apl 2000 2250 interfacetraps

Upload: kurabyqld

Post on 03-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 Apl 2000 2250 Interfacetraps

    1/3

    Interface trap profile near the band edges at the 4H-SiCSiO2 interface

    N. S. Saksa)

    Naval Research Laboratory, Code 6813, Washington, DC 20375

    S. S. Manib) and A. K. Agarwalc)

    Northrop Grumman Corp., Pittsburgh, Pennsylvania 15235

    Received 1 November 1999; accepted for publication 18 February 2000

    The transconductance of SiC metal oxide semiconductor field-effect transistors MOSFETs istypically much lower in devices fabricated on the 4H-SiC polytype compared to 6H. It is believed

    that this behavior is caused by extreme trapping of inversion electrons due to a higher density of

    traps D it at the SiC/SiO2 interface in 4H-SiC. Here we present an approach for profiling D it versus

    energy in the band gap using a modified capacitancevoltage technique on large-area MOSFETs.

    We find that D it increases towards the conduction band edge Ec in both polytypes, and that D it is

    much higher in 4H- compared to 6H-SiC for devices fabricated in the same process lot. 2000

    American Institute of Physics. S0003-6951 00 00516-7

    Experimental transconductance reported for SiC metal

    oxide semiconductor field-effect transistors MOSFETs is

    relatively poor for both 4H- and 6H-SiC polytypes. Further-more, the transconductance is typically worse on 4H com-

    pared to 6H material,1,2 even though 4H-SiC has a higher

    bulk electron mobility. Poor transconductance in SiC MOS-

    FETs is generally believed to be caused by trapping of in-

    version layer electrons at traps D it at the SiO2/SiC interface

    in both polytypes.3 7 Recent values reported for D it on 6H-

    SiC are as low as mid-1010 traps/cm2 eV by using a low-

    temperature reoxidation process.8 Despite this low D it value,

    however, the best reported inversion layer mobilities for 6H-

    SiC MOSFETs are typically only about 40100 cm2/V s,

    which is considerably less than the expected value of about

    half the bulk mobility (bulk300 cm2/V s). Reported in-

    version layer mobilities for 4H-SiC MOSFETs are evenworse, typically 0.125 cm2/V s, despite its higher bulk

    mobility (bulk900 cm2/V s). Recent Hall measurements

    on 6H-SiC MOS inversion layers reveal substantial electron

    trapping which is inconsistent with the apparently low D it(1 21011) in these samples.9 To explain this apparent

    discrepancy, it has been proposed that D it increases very rap-

    idly near Ec .3 7 Here we report measurements of D it near

    both band edges in 4H-SiC MOSFETs using a modified

    GrayBrown technique.10 We find that D it does increase dra-

    matically near Ec in both SiC polytypes, and that D it near Ecis an order of magnitude higher in 4H compared to 6H de-

    vices fabricated in the same process lot, consistent with the

    lower transconductance in 4H.

    N-channel, nonself-aligned polysilicon gate MOSFETs

    were fabricated on 4H- and 6H-SiC wafers. The 4H-SiC wa-

    fers have a 10-m-thick, lightly doped (11016/cm3)

    p-type epitaxial layer grown over a heavily doped p

    substrate.11 n source and drain regions were created by

    multiple energy nitrogen implantation to create a box

    doping profile approximately 0.5 m deep. This implant was

    annealed uncapped at 1300 C 4H-SiC or 1200 C 6H-

    SiC this was the only intentional process difference be-

    tween the two SiC polytypes . After a standard RCA clean, a31 nm gate oxide was grown wet at 1100C for 3 h, fol-

    lowed by a wet reoxidation at 950 C for 2.5 h to obtain a

    lower D it .8 Other details of the fabrication process are de-

    scribed elsewhere.12

    Gate capacitance was measured as a function of gate

    voltage on large area ( 100 m long100 m wide)

    MOSFETs. The MOSFETs were bonded in ceramic pack-

    ages and measured under vacuum at temperatures from 80 to

    570 K. During the CapacitanceVoltage (C V) measure-

    ments, the MOSFET n source and drain are tied to the

    p-type substrate. The source and drain act as a source of

    minority carriers electrons , so that an effective low-

    frequency C V curve can be obtained. This is similar toquasistatic C V13 where a slow direct current dc sweep

    rate is used, wherein equilibrium is maintained by slow ther-

    mal generation of minority carriers. Here, the rate-limiting

    step is electron conduction in the MOS inversion layer from

    the source/drain to the center of the MOSFET channel .

