈1−4πV~1/2+12π2V~+O(V~−3/2),(52)\tilde{E} \approx1-\frac{4}{\pi\tilde{V}^{1/2}}+\frac{12}{\pi^2\tilde{V}}+O\left(\tilde{V}^{-3/2}\right), \tag*{(52)}

so that E~≈0.70\tilde{E} \approx0.70 for a 5 nm QW and E~≈0.77\tilde{E} \approx0.77 for a 7 nm QW. The reduced E~\tilde{E} enters directly into the interface roughness scattering potential Eq. (33) and Eq. (50) and lowers the scattering rate by a factor of E~2\tilde{E}^2. The parentheses in Eq. (50) lead to a further reduction in the scattering rate. As a result, τIR,inf/τIR,fin≈0.34\tau_{\mathrm{IR,inf}}/\tau_{\mathrm{IR,fin}} \approx0.34 for a 5 nm QW and τIR,inf/τIR,fin≈0.45\tau_{\mathrm{IR,inf}}/\tau_{\mathrm{IR,fin}} \approx0.45 for a 7 nm QW. To account for this reduction in the scattering rate due to the finite well potential, the product of ΔΛ\Delta\Lambda must be increased to fit the mobility data, as evident from Eq. (33). Above, we see that the finite potential well approximation significantly alters the interface roughness scattering. The leading order correction to τIR\tau_{\mathrm{IR}} is proportional to V~−1/2\tilde{V}^{-1/2}, which is substantial given that V~\widetilde{V} is around 10 for a 5 nm QW and 20 for a 7 nm QW. Next, we consider how the finite potential well model alters the background impurity scattering rates. The model alters the form factors as per Eqs. (4) and (5). Since the form factors are proportional to the electron density ∣ψ(z)∣2|\psi(z)|^2, the leading correction should be comparable to the leakage probability outside the quantum well

Prob⁡∣z∣>w/2=∫∣z∣>w/2∣ψ(z)∣2dz=C2η−1e−ηw.(53)\operatorname{Prob}_{|z|>w/2} = \int_{|z|>w/2} |\psi(z)|^2 dz = C^2 \eta^{-1} e^{-\eta w}. \tag*{(53)}

In the limit of V~≫1\widetilde{V} \gg1, we have

Prob⁡∣z∣>w/2≈2πV~3/2.(54)\operatorname{Prob}_{|z|>w/2} \approx\frac{2}{\pi\widetilde{V}^{3/2}}. \tag*{(54)}

Hence, Prob⁡∣z∣>w/2≈0.02\operatorname{Prob}_{|z|>w/2} \approx0.02 for a 5 nm QW and Prob⁡∣z∣>w/2≈0.007\operatorname{Prob}_{|z|>w/2} \approx0.007 for a 7 nm QW. This indicates that the finite potential well approximation only marginally modifies the background impurity scattering rates, affecting them by at most a few percent.

Alloy disorder within Si1−xGex\mathrm{Si}_{1-x}\mathrm{Ge}_x barriers is another short-range scattering mechanism that contributes equally to the transport and quantum scattering rate [68, 69, 70].

Image not extracted

Figure 9.

where ΔEc≈0.8\Delta E_c \approx0.8 eV is the conduction band offset between Si and Ge [71], Ω=a3/4\Omega= a^3/4 is the scatter volume with a=5.4a = 5.4 Å is the lattice constant. In the limit of V~≫1\widetilde{V} \gg1, we have ∫∣z∣>w/2∣ψ(z)∣4dz≈2V~−5/2/πw\int_{|z|>w/2} |\psi(z)|^4 dz \approx2\widetilde{V}^{-5/2}/\pi w, and the corresponding mobility is given by

μAD≈w6e(2mz∗V)5/22π4ℏ2m∗2(ΔEc)2Ωx(1−x).(55)\mu_{\mathrm{AD}} \approx\frac{w^6 e(2m_z^*V)^{5/2}}{2\pi^4\hbar^2m^{*2}(\Delta E_c)^2\Omega x(1-x)}. \tag*{(55)}

For a w=5w = 5 nm quantum well, μAD≈1.4×106 cm2/Vs\mu_{\mathrm{AD}} \approx1.4 \times10^6\ \mathrm{cm^2/Vs}, while for a w=7w = 7 nm quantum well, μAD≈1×107 cm2/Vs\mu_{\mathrm{AD}} \approx1 \times10^7\ \mathrm{cm^2/Vs}, which are much larger than the peak mobility reported in Refs. [13, 14]. The complete numerical results for alloy disorder scattering using the full finite potential well approximation are shown in Fig. 9, and we see that the order of magnitude agrees with the rough analytical estimation of Eq. (55). Therefore, AD scattering is negligible in the high-mobility quantum wells reported in Refs. [13, 14], where there is no Ge present inside the well. (For other devices that incorporate Ge concentration within the quantum well [72], AD scattering is much stronger, resulting in an electron mobility of 2−3×104 cm2/Vs.)2 - 3 \times10^{4}\ \mathrm{cm}^{2}/\mathrm{Vs}.) Regarding Eq. (55), we follow the approach of Refs. [68, 73], omitting the dielectric function due to the short-range nature of AD around a lattice constant. Though several other studies have included the dielectric function in AD scattering calculations [48, 51, 74], the inclusion of dielectric screening would only increase mobility, further diminishing the role of AD. Therefore, our conclusion that AD scattering is not a limiting factor in the high-mobility structures reported in Refs. [13, 14] remains valid.

Mobility and quantum mobility due to alloy disorder versus electron density

Figure 9 (PDF p. 12). Mobility and quantum mobility due to alloy disorder μAD\mu_{\mathrm{AD}} versus electron density nen_e for a w=5w = 5 nm (red) and w=7w = 7 nm (blue) quantum well, calculated using Eq. (55) with the finite potential well approximation.

The results of mobility and quantum mobility calculated using the finite potential well approximation without the local field correction are shown in Figure 7. Here, Figure 7 (a) and (Figure 7b) are calculated using a combination of RI, IR, and AD scattering. (c) and (d) are calculated using a combination of BI, IR, and AD scattering. The corresponding results calculated with local field correction are shown in Figure 8.

