I. INTRODUCTION

Gallium nitride-based high-electron mobility transistors (GaN HEMTs) are tailor-made for applications requiring high power and high frequency operations because of the desirable intrinsic properties of GaN ^[1-4]. Spontaneous and piezoelectric polarization fields within the AlGaN/GaN heterostructures result into a default presence of two-dimensional electron gas (2DEG) ^[5], which serves as the current carriers between the drain and source terminals that can be modulated by an external applied voltage on the gate terminal. This inherent presence of 2DEG in the channel even without applied gate voltage renders AlGaN/GaN HEMTs to be normally-on devices. In most power applications, however, normally-off (or enhancement-mode) operation is more desirable considering fail-safe design and requirements. The standard recessed-gate devices are fabricated using the conventional top-down approach of selectively etching the AlGaN layer in the channel to below sub-critical thickness to deplete the underlying 2DEG ^[6]. For the standard recessed-gate MIS-HEMTs, aside from the difficulty of controlling the remaining AlGaN barrier thickness (3 nm in our case), the insulator is usually deposited on dry-etched-damaged AlGaN surface, which has undesirable impact on device performance ^[7,8]. On the other hand, for the proposed devices with ultra-thin regrown barrier, the original AlGaN layer in the channel region is completely removed, and replaced it with a 3-nm-thick regrown AlGaN layer using metal-organic chemical vapor deposition (MOCVD) growth technique. It is to be noted that applying this bottom-up approach, the etched surface is completely buried by a regrown AlGaN layer. Moreover, it is relatively easier to control and obtain a uniform 3-nm-thick AlGaN layer using MOCVD growth than thinning the AlGaN barrier in the channel by dry-etching. Meanwhile, Yamamoto et al. ^[9] developed a technique of regrowing AlGaN layer on dry etched surface. Using this technology, which involves dipping the dry-etched surface to hot HCl solution, annealing in NH$_{3}$ and H$_{2}$ atmosphere, and growing at 950 $^{\circ}$C, Yamamoto and co-workers were able to obtain 2DEG density and mobility values nominally similar to commercial AlGaN/GaN heterostructures, thereby demonstrating the effectiveness of this regrowth technology in healing the damage induced by dry-etching. In a recent report, we have proposed and demonstrated normally-off GaN-based metal-insulator-semiconductor HEMTs (MIS-HEMTs) with superior performance including record combination of threshold voltage (V$_{\mathrm{TH}}$) and maximum drain current (I$_{\mathrm{Dmax}}$) employing recessed-gate structure and ultra-thin ex-situ regrown AlGaN barrier ^[10]. To obtain a deeper understanding of the underlying mechanism contributing to the excellent performance of these devices, it is imperative to have a comparison of their electrical characteristics with conventional recessed-gate GaN based MIS-HEMTs. In this work, we compare the electrical performance of the proposed devices with those of conventional recessed-gate GaN-based MIS-HEMTs for the first time and demonstrate the far more superior attributes of the proposed devices.

