Click for next page ( 16


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 15
2 State of the Art of Wide Bandgap Materials This chapter surveys the state of the art for the three major wide bandgap materials for high-temperature semiconductor devices: silicon carbide (SiC), the nitrides, and diamond. This chapter is not a comprehensive examination of all the properties of the different materials, but does examine closely those properties related to high- temperature operation. The intrinsic properties of the wide bandgap materials versus those of the more common silicon and gallium arsenide (GaAs) materials are compared in Table 2-1.~ Although silicon and GaAs are not considered in this chanter because of their expected -, devices and interconnects of these materials are discussed in Appendices A and B. respectively. limited high-temperature applicability SILICON CARBIDE Materials Description and Properties Of the wide bandgap materials, SiC is by far the most developed. The earliest reported recognition of the silicon- carbon (Si-C) bond is by Berzelius in 1824. SiC has been produced in the United States since 1891 when Eugene G. Acheson (1893) of Monongahela City, Pennsylvania, melted a mass of carbon and aluminum silicate by passing a current through a carbon rod immersed in the mixture. A variety of vapor-transport furnaces have been used in this century to grow boules of single-crystal SiC. In addition, high-purity homo-epitaxial single-crystal films of SiC have been grown in both horizontal and vertical chemical vapor deposition (CVD) reactors. ~ Table 2-1 was developed for comparative purposes using the data that was available during the course of this study. This table should not be considered a definitive tabulation of the properties of these materials, .. ~ . ~ . ~ ~ since new, more accurate data are constantly being accumulated for most of these materials. 15 Moisson reported in 1904 and 1905 that hexagonal crystals of SiC were present in meteoritic specimens from Canyon Diablo, Arizona. Naturally occurring SiC was viewed as exclusively of extraterrestrial origin until 1957. However, SiC has recently been discovered in alluvial sands and in Kimberlite breccia in a number of locations on the earth. SiC forms in a variety of crystal structures, termed polytypes, of which over 175 have been described in the literature (Verma and Krishna, 1966; Pandy and Krishna, 19831. Only simple polytypes are of interest for SiC devices. Their basic crystallographic stacking sequences and most common notations are illustrated in Table 2-2 (Verma and Krishna, 19661. The optical properties of SiC do not differ very much from polytype to polytype (Figure 2-1~. To better understand SiC, a brief discussion of electronic band structure is warranted. Band-structure calculations for SiC have been made for the past 30 years, but theorists have concentrated on the zincblende 3C-SiC polytype and the wurtzite 2H-SiC structure since the other polytypes are much more complicated due to their much larger unit cells. The accuracy of such calculations has recently been considerably improved and currently there is a sizable effort to work on the band structures of 4H-, 6H-, and l5R-SiC. Early band-structure calculations of 3C and 2H are shown in Figures 2-2 and 2-3 to provide a qualitative "feel of the neighborhood" where the maxima in the valence band and the minima in the conduction band are likely to be located. Since both 3C-SiC and 2H- SiC are ircdirect-gap semiconductors, it is reasonable to assume that all polytypes are indirect-gap semiconductors. Indeed, experiment has verified that in addition to 3C- and 2H-SiC, 4H-, 6H-, 8H-, 15R-, 21R-, 27R-, and 33R- SiC are also indirect-gap semiconductors. Figure 2-4 summarizes the experimentally observed excitor bandgaps

OCR for page 15
3 . U. TIC V) C) V, He c . , 0 e .O m En o 0 Cal C) 1_ Go . ~ Cut ~ . Cal -f) ~ O 00 ~O . ~O O - ) O to Of Alp _. O ~ ~ t~ of In to to. c,) O O ~ ~00 To to o ._ Cal Cal P4 ~ O ~ _' ~1 ~I ~A c~ 88 g ~_, ~A o ~ ~ o 0 c~I 0 c~l 0 _4 o - o 0 t 0 ~ o. ~ 0 ~ 0 ~ ~ ~i o 1 ~ 0 ~ g ~o c., 1 1 0 0 ~ 8 o~ ~ ~ ~- ~o 8 o \~, ~ ~C ~ ~o ~ ~4 ~Cq ._ . C> o ~ ~C ;> c~ ~_ 2 ~ t) ~ o ~ ~ o V, ~ ~ 16 g o o 1 o ~ o oo o ~oo ~g ~o o oo mo _4 _I o~ U. P~o 3 ', ._ o~ C) ~._ m ~ ._ .> C~ ._ CO . ~ ._ ~ ._ _ C) ~ ~4 O ~ 5 o ~ ~CO~ =3 ~ ~ _ ~ ._ .> C~ o- - c~ u, 'e 'o .~ o~ o o ~ o ~ ca) 4 - ^ c) - 11 ~ ~. c ~ ~ o ~s

