Method of Forming Heterostructure Field Effect Transistor

The present invention relates to electronic semiconductor devices and, more particularly, to high frequency field effect power transistors.

Continuing efforts are being made to improve the performance of field effect transistors (FETs) for microwave applications. Candidates for future high speed power transistors include metal semiconductor field effect transistors (MESFETS) comprising III-V compound semiconductor materials, e.g., GaAs, AlGaAs or InGaAs, and having double recessed gates to minimize the effects of surface charge.

High electron mobility transistors (HEMTs) are among the fastest microwave power transistors. Low noise performance at high frequencies has been demonstrated with these devices at low temperatures, e.g., 77 K. as well as at room temperature. High transconductance of the HEMT design has been attributed to the relatively thin (less than 100 .ANG.) conduction channel layer, commonly referred to as a two dimensional electron gas (2DEG), which results from a conduction band discontinuity at the interface of a selectively doped heterostructure. For example, in a n-AlGaAs/GaAs heterostructure silicon doped Al.sub.X Ga.sub.(1-X) As (X=0.3) is deposited by molecular beam epitaxy over an undoped layer of GaAs. Because the 2DEG layer is confined to a potential well in the undoped GaAs layer, high electron mobilities are observed.

In order to improve low temperature HEMT performance and avoid formation of DX centers in the Al.sub.X Ga.sub.(1-X) As layer the mole fraction of AL must be reduced to less than 18 percent. Because this effectively limits the conduction band gap between the Al.sub.X Ga.sub.(1-X) As layer and the GaAs layer, InGaAs, having a lower conduction band than GaAs, is chosen as the 2DEG channel material. This allows for a reduction in the mole fraction of Al present in the AlGaAs, e.g., X=0.15. Lattice mismatch is minimized by forming a relatively thin, e.g., 200 .ANG., layer of In.sub.Y Ga.sub.(1-Y) As (Y=0.15) between the doped AlGaAs and an undoped GaAs buffer layer. This MESFET is termed a pseudomorphic HEMT because lattice strain causes the InGaAs layer to distort from a cubic structure to match the lattice constants of each adjacent layer. For further discussion see Swanson, "The pseudomorphic HEMT", Microwaves and RF (March 1987) beginning at page 139.

Although pseudomorphic HEMTs exhibit better high frequency operating characteristics, e.g., 0.43 W/mm at 62 GHz, than other FETs, such performance requires formation of a sharp material transition at the heterostructure interface in order to transfer a large number of electrons into the channel region. In addition, a very thin, e.g., 20-30 .ANG., undoped AlGaAs layer must isolate the heavily doped AlGaAs from the undoped GaAs. Otherwise dopants can easily outdiffuse into the 2DEG region and cause impurity scattering. While stringent requirements such as these can be met in a prototype laboratory environment, i.e., with carefully controlled molecular beam epitaxy techniques, repeatability and uniformity problems must be overcome in order for the device to be manufactured on a monolithic microwave integrated circuit (MMIC) structure or in a volume production environment. Even with metal-organic chemical vapor deposition (MOCVD), it is difficult to abruptly control gas flows in order to repeatably form sharp heterostructure interfaces and thin layers of uniform thickness.

It has been demonstrated that the performance of sub-half-micrometer gate MESFETs can be enhanced by incorporating extremely heavily doped active layers. See Kim et al., "GaAs Power MESFET with 41-Percent Power-Added Efficiency at 35 GHz", I.E.E.E. Electron Device Letters, Vol. 9, No. 2 (February 1988). See also Dambkes et al., "Improved Performance of Micron and Submicron Gate GaAs MESFETS Due to High Electron Concentrations (n=10.sup.18 cm.sup.-3) in the Channel," 1983 GaAs IC Symposium Digest, beginning at page 153.

Schubert et al. have proposed a delta doped MESFET (.delta.FET) composed entirely of GaAs and having a Dirac-delta-function-like profile. See "The Delta-Doped Field-Effect Transistor (.delta.FET)", IEEE Transactions On Electron Devices, Vol. ED-33, No. 5, pp. 625-632, (May 1986). This results in a V-shaped conduction band suitable for confining carrier electrons at a density exceeding 10.sup.18 cm.sup.-3. Theoretically, the impurity layer can be confined to a lattice constant while the electron distribution extends over several lattice constants. In fact, because device operation is in the high field saturation region, formation of such a sharp dopant profile may have a more significant effect on short gate, e.g., 0.25 .mu.m, FET performance than low field mobility characteristics. However, due to high field electron velocity limitations in the GaAs channel region, high frequency performance, e.g., 60 GHz or higher, has not been attainable with pulse-doped GaAs power MESFETs. That is, a gate length on the order of approximately 0.1 .mu.m would be required in order to offset the material limitations of GaAs and thereby provide 60 GHz operation.

