Zhenxing Liu(劉振興) Liuan Li(李柳暗) Jinwei Zhang(張津瑋) Qianshu Wu(吳千樹)Yapeng Wang(王亞朋) Qiuling Qiu(丘秋凌) Zhisheng Wu(吳志盛) and Yang Liu(劉揚)
1School of Electronics and Information Technology,Sun Yat-Sen University,Guangzhou 510275,China
2State Key Laboratory of Optoelectronic Materials and Technologies,Sun Yat-sen University,Guangzhou 510275,China
Keywords: electrical properties and parameters, semiconductor materials, chemical vapor deposition, electronic transport
GaN-based devices have promising applications in the field of power electronics due to their superior properties.[1,2]In order to realize the characteristics and minimize the vertical leakage current of high breakdown voltage, a semiinsulating or high resistivity GaN layer can easily prevent the undesired parallel current paths which are beneath the twodimensional electron gas channel. However, it is already reported that the unintentionally doped-GaN exhibits n-type conductivity. These unintentional donor impurities are O,[3]Si,[4]and nitrogen–vacancy (VN).[5,6]Earlier calculations[7]reveal thatVNmay not be the major participants in the ntype GaN as its formation energy is very high in n-type materials. Recently, a group from the Helsinki University of Technology[5]demonstrated that nitrogen vacancies can be the major participants in the defect kinetics of n-type GaN.This VNis mainly derived from the activation energy of selfdiffusion in the nitrogen sublattice. Conventional synthetic techniques can easily produce VNin the nitrogen sublattice of GaN. Conventional techniques such as thermal decomposition (metal–organic chemical vapor deposition) can occur above 900°C, where the evaporated nitrogen from the sample surface improves the concentration of VN.[5–9]However,the mechanism behind this is still unclear and the major participants in the n-type GaN are still under debate.
Generally, high resistivity of the materials can be achieved through two different techniques such as(i)eliminating the background impurities[10]and(ii)introducing the compensated acceptors. In previous reports, iron (Fe)[11,12]and carbon (C)[13–20]atoms are used as a compensator. Besides,the generation of intentionally edge dislocation in GaN[21,22]are also used as a compensator. Compared to Fe-doped GaN,the incorporation of C atom has produced a significantly high resistive layer. This is due to its wide doping range,better stability, and weak memory effect. The theoretical calculation shows that carbon atoms can preferably substitute the nitrogen site (CN) and form a deep CNacceptor in GaN, verified by many methods.[23–26]The PL spectrum of carbon-doped-GaN shows blue luminescence(BL)and yellow luminescence(YL)peaks at around 2.8 eV and 2.2 eV,along with the band edge (BE) luminescence at around 3.4 eV. The CNacceptors with a negative-to-neutral charge state(?/0)(0.9 eV above valence band)and a 0/+level(0.4 eV above valence band)[25,26]are theoretically calculated which interpret the YL and BL luminescence, respectively. It is also recognized that these deep energy levels are difficult to be electrically ionized[27–31]and the compensation of background carrier is associated with the Fermi level pinning effect at around CNacceptor energy level.[28–31]While the pinning effect occurs when the acceptor trap density is much higher than that of donor trap density,the fermi energy levels and background carrier concentration remain constant as C concentration increases. In this work the background concentration generally varies with the carbon doping concentration.The Fermi level pinning effect at around CNacceptor can be excluded for the reduction of background carrier concentration. Therefore,the formation mechanism of high resistivity layer by C-doping is still under debate.
In this work, we have investigated the relationship between background carrier conduction and carbon doping in GaN on the sapphire substrate. On fitting the temperaturedependent carrier concentration and mobility, it is found that the background conduction at room temperature is dominated by the nitrogen–vacancy (VN). The ionization concentration of CNdeep acceptor is lower,and the donor impurities cannot be sufficiently compensated. So, the incorporation of carbon atom has suppressed the background carrier in GaN,where C occupies the nitrogen site and reduces the VNconcentration.
In order to focus on the impact of carbon doping on the background carrier conduction in GaN, we chose a quasivertical diode for C-doped material instead of high electron mobility transistor(HEMT)structure,as heterojunctions may cause unnecessary influence on the investigation of materials. Additionally, employing a quasi-vertical design results in a better field distribution and the peak electric field is moved away from the surface. The samples were prepared by metal–organic chemical vapor deposition using trimethylgallium (TMGa), trimethylaluminum (TMAl), and ammonia(NH3)as the Ga,Al,and N precursors,respectively.Hydrogen was used as the carrier gas.From the bottom to top,the epitaxy structure consists of a 30-nm-thick low temperature(LT)GaN layer,a 500-nm-thick n+-GaN layer(Si:~2×1019cm?3),a 10-μm-thick carbon-doped GaN(C–GaN)bulk layer. All the epitaxy layers were grown at similar growth conditions except the top GaN bulk layer. Four samples(labeled as A,B,C,and D in Table 1)at different carbon concentrations were prepared by simply adjusting the pressure of the chamber. In our previous report,it has been described that the concentrations of carbon increase along with the decrease in reactor pressure.[14,15]This is also confirmed from the secondary ion mass spectrometry (SIMS) measurement (Fig. 1(c)). A quasi-vertical structure Schottky barrier diode(SBD)was fabricated for the electrical measurements (Fig. 1(a)). The process started with a 10-μm deep etching using inductively coupled plasma etching with a SiO2hard mask. After treatment of 5%tetra methyl ammonium hydroxide solution at 85°C for 60 min,then a 200-nm thick SiO2passivation layer was deposited. A metal stack of Ti/Al/Ni/Au was evaporated and annealed at 870°C for 30 s in a nitrogen ambient, to form a cathode electrode. Finally,Ni/Au(25 nm/65 nm)anode metal was evaporated. The Hall measurements were performed with the help of a common van der Pauw structure with four ohmic contact pads,prepared by annealing the indium(Fig.1(b)). The three conductive layers are connected in shunt during measurement.