OCR for page 15
State of the Art of Wide Bar~gap Materials TABLE 2-2 Notations for Selected SiC Polytypes Ramsdell Notation 3C 2H Stacking Sequence ...ABC... ...AB... Zhdanov Notation 11 4H . ABAC 22 6H . ABCBAC 33 15R . ABCBABCACBCABAC (32)3 and their temperature variation. Experiment has also given an estimate of the binding energy (27 meV) of the excitor in 3C-SiC. Assuming that this value will not be very different in the other polytypes, the actual bandgap, KG, can be estimated by adding 27 meV to the known value of ant no JO o 1n-2 1 no 10-4 - (a) Face plane - (2110) o-1 _ _ _ _ _ _ i 1111"1 ' ' '11' 1''1 1 1 11""1 1 1 1] ~| E no ~ 7! it/ \,, ~ I , v k I! o/ ) Plasma / J Frequency / 1 ~ ~ two ~ ED 10-5 10-l 10 , , . I, , . . . 1o1 Wavelength (pLm) FIGURE 2-1 Average values of Me optical constants of SiC from We vacuum ultraviolet to Me middle infrared. NOTE: n0 = index of refraction; Lo = extinction coefficient. 17 the excitor bandgap, EGX. Estimates of room-temperature values of both EG and EGX are given in Figure 2-4. The thermal conductivity is shown in Figure 2-5. The electrical properties in the various polytypes can be very different because the actual conduction-band minima in the various polytypes will not be in exactly the same positions in the Brillouin Zone. In addition, there is the extra complication of having a different number of nonequivalent sites in different polytypes as a consequence of different size unit cells. This is illustrated for the donor nitrogen in Table 2-3. SiC may be doped e-type with nitrogen up to at least 10~9 cm~3. The acceptors aluminum and boron can be used to dope SiC p-type to at least 5 x 10~8 cm~3. Nitrogen is difficult to keep out of the growth process, and at present unintentional concentrations of nitrogen in the range of 10~4 cm~3 are found in the best epitaxial films. This is sufficiently low not to interfere with current device fabrication. Deep electronic states due to scandium (Tairov et al., 1974), titanium (Patrick and Choyke, 1974), and vanadium (Mater et al., 1992) have been studied in some detail in various polytypes of SiC. Other deep states, termed Do and D,,, due to implantation or radiation damage have also been widely studied. Many other impurity defect complexes have been observed during annealing of irradiated samples from 0-2000 C. Methods of Fabrication Bulk Growth The commercial potential of SiC semiconductor technology has been enhanced by recent significant progress in the growth of large single-crystal SiC boules.

OCR for page 15
Materials for High-Temperature Semiconductor Devices 20 ,2 1 10 8 r15 6 4 r, _ 2 a) 0 F15 >` Al UJ -10 -20 X3 ~1 x1 ~ r,s -~ ~ 1 1 ~ I; K X ~r A L /~L r k A1 / FIGURE 2-2 Calculated band structure of 3C-SiC. SOURCE: Based on Hemstreet and Fong (1974). For many years, the lack of suitable SiC crystal-growth processes inhibited the commercialization of this promising semiconductor material. There are two properties of SiC that make the growth of bulk single crystals more difficult than that of silicon. First, it does not melt under any reasonably attainable pressure; rather it sublimes at temperatures above 1800 C. Thus, conventional growth-from-melt techniques (e.g., as for silicon or GaAs) cannot be used for SiC crystal growth. Second, different polytypes with different electronic characteristics can grow under apparently identical conditions (Knippenberg, 19631. A completely satisfactory model for the formation of the various polytypes does not exist. Despite these difficulties, major progress has recently been made in SiC boule growth. The diameter of commercially grown, single-crystal boules is typically 30 mm, and prototype boule diameters have exceeded 50 mm. Currently, there is interest in at least five of the SiC polytypes: 3C-SiC, 2H-SiC, 4H-SiC, 6H-SiC, and 15R- SiC. Boules of 4H, 6H, and 15R have been grown, and wafers from 4H and 6H boules are commercially available. No s~gn~ficant-s~zed boules of 3C have been 18 reported. To date, 2H has only been grown in the form of 63 small, millimeter-sized needles. ' There are several key review papers that discuss the growth of bulk SiC single crystals (Knippenberg, 1963; Tairov and Tsvetkov, 1983; Powell and Matus, 19891. This section summarizes some of the early work and L3 describes recent developments for which information is publicly available. Much of the current technology is considered to be proprietary and has not been published. Although growth-from-solution techniques have been L3 tried, the most successful growth techniques are based on the sublimation of SiC. Background. Prior to the mid-1950s, small , hexagonally shaped SiC platelet crystals were available through the industrial Acheson process for making abrasive material (Knippenberg, 1963~. In 1955, Lely developed a laboratory sublimation process for growing crystals that were much purer (Lely, 19551. In the Lely process, a hollow cavity was formed inside a charge of polycrystalline SiC. The charge was heated to about 2500 C in a graphite tube furnace at which point the SiC sublimed and condensed on slightly cooler parts of the cavity. Growth took place on a thin, porous graphite cylinder that formed the wall of the cavity. Nucleation was uncontrolled and the resulting crystals were randomly sized, hexagonally shaped or-SiC platelets. These platelets often exhibited a layered structure of various (x polytypes. The predominant polytype (generally more than 75 6 3 2 1 o . ~ -M-am t _A56 1 r \ .. R L U M At: W: M1 r < nit / Band ~ it/ \ ~ 1/ ~ r ~ / H3 As.6 Or\ , 1 K2 .3~: S k FIGURE 2-3 Calculated band structure of 2H-SiC. SOURCE: Based on Hemstreet and Fong (1974).