It is therefore desirable to form a millimeter wave transistor device having a cut off frequency comparable to or better than that of a HEMT, but which does not require the complex manufacturing processes associated with production of pseudomorphic HEMTs.

According to one form of the invention there is provided a field effect transistor having a buffer layer comprising a first compound semiconductor material. A layer of second semiconductor material different from the first material is formed over the buffer layer. The second layer has a total thickness less than 250 .ANG.. A doped third semiconductor layer is formed over the second layer. The net dopant concentration in the second layer exceeds the net dopant concentration in the third layer. A gate layer is positioned over the third layer.

The second layer results in high power characteristics by increasing both the dopant concentration and the high field electron velocity over conventional MESFETs.

There is also provided a method for making a field effect transistor. A doped pseudomorphic semiconductor layer of a first conductivity type is formed between first and second other semiconductor layers, the second layer including a net dopant concentration of the first conductivity type. A schottky gate electrode is formed in contact with the second layer.

The invention may best be understood by reference to the following detailed description of illustrative embodiments when read in conjunction with the accompanying drawing, wherein:

FIG. 1 illustrates in a schematic cross sectional elevation view a heterostructure device according to the present invention;

FIGS. 2 present the conduction band diagram for a device constructed according to the present invention and conduction band diagrams for various prior art devices;

FIGS. 3 illustrate a first preferred method for fabricating a heterostructure device according to the present invention;

FIG. 4 illustrates the I-V characteristics for the heterostructure device of FIG. 1;

FIG. 5 presents the gate-drain reverse bias I-V curve for the device of FIG. 1;

FIG. 6 illustrates the gain compression curve for the device of FIG. 1; and

FIG. 7 is a conduction band diagram for a device having a uniformly doped conduction band layer.

Certain preferred embodiments of the invention are described herein. However, it should be appreciated that the present invention is applicable to field effect transistors generally, and can be formed from a variety of materials depending on the particular circuit applications. The specific semiconductor materials which are disclosed are merely illustrative of ways to embody the invention and do not limit the scope of the invention. It is also recognized that the invention can be formed with a wide variety of dopants as well as impurities of opposite conductivity type to that which is illustrated.

Although the performance of GaAs MESFETs has been improved significantly by using extremely heavily doped active layers, it is now apparent that further improvement can be obtained by forming a heterostructure device which does not operate as a HEMT. FIG. 1 is a schematic cross sectional elevation view of a first preferred embodiment heterostructure device, generally denoted by reference numeral 10, according to the present invention. The device 10 includes a semi-insulating GaAs substrate 12, a 1 .mu.m thick undoped GaAs buffer layer 13; a thin (150 .ANG.) and heavily doped In.sub.X Ga.sub.(1-X) As layer 14; an n-type GaAs layer 16 doped to approximately 10.sup.17 cm.sup.-3 ; a second n-type GaAs layer 18 doped to approximately 10.sup.18 cm.sup.-3 ; a 0.25 .mu.m long Ti/Pt/Au double recessed gate 20; and n+ source and drain ohmic contacts 22 over layer 18 each formed as a gold:germanium/nickel/gold structure. A 500 .ANG. Si.sub.3 N.sub.4 passivation layer (not illustrated) is formed overall.

In the preferred embodiment, X=0.15 for the layer 14 and the sheet carrier level, ranging from 3.times.10.sup.12 cm.sup.-2 to 5.times.10.sup.12 cm.sup.-2, is localized in a mid-level region 26 approximately 50 .ANG. thick. This provides a conduction channel dopant concentration about the region 26 on the order of 10.sup.19 cm.sup.-3. The structure results in a GaAs/InGaAs/GaAs pseudomorphic quantum well which confines the carriers to the relatively high mobility InGaAs layer 14.

