Fermi energy tuning with light to control doping profiles during epitaxy

Controlling interfaces, dopant distributions, and defect formation during semiconductor epitaxy is paramount to realizing device structures with smaller features, enhanced performance, and new functionalities. This has typically been achieved by regulation of growth rates, introduction of surfactants, manipulation of substrate temperature, and use of non-equilibrium growth methods. Yet, as device layer thicknesses shrink and doping concentrations increase, and as alloy compositions and interfaces become more complex, control over dopant distribution has become more difficult, even as it has become more critical.

The Fermi energy, E_F, plays a significant role in epitaxial growth processes. Shifting E_F towards either the conduction or valence band edge by adding dopants of one polarity or the other often reduces the formation energy of compensating defects of the opposite charge.^1–3 These defects have the potential to alter adatom incorporation dynamics or modify atomic diffusion coefficients, among other effects.^4,5 In terms of dopant profiles, it is also understood that the pinning of E_F at the growth surface by mid-gap states can create an electric field in highly doped layers that forces the drift of dopants towards the surface.^6,7 Both of these effects can work in tandem to induce a so-called “carry-forward” process that spreads out an abrupt dopant profile over tens of nanometers and can substantially affect the performance of semiconductor devices with delta-doped layers.^4,8 One method for suppressing the carry-forward process has been to co-dope these layers with dopants of the opposite polarity, which effectively shifts E_F in the bulk back toward the center of the bandgap and reduces the abovementioned mechanisms.^6,9 However, co-doping can be impractical or undesirable in actual applications.

In this letter, we show that ultraviolet (UV) illumination of the semiconductor surface during growth can reduce dopant carry-forward. We propose that this effect is due to the creation of photogenerated carriers that split the quasi Fermi levels and suppress the mechanisms for dopant migration. These results are of key importance, since practical approaches for achieving abrupt dopant profiles are needed for improved device performance and for the realization of next-generation devices containing very thin or even two-dimensional (2D) epitaxial layers. As demonstration of the utility of this approach, we show that UV illumination during growth can be used to improve the low temperature mobility in inverted modulation doped quantum wells.

Light stimulated epitaxy has been used in the past to improve crystalline quality at low growth temperatures and to overcome low doping limits in II–VI semiconductors.^10–12 More recently, UV illumination was shown to reduce compensating defects in Al_xGa_1−xN.¹³ Many of these results have been linked to changes in adatom incorporation rates, defect formation energies, and atomic chemical potentials.^11,13,14 A common denominator among such effects is a photocarrier-induced alteration of the Fermi energy. Thus, photogeneration of electrons and holes can be used as a tool to overcome barriers to growth imposed by E_F.

We have tested the use of UV illumination to reduce dopant carry-forward in molecular beam epitaxial (MBE) growth of Si-doped GaAs films. Ga and Si were supplied by conventional effusion cells; C atoms for co-doping studies were generated with a C sublimation source; and an As₂ flux was produced by a two-zone valved cracker source. Growths were carried out under V:III flux ratios > 6 at a rate of 1 μm/h. UV illumination was generated by a KrF excimer laser (λ = 248 nm, pulse width = 25 ns) and was focused into a uniform 1 cm² spot on the sample surface via a homogenizer and a series of beam shaping optics. The output of the laser was tuned with an attenuator. We investigated two levels of laser fluence at the sample surface—30 mJ and 60 mJ—and we compared these two cases with the zero-fluence (“dark”) case. The “dark” samples were obtained simply by analyzing regions of the same substrates used for the illuminated growths that were far from the focused laser spot. For all samples, we used a laser pulse repetition rate of 10 Hz.

A nominal substrate temperature of 540 °C was chosen for all growths, in order to reduce the effects of thermal diffusion. The temperature was measured both by a thermocouple and by a k-Space BandiT band edge thermometry system aligned to the laser spot. Based on separate measurements, we have found that the levels of laser illumination investigated here heat the bulk of the substrate by only a few degrees. The BandiT was used to verify the sample temperature at the start of each growth but was turned off during the main part of the growth, so as not to introduce extraneous above-bandgap light to the growth surface. It is worthwhile to note that the laser can be expected to induce local heating immediately at the growth surface in the illuminated area that will not be detected by band edge thermometry or by the thermocouple. Thus, the illuminated surface is presumably locally hotter than the nominal substrate value of 540 °C. However, the resulting effects are the opposite of what would be expected due to sample heating and therefore cannot be attributed to local heating by the laser (discussed below). Dopant profiles were measured by secondary ion mass spectrometry (SIMS). Measurements were carried out using a CAMECA IMS-7F instrument, using Cs⁺ as the primary ion beam. Mobility measurements were made in a standard Hall system at 77 K with an accuracy of ±1%.