OCR for page 15
State of the Art of Wide Bandgap Matenals ~[it 3.3063.300 3.327 ---em SiC +~~~~~"~ 2H '4H " " "N OH `;33R _ ~ a' - x (4.2K) (RT) (8T) EGX (2H) 3.33C (4H) 3.265 (OH) 3.023 (33R) 3.003 (15R) 2.986 (21 R) 2.853 c, >` ~ (24R) 2.73 ~ o c' x UJ (OH) 2.80 (3C) 2.39O i5R - ~ ~ ,-_ ' 2~1 R 8H`" '24R~ `` 3C 2.30 "` 2.20 I I , 0 200 400 600 800 Temperature (K) _ .3.235 3.262 3.20 3.10 3.00~2.995 3.022 2.972 2.999 .2.957 2.984 2.90 28o~.82 2.827 277 2.70 -2.71 2.74 2.60 2.40 .2.360 2.387 Energy Gap - EG(eV) = EGx(eV) + (BE)X Exciton Binding Energy = ~ B E) x ~ 0.027 ev FIGURE 2-4 Summary of Me experimentally observed excitor bandgaps and Weir temperature variation for Me different SiC polylypes. percent) was 6H, followed by 4H and 15R. Although much was learned about SiC from investigations of these crystals over the next 30 years, the process was not suitable for commercial development of SiC. In the 1970s, Tairov and Tsvetkov (1978, 1981) developed a modification of the Lely process (now commonly called the modified sublimation process, or the modified Lely process) in which growth occurred on a seed crystal. Although some research groups have been somewhat slow in adopting this process, it is now being developed in many labs in Russia, Germany, Japan, and the United States. The basic elements of the modified sublimation process are shown in Figure 2-6, which is a schematic diagram of the configuration used by Westinghouse. Nucleation takes place on a SiC seed crystal located at one end of a cylindrical cavity. A temperature gradient is established within the cavity such that the polycrystalline SiC is at approximately 2400 C and the seed crystal is at approximately 2200 C. At these temperatures and at reduced pressures (argon at 200 Pa), SiC sublimes from the source SiC and condenses on the seed crystal. Growth rates of a few millimeters per hour can be achieved. Current Status. Cree Research Incorporated of Durham, North Carolina, is the only commercial source in the world of SiC wafers produced from boules. Cree is currently selling 30-mm-diameter wafers of both 4H- and 6H-SiC. Other companies and institutions, known to be producing SiC boules for internal consumption, include Westinghouse, ATM, Siemens, Sanyo, Nippon Steel, 280 Kyoto University, and Kyoto Institute of Technology. Both Cree and Westinghouse have demonstrated boules (and wafers) of up to approximately 50 mm in diameter. Despite the fact that SiC is extremely hard (between sapphire and diamond in hardness), techniques for cutting and polishing wafers are currently in use. However, the capability is far short of that for silicon. As a result, the polished surface of commercial SiC wafers contains many scratches and defects. Some defects introduced into the wafer by cutting and polishing can be removed by suitable pregrowth (i.e., prior to epitaxy) etching processes (Powell et al., 1991~. Currently, SiC boules (and the commercially available wafers) do contain defects and impurities. One of the most significant defects is a distribution of tubular voids, called micropipes, in the order of a micrometer in diameter (Koga et al., 19921. The micropipes are oriented with respect to their long axis and are approximately parallel to the crystal c-axis; density is typically several hundred per square centimeter. In addition, wafers contain line defects (dislocations) intersecting the surface with a density of 104 to 105 cm~2. The most common background impurities are nitrogen, aluminum, boron, and metals that can act as deep-level traps. It has been shown that the micropipes can cause premature reverse breakdown in pen junctions (Neudeck and Powell, 19941. Evidence shows that microplasmas form in the micropipe at reverse voltages of several hundred volts. We current micropipe density limits the area of high-voltage devices to about 3 mm2; hence, this defect must be significantly reduced before high-power devices are practical. Several theories have been proposed 19

OCR for page 15
1o2 _ 1o1 y 5 - Is E a, 1 10-2 1/ ~ _ . _ Materials for High-Temperature Semiconductor Devices fi: l I 1 1 1 ~1 1 1 1 1 ~ 10 1o1 1o2 103 Temperature (K) FIGURE 2-5 Thermal conductivity of two single crystals of SiC. SOURCE: Adapted from Slack (1964). to explain the formation of micropipes. One theory is based on the presence of contaminant particulates during nucleation and boule growth (Yang, 19931. Another theory is based on the presence of super-screw dislocations (Wang et al., 1993~. In this latter theory, hollow cores would form to relieve stress caused by screw dislocations. Progress is being made in reducing the density of micropipes. In a recent paper, growth of boules in the (1010) direction significantly reduced the formation of micropipes (Takahashi et al., 19941. However, the dislocation density is very high in these crystals. Researchers at Cree have reported (J.W. Palmour, personal communication, 1994) that the density of micropipes has been reduced significantly in the last year. It should be noted that the research team directed by Professor Yu Vodakov at the Ioffe Institute in St. Petersburg, Russia, have produced small single polytype SiC boules (1.5 cm diameter and 7 mm thick) that are claimed to have no micropipes (Y. Vodakov, personal communication, 1994~. Another impediment to wide use of SiC technology is the cost of wafers. At present, there is only one commercial supplier of wafers in the world. The current price per 30-mm-diameter wafer is more than $1,000. This high price can be expected to drop considerably during 1995 as other sources enter the market. The primary reason for this price being lower than GaAs is that both silicon and carbon are 100 times cheaper than gallium. Epitaxial Growth Semiconductor-quality a-SiC epitaxial films can now be grown routinely on or-SiC wafers by CVD. In addition, in situ CVD doping processes can produce both e-type and p-type epitaxial films with net carrier concentrations from the 10~4 cm~3 range to greater than 10~9 cm~3. This technology, which has largely been developed in the last few years, has allowed the development of SiC devices with record-setting performance. Background. The growth of epitaxial SiC films has many similarities with the growth of epitaxial silicon; however, it has only been recently that the differences in growth processes have been appreciated. While conventional semiconductors are grown at approximately two-thirds of their melting temperatures, these temperatures are not practical with wide bandgap materials. For this reason, the substrate temperature cannot be used to assure that all components of the activation energy have been exceeded. In addition, only one crystal structure can be produced in silicon, whereas many crystal structures are possible in SiC. Thus, the polytypic structure of the film must be controlled during formation. The factors that control SiC structure are the crystal orientation and perfection of the substrate. The presence of defects and contamination can also significantly affect the resulting structure. In this report, the term "homo-epitaxy" is used for growth in which the film and substrate are the same polytype, and "hetero- epitaxy" is used when the SiC polytype of the film is different from the substrate. With respect to doping, the incorporation of dopants is dependent on the ratio of the silicon and carbon sources during the growth process and also on the crystal orientation. 20

