GaN HEMT: Two-Dimensional Electron Gas (2DEG)

GaN power transistors have pushed switching frequencies in DC-DC converters and RF power amplifiers well beyond what silicon MOSFETs can reach, largely because of one structural feature: a sheet of mobile electrons confined at the interface between two different III-nitride layers, with no dopant atoms in the path of conduction. Understanding how that sheet — the two-dimensional electron gas (2DEG) — forms and why it delivers such high electron mobility and current density is the key to reading a GaN HEMT datasheet intelligently and using the device correctly in a switching or RF design.

Structure: Not a MOSFET, a Heterojunction

A GaN HEMT (High Electron Mobility Transistor) is not built like a MOSFET, even though it's often drop-in-used as a power switch. The critical layers, from bottom to top:

• Substrate: typically silicon, SiC, or sapphire (Si is dominant for cost in power applications).
• GaN buffer layer: several µm of gallium nitride, undoped (intrinsic), grown epitaxially.
• AlGaN barrier layer: a thin (~20–30 nm) layer of aluminum gallium nitride grown directly on top of the GaN.
• Gate, source, drain contacts: metal electrodes on top; source and drain are ohmic contacts, gate is either a Schottky contact (D-mode) or sits over a p-GaN cap (enhancement-mode, E-mode).

The device has no p-n junction forming the channel and no implanted/diffused source-drain regions the way a MOSFET does. The channel is a byproduct of the AlGaN/GaN material interface itself.

Why the 2DEG Forms

AlGaN and GaN have different bandgaps (AlGaN ≈ 3.4–4.0 eV depending on Al fraction, GaN ≈ 3.4 eV) and, critically, different spontaneous and piezoelectric polarization. GaN and AlGaN are wurtzite crystals lacking inversion symmetry, so they carry a built-in electric polarization along the c-axis even with no applied field. When AlGaN (higher Al content, more strained, larger polarization) is grown on relaxed GaN, the mismatch in polarization at the interface produces a very large, fixed sheet of bound positive charge at the AlGaN/GaN boundary — on the order of 10¹³ cm⁻².

To maintain charge neutrality, free electrons accumulate on the GaN side of that interface, pulled in from surface donor states and the AlGaN layer. These electrons are confined in a narrow quantum well formed by the conduction-band discontinuity between AlGaN and GaN — roughly triangular, only a few nanometers thick. Because the electrons are constrained in the vertical direction but free to move in the plane of the interface, this is the two-dimensional electron gas.

Key point: the 2DEG exists with zero gate bias and no intentional doping. This is fundamentally different from a MOSFET inversion layer, which requires an applied gate field to invert a doped body region. Because of this, unbiased AlGaN/GaN HEMTs are naturally normally-on (depletion-mode) devices — the channel conducts by default.

Why the 2DEG Gives High Performance

Three properties make the 2DEG attractive for power and RF switching:

The combination of mobility and carrier density gives a very low sheet resistance, typically 300–500 Ω/□ for a standard AlGaN/GaN stack — comparable to or better than a heavily doped silicon layer, but without the dopant scattering that would otherwise degrade mobility. This is what allows GaN HEMTs to achieve low RDS(on) in a much smaller die area than an equivalent-voltage silicon MOSFET, and to switch with much lower gate charge and output capacitance, enabling multi-MHz switching frequencies.

Normally-On vs. Normally-Off (E-mode) Operation

Because the native 2DEG is present at zero gate bias, a bare AlGaN/GaN HEMT is a depletion-mode (D-mode, normally-on) device — dangerous in most power applications where a normally-off, fail-safe switch is required. Two common approaches convert this into an enhancement-mode (E-mode, normally-off) device:

• p-GaN gate: a p-type GaN layer is grown under the gate metal. Its built-in junction depletes the 2DEG directly beneath the gate at V_GS = 0, pinching off the channel. Applying a positive V_GS (typically +1 to +2 V threshold) forward-biases this junction enough to repopulate the 2DEG under the gate and turn the device on. This is the dominant commercial approach for power GaN HEMTs.
• Cascode configuration: a normally-on GaN HEMT is placed in series with a low-voltage silicon MOSFET, and the combination is packaged/sold as a single normally-off device. The Si MOSFET holds the GaN gate at a negative bias relative to source until commanded on.

