InN/InAlN heterostructures for new generation of fast electronics Open Access

InN has been recognized as a far-reaching candidate for ultra-high-speed electronics since almost two decades ago.¹ Indeed, very recently, these expectations have been supported by us by extracting the electron drift velocity (v_d) of 1 × 10⁸ cm/s in 775 nm thick molecular-beam epitaxy (MBE)-grown InN, the highest ever reported value in any semiconductor material.² High v_d is directly related to transistor cut-off frequency f_T ∼ v_d/2πL_G, where L_G is the gate length.³ However, the mentioned extended layer thickness is not applicable in planar transistor structures and finer designs are needed. Unfortunately, because of the large lattice misfit to typically used GaN templates, thinner InN layers suffer from large density of defects and low electron mobility (μ).⁴ Consequently, no InN-based microwave transistor has been demonstrated yet. Therefore, to mitigate the misfit and InN lattice relaxation, in the past, we suggested using an In-rich InAlN buffer layer instead of the GaN one.^5,6 Highly confined electrons of the InN/InAlN quantum well (QW) and strained 5–10 nm thick InN can be expected in this case.^5,6

Elsewhere, we have demonstrated that In-rich InAlN can be grown by metalorganic chemical vapor deposition (MOCVD) directly on sapphire, when the buffer layer is readily relaxed providing N-polarity and In molar fraction over 0.6.^7,8 In our next experiment, 340 nm thick InAlN was followed by the growth of the InN channel, reaching μ over 700 cm²/V s in an InN layer as thin as 20 nm.⁹ XRD diffractions indicated clear signals of both the InAlN buffer and InN channel layers, while by atomic force microscopy (AFM), the established InN root-mean-square (RMS) surface roughness was only about 1.3 nm. InN coalescence occurred at 10 nm film thickness, and the InN background electron concentration (n) was about 2 × 10¹⁹ cm⁻³,⁹ while an optical bandgap (E_G) determined on an alternative, 110 nm thick InN test structure was 0.73 eV.⁸ We note, however, that because of the high In content and narrow E_G, the InAlN buffers tested by us provided degenerate n-type conduction and an un-intentional doping density in a range of 10¹⁹ cm⁻³.^9,10 On the other hand, a highly resistive buffer is needed for the proper operation of a transistor and channel pinch-off. Therefore, in another study, the InAlN buffer resistivity was modified by an acceptor doping by introducing a Cp₂Mg precursor flow during MOCVD growth.¹⁰ This led to a mobility-controlled resistivity increase via electron scattering on Mg-related complexes of InAlN.¹⁰ Still, InN/InAlN QWs with a resistive Mg-doped buffer were not tested until now.

We assume several particular electron scattering mechanisms to influence the performance of suggested QWs. In the case of the un-intentionally doped InAlN buffer, apart from the alloy disorder, μ of InN two-dimensional electron gas (2DEG) will be modulated also by remote donor Coulomb scattering up to a distance of about 5 nm.^11,12 The quality of the channel/buffer interface will determine interface roughness scattering,^11,13 while the InN channel itself will allocate charged dislocation, ionized impurity, and optical phonon scatterings.^4,13−16  Interestingly, optical phonons could be linked also to premature v_d saturation, particularly in the case of high 2DEG density (N_S) via the accumulation of hot phonons.^11,17 We note that a method of substrate nucleation can also influence 2DEG transport properties by changing the quality of grown layers.^18,19 Earlier, we have shown that sapphire nitridation at 1010 °C for 300 s combined with AlN nucleation at 730 °C for 30 s led to N-polar growth.⁷ On the other hand, the dissociation of the thin AlON nucleation layer to Al-polar AlN islands might be responsible for inversion domain formation.¹⁸ Taking this into account and to access all the mentioned scattering phenomena, in this study, we prepared InN (20 nm)/In_0.61Al_0.39N (Mg) QWs at slightly reduced temperature of nitridation and nucleation and with an optional <1 nm thin AlN interlayer between the buffer and the channel. An additional QW structure without an AlN nucleation step has been also prepared. If our proof-of-concept structures validate expectations for InN unique electron transport properties, they might open a way toward a new generation of fast electronics.