OCR for page 15
State of the Art of Wide Bandgap Materials TABLE 2-3 Exciton Binding, Nitrogen Ionization, and Valley-Orbit Splitting Energies and Effective Mass for SiC Polyn,rpes Exciton Nitrogen Ionization Energy Valley-Orbit Effective Binding to 4D Splitting Mass ED (meV) EvO(meV) EBX SiC L: {) (PL) | (Haynes) (JR) (2EL) (Hall) | (JR) (ERS) | ml, mll (Cyclotron . resonance) 3C 10 57 53.6 20-47 - 8.37 0.247, 0.677 7 47 52.1 45 7.6 0.42, 20 96 91.8 - 100 - 0.29 6H 16 81 81.0 85 12.6 13.0 31 136 137.6 125 60.3 0.42, 32 140 142.4 62.3 2.0 15R 7 47 49.3 9 54 59.6 53 19 91 99 20 96 - Techniques used to produce epitaxial SiC films include CVD (Davis et al., 1991), the "sublimation sandwich" process, and liquid-phase epitaxy (Ivanov and Chelnokov, 1992~. Homo-epitaxial growth of cx-SiC on a- SiC substrates has been achieved by all three techniques. The lack of 3C-SiC substrates has led to a variety of hetero-epitaxial processes to produce 3C-SiC epitaxial films. The 3C-SiC polytype has been grown on silicon, TiC, and o`-SiC substrates. These processes are examined in the following sections. Cal) of ~x-SiC Epitaxial Films. For both ax- and 3C- SiC, the CVD process is the current method of choice because. of the three techniques. it yields better films at - 7 Al ~ ~ _ the lowest temperature. It is also adaptable to commercial production. A typical SiC CVD growth chamber, shown in Figure 2-7, is similar to chambers used for silicon (Powell et al., 19871. The quartz chamber is water-cooled because growth temperatures are generally higher than those used for silicon epitaxy. The substrates are heated by an inductively heated SiC-coated graphite susceptor. Hydrogen is used as a carrier for various process gases. Prior to growth, the substrates are frequently subjected to 21 an etch with hydrochloric acid (HC1) to reduce defects and contamination. Silane (SiH4) and propane (COHN) can be used as sources of silicon and carbon during growth. Important system parameters for growth include the growth temperature, flow rates of the various gases, and the silicon/carbon ratio in the gas. Important substrate parameters include the orientation and polarity of the SiC substrate. Typical growth rates are in the 1- to 5- ~mlh range. In situ doping is achieved by adding nitrogen or phosphorous for e-type and aluminum (trimethyl aluminum , TM A) or boron (diborane) fo r p -type material. Particular growth and doping processes are discussed. SiC Epitaxy in the Claris Direction. An important discovery in SiC epitaxy was that the crystalline orientation of the growth surface is an important growth parameter. In the past, much of the growth was carried out on the "as-grown" (0001) surface (the basal plane) of Lely crystals; that is, growth was in the c-axis direction. The (0001) SiC crystals with polished surfaces have vicinal (0001) orientations, that is, the growth surface may be tilted slightly "off-axis" with respect to the (0001) crystallographic plane. The size of this tilt angle can have

OCR for page 15
Materials for High-Temperature Semiconductor Devices temperatures because deposited atoms cannot migrate to the steps on large terraces. Also, mobility of deposited atoms is reduced at these lower temperatures and deposited atoms do not reach the steps. Work at the NASA Lewis Center demonstrated that homo-epitaxy of 6H-SiC on vicinal (0001) 6H-SiC can be achieved at 1450 C with tilt angles as low as 0.1 (Powell et al., 1991~. As a consequence of this result, it was proposed that the cause of the 3C-SiC nucleation was due to defects and contamination on the growth surface. By a suitable pregrowth etching process, the defects and contamination were reduced or eliminated. In effect, there is a competition between defects and surface steps. At sufficiently large tilt angles (high step density), homo- epitaxy will occur even in the presence of defects. At low tilt angles (low step density), any defects that are present will dominate and act as nucleation sites for 3C-SiC. Thus, growth must occur at atomic steps if homo-epitaxy of 6H-SiC is to be achieved. In addition, suitable pregrowth etches can be effective in reducing or eliminating defects caused by cutting and polishing the SiC substrate. Homo-epitaxial SiC films on vicinal (0001) SiC substrates have been obtained with the 4H-, 6H-, and 15R-SiC polytypes. These films exhibit a variety of surface features that include hillocks and depressions. Structural defects that occur include the micropipes and dislocations that propagate from the substrate into the film (Powell et al., 19941. Although excellent devices have been fabricated using these films, much work remains to improve the surface morphology and to reduce the defect density. The electrical quality achievable in SiC epitaxial CVD films was significantly improved recently by the development of the "site-competition epitaxy" process by Larkin et al. (1993) at the NASA Lewis Center. In this process, the incorporation of nitrogen and aluminum into a SiC epilayer grown on a silicon-face vicinal (0001) plane is controlled by setting the silicon/carbon ratio in the precursor gases to appropriate values. The nitrogen donor atoms that reside on carbon sites in the SiC crystal lattice compete with carbon atoms during growth. Increasing the carbon concentration (i.e., decreasing the silicon/carbon ratio) decreases the nitrogen incorporation in the epilayers. On the other hand, the aluminum acceptor atoms that reside on silicon lattice sites compete with silicon atoms during growth. Increasing the silicon ~ ~ ~ ~[~] ~ ~ 1 ~1~: 3 3 ~ Seed I) 3 Crystal ~ Growth => Cavity 3 sic Charge Crucible -Thermal Insulation -; Quartz Tubes 1 ~ Water Cooling FIGURE 2-6 Schematic showing Me basic elements of He modified sublimation process. SOURCE: Hobgood (1993). Courtesy of Westinghouse, Inc. a dramatic effect on the structure of an epitaxial film. In subsequent discussions in this report, SiC substrates having tilt angles of about 3 are referred to as being "off-axis" and substrates with tilt angles of less than 0.5 are "on-axis." The polarity (i.e., silicon face or carbon face) of the substrate is also an important parameter. In sublimation sandwich growth, it was found that homo-epitaxy of the various polytypes was enhanced if the growth surface of the substrate was polished off-axis by a few degrees from the (0001) basal plane (Tairov and Tsvetkov, 19831. The research team of Matsunami at Kyoto University discovered that the CVD growth temperature required for producing good-quality 6H-SiC epilayers on 6H-SiC substrates could be reduced from about 1750 C to about 1450 C if the growth surface was off-axis by a few degrees from the (0001) plane (Matsunami et al., 1989~. They called this growth "step- controlled" epitaxy because growth occurs at steps on the off-axis surface. The stepped surface automatically provides the stacking sequence of the substrate polytype. Hence, homo-epitaxy takes place. 3C-SiC was found to grow at small tilt angles (e.g., less than 1.5) or at low 22

