The presence of a strong, inherent polar electric field in GaN 1 causes the well-known quantum confined Stark effect 2 3 4 (QCSE) regarding carrier behavior inside a QW (Fig. 1a). The separation of the center of charge between electron and hole wave functions, caused by the polar E-field, reduces mutual overlap and the related emission probability. The lowering of the confined energy levels, relative to the unperturbed square QW, causes red-shifting of the emitted radiation during electron-hole recombination. This effect has been the subject of extensive perturbative 5 as well as non-perturbative analytic treatments 6 7 8 9, including excitonic effects 10 11 12 13 14. In general earlier analytic theories neglected the modifications to the (intrinsic polar or externally applied) E-field caused by the charge separation and the resulting dielectric shielding, assuming in effect very low carrier densities.

At high carrier densities, charge separation and dipole field formation is sufficient to cause shielding of the intrinsic polarization E-field 15 . The resulting potential gradient across the QW is not uniform, and most of the potential drop is localized across charged layers formed at the edges of the QW (Fig. 1b). The electric gradient scale is of the order of the Debye length. For densities near 10¹⁹ cm⁻³ the Debye length shrinks down to nm-scale (Fig. 1c), and the potential drop is mostly localized at the QW edges while the QW interior is nearly field-free (shielding of the intrinsic E-field). This constitutes the ES-shielded QCSE. It has been anticipated 16 that the shielding of the interior E-field would reduce or even eliminate the QCSE at densities 10¹⁹ cm⁻³. Detailed numerical simulations, employing the self-consistent Poisson–Schrodinger equations 17 have showed that a much higher than expected carrier density, near 10²⁰ cm⁻³, is required to eliminate the QCSE for QWs wider than 5 nm. This has been attributed to the persistence of carrier confinement in the potential dips at the QW edges, even when the electric field is screened out from the middle. However, an analytic treatment examining the carrier behavior in the ES-shielded QCSE is so far lacking.

This study focuses in finding solutions for the confined carrier wave functions by solving the one-particle Schrodingers’ equation. To gain insight the following simplifying assumptions are used: (a) The shielded potential has exponentially decaying profile on the Debye length ∼λ_D scale; (b) the peak-to-peak shielded voltage is a given function of the carrier density and the intrinsic polarization strength and (c) excitonic effects are ignored.

The shielded potential results from a self-consistent solution of Poisson’s equation for point-like charges obeying Fermi statistics 15 . Neglecting the charge spreading of the carrier wave function is not too severe when the carrier localization length ∼λ_D is much smaller than the QW width L. When the Fermi level separation from the lowest occupied levels is much larger than κT, i.e., for nearly Maxwellian distributions, the shielded potential is well approximated by a symmetric profile The exponentially decaying profiles remain a reasonable approximation for Fermi–Dirac distributions in general.

We obtain results based on: (a) a second order perturbative expansion; (b) non-perturbative series expansion; and (c) a numerical solution of Scrodinger’s equation for the carrier envelope wave function. The analytic expressions for the energy levels from (a) are evaluated against numerical the results from (c). The infinite λ _D , zero shielding limit reverts to the original (unshielded) QCSE results.

Our analytic models find that increasing the carrier density causes an increase (blue shifting) of the energy levels relative to the unshielded (red-shifted) QCSE values. The confined energy levels asymptote to the values for a flat square QW, and the red shift is effectively eliminated, for densities ≥ 10¹⁹cm⁻³. The perturbative energy levels agree with the numerical values at lowV _p, and become inaccurate when the polarization voltage exceeds the energy of the fundamental confined mode in a square QW. Numerical solutions of the Schrodinger equation for high polarization, relevant to GaN parameters, show that at high V_p the perturbation results overestimate the energy level shifts by a factor of 2, but they provide the correct trends over the entire range.