Recently, N-polar III-N heterostructures have attracted considerable attention due to several advantages offered to high-electron mobility transistors (HEMTs), compared with their metal-polar counterparts.²⁰ Among them are low-resistance ohmic contacts formed directly on the channel, the scalability of the transistor dimensions not limited by the barrier thickness, and reduced short-channel effects provided by their inherent back barrier.²⁰ Similar advantages could be expected also by realizing N-polar InN/InAlN HEMTs. Moreover, the polarity provides a possibility to grow InN as the last layer with a relatively low polarization at the surface and without risking damage during subsequent growth.⁶ However, the MOCVD of N-polar III-N structures is challenging, particularly because of possible impurity incorporation, such as oxygen, and because of possible inversion domains and threading dislocation formation.^20,21 Vicinal sapphire substrates were successfully applied to handle these issues.^20,21 In the case of InN, however, situation might be more favorable for N-polar InN because of the higher temperature of the growth compared with In-polar InN.²² Consequently, while the density of edge-type dislocations of ∼1 μm thick MBE-grown N-polar InN was about 2 × 10¹⁰ cm⁻² in the broad temperature range of the growth between 500 and 600 °C, it could reach a range of 10¹¹ cm⁻² in In-polar InN if the substrate temperature was kept at 440 °C. Similarly, worse structural quality and lower μ = 1050–1850 cm²/V s were recorded in ∼0.8 μm thick MBE-grown In-polar InN compared with μ = 1300–2600 cm²/V s observed for N-polar InN.²³ A relatively high temperature of 640 °C during MOCVD growth was probably beneficial also for the N-polar 20 nm thick InN grown on GaN, reaching μ = 706 cm²/V s at room temperature (RT).²⁴ 

We used an AIXTRON 3 × 2″ flip-top close-coupled showerhead reactor for MOCVD growth; trimethylindium (TMIn), trimethylaluminum (TMAl), and ammonia $( N H 3)$ were utilized as In, Al, and N precursors, respectively; and N₂ was used as the carrier gas. Sapphire substrates were exposed to an NH₃ flow for 300 s at 990 °C, followed by an AlN nucleation spike grown for 35 s at 710 °C as a part of sapphire nucleation. A 340 nm thick Mg-doped InAlN buffer has been grown at the same temperature using the 30 nmol/min Cp₂Mg precursor flow and in situ baked for 20 min, as described elsewhere.¹⁰ Finally, a 20 nm thick InN channel has been grown at 600 °C.⁹ As an alternative approach to sample A, an AlN spike was also introduced as an interlayer between the buffer and the channel (sample B) or has been completely omitted (sample C). We note that the acceptor doping of the InAlN buffer can lead to the redistribution of Mg atoms to the InN channel layer. This can happen either due to the reactor memory effect²⁵ or via various diffusion processes.^26,27 Elsewhere, it was shown that Mg-doped InN electrical and optical properties are only slightly modified if the Cp₂Mg/(TMI) flow ratio during the InN growth is <4 × 10⁻².²⁸ This may implicate that the properties of our nominally undoped InN are not much affected by diffusion from the grown InAlN [when the flow ratio Cp₂Mg/(TMAl + TMI) = 1.9 × 10⁻³], apart from the fact that the Mg-memory effect might be suppressed by 20 min baking before InN growth.

High-resolution x-ray diffraction (HR-XRD) measurements were performed using a Bruker D8 DISCOVER diffractometer equipped with a rotating Cu anode and operating at 12 kW. Diffractions $0004$ and $11 2 ¯4$ were measured to determine the In molar fraction, lattice parameters, and strain states. We estimate the uncertainty of the measured 2θ angles of the InAlN and InN diffractions and of the signal full width half maximum (FWHM) to be 0.0005°, 0.006°, and 0.02°. The density of dislocations (N_dis) with screw (N_disS) and edge components (N_disE) was evaluated from the x-ray rocking curves.⁹ An NT-MDT NTEGRA Prima atomic force microscope (AFM) in the tapping mode and an FEI Quanta 250 FEG scanning electron microscope (SEM) were used to study the surface morphology of the grown samples. High-resolution transmission electron microscopy (HRTEM) was carried out in an FEI CUBED TITAN microscope equipped with a monochromator and a spherical aberration (Cs)-corrected objective lens. The negative Cs mode, which permits to image also light elements, has been used to determine the polarity of the layers. The analysis of the layers’ crystallographic quality was carried out in the weak beam mode using a JEOL F200 microscope. Photoluminescence (PL) spectra of the InAlN buffer layers were measured at 6.5 K using a 488 nm line of an Ar+ laser as a pump and a silicon photodiode for the detection of PL radiation. Hall transport parameters were measured in the temperature (T) range between room temperature (RT) and 4 K in the standard van der Pauw configuration. The energy band diagram and free electron concentration profile of studied heterostructures were calculated using the Schrödinger–Poisson equation solver.²⁹ 