    Since inversion conduction in the SiC MOSFETs is poor, a

    low 10 Hz frequency was used to minimize the effect of the

    high inversion layer series resistance.

    Although both minority and majority carrier densities

    are in equilibrium with the slowly varying applied dc gate

    bias, the occupancy of interface traps far from the band

    edges is not, due to the very wide 3 eV SiC band gap.

    The thermal emission rate for carriers trapped in midgap in-

    terface traps is very long 1016 s at 22 C! . The key to the

    C V measurement technique here is that equilibrium of the

    traps is re-established only in strong accumulation or strong

    inversion when the trapped carriers are removed by recom-

    bination with high densities of free carriers in the

    accumulation/inversion layers which is fast , not by emis-

    sion which is slow . Thus, although the C V data obtained

    in these measurements looks like classic low-frequency data,

    the standard quasistatic analysis13 cannot be applied, since

    most interface traps are not in equilibrium.

    Experimental C V results for 4H MOSFET at 20 and

    a Electronic mail: [email protected] Present address: Sandia National Laboratory, Albuquerque, NM 87185.c Present address: CREE Inc., Durham, NC 27703.

    APPLIED PHYSICS LETTERS VOLUME 76, NUMBER 16 17 APRIL 2000

    22500003-6951/2000/76(16)/2250/3/$17.00 2000 American Institute of Physics

  • 7/28/2019 Apl 2000 2250 Interfacetraps

    2/3

    200 C are shown in Fig. 1. Theoretical curves calculated

    using the known oxide thickness and substrate doping are

    also shown for comparison. The theoretical curves in Fig. 1

    use a flatband voltage Vfb of3.0 V which was obtained by

    matching theory and experiment at 200 C .14 At 20 C, the

    experimental data looks like a good low-frequency curve. A

    larger experimental gate voltage swing is required to move

    from accumulation to inversion compared to theory because

    of charge in interface traps. Good low-frequency data is also

    obtained at 200 C, although some distortion is apparent. The

    average D it in this device can be calculated from the differ-

    ence between theory and experiment using D itavCox(VfbVth)/ (qE), where E is the change in sur-

    face potential from flatband to inversion, Cox is the oxide

    capacitance, and (VfbVth) is the difference in gate volt-

    age swing between experimental and theoretical curves from

    flatband to inversion. In Fig. 1 at 200 C, (VfbVth)

    0.31 V and E2.5 eV. The extracted value for average

    midgap D itav is reasonably low at 8.51010 traps/cm2 eV.

    Note: for convenience, V

    this defined here to occur at the

    same capacitance as flatband. In Fig. 1 at 20 C, this point is

    about 0.07 eV closer to the conduction band edge than the

    normal definition of inversion for a MOSFET, s2*bulk .

    C V curves for the same sample at various temperatures

    are shown in Fig. 2. Both the flatband and inversion points

    move as the temperature changes, but in opposite directions.

    Gray and Brown demonstrated that this temperature depen-

    dence is caused partly by the temperature dependence of the

    flatband capacitance itself, and partly by the changing occu-

    pancy of the interface traps which causes a change in the

    trapped charge. By subtracting the amount of threshold shift

    due to the theoretical temperature dependence, the remainder

    can be used to calculate D it(E) at flatband.10

    Here we alsouse analogous changes in the threshold voltage Vth to deter-

    mine D it(E) at threshold. Changes in Vth can be measured

    here because MOSFETs are used rather than MOS capaci-

    tors . The analysis is essentially the same as at flatband: At

    each temperature, the theoretical capacitance at threshold is

    calculated and subtracted from the measured shift. D it(E) at

    threshold is calculated from the remainder. In this technique,

    at both flatband and threshold, the difference in interfacial

    trapped charge is calculated from two C V curves at suc-

    cessive temperatures. Therefore, the calculated D it(E) is in-

    dependent of unknown constants such fixed oxide charge, so

    long as these values do not depend on temperature.

    Using this approach, analysis of the C V data in Fig. 2

    yields a D it(E) profile for a 4H-SiC MOSFET as shown in

    Fig. 3. Similar data for a 6H-SiC MOSFET from the same

    process lot, cleaned and oxidized at the same time, is also

    shown in Fig. 3.15 Average midgap D it values for both 4H

    and 6H samples, obtained at 200 C as discussed earlier, are

    shown by the dashed lines in Fig. 3. D it is substantially

    higher near the band edges than the average midgap values.