The dependence of the characteristic emission time on the carrier density is computed based on the numerically evaluated eigenfunctions. Despite the adopted simplifications these results reproduce the three order of magnitude increase in the emission rate between densities 10¹⁹ and 10²¹, leading to a complete rectification of the QCSE, as was reported from experimental and detailed computations in Ref. 17 .

Interestingly, it is found that elimination of the QCSE-related energy red-shift clearly precedes the recovery of the radiative emission time: the energy red-shifting is gradually eliminated between densities 10¹⁷cm⁻³ and 10¹⁹cm⁻³ while the emission probability is restored at higher densities between 10¹⁹cm⁻³ and 10²⁰cm⁻³. The first result agrees with the energy recovery behavior obtained in 16 while the emission probability behavior agrees with the results in 17 . The delay in the restoration of the emission probability is explained in terms of carrier trapping at the QW edge.

We investigate the wave function profiles and the structure of the energy spectrum inside QWs in the presence of an ES-shielded polarization potential. It can be shown (Appendix1) that the self-consistent charged layer (plasma sheath) potentials can be reasonably approximated by exponentially decaying

where κ_D = a/λ_D scales as the inverse Debye length and a is of order unity. The peak amplitude V_o here is taken equal to half the intrinsic "polarization voltage" The value Φ_p(0) = 0 at mid-point equals the bottom energy for a polarization-free square well (Fig. 2), and serves as the reference point for electron energy levels. Hole levels are measured from the bottom of the valence well. The above symmetric potential applies for low carrier density and a Fermi level near the mid bandgap. For high doping the reference point x _o defined by Φ_p(x_o) = 0 moves closer to the left (right), with unequal edge potentials −V _p(−L/2) > V_p(L/2) (−V_p(−L/2) > V _p(L/2)) for N-doped (P-doped) materials. For analytic simplicity this study will retain the symmetric potential.

Expressing the slowly varying envelope wave function in separable coordinates as casts the 1-D Schrodinger’s equation along x as

where is the net energy contribution from the motion across the well, and k _y , k _z correspond to the continuous spectrum along the QW. Analytic solutions of (2) are obtained from second order perturbation theory, in terms of an expansion in unperturbed square well modes

with

A change of variable transforms the integral in the rhs of (4) into

Substituting inside (3) yields

In the zero-shielding, infinite Debye length limit when one recovers the unshielded QCSE levels

The mode energy E _n is always measured relative to the middle of the well; the latter always coincides with the bottom energy for the square (un-biased) QW, as shown in Fig. 1.

The shift in energy levels relative to the square QW eigen values, obtained from (7), is plotted in Fig. 2a versus the ratio κ_D L ≡ L/λ_D for the lowest three modes. The chosen parameters are peak-to-peak sheath potential 2V _o = 50 meV, QW width L = 8 nm and m _e ^*/m_e = 0.19 for GaN. For λ_D ≫ L/2 the polarization field is nearly unshielded, the potential profile nearly linear, and the red-shifting hovers near the maximum value, characterizing the ordinary QCSE. Red shifting is however reduced rapidly as the screening range becomes equal or shorter than half the QW width, λ_D ≤ L/2, becoming completely negligible at λ_D < L/4. Beyond this point the energy levels revert to the square QW eigen values and the QCSE is completely "rectified". Using the scaling with the value for the GaN dielectric constant recasts energy shift Fig. 2a in terms of the carrier density N_e, Fig. 2b. Complete shielding of the QCSE occurs at N_e ≥ 10²⁰ cm⁻³. This value agrees well quantitatively with similar results obtained in 17 , based on the observed decrease in the radiative emission time.

As expected, perturbation theory breaks down when the polarization potential exceeds the unperturbed (square QW) energy eigen values e V_o ≥ E₁ ⁽⁰⁾ ∼31 meV. Since the combined inherent and strain-induced polarization fields can reach values up to 5 MeV/cm 18 and up to 2.5 V over a 10 nm QW, numerical solutions of Schrodinger Equation are required for realistic polarization values. For comparison Fig. 3 plots the lowest energy levels obtained from numerical solutions (points) and perturbation theory (curves) versus the ratio L/2λ_D for V _o = 0.250 V. For unshielded or partially shielded QCSE with λ_D ≤ L/4 the perturbation theory overestimates the red-shift by a factor of 2. Good agreement occurs for λ_D < L/8 when the charged layer thickness is much smaller than the QW thickness, and thus the size of the perturbation, parameterized by becomes negligible.