Datasheet threshold voltages for p-GaN E-mode devices are notably low (often +1 to +2 V) and gate drive voltage range is tight (commonly 0 to +6 V absolute max) — the gate is essentially a forward-biased diode junction, not an insulated oxide gate, so overdriving V_GS risks large gate current and damage. This is a direct consequence of how the E-mode threshold is created (a p-n junction gate) rather than a thick oxide as in a MOSFET.

Worked Example: Sheet Resistance and RDS(on) Estimate

Take a typical power GaN HEMT with a 2DEG sheet carrier density n_s = 1×10¹³ cm⁻² and electron mobility µ = 1700 cm²/V·s.

Sheet conductance:

σ_s = q · n_s · µ
    = (1.6×10⁻¹⁹ C)(1×10¹³ cm⁻²)(1700 cm²/V·s)
    = 2.72×10⁻³ S/□  (siemens per square)

Sheet resistance:

R_sheet = 1/σ_s ≈ 368 Ω/□

For a device with a gate-to-drain access region of length L = 10 µm and channel width W = 200 mm (a wide power device built from many parallel fingers), the access resistance contribution is:

R_access = R_sheet · (L/W) = 368 Ω/□ · (10 µm / 200,000 µm) = 368 × 5×10⁻⁵ ≈ 0.018 Ω

Check: this is only the access (drift) region resistance, excluding the channel-under-gate and contact resistance, but it already shows how a modest sheet resistance combined with a wide, multi-finger layout gives sub-50 mΩ total RDS(on) — consistent with commercial GaN power switches rated in the 10–100 mΩ range at 600–650 V. The order of magnitude checks out against published datasheet values.

Practical Design Implications

• Gate drive is unforgiving: p-GaN E-mode gates behave like forward-biased diodes above ~6 V; use dedicated GaN gate drivers with tight voltage clamping, not generic MOSFET drivers with wide swing.
• No body diode in the MOSFET sense: GaN HEMTs conduct in reverse through a different mechanism (channel modulation by V_GD), giving a higher effective reverse voltage drop than a Si MOSFET body diode — relevant for hard-switched vs. soft-switched topology choices.
• Dynamic RDS(on) (current collapse): trapped charge at the AlGaN surface or buffer layer can transiently deplete part of the 2DEG after switching, increasing RDS(on) for microseconds to milliseconds — this must be characterized under the actual switching waveform, not just DC, when budgeting conduction loss.
• Thermal behavior differs from Si: GaN-on-Si devices often have higher thermal resistance from channel to case than comparable Si MOSFETs because of the substrate stack, which affects derating even though conduction losses are lower.

Key Takeaways

• The 2DEG is a high-mobility electron channel formed spontaneously at the AlGaN/GaN interface by polarization-induced charge, with no doping and no applied field required.
• Because it exists at V_GS = 0, native AlGaN/GaN HEMTs are normally-on; commercial power devices convert this to normally-off using a p-GaN gate junction or a cascode Si MOSFET.
• High electron mobility (undoped channel, no dopant scattering) combined with high sheet carrier density (~10¹³ cm⁻²) gives low sheet resistance and enables low RDS(on) with low gate charge, supporting multi-MHz switching.
• p-GaN E-mode gates are junction-like, not oxide-insulated — gate drive voltage windows are narrow and require GaN-specific drivers.
• Dynamic RDS(on) (trapping effects) and reverse-conduction voltage drop are GaN-specific behaviors that must be evaluated separately from static DC datasheet parameters.