GaAS-Gate Semiconductor–Insulator–Semiconductor FET

Paul Solomon; David J. Frank; Steven L. Wright; Frank Canora    IBM Research Dmsion, IBM T. J. Watson Research Center YorkTown Heights, New York

I Introduction—Basic Principles

The SISFET or “GaAs gate FET,” as it was originally called, belongs to an extensive family of heterojunction field effect transistors (Tiwari, 1992; Sze, 1990; Solomon and Morkoç, 1984) built out of III–V compounds, and known collectively as HFETs. These FETs are characterized by high electron mobilities and transport velocities for electrons in the channel formed at the heterojunction boundary between two semiconductors having different electron affinities. As a result, these HFETs achieve high speeds in both digital and analog (microwave) applications and low-noise operation as microwave amplifiers.

It is primarily in analog applications that HFETs have achieved commercial success to date, although the conventional GaAs MESFET (metal semiconductor FET) has begun to penetrate the digital market (e.g., the CONVEX [McDonald, 1991] and CRAY [Kiefer et al., 1987; Wilson et al., 1991] supercomputers). If the digital market becomes established, HFETs would provide for the following generations. The SISFET is intended primarily for digital applications.

The first HFETs were modulation doped FETs (MODFETs) where the wider bandgap semiconductor (AlGaAs) was doped, and electrons spilled over into the narrow bandgap semiconductor (GaAs) forming a channel there. Because of the confinement of the electrons in this channel, coupled with the low effective mass of GaAs, the perpendicular momenta of the electrons in the channel are quantized, with appreciable effects even at room temperature, so that this channel is often referred to as a twodimensional electron gas (2DEG).

With molecular beam epitaxy capable of growing wide bandgap AlGaAs layers on GaAs, with sharp interfaces, the AlGaAs could be used as the gate insulator of an FET. As was pointed out by Solomon (1982a), the existence of an appropriate field boundary condition at the GaAs/AlGaAs interface such that the 2DEG satisfies Gauss’s law

Nch=εFch,

where Nch is the sheet electron concentration, ε the permittivity of GaAs and Fch the field in the GaAs at the GaAs/AlGaAs interface (the field in the bulk GaAs is assumed to be zero). This field can be supplied either by doping in the AlGaAs, as in a MODFET or, as in a conventional Si MOSFET, by an applied gate voltage. Looked at another way, the doping in a MODFET is equivalent to charge in the insulator of a MOSFET.

The SISFET or GaAs gate FET (Solomon et al., 1984; Matsumoto et al., 1984) was derived as an analog of the polysilicon gate MOSFET, where the n+ GaAs, i AlGaAs, and i or p− GaAs of the SISFET replace the n+ polysilicon, SiO2 and p silicon of the MOSFET. The comparison between the devices is illustrated in Fig. 1. The main advantages of the SISFET over the MOSFET are the better electron transport properties in the channel, and the fact that the entire vertical structure is a single crystal, eliminating many problems pertaining to interface charge and interface states that can affect MOSFETs. The major disadvantage of this structure is the low barrier height of the gate “insulator” (0.35 eV at most for GaAs/AlGaAs), which compares unfavorably with ~3 eV for Si/SiO2. This means that gate leakage currents are much higher in SISFETs than in MOSFETs. The positive aspect of this is that there are no long-term charge trapping effects in the AlGaAs, at least down to 77 K, like those that have plagued the MOSFET.

Band diagrams of the SISFET and the MODFET are compared in Fig. 2. The main differences are the replacement of doped with undoped AlGaAs, and the replacement of the gate Schottky barrier with a heterojunction barrier. The threshold voltage for either device may be written as

T=Φg−Φch−eNIt122ε1+εεIFBtI,

where εI, tI and NI are the permittivity, thickness and doping of the gate insulator, Φg and Φch the barrier heights (measured from the quasi-Fermi levels when biased at threshold) at the gate and channel interfaces, and FB the back field originating from behind the channel due to doping, trapped charge, etc. (see Section IV.D for more details).

This equation yields the zero threshold voltage property of the SISFET when NI ≈ 0, and for small back fields, due to the fact that Φg ≈ Φch for the same material in gate and channel. This property is independent of the thickness of the AlGaAs. For use in digital circuits, it is necessary to adjust VT to a small, positive value, and several ways to do this will be discussed. Furthermore, it is necessary to adjust the threshold voltage differently for enhancement- and depletion-mode FETs, necessitating the use of doping in at least one of them. In addition, uncontrolled doping and trapped charge in the barrier or substrate will cause threshold voltage shifts. Given this, the SISFET still retains superior threshold control properties compared to MODFETs or MESFETs. In the MODFET, for example, VT is extremely sensitive to doping and thickness, because of the large difference between Φg and Φch. Tolerances of doping of ~2% and thicknesses of ~1% are required for the MODFET to meet the stringent VT control requirements of digital logic.

Doping of the AlGaAs has introduced another problem for MODFETs, from which the SISFET escapes. Every donor in the AlGaAs can potentially become a deep electron trap when it captures an electron, through a process of lattice deformation. These “DX” centers (Mooney, 1990) have been the bane of MODFETs and have forced MODFET designers to use low AlAs mole fractions (~25%), resulting in poor electron confinement by the heterobarrier. In contrast to this, the SISFET can use a high AlAs mole fraction (~50%) optimized for low leakage current.

This chapter will discuss the SISFET, mainly using examples from IBM’s SISFET program. A review of this work has been given recently (Kiehl et al., 1990), although the present review will be much more detailed in its treatment of device and circuit designs, as well as presenting previously unpublished device and circuit results of the present authors. Much has been learned from this program about SISFETs, and about FET III–V HFETs in general. Looking toward the future, this chapter concludes that the scalability of SISFETs to gate lengths smaller than 14μm will require semiconductor pairs having higher barrier heights than can be obtained with GaAs/AlGaAs. This may be fulfilled by InGaAs and InAlAs.

II History and Development of SISFETs and Related Devices

The family of SISFETs and related devices include the SISFET itself, in both GaAs-gate and Ge-gate implementations; some version of the MISFET; and the doped channel MISFET. Often different acronyms are used by different groups of researchers. All of these devices have the common feature that the gate insulator is an undoped, wide-bandgap, III–V semiconductor. These FETs have also been used for complementary heterostructure logic, which is discussed in a separate chapter of this book entitled “Complementary Heterostructure FET Integrated Circuits,” by Kiehl, and the reader is referred there for details not covered in what follows.

A CONCEPT AND FIRST REALIZATION

The concept of the SISFET as a close analog of the MOSFET was first put forward by Rosenberg (1986). Solomon (1982b) realized the advantages of the zero threshold voltage property of the SISFET and came up with an independent proposal.

The properties of GaAs/AlGaAs/GaAs SIS capacitors consisting of n+ GaAs − i Al0.4Ga0.6As − n GaAs were then investigated, by Solomon and Hickmott (1983), in much the same way as MOS capacitors, to study the charge control properties and gate leakage current characteristics of the basic vertical structure. The results show MOS-like CV curves (Fig. 3) and IV characteristics (Fig. 4) that reveal the basic leakage current mechanisms of thermionic emission at high temperature and Fowler-Nordheim tunneling at low temperature. The GaAs/AlGaAs interface was capable of supporting electron densities of greater than...