1. Introduction

Direct bandgap III-V compound semiconductors are known for their robust and high optical performance and widely used in visible light-emitting diodes (LEDs) and lasers for red, green, blue (RGB) displays and solid state lighting. Recent developments in epitaxy of InGaN blue and green LEDs on Si (111) and AlGaInP red LEDs on Si (100) [1, 2] would enable vertically-stacked RGB LEDs to be deployed in an 8 inch wafer-scale manufacturing environment. Our progress for integrating devices from different substrates is described in Reference [3]. The integration of III-V optoelectronics into microelectronic circuitry also allows other novel applications, such as on-chip optical communications and complementary metal-oxide semiconductor (CMOS)-controlled LED arrays [4].

Due to the large lattice mismatch of 4.1% between GaAs and Si, many applications based on GaAs substrate, such as GaInP/GaAs solar cells, AlGaAs/GaAs laser diodes and AlGaInP/GaAs LEDs are difficult to integrate on a Si substrate. Germanium (Ge) is a good intermediary material between III-V and silicon, because Ge is closely lattice-matched with GaAs. A GeSi graded buffer on Si with a low threading dislocation density (TDD) of 1 × 10⁶ cm⁻² has been developed, which demonstrates the possibility to grow high quality III-V compounds on Si [2]. However, in order to achieve low TDDs, a rather thick GeSi graded buffer (typically 10 μm) is a necessity. This makes the planar integration of III-V optoelectronic devices with CMOS circuitry difficult. The cross-hatched patterns formed during the buffer grading lead to surface roughening. Thus, chemical mechanical polishing (CMP) is needed at the grading composition of Ge_0.5Si_0.5, which interrupts the continuous buffer growth [2].

The direct epitaxial growth of Ge-on-Si (Ge/Si) using a low and high temperature two-step method reduces the buffer thickness to 1 to 2 μm. The surface roughness of the Ge/Si is less than 1 nm, thus a CMP process is not required during the Ge buffer growth. The reported TDD values of Ge/Si after post-annealing are between 3 × 10⁷ cm⁻² to 5 × 10⁷ cm⁻² due to the large lattice mismatch [5]. Though InGaAs high-mobility electron transistors (HEMT) on Ge/Si have been demonstrated [6], minority carrier devices are more sensitive to TDDs and reliable performance of III-V optoelectronic devices on Ge/Si is still questionable. In KH. Lee, et al.’s recent report, the TDD of Ge/Si is reduced to ~4 × 10⁶ cm⁻² using heavy arsenic (As) doping of 4 × 10¹⁸ cm⁻³ in the Ge seed layer on Si [7]. These results show the possibility of using this substrate for the integration of III-V semiconductors on Si.

In this work, we demonstrate the epitaxy of an GaInP/AlInP red LED on an As-doped Ge/Si substrate. As the TDD is critical to the optoelectronics performance, we also chose two alternative substrates for comparison: an undoped Ge/Si substrate and a bulk Ge substrate. The LED junction and GaAs buffer were deposited on the selected substrates in a single epitaxial run. The electrical and optical performance of these samples was studied.

2. Experimental procedures

Ge/Si substrates and Ge (001) substrates were 6° off-cut toward the [110] direction, which was necessary to supress the formation of anti-phase boundaries at the interface between III-V and Group IV materials. The Ge buffer and the III-V materials were grown using an Aixtron Crius metal-organic chemical vapor deposition (MOCVD) reactor. The As-doped Ge/Si substrate had a high n-type doping concentration of 4 × 10¹⁸ cm⁻³ in the Ge seed (~10 nm thick), initiated on the Si surface. The doping concentration was gradually reduced to ~1 × 10¹⁷ cm⁻³ during the subsequent growth of the 1.5 μm Ge buffer. For the undoped Ge/Si substrate, the Ge seed layer and the buffer layer were not intentionally doped.

The Ge/Si substrate samples were prepared in the size of 3 × 3 cm². The substrate samples were cleaned in 10% HF for 2 min, and then they were loaded with epi-ready bulk Ge wafer to the MOCVD reactor. Before the start of the III-V growth, the substrates were baked in H₂ at 630 °C for 10 min in order to form a double-step surface on Ge. Then the temperature remained at 630 °C for III-V growth. After a 30 s AsH₃ pre-exposure, TMGa precursor was introduced to initiate a 100 nm thick GaAs nucleation layer at a reactor pressure of 400 mbar to provide an AsH₃ partial pressure of 5 mbar. Subsequently, the reactor pressure was reduced to 100 mbar and a 1.5 μm high quality GaAs buffer was grown to avoid Ge auto-doping. The GaAs buffer was Te-doped for the LED n-contact with a Te doping concentration of 5 × 10¹⁸ cm⁻³. The p-i-n junction of the GaInP/AlInP LEDs was subsequently grown on the n-GaAs buffer. Bulk Ga_0.51In_0.49P has a bandgap energy of 1.89 eV, corresponding to an emission wavelength of 656 nm. A 300 nm intrinsic Ga_0.51In_0.49P active layer was sandwiched between the lattice-matched n-type and p-type Al_0.52In_0.48P cladding layers. The 300 nm n-AlInP clad was Te-doped with a doping concentration of 1 × 10¹⁸ cm⁻³ and the 500 nm p-AlInP clad was Zn-doped with a doping concentration of 5 × 10¹⁷ cm⁻³. A 100 nm p-type GaAs layer was deposited on the p-AlInP cladding layer as a capping layer to prevent oxidation. Ni/Ge/Au and Ti/Au metal contacts were deposited by electron beam deposition on the n-GaAs buffer and p-GaAs capping layer, respectively. The Ni/Ge/Au metal contact was annealed at 390 °C for 40 s in a rapid thermal annealing furnace. Then a 70 nm silicon nitride (Si₃N₄) layer was deposited at 300 °C in a plasma-enhanced chemical-vapour deposition (PECVD) tool to passivate the device mesa. To ensure that the deposited GaInP active layers have similar material quality, III-V compounds were grown in the same run on the different substrates.

