Comparison of MOCVD and MBE Regrowth for CAVET Fabrication

Abstract

In this paper, we demonstrate the fabrication of current aperture vertical electron transistors (CAVET) realized with two different epitaxial growth methods. Templates with a p-GaN current blocking layer (CBL) were deposited by metal organic chemical vapor deposition (MOCVD). Channel and barrier layers were then regrown by either molecular beam epitaxy (MBE) or MOCVD. Scanning electron microscope (SEM) images and atomic force microscope (AFM) height profiles are used to identify the different regrowth mechanisms. We show that an AlN interlayer below the channel layer was able to reduce Mg diffusion during the high temperature MOCVD regrowth process. For the low-temperature MBE regrowth, Mg diffusion was successfully suppressed. CAVET were realized on the various samples. The devices suffer from high leakage currents, thus further regrowth optimization is needed.

1. Introduction

The research on wide-bandgap semiconductor power devices is driven by the need for efficient transistors. The superior material properties of e.g., GaN or SiC allow for higher reverse bias and elevated operating temperature limits, compared to lower-bandgap Si-based devices. Lateral high electron mobility transistors (HEMT) have already entered the market in the medium power conversion range, achieving blocking voltages of 600 V [1]. For high-power applications, a vertical geometry is preferred over a lateral one, particularly due to better scaling possibilities and an improved breakdown voltage. The peak electric field is located in the bulk of the semiconductor for most vertical designs and a more uniform electric field distribution is achieved [2], reducing the need for complex field-plate structures [3]. Several device concepts are currently investigated, such as vertical trench metal oxide semiconductor field effect transistor (FET) [4,5], finFET [6,7] or vertical junction FET [8,9,10]. Another promising design is the CAVET [11,12,13,14,15,16,17]. Here, a two-dimensional electron gas (2DEG) with high carrier mobility is formed above a CBL with an aperture. It combines the mobility advantage of the HEMT with a vertical device design to increase the blocking voltage without enlarging chip size. However, selective doping or etching [18] is necessary in order to fabricate the aperture. While selective air-gap etching moves the peak electric field back to the semiconductor surface, selective doping can be realized by either implantation or regrowth [19]. In this study, we investigate both MOCVD and MBE regrowth to fabricate CAVET. The growth mechanisms of the two techniques on the patterned aperture region will be compared and evaluated regarding CAVET fabrication and properties.

2. Experiment

Templates for both regrowth techniques were grown on sapphire in an AIXTRON Planetary MOCVD tool. Due to the insulating substrates, a quasi-vertical design has to be employed. Trimethylgallium (TMGa) and ammonia were used as precursors, silane and magnesocene for n- and p-type doping, respectively. The carrier gas was hydrogen. The complete template stack is depicted in Figure 1a. No additional preparation of the sapphire wafers was carried out before growth. After AlN nucleation on sapphire, 2 µm n${}^{+}$-GaN were deposited, forming the backside contact layer. Above, a 700-nm-thick drift layer with a doping concentration of $N_{\mathrm{D}}$ = 5 × 10¹⁶ cm^$-3$ was grown. Thickness and $N_{\mathrm{D}}$ were chosen for low $R_{\mathrm{ON}}$ with respect to the $V_{\mathrm{BR}}$ trade-off [20]. The epitaxial stack was completed with a 250-nm-thick p${}^{+}$-GaN layer, possessing a high acceptor concentration of $N_{\mathrm{A}}$ = 1.3 × 10¹⁹ cm^$-3$. This layer serves as CBL.

After template growth, patterned samples were fabricated by etching the aperture into the p-GaN, utilizing a two-step Cl-based etch process (Figure 1b). At first, a high-power inductively coupled plasma reactive-ion etching (ICP-RIE) is employed to expose the drift layer. It was optimized for steep sidewalls [21]. Subsequently, a low-damage cyclic ICP etch process is used to carefully remove etch-damaged material [22]. This second process leads to a smooth surface with low rms roughness of 1.2 nm. Both processes combined yield an etch depth of 300 nm. The templates were dipped in HCl and BOE consecutively for one minute each in order to remove surface contamination and for regrowth preparation [23].

