⇦Chinese Physics Letters, 2020, Vol. 37, No. 9, Article code 097101 Performance Enhancement of AlGaN/GaN MIS-HEMTs Realized via Supercritical Nitridation Technology Meihua Liu (刘美华), Zhangwei Huang (黄樟伟), Kuanchang Chang (张冠张), Xinnan Lin (林信南), Lei Li (李蕾), and Yufeng Jin (金玉丰)* Affiliations School of Electronic and Computer Engineering, Peking University Shenzhen Graduate School, Shenzhen 518055, China Received 9 May 2020; accepted 24 July 2020; published online 1 September 2020 Supported by the Shenzhen Science and Technology Innovation Committee (Grant Nos. ZDSYS201802061805105, JCYJ20190808155007550K, QJSCX20170728102129176, and JCYJ20170810163407761), and the National Natural Science Foundation of China (Grant No. U1613215).
*Corresponding author. Email: yfjin@pku.edu.cn Citation Text: Liu M H, Huang Z W, Zhang G Z, Lin X N and Li L et al. 2020 Chin. Phys. Lett. 37 097101    Abstract This paper proposes a method of repairing interface defects by supercritical nitridation technology, in order to suppress the threshold voltage shift of AlGaN/GaN metal-insulator-semiconductor high-electron-mobility transistors (MIS-HEMTs). We find that supercritical NH$_{3}$ fluid has the characteristics of both liquid NH$_{3}$ and gaseous NH$_{3}$ simultaneously, i.e., high penetration and high solubility, which penetrate the packaging of MIS-HEMTs. In addition, NH$_{2}^{-}$ produced via the auto coupling ionization of NH$_{3}$ has strong nucleophilic ability, and is able to fill nitrogen vacancies near the GaN surface created by high temperature processes. After supercritical fluid treatment, the threshold voltage shift is reduced from 1 V to 0 V, and the interface trap density is reduced by two orders of magnitude. The results show that the threshold voltage shift of MIS-HEMTs can be effectively suppressed by means of supercritical nitridation technology. DOI:10.1088/0256-307X/37/9/097101 PACS:71.55.Eq, 73.20.-r, 73.50.-h © 2020 Chinese Physics Society Article Text GaN-based high electron mobility transistors (HEMTs) are good candidates for high frequency and high efficiency power switching applications due to their superior material parameters. In order to maintain a relatively large gate swing, a low gate leakage current, metal-insulator-semiconductor (MIS) gate structure is typically adopted. However, there are several reliability issues related to GaN MIS-HEMTs. When a positive gate bias is applied, defects located in the gate stack act as charge trapping sites. This induces a shift of the device transfer characteristics toward more positive values.[1] Generally, the trapping effect in an MIS gate stack could be related to traps at or near the interface, named interface states/border traps, or in the bulk of the insulator.[2,3] To date, various technologies have been employed to suppress threshold voltage ($V_{\rm th}$) shift, including pre-fluorination argon treatment,[4] sputter deposited Al$_{2}$O$_{3} $,[5] and in situ dielectric pre-deposition plasma nitridation,[6] to reduce the trapping effect at/near the insulator/semiconductor interface. Due to the low deposition temperature, traps occur in high densities in the bulk gate insulators deposited by PECVD or ALD. Recently, high temperature deposited gate insulators, such as low-pressure chemical vapor deposition (LPCVD) grown SiN$_{x}$, have demonstrated a robust gate dielectric for both normal-on GaN MIS-HEMTs and normal-off gate recessed hybrid MIS-HEMTs with low bulk trap density. However, the interface quality between LPCVD SiN$_{x}$ and (Al)GaN is degraded, due to the high growth temperature and H erosion. Although low temperature deposited insertion layers or N surface plasma treatments have also been adopted to improve interface quality, the drift of $V_{\rm th}$ is still a persistent problem. Supercritical fluid technology can effectively bring elements into materials by means of supercritical CO$_{2}$ fluid to reduce trap density, because of its capacity for penetration of and damage-free diffusion in devices.[7] Supercritical technology has been applied in the fields of memory and LEDs[8] but no research has yet been conducted regarding the effectiveness of GaN power devices. In this Letter, we firstly propose the application of supercritical nitridation treatment (SNT) to passivate defects and mitigate the shift of $V_{\rm th}$ in GaN MIS-HEMTs. Subsequent to SNT, the interface trap density in the LPCVD Si$_{3}$N$_{4}$/GaN cap layer interface is effectively reduced, and a near zero shift of $V_{\rm th}$ in the transfer curve of a GaN power device is observed, with a bidirectional gate bias sweep of up to 10 V. The devices detailed below were fabricated in a 6-inch silicon foundry, using 30 nm LPCVD grown Si$_{3}$N$_{4}$ as the gate insulator. Details of the device fabrication process is similar to that given in Ref. [9]. Figure 1 shows a schematic cross-section view and optical image of the GaN/AlGaN/GaN MIS-HEMT examined in this work. The device under test has dimensions of $L_{\rm G}$/$L_{\rm GS}$/$L_{\rm GD}=1.5/5/10$ µm. The total gate width is 15 mm. The $V_{\rm th}$ of the fresh device is about $-8$ V, and the output saturation current is 430 mA/mm at a gate bias of 0 V. Figure 2 shows a schematic diagram of the supercritical nitridation system. The MIS-HEMTs were put into the chamber of a supercritical nitridation system at 150℃ for three hours; the chamber was injected with CO$_{2}$ at a pressure of 3000 psi (psi: pound per square inch, 1 psi = 6.895 kPa), which was mixed with NH$_{3}$.
