I. Introduction

Contact behavior is an important issue in device design and performance. In our laboratory, we have made an attempt to study the contact properties on GaN, AlGaN and AlGaN/GaN HFET structures in a systematic manner. We found that the Schottky barrier heights, ϕ_B ⁿ, between n-GaN and n-Al_0.15Ga_0.85N differ by about 0.3 eV. For example, the barrier height of Ni on GaN is ∼0.95 eV and that on Al_0.15Ga_0.85N is ∼1.27 eV^[ Reference Yu, Qiao, Xing, Lau, Boutro and Redwing ^{1 ]}. These are the average values (I-V and C-V) obtained on bulk samples, i.e., the layer thickness of GaN and AlGaN exceeds 1 μm. The barrier height of Ni on HFET structure i.e., Al_0.15Ga_0.85N (300 Å)/3μm GaN (undoped), cannot be ascertained using the conventional I-V and C-V methods due to the presence of the piezoelectric donor charge at the AlGaN/GaN interface^[ Reference Chen, Baier, Arch and Shur ^{2 ]}. Internal photo emission is an alternative way to determine the Schottky barrier height of a metal/AlGaN/GaN heterostructure. In our laboratory, we have measured the barrier height of Ni of Al_0.15Ga_0.85N (300 Å and 500 Å)/GaN using the internal photo emission technique and obtained a barrier height of ∼1.30 eV, independent of the AlGaN layer thickness^[ Reference Yu, Dang, Yu, Asbeck, Lau, Boutros and Redwing ^{3 ]}. This value, well within the experimental scattering range, is considered to be consistent with the value obtained on bulk Al_0.15Ga_0.85N samples. As the mole fraction of Al in the AlGaN top layer changes to 30%, the barrier height is seen to increase further to ∼1.56 eV (see Table 1). These results suggest that the Schottky barrier of HFET structures is largely determined by the upper most AlGaN layer and that the barrier height appears to increase between 0.25 to 0.3 eV for every 15% increment in Al mole fraction in the AlGaN layer up to 30 % of Al. More work is needed to correlate the barrier height and the Al mole fraction in detail.

We have also been investigating the ohmic behavior on the III-V nitrides. Of particular interest to us was understanding how to fabricate low resistance and reproducible ohmic contacts on HFET structures. During the course of the study, we found that the ohmic behavior varied significantly from one wafer to another, even though these wafers were nominally the same, i.e., n- Al_0.15Ga_0.85N (300 Å)/ Al_0.15Ga_0.85N (30 Å, undoped)/i-GaN(1 μm). Table 2 shows the results of the measured contact resistivity on four different wafers with nominally the same structure. The n_sμ product, extracted from Hall effect measurements, is the usual parameter that characterizes the HFET samples (see Table 2). For samples with a large value of n_sμ product, the contact resistance should be low, compared to samples with a small n_sμ value, at least in principle. This is because the n_sμ product is believed to indicate good electrical conduction in the channel region and should lead to low ohmic contact resistance. According to this idea, sample #4, in Table 2 should yield the lowest contact resistance, since it had the largest value of n_sμ and sample 2 with the smallest n_sμ should yield the largest contact resistance. Contact resistance measurement using the TLM method indicated that sample #4 had the highest contact resistivity (2.07E−3 Ω cmReference Chen, Baier, Arch and Shur ² ), and sample #2 had a low contact resistivity of 4E−6 Ω·cm². This behavior was not expected and could not be explained by the n_sμ product value alone.

Structural and chemical analysis of the samples showed that these four nominally identical samples were in fact very different in the Al mole fraction and in thickness in the AlGaN layer. Sample #4 had an Al mole fraction of 22% (SIMS value, believed to be more accurate than EDX) in the AlGaN layer with a thickness of 340 Å (TEM value, believed to be more accurate than SIMS). This sample had the largest contact resistivity (2E−3 Ω·cm²). Sample #1 had an Al mole fraction of 9% in the AlGaN layer with a thickness of 210 Å, and yielded the smallest contact resistivity (8.6E−7 Ω·cm²). Table 2 suggests that the contact behavior is primarily governed by the Al mole fraction and the thickness of the top AlGaN layer in these nominally identical samples. This observation led us to conclude that the control of the growth of the HFET samples is far from satisfactory. We, further, assume that sample non-uniformity is a common problem in almost all nitride growth systems.