OCR for page 15
. ~ ,'~,,, ~ s ',.' ~ '. a ~ B ~ i ~ 9 ~ ~ S ~ ~ ~ S . ~ D 8 ~ ~ ~ D - D - ~ n ~ A ' ~ PHI ~ ~ s D ~ a 7 R ^3 o ~ ~ D ~ ~D ~ ~ 7 S _ ~ 7 S 1 1~ At I C) C/) I C SCOT --act CO ~ - II _ o _. .~m ~ I ~ cn ~ cn ~ pa,, - rD cn ~ :~ I i) O O ~ o o CD I N it.

OCR for page 15
Materials for High-Temperature Semiconductor Devices To eliminate the problem of the large lattice mismatch, titanium carbide (TiCX) with a lattice match within 1 percent was investigated (Parsons, 1987~. Somewhat improved growth of 3C-SiC films was reported, but great difficulties in producing defect-free, single-crystal TiCX has hindered its use as a substrate for SiC growth. In a previous section, it was pointed out that 3C-SiC generally grows on vicinal (0001) ~x-SiC with small tilt angles if there is contamination or defects on the growth surface. Unfortunately, 3C-SiC films grown in this manner typically have a defect known as double- positioning boundaries. This defect arises because there are two possible orientations of the 3C-SiC film that can nucleate on an a-SiC substrate; these two orientations are rotated 180 about the c-axis with respect to each other. When nuclei with both orientations occur on the substrate, the intersection of domains with different orientations are not coherent and they form double-positioning boundaries that are electrically and chemically active. Recent work at Kyoto University has shown that the density of double-positioning boundaries in 3C-SiC films grown on vicinal (0001) 15R-SiC is less than that found in 3C-SiC films grown on 6H-SiC (Chien et al., 1994~. Chien and colleagues presented a model that predicts 3C- SiC films that are tens of micrometers thick and grown on (0001) 15R-SiC should be free of double-positioning boundaries. Unfortunately, the stacking-fault density appears to be very high in these 3C-SiC films. Another approach investigated at the NASA Lewis Center is to limit the epitaxial growth areas to small mesas on vicinal (0001) 6H-SiC substrates and then limit the nucleation of 3C-SiC to the highest atomic planes on the mesa (Powell et al., 1991~. With nucleation limited to a very small region on each mesa, 3C-SiC films will grow laterally and will subsequently cover the mesa with a double-positioning boundary-free 3C-SiC film. This approach has been successful obtaining double-positioning boundary-free 3C-SiC films on 1 mm2 mesas. These films also have a lower stacking-fault density than previously reported 3C-SiC films grown on SiC substrates. Combining this technique with the site-competition epitaxy process for doping SiC epitaxial films, p-njunction diodes with reverse breakdown voltages exceeding 300 V were fabricated (Neudeck et al., 19931. This breakdown voltage is four times that of any previously reported 3C-SiC diode. 24 Other Epita~cial Processes. The sublimation sandwich process (Ivanov and Chelnokov, 1992) is similar to the modified sublimation process. In the sublimation sandwich process, the substrate is placed near a solid SiC source that is sublimed at temperatures greater than 1800 C. The resulting vapor condenses on the substrate that is held at a slightly lower temperature. The high temperature required by this process is its main disadvantage. In the liquid-phase epitaxy technique (Ivanov and Chelnokov, 1992), the substrate is placed in liquid silicon that is saturated with carbon at a temperature in the range of 1500-1700 C. If the temperature is lowered, SiC is deposited from the supersaturated silicon solution onto the substrate. In one version of this process, the liquid silicon solvent is suspended by an electromagnetic field; this "containerless" approach avoids contamination of the solvent by a crucible. The higher temperature required and the difficulty of control are disadvantages of this approach. Summary. Excellent epitaxial films of or-SiC polytypes can now be grown on ~x-SiC substrates. Both e-type and p-type films with net carrier concentrations from 10~4 cm:3 to greater than 10~9 cm~3 can be routinely achieved. The growth of large-area epilayers that are free of micropipes will only be possible when micropipe-free substrates are available. In the future, it will probably be desirable to reduce the growth temperature from the present 1450 C; this may be beneficial for some device fabrication processes. NITRIDE MATERIALS There are four major nitride semiconductors and several minor ones. The four major nitride semiconductors are indium nitride (InN), gallium nitride (GaN), aluminum nitride (A1N), and boron nitride (BN). For high-power electronics applications, there is yet another nitride (iron nitride) that, although not a semiconductor, warrants attention. These materials are composed of cations from Group III of the periodic table and a nitrogen anion from Group V. They are often referred to as III-N materials. A1N, GaAlN, and GaN have been studied for some time, but due to the lack of good single crystals, the electronic, optical, and physical properties of single-crystal nitrides are not extensively