Fig. 1. (a) Schematic cross-section of the fabricated SBD on sapphire. (b)Multilayer conduction model for Hall measurement. σ1, σ2, and σ3 are the 10-μm-thick carbon-doped GaN, the 500-nm-thick n+-GaN layers, and the GaN/sapphire interface layer,respectively. (c)The secondary ion mass spectrometry(SIMS)measurement. The unit 1 bar=105 Pa.
The carrier concentrations (the carrier is electron) are studied with the help of room-temperature Hall measurement which decreases along with the increase in the concentration of carbon doping (Fig. 2(a)). On the other hand, the crystallinity of GaN films is characterized by the full width at half maximum peaks of (002) and (102) planes changed slightly,with the screw- and edge-type dislocation density approximately around~107cm?2and~108cm?2levels, respectively. A 3 μm×3μm scaled atomic force microscopy measurement shows that the root mean square roughness (RMS)of the as-grown samples are comparable. The specific onresistance increases with C concentration extracted from the forward current–voltage (I–V) characteristics. The carrier spatial distribution extracted from capacitance–voltage(C–V)measurement(Fig.2(b))confirms the variation of carrier concentrations as summarized in Table 1.

Table 1. The growth parameters and electrical properties of GaN samples at different methods. Here, Hallmea. (RT) represents the carrier concentration,obtained from the room temperature (RT) Hall measurement and Hallextr. represents the carrier concentration which is extracted from the temperaturedependent Hall,using a bilayers conduction model. The deviation is defined as the ratio between the maximum and the minimum to the mean.

Fig.2. Variation of carrier concentration(con.),dislocation density,and surface roughness versus reactor pressure. (b) The spatial distribution of net carrier concentration is calculated from the C–V curves of SBD. The unit 1 bar=105 Pa. RMS:mean square roughness, RT:room temperature, XRD:x-ray diffraction,AFM:atomic force microscopy.
The results of temperature-dependent Hall show that initially, the concentration of carrier (open circle in Fig. 3(a))decreases along with the increase in temperature, and then it nearly shows no variation at relatively low temperature. This phenomenon is inconsistent with the classic donor frozen out and suggest another origination dominant high carrier concentration at low temperature. As shown in Fig.1(b),the epitaxial structure has three conduction layers such as GaN/sapphire interface layer,[32–35]n+-GaN layer, and the GaN bulk layer.Generally, the GaN/sapphire interface and the n+-GaN are degenerate layers, which add additional parallel conducting channels and affect the conduction of GaN bulk layer at all temperatures.[32–35]As reported in previous works,the donor band would merge with the conduction band and produce zero activation energy. This zero-activation energy is achieved when the density of carrier is higher than 6×1018cm?3. Due to similar conduction behavior,we simplified these two degenerate layers as one single layer and extract the electron concentration and mobility of bulk layer. These are calculated with the help of a bilayers conduction model as reported by Look and Molnar.[8]In this model,the total sheet conductanceσsis considered to be the sum of the sheet conductance in all conductance channels:

It is worth noting that the carrier concentration ofC–Vis nearly one order of magnitude lower than that of Hall. However, they have a similar carrier decreasing trend as the concentration of carbon increases.This concentration discrepancy can be explained in terms of multilayer conduction in the following section.
whereμ1andn1are the mobility and electron concentration of the bulk layer,which can be extracted fromμHandnH(the mobility and electron concentration of measurement by Hall),nS-LT andμLTare the sheet carrier concentration and mobility of the degenerate layer at low temperature and can be obtained from an open circle in Fig.3.