cpl-37-9-097101-fig1.png
Fig. 1. Schematic views of the GaN/AlGaN/GaN MIS-HEMT together with their properties: (a) cross section, (b) overview, (c) transfer curves, (d) output curves.
cpl-37-9-097101-fig2.png
Fig. 2. Schematic diagram of the supercritical nitridation system.
The $I_{\rm D}$–$V_{\rm GS}$ transfer characteristics of the devices, with and without SNT, are shown in Fig. 3. All the curves are swept in bidirectional mode at $V_{\rm DS} = 0.05$ V. For the MIS-HEMTs without SNT, significant hysteresis was observed during the sweep, suggesting a severe trap-induced $V_{\rm th}$ shift. Double-mode gate capacitance-voltage ($C$–$V$) measurements from $-10$ V to 10 V were also conducted on the devices, with shorted source and drain at a frequency of 10 kHz. As shown in Fig. 4, after SNT, the hysteresis of the $C$–$V$ curve in a bidirectional sweep is reduced from 1 V to nearly 0 V, which is consistent with the result in terms of device transfer characteristics.
cpl-37-9-097101-fig3.png
Fig. 3. Transfer curves of AlGaN/GaN MIS-HEMTs in a bidirectional $V_{\rm GS~sweep}$ from $-10$ V to 10 V then back to $-10$ V (a) without SNT and (b) with SNT.
cpl-37-9-097101-fig4.png
Fig. 4. Capacitance measurements for AlGaN/GaN MIS-HEMTs (a) without SNT and (b) with SNT.
Assuming a column of trap levels, $G_{\rm p}$ can be expressed as follows: $$ \frac{G_{\rm p}}{\omega }=\frac{qD_{\rm it}}{2\omega \tau_{\rm it}}\ln [1+\left(\omega \tau_{\rm it} \right)^{2}],~~ \tag {1} $$ where $\omega$ is the radial frequency, and $\tau_{\rm it}$ denotes the trap time constant given by the Shockley–Read–Hall statistics.[10] The interface trap density $D_{\rm it}$ can be determined by $$ { D}_{\rm it}=\frac{C_{\rm ox} \cdot\Delta V_{\rm th}}{q^{2}}.~~ \tag {2} $$ Figure 5(a) shows the measured $G_{\rm p}/\omega$ as a function of $\omega$ at different gate biases. The peak value of $G_{\rm p}/\omega$ in devices with SNT is much smaller, suggesting a lower density of trap states at the interface. Here, $D_{\rm it}$ is extracted by fitting the experimental data using Eqs. (1) and (2). The extracted $D_{\rm it}$ for various time constants is shown in Fig. 5(b). For MIS-HEMTs without SNT, the $D_{\rm it}$ value is in the order of 1–$2\times {10}^{12}$ cm$^{-2}$eV$^{-1}$ with a time constant of 3.4 µs. After SNT, the traps with a time constant of 3.4 µs are greatly suppressed, and only shallow level traps with a time constant close to 0.15 µs are observed. The shallow trap density is in the order 1–$2\times {10}^{10}$ cm$^{-2}$eV$^{-1}$. The reduced trap density and energy level leads to a much smaller hysteresis of the transfer and $C$–$V$ curves as shown in Figs. 3 and 4.
cpl-37-9-097101-fig5.png
Fig. 5. (a) Measured $G_{\rm p}/\omega$ as a function of $\omega$ at different gate biases. (b) Frequency-dependent conductance as a function of radial frequency for MIS-HEMTs without and with SNT, biased at selected gate voltages.
cpl-37-9-097101-fig6.png
Fig. 6. The shift of $V_{\rm th}$ of the Si$_{3}$N$_{4}$/GaN/AlGaN/GaN MIS-HEMT during time-of-fly gate-bias-induced stress and $V_{\rm th}$ measurement (a) without SNT and (b) with SNT.