II. EXPERIMENTAL

Fig. 1 compares the fabrication process flow for proposed (Device A) and reference devices (Device B). The reference device was fabricated using the standard top-down fabrication process, where the 3-nm-thick AlGaN barrier layer was formed by thinning the original 25-nm-thick Al$_{\mathrm{0.25}}$Ga$_{\mathrm{0.75}}$N layer in the channel region by inductively coupled plasma reactive ion etching (ICP-RIE) using Cl$_{2}$ and BCl$_{3}$ mixture gas of the ICP power of 5 W. For the proposed device (Device A), using a bottom-up approach, the channel region was firstly selectively etched to completely remove the original AlGaN layer in the channel region using the same dry etching process recipe. Then, following pre-annealing of the substrates under H$_{2}$ + NH$_{3}$ flow at 850 $^{\circ}$C for 15 min, an approximately 3-nm-thick Al$_{\mathrm{0.25}}$Ga$_{\mathrm{0.75}}$N layer was regrown using Yamamoto regrowth technique ^[9,11,12] by MOCVD at 950 $^{\circ}$C using trimethylaluminium (TMA), trimethylgallium (TMG), and NH$_{3}$ as sources. For both devices, Ti/Al/Mo/Au metal stack was deposited by electron beam evaporation and then followed by rapid thermal annealing at 880$^{\circ}$C to form ohmic drain and source electrodes. As the gate dielectric, 24-nm-thick Al$_{2}$O$_{3}$ was then deposited by atomic layer deposition using TMA and ozone as aluminium and oxygen sources, respectively. The device fabrication was completed by evaporating a gate metal stack of Ni/Au as gate electrodes. For both MIS-HEMT devices A and B, channel length (L$_{\mathrm{ch}}$), channel width (W$_{\mathrm{G}}$), gate-to-source spacing (L$_{\mathrm{GS}}$), gate-to-drain spacing (L$_{\mathrm{GD}}$), gate length (L$_{\mathrm{Metal}}$) were 3, 100, 4, 10 and 5 ${\mu}$m, respectively. Furthermore, we fabricated the circular-shaped MIS-Capacitors with gate diameter ($\phi $$_{\mathrm{G}}$) and channel diameter ($\phi $$_{\mathrm{ch}}$) of 100 and 80 ${\mu}$m, respectively (see Fig. 2).

Fig. 1. Simplified fabrication process flow for (a) proposed device A; (b) reference device B.

Fig. 2. Schematic cross sectional illustration of MIS-HEMTs and MIS-Capacitors of (a) proposed device A; (b) reference device B.

III. RESULTS AND DISCUSSION

Fig. 3 shows the C-V curves comparison of devices A and B measured under the frequency of 1 MHz using an Agilent 4284A LCR meter. The circular symbols represent the measured data while the solid lines represent the C-V characteristics calculated using one-dimensional simulator including self-consistent Poisson-Schrodinger solver developed by Nishiguchi et al. ^[13]. Both C-V profiles showed two plateaus: one related to the field-plate (FP) accumulation capacitance and the other related to channel accumulation capacitance. The extracted FP and channel capacitances for devices A and B are illustrated in Fig. 4(a) and (b), respectively. Compared to device B, device A showed a bit lower accumulation capacitance in the FP region due to the additional 3-nm-thick regrown AlGaN layer. On the other hand, both devices showed equal accumulation channel capacitance evidencing equal thickness values of AlGaN layer in the channel. Fig. 5 compares the extracted interface states density D$_{\mathrm{IT}}$ along the Al$_{2}$O$_{3}$/AlGaN interfaces, showing suppressed D$_{\mathrm{IT}}$ formation for device A. Examining Fig. 4(a) and (b), device A exhibited a more negative V$_{\mathrm{TH}}$ of the FP region and spill-over phenomenon.

Spill-over is the sudden increase in capacitance at positive gate bias due to the to the ``spill-over'' of 2DEG from AlGaN/GaN to Al$_{2}$O$_{3}$/AlGaN interface, and only manifested in high-quality Al$_{2}$O$_{3}$/AlGaN interfaces ^[4]. This observation is consistent with the previously reported effect of regrown AlGaN layer on planar-type MIS-HEMTs ^[14], which can be explained by highly reduced acceptor-like interface states density D$_{\mathrm{IT}}$. Interestingly, considering the channel region, device A exhibited about 4 V higher positive V$_{\mathrm{TH}}$ than that of device B. It is widely known that dry-etching of AlGaN surface generates defect complexes related to nitrogen vacancy V$_{\mathrm{N}}$ ^[15], which can generate positive charges along the Al$_{2}$O$_{3}$/AlGaN interface ^[16-18], limiting the movement of V$_{\mathrm{TH}}$ towards the positive voltage direction. For the device A, since the dry-etched surface was subjected to pre-annealing under H$_{2}$ + NH$_{3}$ flow in the MOCVD chamber and subsequently buried by the AlGaN regrown layer, we believe that the density of V$_{\mathrm{N}}$-related defects in this device was highly reduced. We therefore anticipate that corresponding device A three-terminal transistor will also demonstrate a high positive V$_{\mathrm{TH}}$ shift relative to reference device B. On the final note, we believe that it also beneficial to study the effects of pre-annealing of dry-etched AlGaN surface under NH$_{3}$ and H$_{2}$ atmosphere and study whether the AlGaN surface can be healed by this process using optimized condition.