OCR for page 15
State of the Art of Wide Barcdgap Materials known. Interest in the nitride materials has dramatically increased with the recent introduction of bright blue light- emitting diodes (LEDs) by Nichia, the successful growth of better samples, and the accumulation of more precise data (Strife and Morkoc, 1992; Choyke and Linkov, 1993; Lin et al., 1994; Morkoc et al., 1994~. Properties The most intriguing aspect of the large bandgap nitrides (i.e., A1N, GaN, and InN) is the fact that they form a continuous alloy system with room-temperature direct bandgaps varying from 6.2 eV for A1N to 3.45 for GaN, to 1.9 for InN. In addition, there is a small lattice mismatch ~ < 1 percent) between (wurtzite, 2H) A1N and 2H-SiC, and between cubic EN (cBN) and diamond. The band structure of the hexagonal and cubic modifications of A1N and GaN are given in Figures 2-8 and 2-9 (Lambrecht and Segall, 19921. Boron nitride is the least understood of the nitrides. Most work is directed towards the synthesis and characterization of cBN as it is believed to exhibit an indirect bandgap in excess of 6.4 eV. The relative dielectric constant of cBN is 6.5 and its hardness is 4,500 kg/mm2 compared with 3,980 for SiC and 10,400 for diamond (Davis, 1992~. Its thermal conductivity is believed to be 1,300 W/m C, or more than twice that of SiC and about 60 percent that of diamond. Young's modulus is 5.2 MPa compared with 4.0 MPa for SiC. The Poisson ratio for cBN is 0.2 or equal to that of both diamond and SiC. The thermal expansion coefficient of cBN is 3.7 x 104/C, which is the same as that of SiC but greater than the 2.3 x 10~ of diamond. Unfortunately, cBN has also been the most difficult to synthesize. Aluminum nitride exhibits a direct bandgap of 6.2 eV in its hexagonal form. Like diamond, A1N exhibits negative electron affinity (Benjamin et al., 1994~. While cubic A1N has recently been synthesized as a thin film on cubic (3C) SiC, its bandgap has not been ascertained but is believed to be somewhat less than 6.2 eV and is most probably indirect. The thermal conductivity of polycrystalline A1N is 3.0 W/cm C at room temperature-over twice that of silicon and 60 percent that of SiC. Its relative dielectric constant is 10.0, or 85 percent of that of SiC. The best crystallinity reported to date using X-ray 8/2-8 diffraction data is 90 arc-seconds for films grown hetero-epitaxially on sapphire. While 25 there are reports in the literature of both n- and p-type A1N having been synthesized, these reports are not recent (Chu et al., 1967; Rutz, 1976~. Aluminum in A1N has an affinity for oxygen and oxygen appears at a deep level in A1N. Oxygen is typically found in AlN in concentrations of 102 cm~3, rendering it extremely difficult to obtain AlN with either p- or e-type conductivity. There is currently a considerable amount of work underway addressing the A1N doping issue. Alloys of AlN and SiC have recently been made. These may not be true alloys, however, as there is no measurable interdiffusion at temperatures up to 1900 C. Nevertheless, absorption- band edge measurements on this alloy appear to track the mole-fraction composition. The low mass of nitrogen engenders A1N with a high optical phonon energy; for this reason, the charge-carrier velocitY could be very hiah and approach that of diamond. r ~ The lattice parameters for A1N are a = 3.112 A and c = 4.982 A (293 K). The A1N linear thermal expansion coefficient is a' = 5.27 x 10~ K-~; T = 20-800 C; and a 11 = 4.15 x 10~ K-~. The thermal conductivity of A1N is k = 2 W/cm C at room temperature. The density of A1N is d = 3.244 g/cm3. Phonons for A1N are in the frequency range of 895 cm~~ to 303 cm~~. The dielectric constants for A1N are e(0) = 9.14 and c(~) = 4.84 (300 K). Gallium nitride exhibits a direct bandgap of 3.5 eV in its hexagonal form and apparently slightly less in its cubic form. Its lattice constant is 3.189 A, or about 4 percent greater than that of SiC. Its dielectric constant has been measured at 8.9 to 9.5, or just less than that of SiC. It is thought to have an effective electron mass of 0.20, but this figure should not yet be taken as definitive. Along the a-axis, the coefficient of thermal expansion is 5.5 x 10~ K-~. The thermal conductivity at room temperature of GaN is 1.3 W/cm C, nearly equal to the 1.45 W/cm C of silicon and about three times higher than GaAs. After two decades of research, both p-type and e-type GaN have now been produced in hetero- epitaxial thin films. In the best of this material (e.g., with X-ray 8/2-8 diffraction data exhibiting a full width at half maximum of about 27-28 arc-seconds; Plano et al., 1994), acceptors freeze out at about 205 K and holes exhibit a mobility of 450 cm2/V s. Electron mobilities of 1,200 cm2/V s at room temperature have been observed. Both of these values exceed those of SiC but not of diamond. With better crystallinity, these mobilities may perhaps be