    Electron densities in strong inversion range from mid-1011 to 1013 e/cm2 depending on gate voltage. If D it is rea-

    sonably low (11011), electron trapping should not ma-

    terially reduce MOSFET transconductance in strong inver-

    sion. However, the important D it values are those at

    inversion; i.e., near Ec . The data in Fig. 3 show that D it(E)

    increases dramatically near Ec in both polytypes, and that D itis about an order of magnitude worse higher in 4H com-

    pared to 6H. This result is consistent with reports that the

    FIG. 1. Comparison of experimental and theoretical low frequency C V

    curves at 20 C top and 200 C bottom in a 4H-SiC MOSFET. The total

    number of interface traps between flatband and inversion can be determined

    from the difference between the experimental and theoretical curves be-

    tween flatband and inversion.

    FIG. 2. C V curves for a 4H-SiC MOSFET from 140 to 200 C.

    FIG. 3. Interface trap density D it(E) vs energy in the SiC band gap, refer-

    enced to the band edges, for 4H-SiC symbols and 6H-SiC solid lines ,

    calculated using GrayBrown analysis. Average midgap values dashed

    lines are calculated from 200 C data. D it(E) near Ec is much higher in 4H-

    compared to 6H-SiC, which is consistent with the lower transconductancetypically reported for 4H MOSFETs.

    2251Appl. Phys. Lett., Vol. 76, No. 16, 17 April 2000 Saks, Mani, and Agarwal

  • 7/28/2019 Apl 2000 2250 Interfacetraps

    3/3

    effective mobility is typically a smaller in 4H compared to

    6H MOSFETs, despite the higher bulk mobility in 4H.1,2 In

    Fig. 3, the integrated density of traps close to Ec within 0.2

    eV in the 4H sample is 1.01013 traps/cm2. Thus, most

    electrons in the 4H MOS inversion layer will be trapped

    even at the highest gate biases, leading to a very poor trans-

    conductance. Hall Effect measurements on the 6H MOS-

    FETs from the same process lot reported elsewhere9 reveal

    much less electron trapping ( 1.51012 trapped electrons/cm2 , indicating a much better interface in 6H.

    In Fig. 3, D it near the valence band edge Ev shows a

    broad peak which is similar for both SiC polytypes. D it near

    Ev

    is only about a factor of 2 higher in 4H compared to

    6H-SiC, much less than the difference measured near Ec . It

    has been pointed out that artificial peaks in D it near Ev in a

    p-type semiconductor can be obtained from GrayBrown

    analysis if the traps capture cross-section varies rapidly

    with energy.16 Indeed, recent measurements of using con-

    ductance measurements on n-type 4H-SiC MOS capacitors

    apparently show a rapid decrease near Ec .17 Also, Gray

    Brown analysis is not self-consistent because the contribu-tion of interface trap capacitance itself to the theoretical flat-

    band capacitance is not taken into account.16 These sources

    of error could cause the D it peak observed near Ev in Fig. 3,

    although a detailed calculation is beyond the scope of this

    letter. Finally, both the 4H and 6H data show a sharp mini-

    mum in D it(E) at 0.15 eV below Ec . Although this mini-

    mum in D it has been reproducibly observed in all 4H- and

    6H- SiC MOS devices measured here about six devices , it

    is not known whether this minimum appears in SiC MOS

    devices from other process lots.

    Profiling D it(E) by changing the temperature and using

    GrayBrown analysis is a relatively insensitive technique

    and can only be used effectively on samples with high D it .For the samples here, a 10 C change in temperature at 22 C

    causes only a small change in the surface potential at flat-

    band 9 mV . It is therefore very important that any

    sources ofC V instability, such as caused by ionic contami-

    nation of the oxide e.g., sodium , be negligible. In order to

    investigate this possible error source, C V curves were

    compared for forward and reverse sweep directions of the

    gate voltage at most measurement temperatures. Bias depen-

    dent instabilities should be evident as a C V hysteresis be-

    tween the two sweep directions at temperatures where the

    sodium is mobile. No such hysteresis was apparent in any

    sample reported here up to the highest measurement tem-

    peratures. However, samples from other process lots did

    show significant hysteresis, indicating problems with con-

    tamination, and making these samples useless for Gray

    Brown analysis. Another effect was also observed which

    limits the accuracy of these measurements. After C V mea-

    surements were made at high temperatures, additional traps

    were found to be created due to the bias-temperature cycling.