Then, a GaN HEMT structure consisting of a GaN channel layer and an AlGaN barrier was regrown with MOCVD and MBE, respectively (Figure 1c). The regrowth process fills the previously etched aperture with GaN.

MOCVD regrowth was performed at $T_{\mathrm{growth}}=$ 1050 °C and two samples were prepared (CVD-A and CVD-B). The regrowth of sample CVD-A consists of a 300-nm-thick GaN channel with $N_{\mathrm{D}}$ = 2 × 10¹⁶ cm^$-3$, which was capped by a 25-nm-thick Al${}_{0.25}$Ga${}_{0.75}$N barrier layer. For sample CVD-B, a 5 nm AlN interlayer was regrown below the GaN channel to block Mg-diffusion from the p-GaN CBL [24].

MBE regrowth was carried out at $T_{\mathrm{growth}}=$ 750 °C. The lower growth temperature suppresses Mg diffusion [25,26]. Two samples (MBE-C and MBE-D) with different GaN channel thicknesses, but without AlN interlayer, were regrown. Sample MBE-C had a 300-nm-thick channel layer, whereas sample MBE-D was fabricated with a 450-nm-thick channel. Both were doped with $N_{\mathrm{D}}$ = 2 × 10¹⁶ cm^$-3$ and capped by the same 25-nm-thick Al${}_{0.25}$Ga${}_{0.75}$N barrier. For sample MBE-D, a more conductive 2DEG is expected, due to the increased distance between 2DEG and regrowth interface.

Following the regrowth, CAVET were fabricated (Figure 1d). The backside contact layer was exposed by ICP dry-etching utilizing a 450-nm-thick Cr hard mask. Subsequently, the sample was dipped in diluted HCl before e-beam evaporation of the ohmic contact metals. The drain and source contact stacks, consisting of Ti/Al/Ni/Au (15 nm/100 nm/40 nm/50 nm), were deposited and annealed at 825 °C to ensure an ohmic characteristic. Finally, the Ni/Au (50 nm/200 nm) gate contact was formed on top of the aperture region.

Additionally, three lateral HEMT reference samples without p-GaN CBL were grown. All samples consist of a 300-nm-thick GaN channel layer and a 25 nm Al${}_{0.25}$Ga${}_{0.75}$N barrier layer, analog to the one mentioned above, to serve as reference for a Mg-free (re)growth. The epitaxial layer stack for reference sample 1 was continuously grown with MOCVD. The HEMT structure for reference 2 and 3 was regrown on lowly n-doped GaN templates with MOCVD and MBE, respectively. A brief summary of all fabricated templates is given in

Table 1. Samples overview.

Sample	Technique	${\mathit{T}}_{\mathbf{growth}}$	Remark
CVD-A	MOCVD	1050 °C	-
CVD-B	MOCVD	1050 °C	5 nm AlN interlayer
MBE-C	MBE	750 °C	300 nm channel
MBE-D	MBE	750 °C	450 nm channel
Ref. 1	MOCVD	1050 °C	continuously grown
Ref. 2	MOCVD	1050 °C	regrown on n-GaN
Ref. 3	MBE	750 °C	regrown on n-GaN

.

SEM images and AFM surface profiles of the aperture region after regrowth were used to identify the growth mechanisms of the different regrowth techniques on patterned samples. Eddy-current and, after ohmic contact deposition, Hall measurements were conducted to characterize the 2DEG properties. I-V measurements were carried out to investigate the CAVET performance. The contact resistances were determined with transfer length method (TLM) measurements.

