1. Introduction

Recently, an impressive progress in wide band-gap III-nitride research has led to the succesful development of blue light-emitting diodes (LEDs) and lasers Reference Mohammad, Salvador and Morkoç[1] Reference Nakamura and Fasol[2]. These realisations have been possible because of the improved crystal quality obtained by Metal Organics Vapor Phase Epitaxy (MOVPE), and especially because of the achievement of GaN p-type doping. Improved growth process (use of buffer layers, purification of carrier gases, and so on) have allowed the reduction of the electron residual concentration.

The mechanism of acceptor compensation which prevents obtaining low resistivity GaN:Mg by MOVPE is likely related with Mg-H complexes created during MOVPE growth Reference Nakamura and Fasol[2]. Because of the large activation energy and the formation of these complexes, a post-growth treatment of the p-type layer is also needed, either by low-energy electron-beam irradiation Reference Amano, Kito, Hiramatsu and Akasaki[3] or by thermal treatment in N₂ ambient Reference Nakamura, Mukai, Senoh and Iwasa[4]. In spite of activation treatments Reference Gotz, Johnson, Walker, Bour and Street[5], the presence of deep level centers or complexes have been demonstrated in highly doped p-type GaN by deep level transient spectroscopy Reference Hacke, Nakayama, Detchprohm, Hiramatsu and Sawaki[6], photocapacitance Reference Sanchez, Calle, Basak, Tijero, Sánchez-García, Monroy, Calleja, Muñoz, Beaumont, Gibart, Serrano and Blanco[7], and photoluminescence (PL) Reference Gotz, Johnson, Walker, Bour and Street[5] Reference Sanchez, Calle, Basak, Tijero, Sánchez-García, Monroy, Calleja, Muñoz, Beaumont, Gibart, Serrano and Blanco[7] Reference Smith, Chen, Lin, Jiang, Salvador, Sverdlov, Botchkarev, Haffouz, Morkoc and Goldenberg[8] Reference Beaumont, Leroux, Laugt, Lorenzini, Nakamura, Gibart and Hubert-Pfalzgraf[9]. However, the donor or acceptor nature of these defects is still under debate.

PL spectra of Mg-doped layers show a broad band around 430 nm (2.8 eV) involving these deep centers. The dominant emission in the electroluminescence (EL) spectra of p-n homojunction MOVPE GaN LEDs may actually take place at this wavelength Reference Mukai and Senoh[10] Reference Golenberg, Zook and Ulmer[11] Reference Schauler, Kirchner, Mayer, Pelzmann, Eberhard, Unger, Kamp, Beaumont and Ebeling[12] Reference Haffouz, Gibart, Leroux, Lorenzini, Calleja and Muñoz[13], thus resulting in a much lower efficiency than the intrinsic capabilities of the material.

For comparison, Nichia's codoped InGaN:Zn, Si/AlGaN double heterostructure (DH) diodes Reference Casey, Muth, Krishnankutty and Zavada[14] Reference Perlin, Osinski, Eliseev, Smagley, Mu, Banas and Sartori[15] Reference Eliseev, Perlin, Furioli, Sartori, Mu and Osinski[16] Reference Basrur, Choa, Liu, Sipior, Rao, Carter and Chen[17] also show a deep emission. Nevertheless, due to the use of a different p-dopant and to an efficient activation, these devices provide intense blue and blue-green light Reference Nakamura, Mukai and Senoh[18], because their operation is not much affected by the deleterious effect of deep levels.

Another way to avoid the effect of deep levels in the EL spectra is the fabrication of p-i-n LEDs Reference Schauler, Kirchner, Mayer, Pelzmann, Eberhard, Unger, Kamp, Beaumont and Ebeling[12], in which the energy of the main emission is close to the band-gap of the active material. On the other hand, the EL of GaN homojunctions grown by Molecular Beam Epitaxy is not much affected by those deep levels, due to a higher activation even without any post-growth treatment Reference Vaudo, Goepfert, Moustakas, Beyea, Frey and Meehan[19] Reference Grandjean, Massies, Leroux and Lorenzini[20]. In these devices, the dominant emission at 390 nm also involves shallow acceptors.

In this work, the electrical and optical properties of homojunction GaN LEDs grown by MOVPE will be studied at different temperatures and under different biasing conditions, in order to investigate the nature of the transitions involved in the EL spectra. After a brief description on the fabrication of the devices and their electrical properties, the EL spectral response will be analyzed and related to the light-current characteristics. Next section is devoted to the time response of the blue and UV emissions, by pulsed biasing with controlled pulse durations and frequencies. The results will be discussed and compared with theoretical calculations, in order to discern between the character of the deep levels, either donor or acceptor. Conclusions are summarized in Sect. 7.