Fig. 6 compares the transfer curves of devices A and B at applied drain-to-source voltage V$_{\mathrm{DS}}$ of 15 V. From linear extrapolation of the transfer curves, while device B exhibited a V$_{\mathrm{TH}}$ value of +1 V, device A showed a V$_{\mathrm{TH}}$ value of +5 V, congruent to those observed from corresponding MIS-Capacitors. Moreover, while the maximum drain current of device B is limited to about 300 mA/mm that of device A reached about 450 mA/mm. Furthermore, comparison of the semi-log double-sweep transfer characteristics (see Fig. 7) indicates that our proposed device has smaller hysteresis of 0.8 V versus 3 V of the reference device, evidencing better quality of Al$_{2}$O$_{3}$/AlGaN interface, and therefore suggesting highly reliable operation. The marked hysteresis loop for the reference device is hallmark of presence of high density of Al$_{2}$O$_{3}$/AlGaN interface states, most likely due to etching damage sustained during dry etching. Furthermore, the semi-log transfer curve of device B showed inflection points in subthreshold characteristics probably due to dry-etching induced damage on the side-walls and/or non-uniformity of remaining AlGaN barrier thickness. Interestingly, even using the exact dry-etching recipe, these inflection points vanished in device A, further corroborating our claim of improved Al$_{2}$O$_{3}$/AlGaN interfaces in our proposed devices. The proposed device offers not only ease of controlling the channel thickness by regrowth but also provides fresh AlGaN surface, healing the dry-etching damage and leading to suppressed generation of unwanted Al$_{2}$O$_{3}$/AlGaN interface states.

Fig. 3. Comparison of C-V characteristics of devices A and B.

Fig. 4. Comparison of calculated C-V characteristics of (a) device A; (b) device B.

Fig. 5. Extracted interfaces states density distribution of devices A and B.

Fig. 6. Transfer characteristics comparison of devices A and B.

Fig. 7. Subthreshold characteristics comparison of devices A and B.

IV. CONCLUSIONS

We have performed comparative investigation of electrical performance of proposed normally-off GaN-based MIS-HEMTs using recessed-gate and ultrathin regrown AlGaN barrier and conventional recessed-gate GaN-based MIS-HEMTs. The proposed device exhibited much less density of interface states extracted from the measured capacitance-voltage characteristics, evidencing highly improved Al$_{2}$O$_{3}$/AlGaN interfaces. For corresponding three-terminal transistors, the conventional reference device exhibited poor gate control of drain current with about 3 V hysteresis in the transfer curves, while the proposed device showed well-behaved subthreshold characteristics with only 0.8 V hysteresis. Furthermore, the proposed device showed a much higher V$_{\mathrm{TH}}$ of +5 V compared to +1 V of the conventional reference device. The comparison of the measured C-V and I-V characteristics of proposed and conventional devices had shed light on the critical and important underlying mechanism of high V$_{\mathrm{TH}}$ and I$_{\mathrm{Dmax}}$ for the proposed device. The high V$_{\mathrm{TH}}$ of the proposed device is highly likely to reduced positive donor-like defects due to nitrogen vacancies along the Al$_{2}$O$_{3}$/AlGaN interface while the high I$_{\mathrm{Dmax}}$ as well as reduce hysteresis, are due to reduced interface states density along the Al$_{2}$O$_{3}$/AlGaN interface. These assertions are made possible by the comparative study of the C-V and I-V characteristics of proposed and conventional devices.