The heterostructure delta doped device 10 provides improved operation over the GaAs .delta.FET and is comparable to or better than a pseuedomorphic HEMT having a double sided doped well. FIGS. 2A-2D illustrate, respectively, energy diagrams of the conduction band level across the channel region for each device type, i.e., FIG. 2A qualitatively presents the conduction band energy level along line 2--2 of FIG. 1 and FIGS. 2B through 2D provide analogous representations for the prior art devices. For comparative purposes the GaAs channel region of the FIG. 2B .delta.FET includes a silicon dopant concentration of 10.sup.19 cm.sup.-3 (based on a delta-doped layer with a sheet carrier density ranging between 10.sup.12 and 10.sup.13 cm.sup.-2); and the pseudomorphic HEMT devices of FIGS. 2C and 2D are assumed to comprise InGaAs layers having the same stoichiometry as the device 10.

The heterostructure device 10 is characterized by a heavily doped high mobility InGaAs channel layer. Measurements indicate .mu..sub.n =3500 cm.sup.2 (V-Sec.).sup.-1, approximately, at 77 K. for a silicon dopant concentration of 10.sup.19 cm.sup.-3. In contrast, the mobility level in the GaAs channel region of the FIG. 2B .delta.FET is 2000 cm.sup.2 (V-Sec.).sup.-1. See Shubert et al. at page 630.

Because the pseudomorphic InGaAs HEMTs of FIGS. 2C and 2D do not exhibit impurity scattering, the channel mobility of these devices should exceed that of the device 10. However, the mobility does not have a large effect on device performance and it now appears that HEMT performance is largely limited by the achievable electron density (approximately 4.times.10.sup.12 cm.sup.-2 for the double doped well of FIG. 2D) in the 2DEG layer. That is, the electron density in each InGaAs quantum well formed near the AlGaAs interface is limited by the conduction band gap between the heavily doped AlGaAs layer and the intrinsic InGaAs layer. By way of example, for Al.sub.X Ga.sub.(1-X) As (X=0.15, N.sub.D =2.times.10.sup.18 cm.sup.-3) the band gap is 0.3 eV and the corresponding 2DEG has a carrier density of approximately 2.times.10.sup.12 cm.sup.-2.

In contrast to the HEMT, a higher sheet carrier density is achievable in the pseudomorphic quantum well of the heterostructure device 10. The achievable electron density is confined to the maximimum net n-type dopant concentration. That is, the carrier density is limited to approximately 10.sup.19 cm.sup.-3, the level at which the impurity, e.g., Si, begins replacing As sites and functioning as a counterdopant.

Measurements indicate that the achievable carrier density in the InGaAs quantum well has a significant influence on transistor gain characteristics. Thus, for a short channel device, e.g., 0.25 .mu.m gate, any detrimental effects of high dopant level impurity scattering on mobility are of lesser importance compared with the advantageous confinement of carriers to the relatively high mobility InGaAs layer 14.

Further features of the device 10 are apparent from the following first preferred embodiment method of fabrication as illustrated in FIGS. 3 (not drawn to scale) and beginning with a 600 .mu.m thick semi-insulating GaAs substrate 12 having a planar surface along the <100> orientation on which there is epitaxially grown the undoped GaAs buffer layer 13 to a thickness of 1 .mu.m. The pseudomorphic pulse doped layer 14 is then formed over the layer 13 by depositing a first 75 .ANG. layer 30 of In.sub.X Ga.sub.(1-X) As, X=0.15, followed by a Si deposition 32 of sufficient thickness, e.g., approximately 4 .ANG., to provide a density of 3.times.10.sup.12 cm.sup.-2. A second 75 .ANG. layer 36 of In.sub.X Ga.sub.(1-X) As, X=0.15, is formed over the Si deposition 32. Next a 2000 .ANG. thick gate forming layer 16 of silicon doped GaAs (dopant concentration of 10.sup.17 cm.sup.-3) is deposited over the second In.sub.X Ga.sub.(1- X) As layer 36 followed by a second 1000 .ANG. thick silicon doped GaAs layer 18 (dopant concentration of 10.sup.18 cm.sup.-3) to enhance ohmic contact with the subsequently formed source/drain electrodes 22. It is noted that with MBE, a 1 .mu.m per hour growth rate and the relatively low, 500.degree. to 600.degree. C., temperature required for formation of the layers 14, 16 and 18 are sufficient to effect diffusion of the silicon deposition 32 into adjacent InGaAs layers 30 and 36, this resulting in the n+ pulse doped layer 26 of FIG. 1.