2. Growth and LED fabrication

p-n homojunctions were grown on basal plane sapphire by MOVPE. Sources for GaN growth were trimethyl-gallium and NH₃, as described in detail elsewhere Reference Beaumont, Vaille, Boufaden, el Jani and Gibart[21]. A low temperature GaN buffer was used, on top of which Si and Mg doped GaN layers were deposited, 1.4 μm and 0.5 μm thick respectively. Si was efficiently introduced in the vapour phase as SiH₄ diluted in H₂, to provide free electron density in the 10¹⁸ cm⁻³ range. The p-type precursor was bismethyl-cyclopentadienhyl Mg (MeCp₂Mg). Mg incorporation up to 10²⁰ cm⁻³and thermal treatment in N₂ atmosphere in the MOVPE growth chamber Reference Beaumont, Leroux, Laugt, Lorenzini, Nakamura, Gibart and Hubert-Pfalzgraf[9] Reference Haffouz, Gibart, Leroux, Lorenzini, Calleja and Muñoz[13] allow for hole concentrations close to 5.10¹⁷ cm⁻³.

Circular mesas with 320 μm diameter were performed by dry etching through the p-type layer Reference Beaumont, Leroux, Laugt, Lorenzini, Nakamura, Gibart and Hubert-Pfalzgraf[9] Reference Haffouz, Gibart, Leroux, Lorenzini, Calleja and Muñoz[13] to expose the GaN:Si layer, as shown in the cross-sectional view of the device (Figure 1). Ni/Au bilayer contacts were thermally deposited on both p and n layers. The devices were encapsulated on TO-5 headers, drilled with a 3 mm hole to allow detection from the n-type side through the sapphire substrate. Isolated devices were wirebonded for electrical contacts.

Figure 1.

Cross-section view of the p-n diode.

3. Electrical properties

The capacitance-voltage (C-V) and current-voltage (I-V) characteristics of the devices were measured with an HP4284A LCR-meter and an HP4155A semiconductor parameter analyzer, respectively. Diode capacitances are approximately 200 pF. An effective impurity concentration of 4.10¹⁸ cm⁻³is determined from the slope of the 1/C²-V characteristic. The space charge region (scr) is estimated to be shared between the n- and p-type regions.

Much information on the LED behaviour is usually obtained from a careful study of the I-V characteristics. A typical curve at room temperature (RT) is shown in Figure 2, where two different regimes can be observed depending on the applied bias:

Figure 2.

Typical I-V characteristic in a log-log scale. Two regimes of tunneling and space charge limited current are identified at low and high biasing, respectively (see text).

i) At low voltages, 0.1 V < V < 1.5 V, once the effect of leakage resistance (Rs ≈10 MΩ) can be neglected, the current increases exponentially with V as

where α≈ 4.86 V⁻¹at RT. In fact, α is almost insensitive to temperature in the 4-300 K range. This dependence, usually observed in wide bandgap p-n diodes, either GaN homojunctions Reference Fedison, Chow, Lu and Bhat[22] or InGaN DH LEDs Reference Casey, Muth, Krishnankutty and Zavada[14] Reference Perlin, Osinski, Eliseev, Smagley, Mu, Banas and Sartori[15] Reference Eliseev, Perlin, Furioli, Sartori, Mu and Osinski[16] Reference Martil, Redondo and Ojeda[23], is attributed to tunneling current assisted by shallow or deep traps.

ii) For higher current injection, V>2.5 V, the characteristic behaves as

where V_o= 2.0 V at RT and increases when cooling the device (V_o= 4.75 V at 4K). This behaviour suggests conduction through an isolator with traps once the diode is conducting, i.e., space-charge-limited current is the dominant mechanism at high voltages. This dependency is expected for a wide band-gap material like GaN, and has been reported for ion-implanted p-n junctions in β-SiC Reference Edmond, Das and Davis[24]. In our case, the isolator is likely to lie beneath the non-ideal p-ohmic contact, and its length is estimated to be ~150 Å. In this regime, series resistance decreases with voltage from 50 Ω at 5 V down to 10 Ω at 10 V.

Fedison et al. Reference Fedison, Chow, Lu and Bhat[22] have reported a similar behavior for low currents in homojunction LEDs as those observed in our devices. However, they can observe a conventional Shockley diffusion current before early saturation of the current occurs. Unfortunately, due to this limitation, the I-V characteristics in our devices cannot provide information on the current mechanisms involved in the junction when the diode is emitting light (voltages higher than 3 V), so that either tunneling, generation-recombination and/or diffusion processes may take place.