OCR for page 15
Materials for High-Temperature Semiconductor Devices a) a) Is , ~ - _/ 1 ,3\ 1 ,3 1 ,3_ 2,4~ - _/ 1 ,3 _ _ _ 1 ,3 _ . - _ L M K (Wurtzite) ~ conduction J\ Bands 2 \ 6 ~ ~- //Val~ W. and ~ . 3\ 3 ~ r k you ~: ~s6 M X r A 4 ~Z l ~ 3 (Zincblende) I A\ I Conduction I / L M K r M X k FIGURE 2-8 Band structure of hexagonal and cubic modifications of A1N. SOURCE: Based on Lambrecht and Segall (1992). expected to further improve. The dielectric strength is believed to be about equal to that of SiC and its computed peak electron velocity exceeds 2 x 107 cm/s. A large number of luminescent features have been reported between 1.65 and 3.5 eV in GaN. These have been attributed to a variety of impurities and defects. However, there is a great deal of controversy in the literature as to the various interpretations. A number of articles have reviewed the literature of luminescence and absorption lines in A1N and GaN (Strife and Morkoc, 1992; Choyke and Linkov, 19931. Although there is currently great commercial potential for GaN optical devices, the beneficial impurities and defects of this 26 r A material for luminescent features are detrimental for high- temperature device operation. Indium nitride exhibits a direct bandgap of 1.9 eV and an indirect bandgap only slightly higher. Its thermal conductivity and most other properties have not yet been definitively ascertained. Its lattice constant of 3.5 A considerably exceeds that of SiC, A1N, and GaN. Double heterostructures of GaN/InGaN/GaN currently exhibit the brightest purple, blue, and blue-green LED s ever made. The blue and blue-green devices are commercially available from Japan and exhibit an operating efficiency of about 2.7 percent.

OCR for page 15
State of the Art of Wide Bandgap Materials 1 ,3\ _ 1 ,3 ~ _ >O .1:, _ ~ . L1J 2,4~ . ~ . 1 ,3 -1 ,3 (Wurtzite) ~: my_ at. ~ .; ~ 1 1 1 1 . _ 3 I_ _ _ ~ _ 1 _ _ 3 3 /`Conduction ~ \ Band ~ 2 \: b ~/ 16~ N3/ /Valence \ ,~ Band <\_ _ ; - ~ 3 '4= l ~ ~ 1 K r M X r A k FIGURE 2-9 Band structure of hexagonal and cubic modifications of The nitride materials, like diamond, are very difficult to etch with liquid etchants. This is an active area of research, however, and phosphoric acid and sodium hydroxide have recently been found to work on both A1N and GaN. Also like diamond, the nitride semiconductors can be left exposed to the atmosphere at high humidity for months at a time without becoming oxidized or otherwise having their surface properties changed. Unlike GaAs, the nitrides do not exhibit self-depleting surface states. For this reasons devices employing pen junctions do not exhibit high surface recombination velocities, which should lead to a long laser-operating life and to extremely 27 ,3 5,6 ,3 1 ,3 1 ,3 (Zincblende) 154 <7CK Conduction ~ Band /` _ _ 1 SIX_ Lop ': . ~ . , . ~ Lo Vale rice / Band \ L1 ~ / ~ =_ L1 r, M F< r k ./ M X GaN. SOURCE: Based on Lambrecht and Segall (1992). V tar it, X1~ 1 K\ 1 \ X/~` L1 . _ . . .= 1 - l 1 r A long charge-storage capability (e.g., millennia) and extremely low-leakage devices suitable for applications such as nonvolatile memories. Crystal Growth Very little work has been done in attempting to synthesize boules of the nitrides. Japanese researchers have synthesized small boules of cBN via high- temperature, high-pressure processes and have even made light-emitting pen junctions of the material. The material was so contaminated by impurities, however, that the absorption band edge was not very sharp and its bandgap

OCR for page 15
Materials for High-Temperature Semiconductor Devices was difficult to ascertain. The pen junctions emitted light in both the visible and in the ultraviolet. Most of the cBN research in the United States has been directed at thin-film synthesis on diamond. It is exceedingly difficult to obtain films thicker than 20 nm that are not polycrystalline or fractured. The only high-temperature, high-pressure attempts to synthesize boules of GaN have been in Poland (Perlin, 1993) however, the boules were very small. growth has been attempted with A1N, but these efforts nave generally not been successful to date. Some boule i. ^~ . ~ DIAMOND Materials Description and Properties Diamond has been admired as a jewel since antiquity and has been studied for a very long time. In fact, Sir Isaac Newton made measurements of the index of refraction of diamond some time around 1665. Large single crystals are found in nature, and synthetic diamonds have been made in high-temperature, high-pressure anvil machines for about 40 years. However, from a semiconductor standpoint, only limited impurity and defect control has been possible to date. Low-temperature and low-pressure polycrystalline film growth has been actively pursued in the last 10 years, but no high-quality, single-crystal films have yet been obtained. The major Synthesis properties of crystalline diamond are well understood, and two excellent books chronicle the development of artifact diamond and tabulate its known properties (Davis, 1992; Spear and Dismukes, 1994~. Diamond is an indirect-gap semiconductor, with the lowest minima of the conduction band being located along the delta axes (k = 0.76~1,0,011. The valence band maximum has a structure that is common to all Group IV semiconductors. There are three bands that are degenerate at the It point when spin is neglected. A band calculation by Chelikowsky and Louie (1984) is shown in Figure 2-10. The indirect energy gap at room temperature is 5.5 eV, and between 135 K and 300 K the variation of the bandgap with temperature is given by -5 >< 10-5 eV/K. The excitor binding energy, Ex, is about 80 med. ma = 0.36 me, and ms0 = 0.15 ma. Hall mobilities have been obtained for n- and p-type diamond as follows: He = 2,200 cm2/V s(RT),and~h = 1,600 cm2/V s(RT). Diamond is famous for its excellent thermal conductivity. In the last few years single isotope diamond has been produced, and it has a higher thermal conductivity than natural diamond (i.e., 32 W/cm.K; Anthony, 19941. Ordinary isotopic ratio diamond has a thermal conductivity as shown in Figure 2-11. The dielectric constant of diamond measured at 300 K is 5.5. The lattice parameter a is 3.56683 A at 298 K. Linear thermal expansion coefficients for various temperatures are -lxlo-6 K-1 -3x10-6 K-1 ~4 x 10-6 K~1 -5 x 10-6 K-1 (300 K), (600 K), (900 K), and (1200 K). The density of diamond is 3.51525 g/cm3, as calculated from the lattice constant. The second-order elastic constants for diamond are On = 10.764 x 1O12 dyne/cm2 (296 K), C12 = 1.252 x 10~2 dyne/cm2 (296 K), and c" = 5.774 x 1012 dyne/cm2 (296 K). Methods of Synthesis and Characterization Aside from the high-pressure, high-temperature synthesized boules of diamond, virtually all diamond films are grown by plasma-assisted methods in the presence of an abundance of atomic hydrogen. The feedstock typically consists of 99 percent H2 and 1 percent CH4. There are many variations of this basic method. While conventional (lower bandgap) materials are typically synthesized at substrate temperatures approaching two-thirds of the melting temperature, this is not possible with diamond and some of the higher bandgap materials. The typical substrate temperature of 900 C cannot be used to assure that feedstock species alighting on the growth surface are fully "activated." The activation-energy threshold (Ea.) is generally composed of three parts: (1) energy sufficient to dissociate the molecule of radical, (2) energy sufficient to chemisorb rather than physisorb the feedstock species, and (3) energy sufficient to ensure that the adsorbed species Effective masses have been measured and calculated in high-quality bulk diamond crystals. The electron effective masses are m' = 0.36 ma, and ml = I.4 ma. The effective masses of the holes are mh = 1.08 mO, arrive at a proper lattice site. In diamond-film synthesis, 28