Gate Screening

In our calculations above, the screening effect of the top gate is considered primarily through the cutoff distance dsd_s of the background impurity distribution outside the quantum well. In § 3, we argued that ds=dg/2d_s = d_g/2 should be half the distance from the gate to the center of the quantum well because the charged impurities close to the gate are effectively screened by the image charges. In particular, we assume that the gate does not screen the remote impurities. In this section, we refine our approach by incorporating the Coulomb potential of the image charges induced by the gate and compare the mobility results of the RI and BI scattering. The best-fit mobility and the corresponding quantum mobility curves are depicted in Figure 10, where the fitting concentrations of charge impurities are nr=5.2×1012 cm−2n_r = 5.2 \times10^{12}\ \mathrm{cm}^{-2} and N1=7.5×1015 cm−3N_1 = 7.5 \times10^{15}\ \mathrm{cm}^{-3}. To explain the difference in the peak mobility between the 5 nm and 7 nm quantum wells, the roughness parameters are chosen differently within the range Δ=4.5−8.0\Delta= 4.5 - 8.0 Å and Λ=20−39\Lambda= 20 - 39 Å. Our analysis confirms that ds=dg/2d_s = d_g/2 serves as a reasonably accurate approximation for the cutoff distance of the background impurity distribution within the upper SiGe barrier. Furthermore, we observe that the gate-screened RI scattering results align more closely with the quantum mobility data compared to the RI results that disregard gate screening. Previous results shown in Figure 8 indicate that at a high density of ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2}, the quantum mobility for RI without gate screening is 1.0×104 cm2/Vs1.0 \times10^4\ \mathrm{cm}^2/\mathrm{Vs}. This value is smaller than the experimental data point (3.0±0.5)×104 cm2/Vs(3.0 \pm0.5) \times10^4\ \mathrm{cm}^2/\mathrm{Vs} [21]. On the other hand, as illustrated in Figure 10, when gate screening is included, the quantum mobility for RI increases to 3.0×104 cm2/Vs3.0 \times10^4\ \mathrm{cm}^2/\mathrm{Vs} at ne=6×1011 cm−2n_e = 6 \times10^{11}\ \mathrm{cm}^{-2}, aligning well with the experimental data. This enhancement in quantum mobility due to the inclusion of gate screening can be attributed to the suppression of small momentum forward scattering by image charges, consequently reducing the ratio τ/τq\tau/\tau_q. Therefore, for a given best-fit mobility curve, the corresponding quantum mobility is expected to be higher.

Mobility versus electron density for quantum wells with widths of 5 nm and 7 nm

Figure 10. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the finite potential well approximation with gate screening and the local field correction. Subfigures (a) and (b) incorporate RI, IR, and AD scattering, whereas (c) and (d) consider BI, IR, and AD scattering. Fitting parameters: (a) nr=5.2×1012 cm−2n_r = 5.2 \times10^{12}\ \mathrm{cm}^{-2}, Δ=4.5\Delta= 4.5 Å, and Λ=39\Lambda= 39 Å; (b) nr=5.2×1012 cm−2n_r = 5.2 \times10^{12}\ \mathrm{cm}^{-2}, Δ=8\Delta= 8 Å, and Λ=36\Lambda= 36 Å; (c) N1=7.5×1015 cm−3N_1 = 7.5 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=5\Delta= 5 Å, and Λ=20\Lambda= 20 Å; (d) N1=7.5×1015 cm−3N_1 = 7.5 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=7.5\Delta= 7.5 Å, and Λ=24\Lambda= 24 Å.

In the scattering rate calculations, the gate screening is taken care by including the image charge potential into the remote impurity potential Eq. (28) at ∣z∣>w/2|z| > w/2:

U1(q,z)=2πe2F0(qw)κqϵ(q)(e−q∣z∣−e−q∣2dg−z∣),(56)U_1(q,z) = \frac{2\pi e^2 F_0(qw)}{\kappa q \epsilon(q)}\left(e^{-q|z|} - e^{-q|2d_g-z|}\right), \tag*{(56)}

where 2dg=2d+2do+w2d_g = 2d + 2d_o + w. As a result, the RI scattering rates Eqs. (15) and (16) are replaced by [75]

1τRI=πℏnr8m∗kF3d3(1+d3(2dg−d)3−2d3dg3)(2g)2,(57)\frac{1}{\tau_{\mathrm{RI}}} = \frac{\pi\hbar n_r}{8m^* k_F^3 d^3}\left(1 + \frac{d^3}{(2d_g-d)^3} - \frac{2d^3}{d_g^3}\right)\left(\frac{2}{g}\right)^2, \tag*{(57)}
1τqRI=πℏnr2m∗kFd(1+d2dg−d−2ddg)(2g)2.(58)\frac{1}{\tau_{q\mathrm{RI}}} = \frac{\pi\hbar n_r}{2m^* k_F d}\left(1 + \frac{d}{2d_g-d} - \frac{2d}{d_g}\right)\left(\frac{2}{g}\right)^2. \tag*{(58)}

The ratio of the scattering rates is given by

(ττq)RI=(2kFd)2Ig(d,dg),(59)\left(\frac{\tau}{\tau_q}\right)_{\mathrm{RI}} = (2k_Fd)^2 I_g(d,d_g), \tag*{(59)}

where the dimensionless function Ig(d,dg)I_g(d,d_g) reads

Ig(d,dg)=1+d2dg−d−2ddg1+d3(2dg−d)3−2d3dg3.(60)I_g(d,d_g) = \frac{1+\frac{d}{2d_g-d}-\frac{2d}{d_g}}{1+\frac{d^3}{(2d_g-d)^3}-\frac{2d^3}{d_g^3}}. \tag*{(60)}

A comparison between Eqs. (17) and (59) reveals that the inclusion of gate screening modifies the ratio by a factor of Ig(d,ds)I_g(d,d_s), which is independent of kFk_F. For the high mobility structures reported in Refs. [13, 14], dg≈44d_g \approx44 nm and d≈30d \approx30 nm, which gives Ig(d,dg)≈0.3I_g(d,d_g) \approx0.3. Thus, incorporating gate screening leads to a decrease in the ratio (τ/τq)RI(\tau/\tau_q)_{\mathrm{RI}}, and consequently, the corresponding quantum mobility should increase by a factor of 3. This explains the difference in the quantum mobility values between Figure 8 and Figure 10 (a) (b).

On the other hand, we see that the results in Figure 8 and Figure 10 (c) (d) are very similar to each other, suggesting that our choice of ds=dg/2d_s=d_g/2 describes the gate-screened BI scattering within the upper SiGe barrier reasonably well. We comment on the cutoff distance in the lower SiGe barrier in the context of the high-mobility structures reported in Refs. [13, 14]. When we discuss the gate screening effects for BI in § 3, we assume that the cutoff distances are the same from both sides of the SiGe barriers, although in experiments the heterostructure is not symmetric. The bottom barrier (a few μ\mum) is significantly thicker than the upper spacer (30 nm). Given that the top gate can effectively screen charge impurities only within the upper SiGe barrier, the cutoff distance for the lower SiGe barrier remains an open question. However, as indicated by Eq. (26), at high densities, the cutoff distance dsd_s only logarithmically changes the quantum mobility, without significantly affecting the transport mobility. Therefore, the choice of the cutoff distance should not change the fit to the experimental mobility data. By considering the cutoff distance as an additional fitting parameter, one can always adjust the model to match a specific quantum mobility data point. A more comprehensive future experimental investigation of high-mobility Si quantum wells would be instrumental in enhancing our understanding of the BI distribution within the SiGe barrieres.