Figure 1 shows the Si profiles of samples grown under dark conditions (solid red line) and laser fluences of 30 and 60 mJ (dashed-dotted orange and dashed red lines, respectively), where the Si flux was abruptly stopped by closing the source shutter part way through the layer growth. In the case of dark growth, the dopant concentration decreases exponentially from 1.5 × 10¹⁹cm⁻³ to 1.5 × 10¹⁶ over ∼230 nm. This results in a characteristic segregation length—which we define as the length over which the Si concentration decreases by one order of magnitude—of 76 nm/decade. By contrast, the segregation length drops to 36 nm/decade for the sample grown under laser fluence of 30 mJ and to 23 nm/decade in the sample grown under the higher laser fluence of 60 mJ. Thus, illumination with increasing intensity appears to significantly decrease the magnitude of the Si dopant carry-forward. We emphasize that this reduction in Si carry-forward under illumination contradicts what one would expect from effects due simply to thermal heating by the laser. In that case, we would expect to see an increase in the diffusion of Si.

The background Carbon concentration was also measured by SIMS and is plotted in Fig. 1, for the sake of reference. It is found to be constant in each of the samples, with a value of ∼1 × 10¹⁷cm⁻³, which is two orders of magnitude below Si. It is expected to have little effect on the Si carry-forward mechanism, as will be discussed below.

The Si dopant carry-forward mechanism is understood to be caused by several processes related to the high-doping induced shift of E_F toward the conduction band edge.^6,15 First, pinning of the Fermi energy by mid-gap surface states¹⁵ creates a strong electric field that drives dopants toward the growth surface.⁶ Schubert et al., used a simple model to semi-quantitatively estimate the change in the position of the dopant front, x_d, relative to the position of the growth front, x_s, for a delta-doped layer as

where D is the diffusion coefficient of Si, e is the elementary charge, k is the Boltzmann constant, T is the substrate temperature, ε is the permittivity, N_D^2D is the sheet charge concentration at the nominal doping front, ϕ_B is the built-in electric potential at the growth surface, and x_s is the growth front position.⁶ This model does not take into account screening effects by the dopant atoms and is known to underestimate the total magnitude of the carry-forward,⁶ but provides a good qualitative description of the phenomenon.

Figure 2(a) shows the band diagram calculated for a Si-doped (1.5 × 10¹⁹ cm⁻³) GaAs film under the growth conditions of the present study without illumination. Calculations were performed using PC1D software with GaAs material properties applicable for growth at 540 °C. Surface pinning of E_F at mid-gap¹⁵ leads to a sizeable built-in electric potential of ∼0.6 eV, which will drive Si dopant carry-forward. The built-in potential is substantially reduced under illumination. Figure 2(b) shows the band diagram calculated for the experimental growth conditions during illumination. Specifically, the illumination input was set to 248 nm, and the power was set to a value corresponding that of the lower, 30 mJ/cm², laser pulse condition, assuming a 25 ns pulse width. Photogeneration of both (majority) electrons and (minority) holes causes E_F to split into quasi electron and hole Fermi energies, F_n and F_p, respectively. The relatively small perturbation of the electron density during the pulse does not cause F_n to deviate considerably from the dark condition, but the large change in the minority hole concentration shifts F_p close to the valence band edge. The resulting reduction in the built-in potential to ∼0.2 eV should reduce dopant drift toward the growth surface according to Eq. (1).

Equation (1) also shows that the diffusion coefficient plays a large role in the field-generated carry-forward effect. Schubert et al. used a fixed value for D in their model, but this parameter is ultimately influenced by E_F. The formation enthalpy of a defect α with charge q can be expressed as

where the Constant term encompasses the change in the energy of the system upon creation of the defect that is independent of E_F.¹⁶ The second term is associated with the transfer of charge between the defect and the electron reservoir. E_F is referenced from the valence band maximum, and charge q is positive for acceptor-like defects and negative for donor-like defects. High n-type doping concentrations shift E_F toward the conduction band, lowering ΔH for compensating acceptor-like defects. In Si-doped GaAs, these acceptors are well known to include triply charged Ga vacancies,^5,16 which substantially enhance D for atoms on Group III sites and are attributed to GaAs/AlAs superlattice disordering.^5,17

Non-equilibrium generation of free carriers under illumination changes the charge balance of the defect formation reaction and therefore alters the defect formation enthalpy. Specifically, the second term in Eq. (2) is now governed by the quasi Fermi energies, and ΔH changes by an amount $−q2(Fn−Fp)$⁠.¹³ Figure 2(b) shows that this splitting is large under the doping and illumination conditions used here and will act to significantly inhibit the formation of charged defects that facilitate dopant drift in the film.