To assess QWs’ electrical performance, we prepared specific test structures consisting of tiny planar resistors (6 μm length/2 × 16 μm width, fed with ohmic contacts in a coplanar arrangement) formed by plasma etching and Ohmic contact formation. For extracting current–voltage characteristics, we applied 10 ns long voltage pulses with a duty cycle of 0.01 generated by HP 8114A. Consequently, v_d could be calculated as I/qN_S, where I is the current density and q is the electron charge.² Pulsed probing was necessary to eliminate the self-heating and premature burnout of structures.³⁰ 

Figure 1 shows the XRD 0002 2θ/ω diffractions of the investigated samples. Table I summarizes the material parameters determined also by supplementary XRD scans, with x_InN representing the content of InN and ɛ₁ representing the in-plane strain, with full width half maximum (FWHM) of corresponding diffractions and with calculated dislocation densities. x_InN of the InAlN buffer layer was found to be between 60% and 61%, and x_InN of the InN channel between 98% and 99% may reflect ambiguity of the extraction method and/or possible contamination of the layer.

Based on the data of Table I, we conclude that by inserting the AlN interlayer between the buffer and the channel of sample B, InN crystallographic quality has improved compared with sample A. This was reflected by a smaller InN diffraction peak FWHM and by lower N_dis, down to about 6.5 × 10⁹ and 38 × 10⁹ cm⁻² of N_disS and N_disE components, respectively. Simultaneously, ɛ₁ of InN was reduced down to −0.5%, representing relaxation over 90%.⁹ This is documented also in Fig. 1 by the sample B InN diffraction maximum shift closer to InAlN compared with sample A. On the other hand, in sample C, both InN and InAlN layers’ crystallographic quality deteriorated if the AlN nucleation spike was omitted. This is reflected by the reduced diffraction signal shown in Fig. 1 and by up to 65% increase in N_dis listed in Table I, compared with samples A and B. Nevertheless, Table I indicates that the AlN spike used neither as the interlayer between the buffer and the channel nor as a part of the nucleation has any influence on x_InN, which is about 61% in the buffer layer and about 99% in the channel.

A higher level of crystallographic disorder of sample C is also evidenced by the PL experiment shown in Fig. 2, with a broader FWHM of about 279 meV and reduced intensity of PL radiation, likely due to the additional non-radiative defects, compared with sample A. We note an invariant position of the PL signal maxima at 1.58 eV, in agreement with determined x_InN ∼ 61% and expected bandgap bowing.³¹ 

For better understanding of InN/InAlN QW growth and performance and the role of particular AlN spikes, we further performed SEM, AFM, HRTEM, Hall, and I–V experiments. InN surface inspections by SEM are shown in Fig. 3. In sample A, we noticed a relatively compact surface but with occasional cracks. On the other hand, the surface of sample B, which is less continuous with occasional voids, contains no cracks. We link this phenomenon to the AlN interlayer of sample B, providing faster InN relaxation, as indicated also by reduced ɛ₁ in Table I. On the other hand, sample B could probably suffer from InN less mobile adatoms or less dense nucleation sites. Surprisingly, the highest surface quality can be suggested for sample C without any AlN spikes, even though with the highest N_dis off the less optimized growth. We speculate that dislocations in the buffer of sample C could provide additional relaxation sites for InN, even though as revealed below, high N_dis deteriorates electrical performance. Our conclusions were further supported by AFM 2 × 2 μm² scans as shown in Fig. 4. We noticed a similar root-mean-square (RMS) roughness of about 3.2 nm for all samples; however, InN terraces that indicate the 2D growth mode of highly mobile adatoms became less obvious for sample B. Moreover, tiny islands were observed on the surface of sample B, which is probably also linked to the presence of the AlN interlayer and the modified InN growth mode.

HRTEM experiments shown in Fig. 5 provided vertical inspections. Sample A in Fig. 5(a) suggests a well-defined InN/In_0.61Al_0.39N QW interface, with only occasional misfit dislocations. A similar investigation of sample B, shown in Fig. 5(b), proved the additional presence of approximately two monolayers of the AlN interlayer; however, misfit dislocations became less detectable. Nevertheless, all samples demonstrated N-polarity without inversion domains, as exemplified also by the most magnified InN image shown in Fig. 5(c).