3. Results and Discussion

3.1. Structural Characterization

3.1.1. MOCVD Regrowth

Figure 2a shows the MOCVD-regrown region of an 8-µm-wide aperture on sample CVD-A. The previously 300-nm-deep etched aperture, indicated by the dotted line, is completely filled. A narrower imprint of the opening is still visible. AFM depth profiles of the MOCVD-regrown aperture with various widths ($L_{\mathrm{AP}}$) are plotted in Figure 2b, confirming residual trenches after regrowth. The remaining trench depth increases from 40 nm for $L_{\mathrm{AP}}=2$ µm to 120 nm for $L_{\mathrm{AP}}=8$ µm. Also, the sidewall angle $\theta$ is increased from 3° to 7° for larger $L_{\mathrm{AP}}$. The growth morphology is different on the sidewalls compared to a planar regrown surface in the vicinity of a trench, as shown in Figure 2d, but rms roughness is similar.

2D layer growth at the p-GaN surface and the trench bottom dominates for large $L_{\mathrm{AP}}$, forming an imprint of the original trench. But also, mass transport from the p-GaN surface to the trench sidewall is occurring, as indicated in gray in Figure 2c [27]. This leads to small sidewall angles, a different surface morphology and a planarization of the trench. For smaller $L_{\mathrm{AP}}$, this effect increases due to more overlapping growth from each sidewall and results in a smaller imprint with shallower sidewalls, as observed in the AFM profiles in Figure 2b.

To prevent field crowding and leakage currents after the gate contact deposition on top of the aperture region, a planar surface and therefore a small remaining trench with a small sidewall angle is desired. However, by decreasing $L_{\mathrm{AP}}$, the aperture resistance will be increased and a high $R_{\mathrm{ON}}$ is to be expected.

3.1.2. MBE Regrowth

For the low-temperature MBE regrowth process, less mass transport from the p-GaN surface into the trenches should occur. SEM images of MBE-regrown apertures in Figure 3a–c confirm this expectation. The schematic cross section in Figure 3d illustrates the growth mechanism. In addition to 2D growth (white arrows), lateral overgrowth can be observed (black arrow). Due to the low temperatures during MBE growth, the diffusion length of the atoms is limited. Thus, lateral growth at the edge is enhanced compared to growth at the bottom of the trench and leads to the formation of a protrusion. For small $L_{\mathrm{AP}}$, this inhibits growth inside of the trench and the formation of the channel layer in the aperture is hindered.

Therefore, a large $L_{\mathrm{AP}}$ is necessary, as this will lead to a smaller protrusion and a larger overlap of the material regrown at the aperture edge. Still, the gate contact has to be formed with care to prevent a rupture of the metal layer at the protrusions.

3.2. Electrical Characterization

3.2.1. Planar 2DEG Properties

Eddy-current and Hall measurements were carried out to investigate the 2DEG formation in the layers on top of the CBL. Statistics of these measurements of several positions on each sample are presented in Figure 4. All p-GaN samples showed a sheet resistance $R_{\mathrm{sheet}}$ in the order of 50,000 $\mathsf{\Omega }/sq$ before regrowth. The strong decrease of $R_{\mathrm{sheet}}$ for all samples cannot solely be explained by regrowing the lowly doped n-GaN channel layer and thus the formation of a 2DEG at the AlGaN/GaN interface appears plausible. For the continuously MOCVD-grown reference sample 1, the lowest $R_{\mathrm{sheet}}=340$ $\mathsf{\Omega }/sq$ and highest mobility $\mu =2000$ cm${}^2/$Vs can be extracted. Also, the sheet carrier concentration is $n_{\mathrm{S}}=10$ × 10¹² cm^$-2$, which is expected for the given AlGaN thickness and Al content.

A slightly increased $R_{\mathrm{sheet}}$ is obtained for the regrown reference sample 2, probably due to increased interface roughness or defect scattering, introduced by the regrowth. Low-temperature Hall measurements at 77 K (not shown here) still reveal a reduced mobility compared to reference sample 1 and confirm the proposed scattering mechanisms, as all other are frozen out at 77 K [28,29].