The abovementioned effects of illumination on the electric field and the diffusion coefficient can qualitatively explain the reduction in the Si carry-forward that was observed experimentally. We should emphasize that the band diagram presented in Fig. 2(b) was calculated only for the illuminated conditions during each laser pulse (i.e., assuming continuous illumination). It is interesting to note, however, that suppression of the carry-forward process was actually observed under pulsed conditions (i.e., illumination occurring only for a fraction of the growth time). Several effects may account for our experimental observations. First, although each pulse has a relatively short duration (25 ns), the pulse width is much longer than the expected bulk carrier lifetime (∼1 ns), and a repetition rate of 10 Hz effectively delivers 10 instances of illumination during the growth of each monolayer (ML) (growth rate used here is approximately 1 μm/h). Second, separation of photogenerated carriers immediately within the first few MLs by the small band bending that persists at the surface during illumination (see Fig. 2) may further increase the minority hole lifetime in this region well beyond the pulse width. Because the dopant atom migration mainly occurs only within the first few MLs with a diffusion velocity < ML/s,⁸ partial reduction of both the field and defect concentration that aid diffusion at this rate may be sufficient to suppress carry-forward to the substantial but incomplete level we observe experimentally. Light-induced changes to MBE-grown semiconductor epilayer morphology under pulsed excimer laser irradiation have also been reported in the literature.¹⁸ Further experimental and theoretical investigations are needed to more fully understand the dependence of illumination time on the carry-forward process.

If, as we argue here, the decrease in the Si carry-forward under illuminated growth conditions is caused by a reduction in Fermi energy-driven mechanisms affecting defect formation and dopant diffusion, then manipulation of the Fermi energy by co-doping could be expected also to affect carry-forward. Indeed, co-doping has been previously suggested to reduce the carry-forward effect by screening the electric field at the growth surface and altering the dopant incorporation rate.^4,6,9 To this end, we have co-doped a structure similar to the one presented in Fig. 1 with a constant background of ∼10¹⁸cm⁻³ C atoms. Hall measurements on C-doped GaAs samples without Si doping indicate that most of these C atoms are activated as acceptors. The growth of the co-doped sample was carried out under dark conditions. In order for it to be effective in modifying the Fermi energy, the C acceptor concentration must be high enough to compensate for a substantial proportion of the electrons supplied by the Si donors. We have therefore partially compensated for the Si donor concentration. Figure 3 shows the SIMS profiles of the C and Si dopants. The characteristic segregation length is now ∼19 nm. This value is just slightly lower than that of the Si-doped sample grown under illumination with the higher laser fluence (60 mJ) and is again significantly reduced compared to that of the Si-doped sample (without C co-doping) grown under dark conditions. The fact that both pulsed photon irradiation and partial C co-doping (under dark growth conditions) produce similar results supports the hypothesis that the photogeneration of carriers, and the subsequent alteration of the Fermi energy, is the primary mechanism for the reduction of Si carry-forward in the case of photon irradiation. We point out that, while co-doping and light exposure can both evidently be used as tools for the control of dopant diffusion, the use of light exposure may be the superior choice in the many cases in which co-doping is not a viable option.

To demonstrate the importance of this technique as an approach for reducing the deleterious effects of dopant carry-forward in devices, we show that it can be used to improve the mobility in inverted modulation-doped quantum wells (QW). High electron mobility transistors (HEMTs) designed with an inverted modulation doped structure, where a 2D electron gas (2DEG) is grown beneath the QW, can have higher breakdown voltages than conventional structures in which the 2DEG is grown above the quantum well. Problematically, dopant carry-forward in such device structures often leads to high concentrations of ionized impurities in the QW channel of the inverted structure, and these ionized impurities degrade the transport properties of the device.¹⁹ We have therefore applied the light-stimulated epitaxy technique to the growth of an AlGaAs/GaAs inverted modulation-doped QW. The structure was as follows: GaAs substrate/GaAs buffer/160 nm AlGaAs/Si delta-doped layer (estimated 10¹³cm⁻²)/10 nm AlGaAs/16 nm GaAs QW/170 nm AlGaAs/10 nm GaAs cap. A laser fluence of 30 mJ was used as the illumination condition. Alloyed Au/Ge contacts were fabricated in a van der Pauw geometry, and mobility measurements were carried out at 77 K in order to better assess the impact of ionized impurities in the QW channel. The measured mobility rose from 6054 cm²/V s under dark growth conditions to 6833 cm²/V s under illumination, representing a ∼12% increase. Based on the Si carry-forward reduction discussed above, we attribute this increase to a decrease in the number of ionized Si dopant atoms driven into the QW through the 10 nm AlGaAs spacer between the QW and the delta-doped layer. The observed improvement is sizeable even in our simple and un-optimized structure, and could be still greater in carefully designed HEMT devices. Most importantly, suppression of dopant carry-forward with light-stimulated epitaxy did not require any alteration in the device structure or co-doping, either of which would have changed the device performance.

In conclusion, we have shown that UV light exposure during MBE growth can reduce dopant carry-forward from Si-doped layers in GaAs, with higher laser fluences leading to greater suppression of carry-forward, and we interpret this effect as arising from changes in the Fermi energy due to the light exposure. Demonstration of the use of this technique to achieve mobility improvements in modulation-doped QW structures suggests that light-enhanced MBE is a promising path toward the design and fabrication of device structures in which—particularly, as the size of devices scales progressively downward—good control of doping concentrations and profiles is necessary.

We acknowledge the financial support of the Department of Energy Office of Science, Basic Energy Sciences under Contract No. DE-AC36-08GO28308. We also thank S.-H. Wei and M. A. Scarpulla for helpful conversations.