It is useful, for the discussion that follows, to obtain an analytic estimate of the carrier energy eigen values for arbitrary V _o and λ_D. To that end the eigenfunctions of Eq. 2are obtained in terms of an infinite power series expansion a la Frobenius, Appendix1. The fast convergence of the series solutions allows the calculation of the expectation values of the kinetic energy potential energy 〈eΦ(x)〉 and the total energy expectation value, yielding

where K _n , W_n are functions of eV_o/κT and the quantum number n, and C _o is the wave function normalization constant. The kinetic energy increases with decreasing λ_D, while the potential (“edge-binding”) energy is fixed. For eV_o > 5κT the ratio W ₁/K ₁ for the fundamental mode is nearly constant and hovers close to 1/2, Appendix 1.

The reduction of the red shift with increasing ES shielding and decreasing shielding distance λ_D, manifested experimentally as a blue shift relative to the unscreened QCSE, is qualitatively understood as following. For λ_D < L/2 the sin h(x/λ_D) potential behaves like an edge-well inside the square well, instead of a tilted QW floor. If confinement within the edge-well occurs, the lowest energy level must satisfy 〈E ₁〉 ≤ 0. As long as the confined “kinetic energy” is less than the edge-binding energyeV _o W ₁ then E₁ < 0 and the wave function is trapped at the QW edge. Edge-confinement within a range shorter than the well width, λ_D < L/2, increases the mode energy relative to that for a tilted QW bottom and causes blue shift relative to the unshielded QCSE. The blue-shift increases with increasing carrier density, meaning shorter confinement length λ_D. Eventually, for large enough density with the kinetic energy exceeds the edge-binding energy and 〈E ₁〉 > 0, edge confinement ceases, and the wave function shifts to the center to occupy the full QW width. At the same time most of the well bottom becomes nearly as flat as in a square well, since is excluded from most of the interior. Full “rectification” of the QCSE occurs and the eigen values and eigen modes approach that of a square QW.

Transition from edge-confinement to full QW occupation occurs for either V _o < V _th or where is the threshold under given λ_D, and the threshold under given V _o. This transition is shown in Fig. 4a and b, plotting the fundamental mode profiles Ψ(x) for various values of λ_D/L, for low and high voltages, respectively V _o= 0.250 V and V _o= 2.05 V. As the screening distance decreases, the center of the wave function moves from the left edge towards the center of the well. The transition to full QW occupancy occurs at shorter screening length λ_D for higher V _o (Fig. 4b).

Figure 5a plots the lower two eigen values versus sheath potential, for given λ_D = L/8. The fundamental E₁ becomes positive at about For V_o < V _th the value E ₁ increases and tends to the square well limit as V _o ≃ 0. Figure 5b shows the fundamental eigen value E ₁ versus L/λ_D for two different voltages V _o. The eigen values asymptote to the square QW limit at shorter screening distance for the case of higher polarization V _o.

The changes in the wave function profiles have a profound influence in the e−h transition probability during radiative emission, proportional to the dipole moment overlap integral

where are the lattice-periodic parts and the slowly varying envelope functions obtained from (2). Employing, as usual, the space-scale separation between the rapidly varying, on the lattice-constant scale, u _c u _v, and the slowly varying envelopes, valid for as long as L, λ_D ≫ a, the above is approximated by

Orthogonality among the lattice functions u _c u _v was used in arriving at (10). The last integral over the unit lattice unit cell volume C is independent of the polarization. For “vertical transitions” with (given that ) the dependence on the polarization voltage V _o and screening distance λ_D is carried entirely in the overlapping between electron-hole envelopes

with a constant. Here we will assume, due to the symmetry in the sinh potential, that Ψ_h (x) = Ψ _e(L−x). Taking the transition probability for a flat QW with as reference, and since the emission time one has