In summary, the results of our model, which incorporates gate screening, suggest that RI scattering is likely the predominant scattering mechanism for transport mobility at low densities, whereas at high densities the transport mobility is restricted by IR scattering. The quantum mobility predicted for gate-screened RI scattering aligns closely with experimental data within the margin of error at the highest density examined. This observation suggests that RI scattering could be the limiting factor for quantum mobility across all densities. On the other hand, since the cutoff distance of the BI distribution inside the bottom SiGe barrier remains an open question, we cannot rule out the possibility that some distant background impurities also contribute to the quantum mobility. Future experiments focusing on the quantum mobility over a broader range of densities could yield more quantitative information on the concentration and cutoff distance of background impurities inside the bottom SiGe barrier.

Metal Insulator Transition

In this section, we discuss the mobility data in the low-density regime, and estimate the critical density of the metal-insulator transition (MIT) using the Anderson-Ioffe-Regel

(AIR) condition [31, 32]. All our results in the previous sections are calculated using the Boltzmann-Born transport theory, which is valid at high densities such that kFl≫1k_Fl\gg1, where l=vFτl=v_F\tau is the transport mean free path. One might expect that the Boltzmann-Born theory fails at a critical density ne=ncn_e=n_c where kFl=1k_Fl=1. This is the well-known AIR criterion of MIT. However, deviations in the metallic regime of our results, as shown in Figure 10, suggest that the theoretical mobility curves, fitting well at high densities, start overestimating mobility at low densities compared to the experimental data. It is crucial to emphasize that these low-density deviations are not due to any neglected scattering mechanism in the theory. The dominant scattering mechanism at low densities is RI (or BI, depending on the choice of the impurity model). For RI-dominant cases, as depicted in Figure 10 (a) and (b), the deviation is observed at ne≲1.8×1011 cm−2n_e\lesssim1.8\times10^{11}\ \mathrm{cm}^{-2}, where kFl≲30k_Fl\lesssim30 or kFlq≲6k_Fl_q\lesssim6 with lq=vFτql_q=v_F\tau_q [76]. On the other hand, for BI dominant scenarios, shown in Figure 10 (c) and (d), the deviation occurs at a higher density ne≲2.8×1011 cm−2n_e\lesssim2.8\times10^{11}\ \mathrm{cm}^{-2}, around kFl≲100k_Fl\lesssim100 or kFlq≲60k_Fl_q\lesssim60. The failure of the Boltzmann-Born theory at low densities is due to the following. Long-range Coulomb disorder leads to the fragmentation of the 2DEG into electron puddles, separated by disorder potential barriers, creating an inhomogeneous 2DEG landscape. Although kFl≫1k_Fl\gg1 is locally satisfied within each puddle, the uniform 2DEG assumption, fundamental to linear screening theory, is no longer valid. Instead, transport properties in the low-density region ne≥ncn_e \ge n_c near MIT should be described by percolation through charged puddles, following σ∝(ne−nc)p\sigma\propto(n_e-n_c)^p with p=1.31p=1.31 [23, 24, 25, 26, 27, 28, 29, 30], and cannot be captured by the Boltzmann-Born theory even including all high-order scattering processes. The failure of the Boltzmann-Born theory at low densities also indicates that the critical density estimated from the AIR criterion should be smaller than the actual critical density (percolation threshold) observed in experiments and should be used as a lower bound estimation of the percolation threshold.

Nevertheless, we apply the AIR criterion, characterized by kFl=1k_F l = 1 and kFlq=1k_F l_q = 1, to estimate the critical densities ncn_c and ncqn_{cq} [31, 32]. Utilizing the delta-layer 2DEG approximation, as discussed in § 3 and VI, we can analytically estimate the critical density of the metal-insulator transition (MIT). By substituting the gate-screened remote impurity scattering rates from Eqs. (57) and (58) into the AIR criterion, we derive:

nc(RI)=g1/5[1+d3(2dg−d)3−2dg3dg3]2/522/5π3/5(nrd3)2/5(61)n_c^{(\mathrm{RI})} = \frac{g^{1/5}\left[1+\frac{d^3}{(2d_g-d)^3}-\frac{2d_g^3}{d_g^3}\right]^{2/5}}{2^{2/5}\pi^{3/5}}\left(\frac{n_r}{d^3}\right)^{2/5} \tag*{(61)}
ncq(RI)=g−1/3[1+d2dg−d−2ddg]2/324/3π1/3(nrd)2/3(62)n_{cq}^{(\mathrm{RI})} = \frac{g^{-1/3}\left[1+\frac{d}{2d_g-d}-\frac{2d}{d_g}\right]^{2/3}}{2^{4/3}\pi^{1/3}}\left(\frac{n_r}{d}\right)^{2/3} \tag*{(62)}

Eqs. (61) and (62) are applicable if nc(RI)n_c^{(\mathrm{RI})} and ncq(RI)n_{cq}^{(\mathrm{RI})} exceed g/16πd2g/16\pi d^2. This condition is met in high-mobility quantum wells [13, 14] where g/16πd2≈1010 cm−2g/16\pi d^2 \approx10^{10}\ \mathrm{cm}^{-2}. For BI scattering, substituting the rates from Eqs. (25) and (26) into the AIR criterion yields:

nc(BI)=g1/52−8/5π−3/5(N1w2)2/5(63)n_c^{(\mathrm{BI})} = g^{1/5}2^{-8/5}\pi^{-3/5}\left(\frac{N_1}{w^2}\right)^{2/5} \tag*{(63)}
ncq(BI)=g−1/32−2/3π−1/3[ln⁡(2ds/w)]2/3N12/3(64)n_{cq}^{(\mathrm{BI})} = g^{-1/3}2^{-2/3}\pi^{-1/3}[\ln(2d_s/w)]^{2/3}N_1^{2/3} \tag*{(64)}

which are valid if nc(BI)n_c^{(\mathrm{BI})} and ncq(BI)n_{cq}^{(\mathrm{BI})} are greater than g/16πw2g/16\pi w^2. Otherwise, we use Eq. (27) and obtain

nc(BI)=ncq(BI)=πgN1(2ds−w)(65)n_c^{(\mathrm{BI})}=n_{cq}^{(\mathrm{BI})}=\frac{\pi}{g}N_1(2d_s-w) \tag*{(65)}