During the transfer and $C$–$V$ measurement, the sweep time is about 1.5 s. To further study the effect of SNT on the shift in $V_{\rm th}$ after various time intervals and different gate-bias-induced stresses, time-of-fly gate-bias-induced stress and $V_{\rm th}$ measurements were performed on the device, with both positive and negative gate biases. The shift in $V_{\rm th}$ with bias-induced stress times from 1 ms to 1000 s at different voltages is shown in Fig. 6. The shift in $V_{\rm th}$ shows a nearly linear relationship with the logarithmic time scale, suggesting a broad distribution of the time constant of deep traps in the devices without SNT. On the other hand, the shift in $V_{\rm th}$ in devices with SNT after the stress test is much smaller. The maximum shift of $V_{\rm th}$ at 8 V gate stress for 1000 s is only 0.5 V. CO$_{2}$ is a double bond structure with large activation energy and stable chemical properties. It does not participate in the reaction during supercritical treatment. It acts only as a solvent for supercritical NH$_{3}$, avoiding the supercritical NH$_{3}$ reaction and eroding the device electrodes. As mentioned above, due to the high growth temperature in LPCVD, the surface of the GaN cap layer itself, and the interface between the Si$_{3}$N$_{4}$ and the GaN cap layer may be relatively defective, which could lead to a trap-induced $V_{\rm th}$ shift phenomena in GaN MIS-HEMTs.[11,12]
cpl-37-9-097101-fig7.png
Fig. 7. Ga $3d$ (left) and Si $2p$ (right) core-level spectra at the Si$_{3}$N$_{4}$/GaN cap interface in 8-nm-Si$_{3}$N$_{4}$/GaN/AlGaN/GaN samples (a) without SNT and (b) with SNT. Each measured spectrum (symbol) is resolved into two Gaussian functions,t corresponding to $M$ in $M$–N (solid magenta line) and $M$–O (solid blue line) bonds. $M$ represents Ga (left) or Si (right). The solid blue magenta line is a superposition of the two fitting functions.
Figure 7 shows the Ga $3d$ (left) and Si $2p$ (right) core-level spectra at the Si$_{3}$N$_{4}$/GaN cap interface in 8-nm Si$_{3}$N$_{4}$/GaN/AlGaN/GaN samples. For the control sample without SNT, the native oxide exhibits a large shoulder (Ga–O bonds) on the high binding energy side of the Ga–N peak [Fig. 7(a) left]. The Ga $3d$ core-level spectrum in Fig. 7(a) indicates the existence of amorphous native oxide at the Si$_{3}$N$_{4}$/GaN-cap interface without SNT. Ga–O/Si–O bonds may also act as a kind of interface state,[12] which could lead to the $V_{\rm th}$ shift phenomena in GaN MIS-HEMTs. During the formation of Ga–O/Si–O bonds, some are due to natural oxidation and some are O filled with N vacancies. With SNT, a higher intensity of Si–N bonds is observed at the Si$_{3}$N$_{4}$/GaN-cap interface [Fig. 7(b) right]. The proportion of Ga–N bonds increases from 92.1% to 93.42%. The proportion of Si–N bonds increases from 87.9% to 92.51%. At the same time, the proportion of Ga dangling bonds decreases from 0.11% to 0.09%. The proportion of Si dangling bonds decreases from 0.19% to 0.29%. This change in the proportion of chemical bonds indicates that some of the N vacancies are filled, and some O atoms in Ga–O bonds are replaced by N during SNT treatment. Furthermore, the full widths at half maximum of Ga–O and Si–O decrease from 1.04 eV and 2.67 eV to 0.96 eV and 2.08 eV, respectively. In fact, the narrowing of half peak width, and the improvement of symmetry of peak shape, can be directly ascribed to the reduction of defects.