The ratio is potted in Fig. 6a versus L/λ_D for various peak voltages V _o, using the wave function profiles obtained from numerical solutions. Characteristic emission times tend to increase with increasing applied polarization voltage V _o, and decrease with decreasing screening distance λ_D. The results of Fig. 6a are plotted verusus the corresponding carrier density N in Fig. 6b, for QW width 8 nm. These results reproduce the three order of magnitude emission increase between densities 10¹⁹ and 10²¹, resulting in complete rectification of the QCSE, that was first obtained using detailed Poisson–Schrodinger simulations in Ref. 17 for a 7 nm QW.

A careful comparison between the energy blue-shifting with increasing density (screening), Fig. 7a, and the decrease in recombination time, Fig. 7b, shows that the rectification of the QCSE red-shift occurs before the recovery of the radiative emission time: the energy red-shifting is gradually eliminated first, between densities 10¹⁷cm⁻³ and 10¹⁹ cm⁻³, though the radiative emission time remains almost constant there. The emission probability is restored, rather abruptly, at higher densities between 10¹⁹ cm⁻³ and 10²⁰ cm⁻³. This lagging in restoring the emission probability is explained via edge-carrier trapping, mentioned in the previous discussion. As carrier density increases and the edge-potential range λ_D narrows down, the increasing edge-confinement of the wave function causes the energy level to increase. As long as the “confinement energy” is smaller than the edge potential depth eV _o electron and hole wave functions remain edge-localized and no significant change in overlap and in recombination time occurs. The abrupt decrease in the radiative emission time (increase in the radiative emission rate) occurs after since at this point the wave function moves from edge-confinement to full QW occupancy. Practically this means that the QCSE-related energy red-shift has already been eliminated before the radiative emission time recovers. This behavior agrees with the results in 17 .

It has so far been tacitly assumed that the charged layer peak-voltage V_o is independent of the screening carrier density N _{e, h} and the peak-to-peak voltage 2V _o was taken equal to the “polarization voltage” for an unscreened QW, Fig. 2a. In other words the shielding only modified the potential profile across the QW. However, for given applied and L, the shielded V_o does depend on the carrier density, and in fact V _o is reduced below V_p at high carrier densities. The shielding of the peak voltage is summarized below, based on results from earlier studies 15 .

Self-consistent charged layer solutions under Fermi–Dirac thermodynamic equilibrium 15 show that as the QW thickness L increases well beyond λ_D the peak-to-peak voltage asymptotes rapidly to a maximum saturation value Figure 8a plots 2V _o versus L for various polarization strength values and shows the saturation for L/λ_D ≫ 1. Clearly V _s increases with polarization strength The dependence of V _s on density is given in Fig. 8b. The fact that V _s decreases with increasing density stems from Gausses law: it takes a given amount of surface charge to screen a given field. Applying scaling arguments the charge layer thickness is (half of the electric field screened at each QW edge) and the sheath voltage Thus for given polarization the voltage V _s scales roughly as when L > 2λ_D .

The screened voltage value is always less or equal to the intrinsic “polarization voltage”, This is shown in Fig. 8c, plotting the ratio of the peak-to-peak voltage 2 V _o to V_p, versus sheath length, for given doping density N _D= 10¹⁸ cm⁻³. For as long as L≤ 2λ_D one has unsaturated behavior Once saturation is reached for L > 2λ_D the peak-to-peak voltage is pinned at V_s, independent of L. This is because when L > 2λ_D the polarization field is screened-out from the QW interior length L− 2λ_D that yields a negligible contribution to the voltage difference; V _s comes entirely from two charged layers of width λ_D. Hence, for wide QWs the peak-to-peak voltage turns out much smaller than the polarization voltage, and the ratio 2V _o/V_p goes as 1/L. Notice that the saturation length L _s where 2V _o dips below V _p depends also on the field strength; letting and yields thus saturation occurs at smaller QW thickness with increasing . According to Fig. 8c, one may apply unsaturated values for QW thicknessL < 10 nm and for up to doping densities 10¹⁹ cm⁻³. This is illustrated in Fig. 9, plotting the ratio 2V _o/V_p versus doping density N_D for fixed QWL = 8 nm and for various strengths