which is applicable if nc(BI)<g/16πw2n_c^{(\mathrm{BI})}<g/16\pi w^2. Given that the crossover density g/16πw2≈2.4×1011 cm−2g/16\pi w^2 \approx2.4\times10^{11}\ \mathrm{cm}^{-2} is significantly higher than the MIT critical density in high-mobility quantum wells [13, 14], we should use Eq. (65) for estimating ncn_c in BI-dominated scenarios. Using d=30 nmd=30\ \mathrm{nm}, ds=dg/2=22 nmd_s=d_g/2=22\ \mathrm{nm}, and the impurity concentration nr≈5.2×1012 cm−2n_r\approx5.2\times10^{12}\ \mathrm{cm}^{-2} and N1≈7.5×1015 cm−3N_1\approx7.5\times10^{15}\ \mathrm{cm}^{-3} from Figure 10, we obtain the critical densities nc(RI)≈5×1010 cm−2n_c^{(\mathrm{RI})}\approx5\times10^{10}\ \mathrm{cm}^{-2}, ncq(RI)≈7×1010 cm−2n_{cq}^{(\mathrm{RI})}\approx7\times10^{10}\ \mathrm{cm}^{-2}, nc(BI)=ncq(BI)≈2×1010 cm−2n_c^{(\mathrm{BI})}=n_{cq}^{(\mathrm{BI})}\approx2\times10^{10}\ \mathrm{cm}^{-2}.

Next, we estimate the critical density of MIT by solving the AIR conditions kFl=(2/g)(hneμ/e)=1k_F l=(2/g)(hn_e\mu/e)=1 and kFlq=(2/g)(hneμq/e)=1k_F l_q=(2/g)(hn_e\mu_q/e)=1 with μ\mu and μq\mu_q numerically calculated from the best fits shown in Figure 10. The complete numerical results for ncn_c and ncqn_{cq} are summarized in the first two rows of Table I. These values closely align with the analytical results from Eqs. (61), (62), and (65). Furthermore, Table I includes results for various quantum degeneracies g=4g=4, 22,

RI (5 nm)RI (7 nm)BI (5 nm)BI (7 nm)
ncn_c (g=4)0.520.490.150.14
ncqn_{cq} (g=4)0.650.620.120.11
ncn_c (g=2)0.500.480.280.26
ncqn_{cq} (g=2)0.830.790.280.25
ncn_c (g=1)0.490.470.420.38
ncqn_{cq} (g=1)1.11.00.480.43

Table I. Critical density ncn_c and ncqn_{cq} in units of 1011 cm−210^{11}\ \mathrm{cm}^{-2} for different quantum degeneracy g=4g=4, 2, and 1. The results are obtained from kFl=1k_F l=1 and kFlq=1k_F l_q=1 using the same impurity parameters from Figure 10 calculated in a finite potential well approximation with gate screening and local field correction. RI (BI) indicates the dominant scattering mechanism at low densities. 5 nm (7 nm) means the calculation for a 5 nm (7 nm) QW.

and 11, while keeping the same disorder parameters as used in Figure 10. The changes in ncn_c and ncqn_{cq} as functions of gg are semi-quantitatively consistent with Eqs. (61), (62), and (65). We should emphasize that our estimate of ncn_c and ncqn_{cq} using the AIR criterion shown in Table I should serve as a lower bound for the experimental critical density of MIT.

Considering that the nature of 2D MIT in high-mobility Si quantum wells likely resembles a density-inhomogeneity-driven percolation transition similar to Si MOSFETs [27], we performed a percolation fit of the experimental data to estimate the experimental critical density. Refs. [13, 14] reported the percolation density nc=7×1010 cm−2n_c=7\times10^{10}\ \mathrm{cm}^{-2} (nc=8×1010 cm−2n_c=8\times10^{10}\ \mathrm{cm}^{-2}) for their high-mobility 7 nm (5 nm) Si quantum wells. The percolation density can also be estimated approximately by considering inhomogeneous density fluctuations in the system [27, 29, 33, 34, 77, 78], leading to nc≈0.1nr/d≈7.5×1010 cm−2n_c\approx0.1\sqrt{n_r}/d\approx7.5\times10^{10}\ \mathrm{cm}^{-2}, where we used d=30 nmd=30\ \mathrm{nm} and nr≈5×1012 cm−2n_r\approx5\times10^{12}\ \mathrm{cm}^{-2}. This ncn_c is higher than the one obtained from the AIR criterion. Furthermore, we individually performed the percolation fit σ∝(ne−nc)p\sigma\propto(n_e-n_c)^p using the conductivity data σ=neμ\sigma=n_e\mu at low densities ne=1−2×1011 cm−2n_e=1-2\times10^{11}\ \mathrm{cm}^{-2} reported in Refs. [13, 14]. The results of the best-fit ncn_c and pp are presented in Table II. The percolation fit with fixed p=1.31p=1.31 yields nc≈0.85×1011 cm−2n_c\approx0.85\times10^{11}\ \mathrm{cm}^{-2}. Allowing pp as a fitting parameter results in p≈2.4p\approx2.4 and nc≈0.5×1011 cm−2n_c\approx0.5\times10^{11}\ \mathrm{cm}^{-2}. Our percolation fit results are in reasonable agreement with those reported in Refs. [13, 14]. Comparing Table I and Table II, we observe that the RI results provide a more stringent lower bound of the critical density, as nc(RI)n_c^{(\mathrm{RI})} and ncq(RI)n_{cq}^{(\mathrm{RI})} are closer to the percolation densities reported in Refs. [13, 14]. This suggests that RI scattering, rather than BI, is likely the limiting mechanism at low densities in the studied high-mobility Si quantum wells. Given that ncq(RI)>nc(RI)n_{cq}^{(\mathrm{RI})}>n_c^{(\mathrm{RI})}, we find that the AIR criterion, which includes the single-particle scattering time τq\tau_q [76], provides a more accurate estimate of the percolation density. In the future, a more systematic experimental study of the temperature and density dependence of the conductivity is needed to gain a more comprehensive understanding of the 2D MIT in high-quality Si quantum wells.

ncn_c (p=1.31p=1.31)ncn_cpp
5 nm(0.837, 0.855)(0.436, 0.514)(2.20, 2.41)
7 nm(0.869, 0.889)(0.447, 0.515)(2.34, 2.52)

Table II. Critical density ncn_c in units of 1011 cm−210^{11}\ \mathrm{cm}^{-2} obtained by the power law fit σ∝(ne−nc)p\sigma\propto(n_e-n_c)^p to the conductivity data σ=eneμ\sigma=e n_e\mu at low densities ne=1n_e=1–2×1011 cm−22\times10^{11}\ \mathrm{cm}^{-2} reported in Refs. [13, 14]. The parenthesis show the 95% confidence interval of the fitting parameters. The first column fixed p=1.31p=1.31 assuming the 2D percolation. The second and third columns are results treating pp as a free parameter.