At this point a first layer 38 of photoresist is spun over the second GaAs layer 18, masked and patterned. The active region of the device 10 is defined with a boron implant of 5.times.10.sup.11 cm.sup.-2 flux at an implant energy of 100 KeV. This is sufficient energy to penetrate into the buffer layer 13. The photoresist is then removed, resulting in the intermediate structure of FIG. 3A. Next a second layer of photoresist (not illustrated) is spun on to the layered structure and patterned to define source/drain regions on the heavily doped GaAs layer 18. With a standard E beam evaporation/liftoff technique gold:germanium/nickel/gold source/drain electrodes 22 are formed and then alloyed into the GaAs layer 18.

Next a first layer of PMMA is spun onto the layer 18 and patterned to define an opening for a one .mu.m wide recessed gate region 40. The GaAs layer 18 is vertically etched 2,000 .ANG. with H.sub.2 SO.sub.4 :H.sub.2 O-2:H.sub.2 O, i.e., about midway into the GaAs layer 16. The first layer of PMMA is removed and a second layer of PMMA is spun on and patterned to define a gate opening, concentric with the first region 40, for the second gate recess. Accordingly, the GaAs layer 16 is vertically etched an additional 700 .ANG. to provide the intermediate structure of FIG. 3B. Then a titanium/platinum/gold multilayer gate 20 is deposited and patterned in the double recessed opening with a standard E beam evaporation/liftoff technique to provide the structure illustrated in FIG. 1. Subsequently a Si.sub.3 N.sub.4 passivation layer is deposited overall. The underside of the GaAs substate 12 may be lapped down to a 100 .mu.m thickness, metallized and bonded to a carrier.

FIG. 4 illustrates the I-V characteristics for the heterostructure device 10 and FIG. 5 presents the gate-drain reverse bias I-V curve. A peak transconductance of 520 mS/mm and a maximum current density of 700 mA/mm have been achieved. The breakdown voltage is in the range of 8 to 9 volts and the current cut off frequency, evaluated from s-parameter data, is near 55 GHz, which is significantly higher than that of a standard MESFET and comparable to that of a pseudomorphic HEMT. Thus, although the layer 14 is heavily doped, it is apparent from the invention that the transport properties of the device are not degraded.

At 60 GHz the device 10 exhibits superior performance over the pseudomorphic HEMT. The small signal gain is 8 dB and a 0.6 W/mm power density has been achieved with 3.5 dB gain and 14% power added efficiency. FIG. 6 shows the gain compression curve for the device 10 having a 0.25 .mu.m.times.75 .mu.m.sup.2 gate.

Another feature of the invention is that the GaAs layer 16 comprises a relatively low dopant concentration thus providing improved Schottky gate properties over MESFETs formed with pulse doped GaAs layers as well as HEMT designs.

Various modifications of the preferred embodiment heterostructure device 10 and method of fabrication may be made while retaining advantageous features thereof. For example, the entire In.sub.X Ga.sub.(1-X) As layer 14 may be doped to form a 2DEG the width of the pseudomorphic layer. A uniform dopant distribution of 5.times.10.sup.18 cm.sup.-3 has been observed to provide a sheet carrier density comparable to that of the pulse doped heterostructure device 10 while providing equivalent performance.

The uniformly doped embodiment may be advantageous over the pulse doped structure in that it requires a simpler growth process. Furthermore, because net n-type dopant concentrations higher than 10.sup.18 cm.sup.-3 can be provided throughout the layer 10, the uniformly doped device is expected to provide a higher current density than the pulse doped device. A conduction band diagram for the device 10 having a uniformly doped pseudomorphic layer 14 is illustrated in FIG. 7. Note the small perturbation in the InGaAs layer near the n-GaAs interface due to a slight HEMT effect.

As a further example, the In.sub.X Ga.sub.(1-X) As stoichiometry may be varied to provide a very large, e.g., up to X=0.5, mole fraction of Indium. This is believed to provide higher mobility in the active region. On the other hand, the increased Indium concentration will result in a greater lattice constant disparity, this possibly limiting the thickness of the pseudomorphic layer.

The semiconductor materials disclosed herein are exemplary of the compounds which may be used to form a high mobility high frequency heterostructure MESFET device. Other materials may be substituted. By way of example, the layer 16 may be formed of ZnSe and the InGaAs layer may be replaced with InAs, GaAs or HgCdTe. Other materials having a closer lattice match to InGaAs may be chosen for the substrate, e.g., InP. The doping levels may be varied from those disclosed herein and p-type doping with 2 dimensional hole gases may be used with all embodiments and variations of the invention. Further enhancements are expected with modifications of the disclosed structure and the scope of the invention is limited only by the claims which follow.