For given L = 8 nm, the values 2V_o assume their saturation values and the shielded voltage falls significantly below V_p when doping densities exceed ≥10²⁰ cm⁻³. This is illustrated in Fig. 10, showing the screened potential profiles, 10a, and electric fields, 10b, for various doping levels N _D across an 8-nm QW for The peak-to-peak voltage decreases well below V_p with increasing N _D. In addition, the electron and hole charged layers become asymmetric: V _e across the negative charged layer is different than V _h across the positive charged layer. In general, reduction of the peak-to-peak voltage, as well as asymmetric electron-hole profiles should be considered for a more accurate description of the ES shielded QCSE. In particular, the drop in V _s < V_p with increasing density could accelerate the cancellation of the QCSE and the blue shifting of the energy levels. For the relevant to our GaN experiments parameters, however, the red-shifting is all but cancelled out at density 10¹⁹ cm⁻³, just before such effects become significant. Thus it appears that energy level blue-shifting caused by the sin h effect in the potential profile cancels to a large degree the QCSE effect, before shielding of the peak amplitude itself becomes important.

A simplified model employing ES-shielded, exponentially-decaying polarization potentials localized at the QW edges, was employed to study the QCSE at high doping densities. Blue shifting of energy levels relative to the unshielded QCSE occurs with increasing carrier density, due to the wave function constriction within scale length λ_D < L/2. When the “edge-localization energy” exceeds the peak-voltage of the charged layer eV _o the wave function center shifts to the middle of the QW and behavior becomes similar to that of a square (unbiased) QW. In addition, at very high doping the shielded peak voltage is reduced well below the original unshielded “polarization voltage” V _p. Both effects cause gradual elimination of the QCSE red-shifting, an increase in the e−h wave function overlap and a decrease of the radiative emission time. A significant reduction of the peak polarization voltage requires higher carrier densities than most practical situations, and screening effects stem mainly from the interior-screening and the localization of the polarization voltage within QW edge-layers. Our theoretical estimates show that the elimination of the QCSE related red-shift in energy precedes the recovery in the radiative emission time, in quantitative agreement with experimental measurements in 17 .

Section "QW Eigen Modes with ES-shielded Polar Potential" derived a perturbative solution for the edge-confined modes in terms of the square well eigen modes. Another approach, involving an infinite series polynomial expansion, will be given here and used to derive the scaling of the edge-confined expectation values for the kinetic and potential energy. First, for λ_D ≪ L/2 one may approximate the sinh potential for x < 0, , as where ζ the distance from the edge The sinh Schrodinger Equation 2 is then approximated by one for an exponential potential which has been analyzed elsewhere.¹ A dimensionless scaling measuring length in units of and energy in units of yields

where n labels the energy quantum number A change of variable for ζ > 0 with removes the exponential term and reduces (13) to

The boundary conditions at correspond to w = 1, 0, and are given by A series expansion

inside (14) yields the coefficient recurrence relation or,

where and is found from the normalization condition. Substitution into the series solution and application of the boundary conditions at yields the eigen values from the roots of the following indicial equation

Switching (15) back to the original variables yields the corresponding eigenfunctions as

making use of The leading term goes as and gives the asymptotic behavior at For practical purposes is suffices to keep polynomial terms up to order M equal to twice the integer part inside the infinite sum in (17).