Effects of spin/valley degeneracy

In earlier sections, our numerical calculations considered a total quantum degeneracy g=gsgv=4g=g_sg_v=4. This assumption is valid under two conditions: first, the Fermi energy must substantially surpass the valley splitting, and second, the external magnetic field should be weak enough to avoid inducing spin polarization. In this scenario, both the valley and spin degrees of freedom maintain a twofold degeneracy, indicated by gv=gs=2g_v=g_s=2. This section theoretically addresses how the total quantum degeneracy gg affects the electron transport properties in the high-mobility Si/SiGe structures for a given disorder configuration. For example, a decrease in degeneracy from g=4g=4 to g=2g=2 can be achieved by applying an in-plane magnetic field BB parallel to the 2DEG, which is assumed to only affect the spin degeneracy of the 2DEG without orbital effect if the thickness of the 2DEG is much smaller than the cyclotron radius w≪ℏkF/eBw \ll\hbar k_{\mathrm{F}}/eB [40, 79, 80, 81]. Such a reduced degeneracy is also possible without any spin polarization if the valley splitting happens to be larger than the Fermi energy so that gv=1g_v = 1 in the system.

In the strong screening limit qTF/2kF≫1q_{\mathrm{TF}}/2k_{\mathrm{F}} \gg1 (for the density range of interest to us ne≪g3/16πaB2n_e \ll g^3/16\pi a_B^2), mobility increases monotonically with increasing gg for a given impurity configuration. This monotonic behavior of mobility versus gg has been shown in a comprehensive analysis of valley-dependent 2D transport in (100), (110), and (111) Si inversion layers with the same bare Coulomb disorder [39, 40]. A similar phenomenon was observed in an ambipolar gate-controlled Si(111)-vacuum field effect transistor, where by tuning the external gate voltage the 2D electron and hole transport can be studied within the same device against the same background Coulomb disorder [42]. For the ambipolar transistor discussed in Ref. [42], the effective valley degeneracy is 6 for a 2D electron system, whereas it is 1 for a 2D hole system, resulting in the peak electron mobility being approximately 20 times greater than the peak hole mobility at low temperatures. This can be qualitatively understood from the expression of the Thomas-Fermi screening wave vector qTF=g/aBq_{\mathrm{TF}} = g/a_B, implying that higher gg leads to improved screening and consequently, higher mobility. However, the relationship between mobility and gg is generally nontrivial, as gg influences both qTF∝gq_{\mathrm{TF}} \propto g and kF∝g−1/2k_{\mathrm{F}} \propto g^{-1/2}. For remote impurity scattering, it is evident from Eqs. (15), (16), (57), and (58) that μRI∝g1/2\mu_{\mathrm{RI}} \propto g^{1/2} and μqRI∝g3/2\mu_{q\mathrm{RI}} \propto g^{3/2}. On the other hand, for short-range impurity scattering, μS=μqS∝g2\mu_{\mathrm{S}} = \mu_{q\mathrm{S}} \propto g^2, as indicated by Eq. (20). As discussed in § 3, the gg-dependence of background impurity scattering crosses over from short-range to remote impurity scattering behavior with increasing electron density, as detailed in Eqs. (25)-(27). The crossover density is at ne=g/16πw2n_e = g/16\pi w^2. For interface roughness scattering, the mobilities μIR\mu_{\mathrm{IR}}, μqIR\mu_{q\mathrm{IR}} are proportional to g3g^3 at low densities ne<g/4πΛ2n_e < g/4\pi\Lambda^2, transitioning to a long-range behavior where μIR∝g1/2\mu_{\mathrm{IR}} \propto g^{1/2} and μqIR∝g3/2\mu_{q\mathrm{IR}} \propto g^{3/2} at higher densities, as shown in Eqs. (34)-(38). However, IR scattering at low densities ne<g/4πΛ2n_e < g/4\pi\Lambda^2 is typically overshadowed by the dominant scattering mechanism at these densities, which is usually remote Coulomb disorder. At ne=g/4πΛ2n_e = g/4\pi\Lambda^2, the minimum mobility due to IR scattering μIR,min∝g2\mu_{\mathrm{IR,min}} \propto g^2, as per Eq. (39). Since the unscreened alloy disorder scattering rate, as given by Eq. (55), depends on kFk_{\mathrm{F}} only through V=V0−EFV = V_0 - E_{\mathrm{F}}, the mobility μAD\mu_{\mathrm{AD}} is not influenced by gg provided V0≫EFV_0 \gg E_{\mathrm{F}} is fulfilled. The results of the above analysis suggest that if a parallel magnetic field is used to lift the spin degeneracy without any orbital effect, the metallic resistance can at most be increased by a factor of 4 [38, 80, 81, 82, 83, 84].

Figure 11 presents the results of mobility μ(ne)\mu(n_e) for different quantum degeneracies g=4g = 4, 2, and 1. For these calculations, the finite potential well approximation with gate screening and local field correction is utilized, employing the same disorder parameters as in Figure 10. As expected, we see that the mobility decreases monotonically with the reduction in gg. The scattering mechanism that limits the transport mobility is IR at high densities, while the limiting factor of mobility at low densities is RI (or BI, depending on the choice of the impurity model). The dashed curves in Figure 11 replicate the g=4g = 4 mobility results from Figure 10. The solid curves represent our prediction for the mobility in the presence of a parallel magnetic field without orbital effect. In this scenario, the spin is fully polarized (gs=1g_s = 1), resulting in a total quantum degeneracy of g=gv=2g = g_v = 2. The observed ratio of μg=4/μg=2\mu_{g=4}/\mu_{g=2} falls within the range from 2\sqrt{2} to 4, which aligns with the predicted g1/2g^{1/2} to g2g^2 scaling of μ\mu across various scattering mechanisms. Our choice of the valley degeneracy gv=2g_v = 2 is appropriate for the high mobility structures reported in Refs. [13, 14], where the Fermi energy in the density range ne=1−6×1011 cm−2n_e = 1 - 6 \times10^{11}\ \mathrm{cm}^{-2} is EF∈(0.6,3.8)×(4/g)E_F \in(0.6, 3.8) \times(4/g) meV. This Fermi energy is significantly greater than the valley splitting of 0.2 meV. Should the valleys split further such that only one valley is occupied, combined with a strong parallel magnetic field, one could achieve a quantum degeneracy of g=1g = 1. The dot-dashed curves in Figure 11 are our predictions in this scenario, indicating a further reduction in mobility.

Figure 11. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The dashed, solid, and dot-dashed curves represent the theoretical mobility assuming the total quantum degeneracy g=4g = 4, 2, and 1 respectively. The calculations are done using the finite potential well approximation with gate screening and local field correction. Subfigures (a) and (b) incorporate RI, IR, and AD scattering, whereas (c) and (d) consider BI, IR, and AD scattering. The corresponding disorder parameters are the same as in Figure 10.