For the MOCVD-regrown samples on p-GaN, a dramatic increase in $R_{\mathrm{sheet}}$ is observed. The highest $R_{\mathrm{sheet}}=5900$ $\mathsf{\Omega }/sq$ is obtained for sample CVD-A, as no measures were taken to inhibit Mg diffusion. The AlN interlayer introduced in sample CVD-B is also not able to suppress the Mg diffusion completely, resulting in an increased sheet resistance of 3200 $\mathsf{\Omega }/sq$ and reduced $n_{\mathrm{S}}=2.2$ × 10¹² cm^$-2$. For sample CVD-A, no Hall data is available.

All MBE-regrown samples (reference 3, MBE-C and MBE-D) achieve the expected $n_{\mathrm{S}}=10$ × 10¹² cm^$-2$ and are comparable to reference sample 2 in terms of $R_{\mathrm{sheet}}$. This indicates a successful suppression of Mg diffusion for the low temperature MBE regrowth. The increased mobility for sample MBE-D could be attributed to an improved AlGaN/GaN interface quality due to a larger distance to the regrowth interface. However, both samples show the same rms surface roughness of 1 nm.

Overall, improved 2DEG properties are achieved with the MBE regrowth, compared to the high-temperature MOCVD regrowth process. The lower $T_{\mathrm{growth}}$ suppresses Mg diffusion effectively. Hence, the acceptor concentration near the 2DEG is supposed to be low. The AlN interlayer in CVD-B did improve $R_{\mathrm{sheet}}$ and may be a viable option to enable an MOCVD regrowth for CAVET after further optimization. However, the AlN interlayer will introduce an additional energy barrier for the drain current through the aperture when regrown below the channel layer. Therefore, sample CVD-A and MBE-D were chosen to compare the properties of CAVET fabricated with each regrowth technique and are analyzed in the following.

3.2.2. MOCVD-Regrown CAVET (CVD-A)

Deduced from the Hall measurements, a low maximum available drain current is to be expected for the MOCVD-regrown CAVET. Source contacts on top of the AlGaN barrier show a contact resistance of $R_{\mathrm{C}},\mathrm{S}=7$ $\mathsf{\Omega }\mathrm{mm}$, the drain contact resistance $R_{\mathrm{C}},\mathrm{D}$ is $1.7$ $\mathsf{\Omega }\mathrm{mm}$. The high $R_{\mathrm{C}},\mathrm{S}$ may be caused by Mg diffusion into the regrown n-GaN.

A CAVET with a gate length of $L_{\mathrm{G}}=2$ µm and a $L_{\mathrm{AP}}=1$ µm is discussed here (cf. Figure 1 for geometry). All currents are normalized to the active area, which is the distance between the source contacts multiplied by the gate width of 54 µm.

The output characteristic of the transistor is presented in Figure 5a. $V_{\mathrm{GS}}$ was varied from $+2$ V to $-8$ V in 1 V steps. No drain current saturation and a delayed turn-on of the transistor can be observed. Both effects are most likely caused by parasitic series resistances of the 2DEG, aperture and source contacts. A simple DC equivalent circuit model of a CAVET is depicted in Figure 5b and comprises the following resistances: the contact resistances ($R_{\mathrm{C}},\mathrm{S}$, $R_{\mathrm{C}},\mathrm{D}$), the lateral 2DEG resistance $R_2\mathrm{DEG}$, the aperture resistance $R_{\mathrm{ap}}$, the drift layer resistance $R_{\mathrm{drift}}$ and the backside contact layer resistance $R_{\mathrm{bcl}}$. Our calculations, based on measurements of lumped-element model test structures (e.g., TLM structures and lateral HFETs regrown on CBL with and without aperture), led to an aperture resistance $R_{\mathrm{ap}}$ of approx. 10 M$\mathsf{\Omega }$. The reason for this high $R_{\mathrm{ap}}$ may be, that the aperture is partially depleted from the lateral p-n junction, formed by the p-GaN CBL and the n-GaN aperture, or Mg has diffused into the regrown aperture during the high temperature MOCVD regrowth, compensating the n-doping. However, the large $R_{\mathrm{ap}}$ and low 2DEG conductivity limit the maximum available drain current to $I_{\mathrm{D}},\max =1.3$ A/cm${}^2$ and cause a high turn-on voltage. The weak pinch-off can be improved by increasing $L_{\mathrm{G}}$, as shown in the inset of Figure 5a. $L_{\mathrm{AP}}$ is 6 µm for these devices. With a larger $L_{\mathrm{G}}$, the gate control of the regrown GaN is enhanced and the source-drain leakage current is effectively suppressed.