3. Results and discussion

The TDD of undoped Ge/Si and As-doped Ge/Si measured by etch-pit density (EPD) were 6 × 10⁷ cm⁻² and 5 × 10⁶ cm⁻², respectively. The TDD of bulk Ge substrate was expected to be extremely low. The introduction of a high As-doping in the Ge seed reduced the TDD of Ge/Si by one order of magnitude. The mechanisms of the TDD reduction is not yet well understood, one possible explanation is the enhanced Si-Ge interdiffusion due to arsenic doping in Ge buffer [8]. Electron beam induced current (EBIC) measurements were conducted to determine the non-radiative recombination sites on the top region of the LED p-i-n structures. The EBIC images of Fig. 1(a) and 1(b) show that the TDDs of the LED on As-doped Ge/Si and the LED on undoped Ge/Si are 5 ± 0.5 × 10⁶ cm⁻² and 3 ± 0.5 × 10⁷ cm⁻², respectively. As Fig. 1(c) shows only one dark pit in the EBIC image, it indicates a low TDD of LED on bulk Ge (TDD<1 × 10⁵ cm⁻²) [9].

 

Fig. 1 EBIC images (40 × 40 µm²) of LEDs (a) on As-doped Ge/Si, (b) on undoped Ge/Si and (c) on bulk Ge with TDD of 5 ± 0.5 × 10⁶ cm⁻², 3 ± 0.5 × 10⁶ cm⁻² and 1 × 10⁵ cm⁻², repectively.

Download Full Size | PDF

Figure 2 shows the current density vs. voltage (J-V) characteristics of LEDs on As-doped Ge/Si, undoped Ge/Si and bulk Ge using the device geometries of 300 × 300 μm², 500 × 500 μm² and 1000 × 1000 μm². A 70 nm Si₃N₄ passivation layer was deposited on the mesa to reduce the perimeter leakage of the LEDs. In Fig. 2(a), LEDs on As-doped Ge/Si show low leakage current densities varying from 5 × 10⁻⁷ A/cm² to 1 × 10⁻⁵ A/cm² at near zero bias condition (−0.5 V). The leakage current density differs for the samples with different mesa areas, which indicates that the leakage current is associated with the perimeter leakage. This leakage is likely dependent on the quality of Si₃N₄ passivation layer. It is noticed that the current at the forward bias below 1 V is also independent of the mesa area. This can be explained by the charged carrier injection difference between the central and edge device regions as the mesa size shrinks [10]. Figure 2(b) shows high revere bias current density of LEDs on undoped Ge/Si. The dark current densities are three to four orders of magnitudes higher than for LEDs on As-doped Ge/Si at near zero bias condition. The leakage current scales with the mesa area, which suggests that the main mechanism is bulk leakage. Since the bulk leakage is an area dependent parameter, it is likely to be caused by an area-dependent defect. Thus, the threading dislocaitons are the possible sources responsible for the bulk leakage.

 

Fig. 2 J-V characteristics of (a) LED on As-doped Ge/Si, and (b) LED on undoped Ge/Si. Device geometries are 300 × 300 μm², 500 × 500 μm² and 1000 × 1000 μm².

Download Full Size | PDF

The J-V curves can be modelled by using a two-diode model [11]:

Figure 3(a) shows the room temperature photoluminescence (PL) intensity vs. excitation power of the LED structure on bulk Ge, As-doped Ge/Si and undoped Ge/Si, using a 488 nm laser. The PL intensity linearly increases with increasing excitation power density. As the p-i-n junctions were identical, the carrier confinement was assumed to be the same. Thus, the slopes of the fitted dashed lines indicate the radiative recombination probability of the electron-hole pairs. The LED on As-doped Ge/Si shows higher PL intensity than the LED on undoped Ge/Si, but its PL intensity is lower than that of the LED on bulk Ge at the same excitation power. Previous studies also reported that the luminescence efficiency degradation of InGaAs quantum well (QW) grown on GaAs/Ge/GeSi/Si (TDD = 2 × 10⁶ cm⁻²) compared to the QW on GaAs/Ge (TDD<1 × 10⁴ cm⁻²). This degradation was related to the defect limited carrier lifetime and carrier diffusion length [9, 12]. To have better understanding of the luminescence efficiency degradation, the carrier diffusion length, Ld, can be modelled as a function of TDD by the following equations [13]:

 

Fig. 3 (a) PL intensity vs. laser excitation power density for LEDs on bulk Ge, As-doped Ge/Si and undoped Ge/Si. The dashed lines fit a linear relationship of PL intensity with excitation power. (b) Normalized PL intensity and minority carrier diffusion lengths of LEDs on different substrates vs. TDD in the junction.