To study the role of all above-mentioned structural peculiarities in the InN/InAlN QW electron transport properties, next we proceeded with the Hall effect analysis. We note that even though the present experiment uses resistive Mg-doped InAlN buffers, some parasitic conduction cannot be completely excluded. Therefore, to determine conduction through the InN channel alone, the initial characterization of sole buffer layers followed by a parallel conduction analysis of the whole QW might be needed, similarly as performed elsewhere for the QW with undoped InAlN buffers.⁹ As for the buffer layer of samples A and B, here we use our earlier data, i.e., μ ∼ 5.6 cm²/V s and n ∼ 6.3 × 10¹⁹ cm⁻³ at RT, down to μ ∼ 4.7 cm²/V s and n ∼ 6.5 × 10¹⁹ cm⁻³ at 4 K.¹⁰ The analysis of the buffer layer grown without AlN nucleation (sample C) reveals an additional ∼15% decrease in μ and ∼50% decrease in n (not shown), which provided an increase in the buffer sheet resistance from about 0.7 kΩ/□ of samples A and B to 1.7 kΩ/□ of sample C. Taking this into account and after finalizing the above-mentioned parallel conduction analysis, we could extract InN/InAlN QW μ and N_S quantities of all samples, as shown in Fig. 6. We may ascertain that μ of all samples increases with decreased T, with a tendency to saturate at T < 100 K. This suggests ionized impurity, charge dislocation, and interface roughness scattering at low T, with gradual contribution from the polar optical phonon scattering at higher T, similarly as observed for InN elsewhere.^13,15 From the other point of view, when comparing samples A and B, we noticed a substantial (by ∼30%) improvement of μ if the InN channel was separated from the buffer by using the AlN interlayer. Therefore, specifically for our case of the ternary InAlN buffer layer with high unintentional doping, we may list InAlN alloy disorder and remote donor scattering among other important aspects of proposed QWs. Interestingly, for sample C, μ remains high even though crystallographic quality suffered from missing AlN nucleation spike, documented by the highest N_dis (see Table I) and degraded PL, as shown in Fig. 2. On the other hand, elsewhere it was suggested that dislocations may act also as acceptor states.¹⁶ If we assume the same, dislocations in sample C may compensate for the unintentional doping of InAlN, as suggested by the above observed 50% drop in n. Consequently, reduced remote donor scattering can also explain the high μ of sample C.

We noticed almost T-independent N_S values, as shown in Fig. 6(b), which were ∼4–5 × 10¹³ cm⁻² for samples A and C without the AlN interlayer. N_S surprisingly dropped to ∼3 × 10¹³ cm⁻² for sample B, even though better 2DEG confinement should be provided with the interlayer insertion.⁵ Next, for a better understanding of all phenomena, we calculated the energy band diagrams and electron concentration profiles of suggested QWs, as shown in Fig. 7. We note that the calculations of sample C are not shown here as being almost identical to sample A. At the surface of all samples, our model reflects the expected Fermi level (E_F) pinning of ∼1.6 eV above the valence band (E_V) and consecutive electron accumulation.³² Strongly confined and from the buffer separated 2DEG is shown in Fig. 7(b) for sample B. That is in agreement and explains our earlier comments on reduced remote donor and alloy disorder scatterings in this case. Another interesting aspect of sample B is that the AlN interlayer provides strong lifting of the buffer E_C above E_F close to the interface [see Fig. 7(b)]. It may on one side provide a beneficial ∼5 nm thick energy barrier for the electron injection from the InN channel to InAlN but also an ambiguity in justifying applied parallel conduction analysis. Consequently, N_S values of sample B shown in Fig. 6(b) might be underestimated.

In Fig. 8, we show the I–V characteristics of InN/InAlN QW test resistors obtained by using 10 ns long pulses. High current densities up to 17 A/mm were obtained, with a tendency to saturate at an electric field (E_sat) of ∼35 kV/cm and with a breakdown field (E_break) at ∼70 kV/cm. We note that the observed E_sat corresponds well with earlier Monte Carlo simulations on electron transport in InN,^33,34 while to our knowledge, there are practically no data on E_break in InN. We consider the observed E_break to be very promising, compared to a significantly lower E_break of 20 kV/cm of an analogous “high-speed” semiconductor material of InAs.³⁵ Moreover, taking into account the dissipated power of ∼20 W, duration of the pulse, resistor area, and our previous self-heating extractions in GaN structures,³⁰ we can estimate the temperature to reach >400 °C at the end of the pulse. This already approaches the thermal stability limit of InN,³⁶ indicating that even higher E_break could be expected with better cooling. We note also that despite high N_S, the lowest I ∼ 10 A/mm has been observed for sample A. The reason for this is unknown, and we speculate that enhanced accumulation of hot phonons due to high N_S combined with high n in the buffer of sample A might be responsible.