Fig.3. The measured and extracted temperature-dependent electron concentration(a)and mobility(b)of the GaN bulk material based on the two-layer Hall model. nH and n1 are measured and extracted data,bulk carrier nb-LT is the bulk carrier concentration of degenerate layer at low temperatures.
Generally,the variation of background concentration can be affected by donor impurity (Si, O), point defect (VN, CN,CGa,etc.),and dislocations(change insignificantly here). The SIMS measurements are shown in Fig.1(c),where we can see that the Si concentration is close to the detector limitation,and oxygen is kept at approximately 5×1016cm?3for all samples. Both the Si and O concentrations have no obvious variation with the reactor pressure. On the other hand,the variation in concentration of carbon has a direct relation with the reactor pressure. Therefore, we suggest that the most possible source for the background carrier is a point defect. PL measurement has been employed(Fig.4(a))to confirm the carbon acceptor levels, with BL, YL, and BE peaks at around 2.8,2.2,and 3.4 eV,respectively. Comparative results of the intensity ratio of BL and YL with the BE are shown in Fig. 4(b).From Fig.4(b), it can be inferred that the intensity of YL increases with carbon doping,and the BL is enhanced when the carbon exceeded 1×1017cm?3. This result confirms that the concentration of CNin GaN layer increases along with the increase in carbon concentration. However,as reported in previous works,only approximately 1%of the incorporated carbon atoms can be transformed into the active CNacceptor(approximately 1×1016cm?3in our experiment).[31]This concentration is not enough to pin the Fermi level at around CNacceptor,resulting in an n-type conduction. So,there might be other reasons which can decrease the carrier concentration.

Fig.4. (a)Room-temperature photoluminescence spectrum. (b)The ratio of intensity of blue luminescence and yellow luminescence with the band edge luminescence.
To evaluate the contribution of possible impurities(O,C,VN,etc.)on the conduction of GaN,we fitted the temperaturedependent carrier concentration with the following ionization theory.[27,36]Here,we considered the carrier concentration as the sum of all impurities, which are ionized simultaneouslyversustemperature,as shown in Fig.5

In this equation,nis the electron concentration,Ncompis the hole concentration of background compensating impurities,MandNare the numbers of donors and acceptor, ?EDi(?EAi)andNDi(NAi) are the activation energy and concentration of the donor (acceptor),giis the degeneracy factor of the electronic states in the band gap.giis considered to be 2 and 4 for donor and acceptor,respectively. The expressions


Fig.5.(a)Typical numerical fitting of the temperature-dependent carrier concentration of sample A,short dash lines are fitted data(orange: total ionized net electron concentration,red: ionized oxygen concentration,pink: ionized CN acceptor concentration, blue: ionized nitrogen vacancy concentration),the black square is the experimental data. (b) The extracted parameters of samples at different carbon dopings.
are the effective state density of conduction and valence band.The effective electron mass of 0.2m0(m0is the free electron mass).level is also considered. However, the ionized CNconcentration is lower by a few orders of magnitude. This result is also consistent with the already reported work that the CNcannot produce p-type conductivity at room temperature. From Fig.5(b),we can observe that both the nitrogen–vacancy and ionized oxygen decrease with the increasing of carbon concentration.However,the concentration of oxygen does not change along with the concentration of carbon doping,as shown in the SIMS measurement. This may be related to the formation of CNON[24]complex which is a deep trap (0.75 eV above the valence band)and also contributes to the YL band.