4. Electroluminescence spectral response and L-I characteristics

In order to investigate on their electroluminescent properties, the diodes were excited by either a Keithley 220 dc programmable current source, and an HP8116A function generator able to provide voltage pulses with a 50 MHz bandwidth. EL spectra were measured with a high resolution HTR-1000 Jobin Yvon monochromator and a GaAs photomultiplier, using lock-in techniques. White light was also detected by a UV-enhanced Si photodetector.

The evolution of EL spectra under increasing voltage has been studied for dc and ac excitations, both producing a similar behaviour except for the more pronounced thermal effects under dc operation. EL spectra recorded at different ac biasing conditions are shown in Figure 3. At low bias, a broad blue-shifting visible band is observed, while a sharp UV peak (370 nm, FWHM=15 meV) becomes dominant at high voltages. In addition, the typical yellow band (YB) at 560 nm, completely absent at low bias, is enhanced by current injection. The oscillatory pattern seen in the spectra is due to Fabry-Perot interference. The lineshape features an abrupt decrease at high energies due to internal optical absorption. EL spectra have been recorded from the p-side and the n-side of the junction, the latter been more intense because the light is mainly produced below the upper contact. A slight reabsorption of UV light by p-GaN also takes place, since the UV-to-visible ratio is somewhat larger for n-side emission.

Figure 3.

EL spectra performed under 1 MHz pulse biasing with several pulse amplitudes.

The visible emission, which appears at 3 V excitation as a broad and weak band centered around 550 nm, increases in intensity and shifts down to 430 nm for high injection (or pulse amplitude). It should be clear that this recombination involves deep centers and not shallow acceptors, since no emission at 390 nm is observed. The transition does not behave as a typical DAP, since it experiences a very large shift, indicating that the defects are located in a broad, multi-level band. The energy shift is due to saturation of the lowest energy levels, either if they behave as deep donors or as deep acceptors. The shifting behaviour of the visible band under increasing bias has also been observed in InGaN Reference Casey, Muth, Krishnankutty and Zavada[14] Reference Perlin, Osinski, Eliseev, Smagley, Mu, Banas and Sartori[15] Reference Eliseev, Perlin, Furioli, Sartori, Mu and Osinski[16] and homojunction MOCVD GaN LEDs Reference Nakamura, Mukai, Senoh and Iwasa[4] Reference Golenberg, Zook and Ulmer[11] Reference Casey, Muth, Krishnankutty and Zavada[14]. In most cases, it is interpreted as due to tunneling transport across the junction through one or several intermediate states.

The broad blue-violet emission has also been observed by low temperature PL in highly Mg-doped GaN layers Reference Sanchez, Calle, Basak, Tijero, Sánchez-García, Monroy, Calleja, Muñoz, Beaumont, Gibart, Serrano and Blanco[7]. This band blue-shifts by around 100 meV for a 2-decade increasing excitation intensity Reference Beaumont, Leroux, Laugt, Lorenzini, Nakamura, Gibart and Hubert-Pfalzgraf[9] Reference Haffouz, Gibart, Leroux, Lorenzini, Calleja and Muñoz[13], in the same way as observed for the visible emission in EL spectra for increasing current. Whatever the current mechanism involved, the blue recombination process takes place at the p-side of the junction.

EL spectra recorded at different temperatures are shown in Figure 4 for a relatively low current injection (~10 μA at 4K, ~10 mA at RT). The dominant emission at 4K peaks at 388 nm (3.20 eV), being due to recombination between residual donors and the shallow acceptor (DAP). However, this emission is thermally quenched and the band at 430 nm (2.88 eV) involving deep levels is observed at RT. This significant change in the LED operation might be explained by the ionization of the shallow acceptor and compensation by the presence of donors, either residual or diffused Si.

Figure 4.

dc EL spectra at several temperatures, under low bias to avoid heating of the device.

The light-current characteristics have been obtained both from a direct measurement of the light (full symbols in Figure 5) and by integration of the EL spectra obtained at several voltages (open circles). In agreement with the spectral evolution, the L-I curve shows a superlinear dependence at low current (L ~ I ^m, with m=2.3), which becomes almost linear (m=0.9) when the defects are saturated. This behaviour has also been observed in Nichia LEDs Reference Martil, Redondo and Ojeda[23]. The change in slope corresponds to the bias for which the blue emission is stacked at 430 nm, and its intensity is also saturated. The relative efficiency, dL/dI vs I, is shown in the inset to Figure 5. A slight decrease is observed for I>50 mA, which likely corresponds to some heating of the LED as also deduced from the small red-shift of the UV band at increasing currents.

Figure 5.

L-I log-log characteristics. Inset: Relative efficiency characteristics, dL/dI vs I. Full and open symbols represent the light measured by a Si photodetector and the integrated EL spectra like those shown in Figure 3, respectively. Solid curves are linear fits in the low and high current injection regimes.