One may now compute expectation values with direct integration of (18). First, orthonormalization yields the normalization constant c_o from

The expectation potential energy yields with

and the expectation kinetic energy yields

Thus the energy expectation value 〈E_n〉 is

where the normalization factor from (19). Thus edge detrapping at about 〈E₁〉 > 0 occurs for Both K and W depend on and on the energy eigen value -ε₁ where ε₁ = ξ₁. The ratio W₁/K₁ is plotted in Fig. 11 versus the peak voltage (normalized in units of κT) using the lowest mode energy n = 1 inside (20) and (21). Note that for V_o > 5κT the ratio hovers near 1/2 and thus detrapping occurs at

The self-consistent Poisson's equation, including the influence of the charged layer (plasma sheath) potential Φ(x) on the Fermi-Dirac occupation number f in determining the local carrier density is

subject to the boundary conditions This means that equals the unshielded value at each QW edge. Above we have normalized and where N_o is a reference carrier density and the corresponding Debye length which includes the dielectric shielding ε from core (bound) electrons. The sum of the electron, hole and charged donor charge densities (N-doping is assumed without loss of generality) on the right-hand side follows from the equilibrium Fermi-Dirac occupation numbers,

with E_C, E_V, F being respectively the conduction, valence, and Fermi levels, G_e,h(E) the electron (hole) density of states and N_D the dopant density (normalized to N_o), and The Fermi level F is obtained from the condition ρ[χ_o|_Φ=0] = 0 at the neutral point Φ(x_o) = 0. This automatically guarantees total charge neutrality over the QW as follows. The point x_o where ρ(x_o) = 0 is also the location of the minimum of the screened electric field, since there. Now, from and Gausses law follows and Q_-=-Q₊. The sheath Eqs. 23 and 24 yield the free carrier dielectric shielding inside a plasma-filled QW capacitor of plate charge under the nonlinear response ρ[Φ].

Analytic solutions of (23) and (24) in terms of the polarization field strength exist for certain degenerate and non-degenerate limits. The simplest treatment illustrating all the salient features is the undoped (intrinsic semiconductor) limit N_D = 0. Since the Fermi level in this case lies close to mid-bandgap and the non-degenerate Maxwellian limit applies for the carrier statistics. The carrier density is simply given by where is the zero polarization electron and hold density. Three dimensional density of states is assumed for large enough QW width with small energy spacing Poisson's equation is then simplified to

It has exact analytic solutions, since x = X(Φ) is given in terms of elliptic integrals of complex argument, and hence Φ(x) follows in terms of the elliptic amplitude (Jacobi ) function,

where is the potential drop over half the QW length L and (Different profiles apply for given applied voltages 19 across the sheaths.) The field and voltage profiles have respectively even/odd symmetry about the middle of the QW, reflecting the opposite electron and hole densities for an undoped material. The opposite polarity electron and hole sheath potentials V_e = -V_h = V_o are respectively defined by V_e ≡ Φ(0) - Φ(L/2) and V_h ≡ Φ(L/2) - Φ(L). The corresponding nominal sheath lengths are L_e = L_h = L/2. However, when L_{e, h} ≫ λ_D, the field in each sheath is essentially localized within a few λ_D while the rest of the length is almost field-free.

Solutions and shielded voltage profiles for both Maxwellian, Eq. 26, as well as Fermi-Dirac distributions in general, Eqs. 23, 24, have been given in 15. Maxwellian profiles are reasonably well fitted with sinh-profiles employed in the present analysis, such as the bottom of the QW Fig. 2a. The screened profiles remain essentially similar for Fermi-Dirac distributions in general, as shown in Fig. 9a, with one difference: the symmetry between the electron and hole charged-layers is broken, V_e ≠ -V_h. In addition, F-D statistics yields higher saturation voltages V_S under given parameters. The saturation values shown in Fig. 7 correspond to general F-D solutions. Finally, for sufficiently small potentials any sheath profiles, including (26), are reduced to exponential profiles 15 , solutions of the linear differential equation

¹ The solutions with are the odd-symmetry eigenfunctions of the general attractive potential