II. Approaches to Fabricate Low Resistance-Contact in a More Consistent Manner

We considered two approaches to improve the consistency of low resistance contact behavior in HFET structures where sample non-uniformity is expected. The first approach was to use Si implantation into the HFETs to increase the electron concentration to facilitate carrier tunneling across the contact^[ Reference Zopler, Rieger, Baca, Pearton, Lee and Stall ^{4 -} Reference Kobayashi and Gibson ^{8 ]}. We picked sample #4 (the worst case) in Table 2 as a test vehicle to examine the implantation approach; we assumed that if the contact behavior on sample #4 could be improved, then all other samples could be improved using the same approach. In using this approach, we divided sample #4 into two groups. For group 1 (samples #4C−1 and #4A−1), Si²⁸ was directly implanted into the HFET structure at 40 keV with a dose of 1E16 cm⁻². The projected range, R_p, was estimated, using TRIM96, to be about 600 Å into the sample with a peak concentration of about 1.4E21cm⁻³. After implantation, a layer of AlN, used as a capping layer for dopant activation, with a thickness of about 1800 Å was sputter-deposited onto the samples. For group 2 (samples #4C−2 and #4A−2), a layer of 1800 Å thick AlN was first deposited onto the samples, followed by Si implantation through the AlN layer at 120 keV with a dose of 1E16cm⁻². The estimated location of the peak concentration (6.2E20cm⁻³) was about 340 Å into the HFET sample. The advantage of implanting through the AlN capping layer was the ability to place the R_p closer to the HFET surface region; the disadvantages include the loss of some implanted Si ions when the AlN capping layer is removed and the possibility of ion-mixing some Al into the AlGaN layer, thus changing the HFET top layer composition. Samples #4C−1, 4C−2, 4A−1 and 4A−2 were then annealed at 1150 °C for 30 seconds to activate the implanted Si, followed by removing the AlN capping layer using hot phosphoric acid. TLM patterns were then fabricated for contact resistance measurements. A conventional Al (710 Å)/Ti(300 Å)/HFET^[ Reference Ruvimov, Liliental-Weber, Washburn, Qiao and Lau ^{9 ]} metallization was e-beam deposited onto the two groups of implanted samples. The samples were coated with AlN (∼1000 Å thick) as an encapsulation layer to prevent the oxidation during ohmic annealing.

It has been shown that Ti-based metallization schemes reduce contact resistance by forming a metallic AlTi₂N layer with AlGaN, leaving an N-deficient AlGaN region, believed to be heavily n-type, beneath the AlTi₂N contact layer^[ Reference Ruvimov, Liliental-Weber, Washburn, Qiao and Lau ^{9 ]}. Fig. 1(a) shows a cross-sectional electron microscopy image of sample #4C. The contact layer contains two sublayers of different contrast due to different Al/Ti ratio in the sublayers. EDX analysis indicates that the composition of the top sublayer is close to Al3Ti, in good agreement with previous results, while the interfacial layer (with darker shading) is rich in Ti (with Ti/Al ∼ 2). Table 3 shows the contact resistance for the conventional metallization (Al (710 Å)/Ti (300 Å)/HFET) on non-implanted samples (sample #4C), direct implanted samples (sample #4C−1) and implanted through the AlN (sample #4C−2).

Fig. 1. TEM cross sectional view of samples (a) 4C, (b) 4A, (C) 4A−1 and (d) 4A−2.

It is clear that direct implantation and activation at 1150 ^°C for 30 seconds significantly reduced the sheet resistance of the HFET from 1200 Ω/ to 90 Ω/ , in spite of the relatively low activation temperature of 1150 ^°C. As a result, the contact resistance R_c reduced from ∼15 Ω·mm (non-implanted sample) to ∼1.1 Ω·mm for directly implanted samples. This is because R_c is related to the sheet resistance by the following relationship:

where R_c (Ω·mm) is the contact resistance, R_s (Ω/ ) is the sheet resistance of the semiconductor beneath the contact and ρ_s (Ω·cm²) is the specific contact resistivity at the metal/semiconductor interface. While the values of R_c for the implanted samples seemed to decrease substantially, depending on the implantation scheme, compared to those of non-implanted samples, the specific contact resistivity, ρ_s, did not decrease as impressively. This is primarily due to the inability of carrier tunneling through the remaining un-reacted AlGaN layer, using a relatively thin Ti(300Å) in the metallization scheme. Increasing the annealing time at 950 ^°C with and without an AlN capping layer did not improve the contact further beyond 80 seconds of annealing, apparently the reaction has reached an end point after 80 seconds of annealing.

To improve the specific contact resistivity, ρ_s, it is necessary to reduce the thickness of the un-reacted AlGaN layer for easier access to the GaN layer underneath. Based on this concept, we used a different Al(200 Å)/Ti(1500 Å) ratio for the contact formation. In this case, the Ti layer was much thicker and would consume more AlGaN to form AlTi₂N, thereby resulting in a much thinner (or none at all) un-reacted AlGaN for easy carrier tunneling. This scheme is referred to as the “advancing metallization” here. In the conventional scheme, 710 Å of Al reacts with 250 Å of Ti to form Al₃Ti at 250 ^°C to 300 ^°C, leaving an excess Ti layer 50 Å thick to react with the AlGaN layer. For samples with thicker AlGaN layers (≥300 Å) and with a high Al fraction (≥20%), the conventional Al(710 Å)/Ti(300 Å) metallization does not yield satisfactory contact resistance. There may be two possible reasons for this: (1) the chemical reactivity decreases with increasing Al fraction in the AlGaN layer, since AlN appears to be more stable at high temperatures than GaN; (2) for thick AlGaN with a high Al fraction, 50 Å of excess Ti is not enough to consume most of the AlGaN top layer, leaving a relatively thick un-reacted AlGaN layer at the interface to hinder carrier tunneling due to its high Schottky barrier height and thickness.