OCR for page 15
State of the Art of Wide Bandgap Materials _ I _ _ f l _ /Valer ce Band ~ > - o, LL 10 5 o -10 -20 Kin k FIGURE 2-10 Band-structure calculation of diamond. SOURCE: Based on Berman and Martinez (1976). / Conduction Band r'S ~ ~ r25 it\ C/~ X U,K r this is typically accomplished by providing kinetic energy to the feedstock species and to the atomic hydrogen. This kinetic energy is typically imparted via a plasma. These plasmas are generated by direct current (including arc jets), radio frequency fields, microwave fields, or by combustion (e.g., oxy-acetylene torch). Diamond-film synthesis has an additional complication unknown to the synthesis of any other semiconductor. The lowest energy form of its surface is that of an sp2-configured or bond that is graphitic in character. When the diamond surface is so constructed, only graphite can be grown on it. To prevent this unwanted surface construction, the surface is terminated by hydrogen, but this hydrogen must be removed and quickly replaced by carbon to grow the diamond. Since hydrogen bonds to the diamond surface with an energy of 104 kcal/mol versus the 88 kcal/mol of the carbon-carbon bond, it is not easy to remove. Removal requires energetic hydrogen in a "sea" of hydrogen atoms and (typically) methyl radicals or acetylene. For every 104 hydrogen atoms removed from the diamond surface, only one Is replaced by a carbonaceous radical; the remainder are replaced by another hydrogen. The growth process is thus slow and relatively inefficient, although DC arc jets and combustion jets have grown diamond at rates exceeding 100 micron/in. 29 The cost and safety aspects of methane and gaseous hydrogen storage are in some cases circumvented by an alcohol and water plasma process. Several halogen-based techniques are also currently under investigation to reduce the synthesis cost. The target for diamond is to grow large areas of single crystals. This has not yet happened. Single crystalline diamond has not yet been grown on any substrate except natural diamond and cBN. Cubic BN is much less plentiful than natural diamond and available only in very small sizes. Attempts to synthesize diamond on all other foreign substrates has resulted in films characterized by numerous low-angle grain boundaries and much worse. Large area arrays of seeded natural diamond on silicon has resulted in mosaic diamond films. Characterization The preferred method of characterization of the diamond films is by Raman spectroscopy. An unstrained diamond film exhibits a Raman signal at 1,332 cud. The full width at half maximum of this Raman signal is an 1o2 _ _ y 10 E 5 ._ 10 o C: Ct BE 10-1 10-2 ., --~ 1 1 1 11 1 1 1 11 1 1 1 11 10 101 1o2 103 Temperature (K) FIGURE 2-11 Thermal conductivity of two Type IIa diamonds. SOURCE: Based on Berman and Martinez (1976).

OCR for page 15
Materials for High-Temperature Semiconductor Devices indication of the quality of the film. The best of type IIa natural diamond (diamond with no active optical centers) typically exhibits a full width at half maximum of 2.2 cm~i or slightly less. The best of artifact homo-epitaxial diamond films was grown by microwave plasma-enhanced CVD and was characterized by a Raman signal of 1.7 cod. The full width at half maximum Raman signature has been correlated with the thermal conductivity of diamond. In polycrystalline diamond films, a full width at half maximum Raman signature of 3.2 cm~~ generally ensures that the thermal conductivity (in the direction of growth) exceeds 17 W/cm K and the electrical resistivity exceeds 106 Q cm. Diamond films can be grown with intrinsic resistivities of 10~ Q cm and larger, but at slower growth rates (e.g., ~ 1 ,um/h). 30 Diamond Processing Diamond is not etched by boiling acids or bases. There are two preferred methods of processing diamond. The first is the use of kinetic energy beams of oxygen or oxygen-containing molecules or radicals. The second method is by electrolytic etching. The electrolytic etching is generally limited to removal of defect-ridden or otherwise conducting regions of diamond. Boron is the only universally recognized acceptor impurity that can be controllably introduced into diamond. Until recently, only a small portion of the boron in diamond was electrically activated. It can now be introduced and nearly 100 percent electrically activated by a series of implantation processes to concentrations of 1 x 10~9/cm3. By a similar procedure the same investigator stated that he has activated phosphorous in diamond at 80 MeV (Prinz, 19941.