In Fig. 9, we show the calculated v_d of all samples; the buffer conduction was subtracted from I values before each v_d extraction. Samples B and C behave rather similarly, reaching the best v_d ∼ 2 × 10⁷ cm/s. We account for these comparable improvements to AlN spacer insertion in sample B and to reduced n in the buffer of sample C. None of these improvements were applied in sample A, and, thus, only limited v_d of ∼1 × 10⁷ cm/s was extracted in this case. Nevertheless, obtained v_d ∼ 2 × 10⁷ cm/s is very promising, taking into account much lower v_d ∼ 1 × 10⁶–1 × 10⁷ cm/s of AlGaN/GaN QWs.³⁷ Still, we were not reaching v_d ∼ 1 × 10⁸ cm/s as reported earlier by us for 770 nm thick InN.² This might account for several phenomena of thin films, such as pronounced surface effects, higher background doping, higher N_dis, and much lower μ compared to the earlier reported value of thick InN, i.e., 1940 cm²/V s at RT. We may also point out some ambiguity in v_d extraction related to different conditions of the Hall effect measurement, performed at ∼200 meV bias and in the steady state, versus up to 70 kV/cm transients of pulsed I–Vs. Earlier, we suggested that carrier injection in AlGaN/GaN QWs might lead to gradual surface charging and v_d reduction,³⁷ which was not analyzed in our study. Similarly, surface/interface electron accumulation in InN (shown in Fig. 7) might also lead to apparent v_d reduction if additional parallel conduction analysis is neglected.² 

In conclusion, we report the proof-of-concept N-polar InN/In_0.61Al_0.39N QW heterostructures. 20 nm thick InN has been MOCVD grown on an Mg-doped InAlN buffer with an optional AlN spike as an interlayer between the InAlN buffer and the InN channel or as a part of InAlN nucleation on sapphire. AlN inclusion as an interlayer between the buffer and the channel brought clear improvements, reaching N_S ∼ 4 × 10¹³ cm⁻² and μ ∼ 620 cm²/V s at RT, and N_disS and N_disE of InN as low as ∼6.5 × 10⁹ and 38 × 10⁹ cm⁻², respectively. Remote donors and alloy disorder have been determined as major scattering sources originating in the InAlN buffer, while dislocations and optical phonon scattering were suggested to be decisive in the InN channel itself. If the nucleation AlN spike was skipped, InAlN structural quality deteriorated; however, buffer resistivity increased. Pulsed I–V characterizations indicated the current saturation of ∼35 kV/cm, breakdown field of ∼70 kV/cm, and v_d of InN electrons to reach at least 2 × 10⁷ cm/s. Our findings prove the potential of the InN/InAlN QW heterostructures for the future generation of fast electronics. Further developments call for improved InAlN epitaxy with less free electrons and/or better donor compensation, for InN with less N_dis, and for surface accumulation suppression.

This work was supported by VEGA under Grant Nos. 2/005/22 and 2/0068/21 and by APVV Agency under Grant No. APVV-21-0008.

The authors have no conflicts to disclose.

J. Kuzmik: Conceptualization (lead); Funding acquisition (lead); Methodology (equal); Project administration (lead); Resources (lead); Supervision (lead); Validation (equal); Writing – original draft (lead). R. Stoklas: Data curation (lead); Investigation (lead). S. Hasenöhrl: Data curation (equal); Investigation (equal). E. Dobročka: Data curation (equal); Investigation (equal). M. Kučera: Data curation (supporting); Investigation (supporting). P. Eliáš: Data curation (supporting); Investigation (supporting). F. Gucmann: Data curation (supporting); Investigation (supporting). D. Gregušová: Data curation (supporting); Investigation (supporting). Š. Haščík: Data curation (supporting); Investigation (supporting). A. Kaleta: Data curation (supporting); Investigation (supporting). M. P. Chauvat: Data curation (supporting); Investigation (supporting). S. Kret: Data curation (supporting); Investigation (supporting). P. Ruterana: Data curation (lead); Investigation (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.