Using the “advancing” scheme, Al and Ti react to form Ti₃Al (not Al₃Ti as in the conventional scheme) at 200−400 ^°C, leaving 850 Å of Ti in excess to fully react with the AlGaN layer to form the AlTi₂N phase. This reaction would leave little or no AlGaN layer left in the HFET source and drain region, thus resulting in efficient carrier tunneling and much reduced specific contact resistivity, ρ_s.

Since the upper most contact layer is probably Ti₃Al (not Al₃Ti), a capping layer of AlN is required for the ohmic annealing at 950 ^°C for 10 minutes to prevent oxidation (5 minutes was found to be insufficient to react fully). Pure Ti capped with an AlN layer was not practical due to the ease of oxidation of un-reacted Ti after the removal of AlN even at room temperature in air. For the conventional Al(710 Å)/Ti(300 Å) scheme, the top Al₃Ti layers has been found to be stable when annealed at 950 ^oC for 80 seconds in N₂ with little oxidation.

Fig. 1 (b) shows the cross-sectional TEM image of the sample with the "advancing" Al(200 Å)/Ti(1500 Å) contact (#4A). It is clear that the AlGaN layer is partially and sometimes completely consumed due to the reaction with Ti. It also can be found that the reaction is enhanced at the dislocations intersecting the AlGaN layer. Fig. 1(c) shows the cross-sectional TEM image of the sample with the "advancing” contact and the direct Si implantation prior to the metallization (#4A−1). Fig. 1(d) shows the image of the sample with the "advancing” contact and the Si implantation through a layer of AlN prior to the metallization (#4A−2). For these two samples, the contact layer contains two sublayers of different structure and composition, suggesting the reaction of the metal (presumably Ti) with AlGaN. The AlGaN layer is hardly visible on the TEM images of Figures 1(c) and (d) due to a high defect density in the HFET structure caused by the ion implantation. However, it seems that the AlGaN layer is less consumed in the samples #4A−1 (directly implanted) and #4A−2 (implanted through a layer of AlN) compared to sample #4A (not implanted). The damaged region, containing a high density of dislocation loops, is almost polycrystalline in the region close to the metal contact. This polycrystalline region differs in length and in average grain size for the direct and through-AlN implantation. The average grain size is about 50 Å and 110 Å for samples #4A−1 and #4A−2, respectively. TEM results suggest that both the sheet resistance, R_s, and the specific contact resistivity, ρ_s, may be affected by the remaining AlGaN layer and by the ion-implantation damage of crystalline lattice and possible diffusion of Ti through grain boundaries.

Table 4 summarizes the TLM results obtained on the samples with the “advancing” Al(200 Å)/Ti(1500 Å) metallization scheme. For the advancing scheme, the contact resistance and the specific contact resistivity were much smaller compared to those obtained with the conventional metallization scheme. The samples using a combination of direct implantation and the “advancing” metallization gave the lowest contact resistance, R_c, of 0.25 Ω·mm or 5.6 ×10⁻⁶ Ω·cm².

The drastic decrease in R_c is due to the reduction in sheet resistance R_s caused by the Si implantation and the reduction of ρ_s caused by the “advancing” metallization scheme. Both of these factors contribute to the reduction of R_c.

A comparison of Table 3 and Table 4 shows the advantage of using the “advancing” metallization scheme to reduce contact resistance. However, results shown in Table 2 indicate that the conventional Al(710 Å)/Ti(300 Å) system can form low resistance contacts quite readily on HFET structures with a thin AlGaN top layer and a low Al fraction.

In summary, Si implantation into HFET structures was found effective in reducing the sheet resistance, R_S, of the structure. With annealing temperatures higher than 1150 °C, the sheet resistance is expected to decrease even more. The “advancing” metallization scheme of Al(200 Å)/Ti(1500 Å) reduced the specific contact resistivity, ρ_s. Combining direct implantation of Si and the “advancing” metallization, very low contact resistance (∼0.25 Ω·mm) and low specific contact resistivity, ρ_s, (∼5.6×10⁻⁶ Ω·cm²) was achieved on HFET structures with an AlGaN layer at least as thick as 340 Å and with an Al fraction at least as large as 22%.