Fig.6. Typically mobility fitting curves of sample A using Matthiesen’s rule(a), “ii” represents the ionized impurity scattering, “dis” represents the dislocation scattering, “VN” represents the nitrogen–vacancy scattering, “pz”represents the piezoelectric potential scattering, “ac” represents the acoustic deformation potential scattering,“po”represents the polar optical phonon scattering. The extracted nitrogen vacancy values versus the carbon concentration(b).
As shown in Fig.5(a), the Hall data of sample A is well fitted with three kinds of impurities based on the previous discussion.The oxygen impurity(ON)[3]with a relatively smaller activation energy of 41 meV(much greater than that of Si)is obtained from the Arrhenius equation(Fig.5(a))which dominates the carrier at low temperature,and finally,it is saturated at room temperature(red dotted line in Fig.5(a)). Therefore,nitrogen–vacancy (VN)[8]with a higher activation energy of 65 meV is obtained from the Arrhenius equation which plays a key role at room temperature.The CNacceptor with a 0.4-eV
The contribution of ionized impurities to the conduction can also be obtained from the temperature-dependent mobility curves. Generally, the mobility can be analyzed from Matthiesen’s rule (1/μ=∑μi), assuming the scattering events are independent of each other. In our fitting, the key scattering items are comprised of ionized impurity scattering with the donor and acceptor concentration which are derived from carrier concentration fitting, polar optical phonon scattering(po),acoustic deformation potential scattering(ac),piezoelectric potential scattering(pz),and dislocation scattering(dis)as reported in previous literature.[8,37]The concentration of VNis chosen as the only variable for the numerical fitting of mobility(Fig.6)with its scattering contribution which is described[35]

Based on a simple square-well scattering potential with well depthU0=2 eV and widtha=c=5.1 °A, the fitting results are shown in Fig.6. Analysis of Fig.6 demonstrates that the nitrogen–vacancy decreases with the increase in carbon concentration. This result is consistent with the carrier fitting data as shown in Fig.5(b)and also summarized in Table 1.

Fig. 7. Temperaturedependence of(a)resistivitycalculatedbyextractedn and μ,variable rangehopping ftiting formula(5)wasused,(b)T?1/4 dependence of resistivity. The hollow circle is data point,short dashes are the ftiting data.
When the ionization of impurities is suppressed and the carrier is frozen to the donor level,then the donor energy levels contribute to the hopping conduction mode in GaN at the lowtemperature region. So,the conduction cannot be determined by the conduction band but the shallow donor such as ONand VN. Therefore, the donor density can be extracted from the hopping conduction fitting. Hopping conduction transportation is commonly found at low temperatures which show a clear non-zero temperature dependence law ofT?1/4.[9,38]The donor density NDconcentration is calculated from the variable range hopping fitting:



Fig. 8. Comparison of VN and carrier concentration extracted from the experimental and numerical fitting at room temperature.
The nitrogen–vacancy concentration is calculated from different methods as well as the background net carrier concentrationversusthe carbon concentration,as shown in Fig.8.It is found that the background carrier density and VNconcentration are comparable, and show a similar decreasing trend along with the increase in carbon concentration. Besides, the ionized CNis 1–2 order of magnitude lower than that of donor impurity in the case of samples A and B with high carbon concentration and above 6 order of magnitude lower in samples C and D(Fig.5(b)).The reason behind this is the high ionization energy of CN. Therefore, it is proposed that the reduction of carrier concentration at room temperature is because of acceptor compensation but the decreasing of VNalong with carbon doping. It is worth noting that the decreasing of carbon concentration is nearly one order of magnitude larger than that of the variation of background carrier density. This may be due to the nitrogen sites of the GaN lattice,which can be filled by nitrogen(Ga–N bond), empty(VN), and occupied by oxygen(ON). The reduction in VNis obtained when the carbon atoms fill the empty nitrogen sites and consequently reduce the carrier density. However, the carbon atoms can also substitute the nitrogen atom (Ga–C bond) and/or form CNONcomplex.So, only part of the carbon atoms can reduce the background concentration. For the sample at a higher concentration of carbon doping,the difference between the VNand ONconcentrations(sample A of Fig.5(b))became smaller and enhance the contribution of ONto the conduction. Finally, this causes the undesired deviation of NDwhich is extracted from different methods(as shown in Fig.8 and Table 1).
In summary,the mechanism of high resistive GaN by applying carbon doping has been studied to investigate the impact of carbon concentration on the background carrier conduction.As expected,the incorporation of carbon atom has effectively suppressed the n-type background carrier concentration. By fitting the carrier concentration and mobility curves,it is found that the n-type conduction of GaN is mainly originated from the nitrogen–vacancy (VN) at room temperature.The Fermi level pinning effect at around CNacceptor can be excluded for the reduction of background carrier concentration. The PL characterizations showed that carbon is a preferable dopant and occupy the nitrogen site, and form a deep CNacceptor in the resulted GaN.Later, we found that an extremely low hole concentration can be ionized at room temperature, which is insufficient to compensate the background carriers. So,it is concluded that the carrier suppression is because of the substitution of VNby carbon doping and not by the acceptor trap.