5. Electroluminescence time response

A more careful study of the EL spectra is needed in order to elucidate the nature of the deep levels. Some EL experiments have been performed under different conditions of pulsed excitation, by changing the pulse duration t_ON or the duty cycle δ at fixed pulse amplitude.

Figure 6 shows EL spectra recorded under pulsed excitation, with 7 V pulses of increasing duration. Each spectrum has been renormalized to give the average spectral intensity during the time while the pulse is in the ON state. It is assumed that the device was switched off for enough time before the pulse is active (long t_OFF). The small red-shift from 370 to 375 nm observed for increasing t_ON is likely due to heating of the device when it is emitting for a long time. As observed in Figure 6, the UV emission is constant with time, indicating the continuous supply of both electrons and holes. On the contrary, the blue emission at 430 nm shows a clear decay, which is attributed to an inhibition of this recombination path due to a long relaxation time of the acceptor. The EL intensity of this band is plotted in Figure 7 vs the pulse duration, from which the exponential fit

Figure 6.

EL spectra recorded at 7V amplitude pulses of different duration.

Figure 7.

Time evolution of the violet EL intensity.

provides the saturation time of the acceptor level τ_sat≈ 135 ns.

In order to obtain direct information on the relaxation time of the acceptor τ₁₀, the diode has been excited by pulses of fixed duration and variable frequency f or duty cycle δ = f. t_ON/ (t_NO + t_OFF). EL spectra obtained under 7 V, 200 ns pulses for frequencies between 200 kHz and 4 MHz (duty cycles δ from 0.04 to 0.8, respectively), are shown in Figure 8. The measured intensity has been corrected to give the average light emitted in each pulse. The UV radiation does not depend on δ (i.e. on the rest time t_OFF), whereas the visible intensity clearly decreases with δ. In other words, a long time is needed so that the involved acceptors are recharged with holes. This relaxation time is obtained by fitting the intensity vs t (= t_OFF) to the expression

Figure 8.

EL spectra recorded at 7V amplitude, 200 ns pulses, at several frequencies (duty cycles).

as illustrated in Figure 9, with the result τ₁₀ ≈ 1.06 μs.

Figure 9.

Violet EL intensity observed after different rest time t_OFF.

6. Discussion

The deep levels which dominate the EL response (broad and red-shifting violet-blue band) are due to the presence of Mg-related defects, which form a continuum of states, as suggested above, whereas the UV emission is grossly due to band-to-band recombination. The UV-to-blue ratio increases with pulse amplitude, as shown in Figure 3, pointing to the saturation of the deep defects involved in the visible emission. Additionally, for a fixed amplitude, the longer the pulse the higher UV/visible ratio, as observed in Figure 6. In other words, the injected carriers are trapped first by the states which generate the blue band, as also reported for the Nichia DH-LED by Basrur et al Reference Basrur, Choa, Liu, Sipior, Rao, Carter and Chen[17]. Besides, the UV-to-visible emission ratio at constant pulse duration is optimized by using high duty cycles (Figure 8), in order to block the visible band emission. All this requirements for the amplitude, duration and duty cycle of the pulsed biasing in order to improve the UV emission contribute to heat the device.

Basrur et al. describe the visible and UV generation mechanism by a four-level system, which implies that visible emission involves transitions from donors to acceptors in the depletion layer. Lifetimes of 5 and 115 ns for the UV and blue bands are obtained, with a relaxation time for the acceptor as low as 0.8 ns (more than one thousand times smaller than in our device). Therefore, the visible emission in these candela-class DH-LEDs involves shallower levels instead than the deep centers introduced by Mg-doping.

7. Conclusions

A detailed characterization of the electrical and optical properties of MOVPE GaN p-n homojunction light emitting diodes has been performed. The EL spectra consist of a blue-violet emission (430 nm) involving acceptor deep states generated by Mg doping located 0.55 eV above the valence band, a UV band-to-band recombination at 370 nm, and the YB. The blue and YB emissions take place at the p-side and n-side of the p-n junction respectively, whereas the UV recombination may be produced at both regions. The p-type emissions can be explained by a simple 3-level model, with intrinsic recombination times τ′₂₁ ≈ 70 ns and τ₁₀ ≈ 1.0 μs. The UV-to-visible emission ratio is optimized by using intense and long pulses, at high duty cycles, in compromise with heating of the device. The UV emission is therefore much more efficient and monochromatic than the blue-violet emission. In order to optimize the LED performance, a better activation of the p-dopant species or a new dopant (like Zn) are needed. Furthermore, the optimum blueLED should be achieved by using an InGaN active layer with the proper In content which gap energy matches the required wavelength emission.