[13–15] In the process of SNT, NH$_{2}^{-}$ produced via the auto coupling ionization of NH$_{3} $[16] which has strong nucleophilic ability and can fill nitrogen vacancies near the GaN surface caused by high temperature processes. Furthermore, during the nucleophilic reaction,[17] NH$_{2}^{-}$ may react with Ga–O/Si–O to replace O and form a Ga–N bond. The exact passivation mechanism during SNT is still under investigation. In summary, high performance AlGaN/GaN MIS-HEMTs, realized via supercritical nitridation technology have been designed, fabricated, and measured. By comparing the devices before and after SNT, we find that supercritical nitridation technology can effectively repair defects and suppress shifts in threshold voltage. After SNT, the optimized MIS-HEMTs demonstrate a low $V_{\rm th}$ hysteresis decrease of about 1 V at $V_{\rm G-sweep}$ from $-10$ V to 10 V, and a decrease in terms of low interface traps $D_{\rm it}$ of almost two orders of magnitude. Our demonstration of the supercritical nitridation technology to repair defects of wide-bandgap family of semiconductors may bring about great changes in the field of device fabrication. References Suppression of Dynamic On-Resistance Increase and Gate Charge Measurements in High-Voltage GaN-HEMTs With Optimized Field-Plate StructureCorrelation of interface states/border traps and threshold voltage shift on AlGaN/GaN metal-insulator-semiconductor high-electron-mobility transistorsTrap behaviours characterization of AlGaN/GaN high electron mobility transistors by room-temperature transient capacitance measurementGate Injection Transistor (GIT)—A Normally-Off AlGaN/GaN Power Transistor Using Conductivity ModulationEffect of Sputtered-Al 2 O 3 Layer Thickness on the Threshold Voltage of III-Nitride MIS-HEMTsAlGaN/GaN MIS-HEMTs of Very-Low ${V}_{\sf {{th}}}$ Hysteresis and Current Collapse With In-Situ Pre-Deposition Plasma Nitridation and LPCVD-Si3N4 Gate InsulatorLow-temperature method for enhancing sputter-deposited HfO2 films with complete oxidizationProduction of griseofulvin nanoparticles using supercritical CO2 antisolvent with enhanced mass transferOptimization of Au-Free Ohmic Contact Based on the Gate-First Double-Metal AlGaN/GaN MIS-HEMTs and SBDs ProcessInvestigation of trapping effects in AlGaN/GaN/Si field-effect transistors by frequency dependent capacitance and conductance analysisForewordModel of interface states at III-V oxide interfacesElectrochemically deposited ZnO films: an XPS study on the evolution of their surface hydroxide and defect composition upon thermal annealingCalculation of the electronic structure of the intermetallic compounds ErNi5 − x Al x (x = 0, 1, 2)Analysis of heat-treated graphite oxide by X-ray photoelectron spectroscopyHighly Efficient “On Water” Catalyst-Free Nucleophilic Addition Reactions Using Difluoroenoxysilanes: Dramatic Fluorine Effects
[1] Saito W, Nitta T, Kakiuchi Y and Saito Y 2007 IEEE Trans. Electron Devices 54 1825
[2] Wu T, Marcon D, Bakeroot B and Jaeger B D 2015 Appl. Phys. Lett. 107 093507
[3] Dong B, Lin J, Wang N and Jiang L 2016 AIP Adv. 6 095021
[4] Uemoto Y, Hikita M, Ueno H and Matsuo H 2007 IEEE Trans. Electron Devices 54 3393
[5] Dutta G, Dasgupta N and Dasgupta A 2016 IEEE Trans. Electron Devices 63 1450
[6] Zhang Z L, Li W Y, Fu K et al. 2017 IEEE Electron Device Lett. 38 236
[7] Tsai C T, Chang K M, Liu P T, Yang and P Y 2007 Appl. Phys. Lett. 91 12109
[8] Chattopadhyay P and Gupta R B 2001 Int. J. Pharm. 228 19
[9] Sun H, Liu M, Liu P and Lin X 2017 IEEE Trans. Electron Devices 65 622
[10] Stoklas R, Gregušová D and Novák J 2008 Appl. Phys. Lett. 93 124103
[11] Hashizume T and Hasegawa H 2004 Appl. Surf. Sci. 234 1
[12] Robertson J 2009 Appl. Phys. Lett. 94 152104
[13] Marrani A G, Caprioli F, Boccia A, Zanoni R 2014 J. Solid State Electrochem. 18 505
[14] Chuvenkova O A, Domashevskaya E P, Ryabtsev S V 2015 Phys. Solid State 57 1
[15] Yamada Y, Yasuda H, Murota K and Nakamura M 2013 J. Mater. Sci. 48 8171
[16]George M M, Craig A K and Manuel U O 2004 US Patent 2004/0049079 A1
[17] Yu J, Liu Y, Tang J, Wang X and Zhou J 2014 Angew. Chem. 53 9512