Download Full Size | PDF

The electroluminescence (EL) measurement was conducted using a micro-PL setup with electrical probes. The samples were placed on a metal holder at the focal point of the microscope. Since the same lens was used for the light collection, it was assumed that the EL intensity of different samples could be compared with each other. Figure 4(a) plots the EL intensity against the injection current density of LEDs on bulk Ge, As-doped Ge/Si and undoped Ge/Si using the same device geometry of 500 × 500 μm². The dashed lines fit the LED linear operation regions, where the EL intensity is controlled by the injected electron-hole pairs. From both PL and EL measurements, the LED on bulk Ge shows the highest efficiency. However, the EL intensity of the LED on bulk Ge starts to saturate at 20 A/cm², while no saturation is observed for the LEDs on Ge/Si. For the LED on undoped Ge/Si, the emission peak appears only when J > 12 A/cm². The light emission from the LED on As-doped Ge/Si shows 20% brightness of the LED on bulk Ge at 8 A/cm². The EL emission peak of the LED on undoped Ge/Si is not observable at this current density. The light output of the LED on As-doped Ge/Si increases to 43% brightness of the LED on bulk Ge at 40 A/cm², while the brightness of the LED on undoped Ge/Si is limited to 7% of the brightness of the LED on bulk Ge.

 

Fig. 4 (a) EL intensity vs. injection current density for LEDs on bulk Ge, As-doped Ge/Si and undoped Ge/Si. The dashed lines fit a linear relationship of EL intensity with injection current density. (b) Junction temperature vs. injection current density for the LEDs on bulk Ge and As-doped Ge/Si. (c) The EL spectra at 2 A/cm² and 40 A/cm². (d) An example of fitted EL spectrum according to Eq. (5) and (6).

Download Full Size | PDF

EL intensity saturation only appears in the LED on bulk Ge, while the PL intensity of this device shows good linearity at high excitation power. These observations indicate that the EL saturation is likely caused by thermal quenching at high current injection. The reason that the saturation of the EL intensity is not observed for LEDs on Ge/Si can be explained by the difference in thermal conductivity between Ge/Si and bulk Ge substrates. The thermal conductivity of Si and Ge at room temperature are 1.5 W/cm°C and 0.6 W/cm°C, respectively [15]. We expect the thermal conductivity of Ge/Si is similar to that of Si since the Ge buffer is thin and it should not reduce the thermal conductivity of the substrate significantly.

Figure 4(c) shows the EL spectra of LED on bulk Ge. As the current density increases from 2 A/cm² to 40 A/cm², the peak positions of EL spectra show a red-shift of 3.3 nm. The peak positon energies at different current densities can be identified from the EL spectra. The bandgap energy, E_g can be expressed as a function of the peak position enegy, E_peak, and the junction temperature, T. Then the EL spectrum A($\hslash \omega$) for a photon energy larger than the bandgap energy can be modelled by substituting E_g into the following equations [16]:

Figure 4(b) plots the calculated junction temperature as a function of injection current density for the LEDs on bulk Ge and on As-doped Ge/Si. The junction temperature of the LED on bulk Ge increases faster and is 25 ± 5 °C higher than the temperature of the LED on As-doped Ge/Si at 40 A/cm². This may explain the significant thermal quenching of the EL intensity for LED on bulk Ge, while the LED on As-doped Ge/Si maintains good linearity as current density increases. Other mechanisms, such as carrier leakage can also contribute to the EL saturation. Since the calculated conduction band offset between GaInP and AlInP was only 70 meV, the rise of junction temperature would induce the carrier leakage over the AlInP cladding layer.

4. Conclusions

In summary, the electrical and the optical performance of GaInP/AlInP red LEDs on Ge/Si and bulk Ge substrates have been studied. For the electrical performance, the LED on As-doped Ge/Si shows low leakage current density compared to the LED on undoped Ge/Si. The two-diode model fitting indicates the presence of high TDD in the LED on undoped Ge/Si and the low shunt resistance causes the high leakage. The LED optical performance is correlated with the defect limited carrier diffusion length. The EL measurement shows that the light intensity of the LEDs on As-doped Ge/Si and on undoped Ge/Si are 43% and 7% brightness of the LED on bulk Ge, respectively. Due to better substrate thermal conductivity, LED on As-doped Ge/Si maintains good linearity at high current injection without EL saturation. These results demonstrate the viability of As-doped Ge/Si substrate for III-V photonics integration on Si.