The quantum degeneracy gg also has an effect on the critical density ncn_c estimated from the AIR criterion. The specific relationship between ncn_c and gg depends on the dominant scattering mechanism responsible for MIT and the dimensionless parameters such as qTF/2kFq_{\mathrm{TF}}/2k_F, kFdk_F d for RI, kFwk_F w for BI, and kFΛk_F \Lambda for IR. These parameters determine the effective screened potential that scatters electrons, thereby altering the MIT critical density. In the strong screening limit qTF/2kF≫1q_{\mathrm{TF}}/2k_F \gg1 or ne≪g3/16πaB2n_e \ll g^3/16\pi a_B^2, which is of interest to us for the high-mobility structures reported in Refs. [13, 14], the critical densities estimated from the AIR criterion for the RI and BI scatterings are summarized in Eqs. (61)-(65). We see that for long-range Coulomb impurity scattering, nc∝g1/5n_c \propto g^{1/5} while ncq∝g−1/3n_{cq} \propto g^{-1/3}. For MIT induced by short-range impurity scattering, the critical density is given by

nc(S)=ncq(S)=πgn2,(66)n_c^{(\mathrm{S})} = n_{cq}^{(\mathrm{S})} = \frac{\pi}{g} n_2, \tag*{(66)}

which is the same as Eq. (65) with N1(2ds−w)N_1(2d_s - w) replaced by the effective 2D short-range impurity concentration n2n_2 nearby the 2DEG. For example, Eq. (66) describes the case in Si MOSFETs where ncn_c increases as gg (and μ\mu) decreases [76, 77, 85]. Table I shows the value of ncn_c and ncqn_{cq} for the same disorder parameters taken from Figure 10 but with different gg. We should emphasize that the disorder parameters in Figure 10 are obtained by fitting the experimental data with the theoretical mobility curves using g=4g = 4, which is justified since EFE_F is much larger than the valley splitting in the high-mobility Si quantum wells. For completeness, Appendix § 10.1 presents the best-fit results of the mobility curves and the corresponding critical densities assuming g=2g = 2. In this scenario, the fitting disorder parameters are lower than those in Figure 10 due to weaker screening.

Conclusion

In summary, as the main goal of our theoretical work in the context of the importance of the Si/Ge quantum computing platform, we have systematically examined the roles of different scattering sources on the transport and quantum mobilities in the Si/Si0.69_{0.69}Ge0.31_{0.31} quantum wells with a co-design of high electron mobility and large valley splitting reported in Refs. [13, 14]. We identify remote charged impurities and interface roughness as primary limitations for transport mobility at low and high electron densities, respectively. Quantum mobility, on the other hand, influenced by both remote and distant background impurities, requires further experimental investigation to understand the charged impurity distribution in SiGe barriers. Our analysis finds alloy disorder scattering to be negligible in high-mobility structures. Other scattering sources, such as inhomogeneous strain distributions and threading dislocations [36] that originated from the bottom metamorphic substrate of several μ\mum, are also quantitatively irrelevant for the reported mobility using similar estimations from Ref. [52, 86]. We estimate the metal-insulator transition critical density using the Anderson-Ioffe-Regel criterion and qualitatively explain the breakdown of the Boltzmann-Born theory at low densities due to the long-range Coulomb disorder. We also estimate the critical density by considering inhomogeneous density fluctuations induced by long-range Coulomb disorder in the system and find a larger critical density compared to the one obtained from the Anderson-Ioffe-Regel criterion. To gain a more comprehensive understanding of the metal-insulator transition, more extensive future experimental studies of the temperature and density dependence of the conductivity should be conducted. We also predict the electron mobility with reduced quantum degeneracy, where the spin is fully polarized by an external parallel magnetic field, for future experimental validation.

Acknowledgments

The authors thank G. Scappucci and D.D. Esposti for discussions and clarification of the experimental details. This work is supported by the Laboratory for Physical Sciences.

Best-fit results assuming g=2g = 2

In this appendix, we present the best-fit results of mobility curves and the corresponding MIT critical densities assuming the total quantum degeneracy g=2g = 2. Figure 12 shows the mobility and quantum mobility as a function of the electron density. By comparing Figure 10 and Figure 12, it is evident that the disorder parameters for g=4g = 4 are generally higher than those for g=2g = 2, due to the fact that screening increases monotonically as gg increases. Table III shows the corresponding results of MIT critical densities. We see that ncn_c and ncqn_{cq} are in general smaller than the g=2g = 2 results shown in Table I, because the values of best-fit mobility and quantum mobility are higher.

Figure 12. Mobility (μ\mu) versus electron density (nen_e) for quantum wells with widths of 5 nm (red) and 7 nm (blue). The experimental data are depicted by red and blue curves, with corresponding error bars shown as shaded areas. The solid black curves represent the theoretical transport mobility, while the dashed black curves denote the quantum mobility, both calculated using the finite potential well approximation with gate screening and the local field correction, assuming the total quantum degeneracy g=2g=2. Subfigures (a) and (b) incorporate RI, IR, and AD scattering, whereas (c) and (d) consider BI, IR, and AD scattering. Fitting parameters: (a) nr=3.6×1012 cm−2n_r = 3.6 \times10^{12}\ \mathrm{cm}^{-2}, Δ=3.5\Delta= 3.5 Å, and Λ=20\Lambda= 20 Å; (b) nr=3.6×1012 cm−2n_r = 3.6 \times10^{12}\ \mathrm{cm}^{-2}, Δ=8\Delta= 8 Å, and Λ=15\Lambda= 15 Å; (c) N1=3.3×1015 cm−3N_1 = 3.3 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=4\Delta= 4 Å, and Λ=15\Lambda= 15 Å; (d) N1=3.3×1015 cm−3N_1 = 3.3 \times10^{15}\ \mathrm{cm}^{-3}, N2=0N_2 = 0, Δ=6.5\Delta= 6.5 Å, and Λ=18\Lambda= 18 Å.

RI (5 nm)RI (7 nm)BI (5 nm)BI (7 nm)
ncn_c (g=2g=2)0.440.410.160.15
ncqn_{cq} (g=2g=2)0.680.630.150.14

Table III. Critical density ncn_c and ncqn_{cq} in units of 1011 cm−210^{11}\ \mathrm{cm}^{-2}. The results are obtained from kFl=1k_F l = 1 and kFlq=1k_F l_q = 1 using the same impurity parameters as in Figure 12 calculated in the finite potential well approximation with gate screening and local field correction. RI (BI) indicates the dominant scattering mechanism at low densities. 5 nm (7 nm) means the calculation for a 5 nm (7 nm) QW.