    Consequently, the maximum measurement temperature was

    limited to 200 C. This effect is discussed in more detail

    elsewhere.15

    In summary, D it has been profiled near the band edges

    using a modified GrayBrown C V technique on large

    MOSFETs. This measurement approach has several advan-tages: it is simple to implement, D it profiles are obtained near

    both band edges with a single sample, and it is possible to

    obtain profiles very close to the band edges. Perhaps most

    important, it is possible to obtain D it and mobility data on the

    same device, which is a powerful advantage in understanding

    mobility issues. Drawbacks include that MOSFETs are more

    difficult to fabricate than capacitors, and GrayBrown analy-

    sis is relatively insensitive and cannot be used if C V insta-

    bilities are present. Also, the technique has inherent errorswhich limit the techniques accuracy and which need to be

    evaluated more thoroughly.

    D it(E) is found to increase rapidly close to the conduc-

    tion band edge in 4H-SiC mid-1013 traps/cm2 eV compared

    to a low average midgap value low 1011 range . This behav-

    ior is qualitatively similar to that found previously in 6H-

    SiC, but D it(E) near Ec is much higher in 4H. These results

    clearly demonstrate that severe electron trapping at the

    SiC/SiO2 interface is the major factor causing the low trans-

    conductance reported for 4H-SiC MOSFETs.

    The authors thank the staff of the Northrop Grumman

    Corp., Pittsburgh, PA, for fabricating the samples, M. White

    and V. Vathulya for growing the gate oxides, and E. Arnold

    for pointing out Ref. 16. This work was partially supported

    by the Office of Naval Research.

    1 H. Yano, T. Kimoto, H. Matsunami, M. Bassler, and G. Pensl, Interna-

    tional Conference on SiC and Related Materials 1999, Research Triangle

    Park, NC., 1015 Oct. 1999.2 M. K. Das, M. A. Capano, J. A. Cooper, Jr., and M. R. Melloch, Elec-

    tronic Materials Conference, Santa Barbara, CA, June 30July 2, 1999.3 R. Schorner, P. Friedrichs, and D. Peters, IEEE Trans. Electron Devices

    46, 533 1999 .4 D. M. Brown, E. Downey, M. Ghezzo, J. Kretchmer, V. Krishnamurthy,

    W. Hennessy, and G. Michon, Solid-State Electron. 39, 1531 1996 .5 M. K. Das, and J. A. Cooper, Jr., International Conference on SiC and

    Related Materials 1999, Research Triangle Park, NC, 1015 Oct. 1999.6 E. Arnold, IEEE Trans. Electron Devices 46, 497 1999 .7 V. V. Afanasev, M. Bassler, G. Pensl, and M. Schulz, Phys. Status Solidi

    A 162, 321 1997 .8 L. A. Lipkin, D. B. Slater, Jr., and J. W. Palmour, Mater. Forum 2648,

    853 1998 .9 N. S. Saks, S. S. Mani, V. S. Hedge, and A. K. Agarwal, International

    Conference on SiC and Related Materials 1999, Research Triangle Park,

    NC, 1015, Oct. 1999.10 P. V. Gray and D. M. Brown, Appl. Phys. Lett. 8, 31

    1966

    ; D. M. Brown

    and D. V. Gray, J. Electrochem. Soc. 115, 760 1968 .11 Wafers with epitaxial layers were purchased from CREE Research, Re-

    search Triangle Park, NC.12 N. S. Saks, S. S. Mani, A. K. Agarwal, and M. G. Ancona, IEEE Electron

    Device Lett. 20, 431 1999 .13 M. Kuhn, Solid-State Electron. 13, 873 1970 .14 Because the amount of oxide fixed charge is unknown, the flatband volt-

    age Vfb cannot be known a priori, and Vfb must be obtained from experi-

    mental data. However, we emphasize that the value extracted here for Vfb

    is not used in, and does not affect, later calculations of D it(E).15 N. S. Saks, S. S. Mani, and A. K. Agarwal, International Conference on

    SiC and Related Materials 1999, Research Triangle Park, NC, 1015 Oct.

    1999.16 M. R. Boudry, Appl. Phys. Lett. 22, 530 1973 .17

    M. K. Das and J. A. Cooper, Jr., Purdue University and Lori Lipkin, CreeResearch. private communications .

    2252 Appl. Phys. Lett., Vol. 76, No. 16, 17 April 2000 Saks, Mani, and Agarwal