Destructive breakdown occurs below the gate at $V_{\mathrm{DS}}=-140$ V. This results in a critical electrical field of $E_{\mathrm{crit}}=2.8$ MV/cm, by assuming a complete depletion of the 1-µm-long gate-drain distance.

3.2.3. MBE-Regrown CAVET (MBE-D)

An improved regrowth quality and lower Mg diffusion are expected for the MBE-regrown CAVET, compared to the CAVET with MOCVD regrowth. The source contact resistance of the MBE-regrown device is $R_{\mathrm{C}},\mathrm{S}=0.4$ $\mathsf{\Omega }\mathrm{mm}$, already indicating a lower Mg concentration at the surface. The drain contact resistance is $R_{\mathrm{C}},\mathrm{D}=2$ $\mathsf{\Omega }\mathrm{mm}$.

The output characteristic of devices with $L_{\mathrm{G}}=2$ µm and $L_{\mathrm{AP}}=8$ µm, 6 µm and 2 µm are depicted in Figure 6a. In contrast to the MOCVD-regrown CAVET, current saturation, a larger maximum drain current density of $I_{\mathrm{D}},\max =4$ kA/cm${}^2$ for $V_{\mathrm{GS}}=2$ V and low $R_{\mathrm{ON}}=2$ m$\mathsf{\Omega }$ cm${}^2$ are achieved.

Figure 6b shows the transfer characteristics of the same devices. The transistors suffer from high off-state and gate leakage currents. Possible off-state drain leakage current paths, i.e., along the etched sidewall to the quasi-backside contact and through the CBL, can be excluded, as previous experiments on these building blocks show the suppression of leakage currents [21]. The most probable drain leakage current path for the transitor with $L_{\mathrm{AP}}=8$ µm is through the lowly doped GaN channel and aperture as indicated in Figure 3d by the dashed arrow. As observed in the previous Section 3.1.2, the overlap between the regrown GaN channel in- and alongside the aperture is reduced for smaller apertures. This cuts off this drain leakage current path at the cost of a smaller $I_{\mathrm{D}},\max$, as can be seen in Figure 6b and increases the on/off ratio to 2.3 × ${10}^3$ [$I_{\mathrm{D}}(V_{\mathrm{GS}}=0$ V)$/I_{\mathrm{D}}(V_{\mathrm{GS}}=-8$V)] for the device with $L_{\mathrm{AP}}=2$ µm. However, the transistors are still limited by gate leakage currents, probably due to the gate deposition on the non-planarized regrown-aperture region.

4. Summary and Conclusions

Two different regrowth techniques were used to fabricate CAVET. The regrowth was conducted after aperture patterning on MOCVD-grown CBL with MBE and MOCVD, respectively. MOCVD-regrown aperture regions showed small imprints of the etched aperture opening with small sidewall angles. This was correlated with mass transport into the trench during regrowth. On the other hand, protrusions were formed at apertures after the low-temperature MBE regrowth.

For the high-temperature MOCVD-regrown transistors, no current saturation and low $I_{\mathrm{D}},\max$ were measured, which is probably due to Mg diffusion into the regrown aperture.

The MBE-regrown CAVET showed improved on characteristics but suffered from high gate and off-state leakage currents. Regrowth thickness and channel layer doping concentration should be decreased to allow for low drain leakage currents. A planarization of the aperture region or an optimized aperture width to height ratio may also help to suppress leakage currents.