References

  1. [1]L. M. K. Vandersypen, H. Bluhm, J. S. Clarke, A. S. Dzurak, R. Ishihara, A. Morello, D. J. Reilly, L. R. Schreiber, and M. Veldhorst, npj Quantum Information 3, 34 (2017).
  2. [2]G. Scappucci, P. J. Taylor, J. R. Williams, T. Ginley, and S. Law, MRS Bulletin 46, 596 (2021).
  3. [3]G. Burkard, T. D. Ladd, A. Pan, J. M. Nichol, and J. R. Petta, Rev. Mod. Phys. 95, 025003 (2023).
  4. [4]X. Hu and S. Das Sarma, Phys. Rev. Lett. 96, 100501 (2006).
  5. [5]D. Culcer, X. Hu, and S. Das Sarma, Applied Physics Letters 95, 073102 (2009).
  6. [6]S. Ahn, S. Das Sarma, and J. P. Kestner, Phys. Rev. B 103, L041304 (2021).
  7. [7]G. Cywiński, Q. Li, X. Hu, and S. Das Sarma, Phys. Rev. B 82, 155312 (2010).
  8. [8]D. Buterakos and S. Das Sarma, PRX Quantum 2, 040358 (2021).
  9. [9]F. A. Zwanenburg, A. S. Dzurak, A. Morello, M. Y. Simmons, L. C. L. Hollenberg, G. Klimeck, S. Rogge, S. N. Coppersmith, and M. A. Eriksson, Rev. Mod. Phys. 85, 961 (2013).
  10. [10]P. Huang and X. Hu, Phys. Rev. B 90, 235315 (2014).
  11. [11]M. L. V. Tagliaferri, P. L. Baydaz, W. Huang, A. S. Dzurak, D. Culcer, and M. Veldhorst, Phys. Rev. B 97, 245412 (2018).
  12. [12]A. Zwerver, S. Amitonov, S. de Snoo, M. Madzik, M. Rimbach-Russ, A. Sammak, G. Scappucci, and L. Vandersypen, PRX Quantum 4, 030303 (2023).
  13. [13]B. Paquelet Wuetz, D. Degli Esposti, A.-M. J. Zwerver, S. V. Amitonov, M. Botifoll, J. Arbiol, A. Sammak, L. M. K. Vandersypen, M. Russ, and G. Scappucci, Nature Communications 14, 1385 (2023).
  14. [14]D. D. Esposti, L. E. A. Stehouwer, Önder Gül, N. Samkharadze, C. Déprez, M. Meyer, I. N. Meijer, L. Tryputen, S. Karwal, M. Botifoll, J. Arbiol, S. V. Amitonov, L. M. K. Vandersypen, A. Sammak, M. Veldhorst, and G. Scappucci, “Low disorder and high valley splitting in silicon,” (2023), arXiv:2309.02832 [cond-mat.mes-hall].
  15. [16]M. G. Borselli, R. S. Ross, A. Kiselev, E. T. Croke, K. S. Holabird, P. W. Deelman, L. D. Warren, I. Alvarado-Rodriguez, I. Milosavljevic, F. C. Ku, W. S. Wong, A. E. Schmitz, M. Sokolich, M. F. Gyure, and A. T. Hunter, Applied Physics Letters 98, 123118 (2011).
  16. [17]A. Hollmann, T. Struck, V. Langrock, A. Schmidbauer, F. Schauer, T. Leonhardt, K. Sawano, H. Riemann, N. V. Abrosi-mov, D. Bougeard, and L. R. Schreiber, Phys. Rev. Appl. 13, 034068 (2020).
  17. [18]T. McJunkin, B. Harpt, Y. Feng, M. P. Losert, R. Rahman, J. P. Dodson, M. A. Wolfe, D. E. Savage, M. G. Lagally, S. N. Coppersmith, M. Friesen, R. Joynt, and M. A. Eriksson, Nature Communications 13, 7777 (2022).
  18. [19]X. Mi, J. V. Cady, D. M. Zajac, J. Stehlik, L. F. Edge, and J. R. Petta, Applied Physics Letters 110, 043502 (2017).
  19. [20]T. Ando, A. B. Fowler, and F. Stern, Rev. Mod. Phys. 54, 437 (1982).
  20. [21]G. Scappucci and D. D. Esposti (private communication).
  21. [22]S. Das Sarma and F. Stern, Phys. Rev. B 32, 8442 (1985).
  22. [23]B. J. Last and D. J. Thouless, Phys. Rev. Lett. 27, 1719 (1971).
  23. [24]S. Kirkpatrick, Rev. Mod. Phys. 45, 574 (1973).
  24. [25]B. I. Shklovskii and A. L. Efros, Soviet Physics Uspekhi 18, 845 (1975).
  25. [26]S. D. Sarma and E. Hwang, Solid State Communications 135, 579 (2005).
  26. [27]L. A. Tracy, E. H. Hwang, K. Eng, G. A. Ten Eyck, E. P. Nordberg, K. Childs, M. S. Carroll, M. P. Lilly, and S. Das Sarma, Phys. Rev. B 79, 235307 (2009).
  27. [28]M. J. Manfra, E. H. Hwang, S. Das Sarma, L. N. Pfeiffer, K. W. West, and A. M. Sergent, Phys. Rev. Lett. 99, 236402 (2007).
  28. [29]S. Das Sarma, E. H. Hwang, and Q. Li, Phys. Rev. B 88, 155310 (2013).
  29. [30]S. Das Sarma, E. H. Hwang, K. Kechedzhi, and L. A. Tracy, Phys. Rev. B 90, 125410 (2014).
  30. [31]P. W. Anderson, Phys. Rev. 109, 1492 (1958).
  31. [32]A. F. Ioffe and A. R. Regel, Prog. Semicond. 4, 237 (1960).
  32. [33]A. Efros, Solid State Communications 67, 1019 (1988).
  33. [34]F. G. Pikus and A. L. Efros, Zh. Eksp. Teor. Fiz. 96, 985 (1989), [Sov. Phys. - JETP 69, 558 (1989)].
  34. [35]F. Stern and S. E. Laux, Applied Physics Letters 61, 1110 (1992).
  35. [36]The STEM images of the step-graded SiGe buffer layer beneath the quantum well, as reported in Refs. [13, 14], demonstrate that strain-release defects and dislocations are limited to the multiple and sharp interfaces within the compositionally graded buffer layer, which emphasizes the overall crystalline quality of the SiGe strain-released barrier below the quantum well.
  36. [37]S. Das Sarma and E. H. Hwang, Phys. Rev. Lett. 83, 164 (1999).
  37. [38]E. H. Hwang and S. Das Sarma, Phys. Rev. B 72, 085455 (2005).
  38. [39]E. H. Hwang and S. Das Sarma, Phys. Rev. B 75, 073301 (2007).
  39. [40]E. H. Hwang and S. Das Sarma, Phys. Rev. B 87, 075306 (2013).
  40. [41]S. Das Sarma and E. H. Hwang, Phys. Rev. B 90, 035425 (2014).
  41. [42]B. Hu, M. M. Yazdanpanah, B. E. Kane, E. H. Hwang, and S. Das Sarma, Phys. Rev. Lett. 115, 036801 (2015).
  42. [43]M. Jonson, J. of Phys. C: Solid State Phys. 9, 3055 (1976).
  43. [44]A. Gold, Phys. Rev. B 50, 4297 (1994).
  44. [45]E. H. Hwang and S. Das Sarma, Phys. Rev. B 87, 125411 (2013).
  45. [46]X. Mi, T. M. Hazard, C. Payette, K. Wang, D. M. Zajac, J. V. Cady, and J. R. Petta, Phys. Rev. B 92, 035304 (2015).
  46. [47]S.-H. Huang, T.-M. Lu, S.-C. Lu, C.-H. Lee, C. W. Liu, and D. C. Tsui, Applied Physics Letters 101, 042111 (2012).
  47. [48]S. Das Sarma and E. H. Hwang, Phys. Rev. B 89, 121413 (2014).
  48. [49]These results are also true even in the weak screening limit q_TF/2k_F ≪ 1 where the density is high g/16πa_B^2 ≪ g^3/16πa_B^2 ≪ n_e, as long as q_TF d ≫ 1 is satisfied.
  49. [50]P. J. Price, Journal of Vacuum Science and Technology 19, 599 (1981).
  50. [51]A. Gold, Phys. Rev. B 38, 10798 (1988).
  51. [52]D. Monroe, Y. H. Xie, E. A. Fitzgerald, P. J. Silverman, and G. P. Watson, Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena 11, 1731 (1993).
  52. [53]I. A. Dmitriev, A. D. Mirlin, D. G. Polyakov, and M. A. Zudov, Rev. Mod. Phys. 84, 1709 (2012).
  53. [54]M. Sammon, M. A. Zudov, and B. I. Shklovskii, Phys. Rev. Mater. 2, 064604 (2018).
  54. [55]C. M. Tanner, Y.-C. Peng, C. Frewin, S. E. Saddow, and J. P. Chang, Applied Physics Letters 91, 203510 (2007).
  55. [56]A. Gold, Europhysics Letters 92, 67002 (2011).
  56. [57]D. Laroche, S.-H. Huang, E. Nielsen, Y. Chuang, J.-Y. Li, C. W. Liu, and T. M. Lu, AIP Advances 5, 107106 (2015).
  57. [58]S. Ahn and S. Das Sarma, Phys. Rev. Mater. 6, L091001 (2022).
  58. [59]A. Gold, Journal of Applied Physics 108, 063710 (2010).
  59. [60]R. M. Feenstra, M. A. Lutz, F. Stern, K. Ismail, P. Mooney, F. K. LeGoues, C. Stanis, J. O. Chu, and B. S. Meyerson, Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures Processing, Measurement, and Phenomena 13, 1608 (1995).
  60. [61]U. Penner, H. Rücker, and I. N. Yassievich, Semiconductor Science and Technology 13, 709 (1998).
  61. [62]S. Das Sarma and E. H. Hwang, Phys. Rev. B 69, 195305 (2004).
  62. [63]These results are also true even in the weak screening limit q_TF/2k_F ≪ 1 where the density is high g/4πΛ^2 ≪ g^3/16πa_B^2 ≪ n_e, as long as q_TFΛ ≫ 1 is satisfied.
  63. [64]F. F. Fang and W. E. Howard, Phys. Rev. Lett. 16, 797 (1966).
  64. [65]F. Schäffler, Semiconductor Science and Technology 12, 1515 (1997).
  65. [66]S. F., in Properties of Advanced Semiconductor Materials: GaN, AlN, InN, BN, SiC, SiGe, edited by M. E. Levinshtein, S. L. Rumyantsev, and M. S. Shur (John Wiley & Sons, Inc., New York, 2001) Chap. 6, pp. 149–188.
  66. [67]See for example, Eq. (5c) in Ref. 87 or Section 4.9 in Ref. 88.
  67. [68]G. Bastard, Applied Physics Letters 43, 591 (1983).
  68. [69]B. Laikhtman and R. A. Kiehl, Phys. Rev. B 47, 10515 (1993).
  69. [70]G. Huang, B. I. Shklovskii, and M. A. Zudov, Phys. Rev. Mater. 6, L061001 (2022).
  70. [71]V. Venkataraman, C. W. Liu, and J. C. Sturm, Applied Physics Letters 63, 2795 (1993).
  71. [72]T. McJunkin, B. Harty, Y. Feng, M. P. Losert, R. Rahman, J. P. Dodson, M. A. Wolfe, D. E. Savage, M. G. Lagally, S. N. Coppersmith, M. Friesen, R. Joynt, and M. A. Eriksson, Nature Communications 13, 7777 (2022).
  72. [73]G. Bastard, Wave mechanics applied to semiconductor heterostructures (Halsted, New York, 1988) chapter. 6.
  73. [74]T. Ando, Journal of the Physical Society of Japan 51, 3900 (1982).
  74. [75]We assume the delta-layer approximation w = 0 to get the analytical expressions.
  75. [76]S. Ahn and S. Das Sarma, Phys. Rev. Mater. 6, L091001 (2022).
  76. [77]S. Das Sarma and E. H. Hwang, Phys. Rev. B 89, 235423 (2014).
  77. [78]S. Ahn and S. Das Sarma, Phys. Rev. B 107, 195435 (2023).
  78. [79]S. Das Sarma and E. H. Hwang, Phys. Rev. Lett. 84, 5596 (2000).
  79. [80]S. Das Sarma and E. H. Hwang, Phys. Rev. B 72, 035311 (2005).
  80. [81]S. Das Sarma and E. H. Hwang, Phys. Rev. B 72, 205303 (2005).
  81. [82]T. Okamoto, K. Hosoya, S. Kawaji, and A. Yagi, Phys. Rev. Lett. 82, 3875 (1999).
  82. [83]V. T. Dolgopolov and A. Gold, Journal of Experimental and Theoretical Physics Letters 71, 27 (2000).
  83. [84]I. F. Herbut, Phys. Rev. B 63, 113102 (2001).
  84. [85]M. P. Sarachik, Europhysics Letters 57, 546 (2002).
  85. [86]K. Ismail, F. K. LeGoues, K. L. Saenger, M. Arafa, J. O. Chu, P. M. Mooney, and B. S. Meyerson, Phys. Rev. Lett. 73, 3447 (1994).

Paper details

vlnc:W_h7dstd4u
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents