AlGaN polarization-doped field effect transistor with compositionally graded channel from Al_0.6Ga_0.4N to AlN

Ultra-wide bandgap (UWBG) AlGaN transistors are candidates for next-generation high voltage power electronics^1–6 and high power rf electronics because the power density of such devices is expected to scale aggressively with bandgap energy (E_g). Al-rich Al_xGa_1-xN with x > 0.6 is an attractive alloy region because the electron mobility (μ) increases as the Al composition approaches the AlN binary.^5,7 Enhancing channel μ is important for both AlGaN rf and power electronics. Maximizing μ for rf devices enables carriers to achieve saturation velocity when transiting short gate lengths, and maximizing μ for power devices enables a transistor to achieve a given on-resistance (R_on) at a lower sheet charge density (n_s), which makes lateral electric field management and threshold voltage (V_th) control easier.

A challenge for Al-rich AlGaN transistors is realizing simultaneously high μ and n_s. Al_xGa_1-xN metal-semiconductor field effect transistors (MESFETs) have been demonstrated for x = 0.65⁸ and 0.70,^9,10 with a current density >0.5 A/mm,¹⁰ but μ has yet to exceed 100 cm²/V s for these devices due to ionized impurity scattering in the heavily doped channel and alloy scattering at lower channel doping levels. Al-rich AlGaN heterojunction field effects (HFETs)^{1–3,11–13} have been reported with μ as large as 284 cm²/V s for n_s = 1.1 × 10¹³ cm⁻²,^11,12 but AlGaN HFETs also face a tradeoff between μ and n_s. In an AlGaN HFET, increasing n_s requires increasing the Al contrast, i.e., reducing the channel Al composition relative to the barrier, which decreases channel μ.

AlGaN PolFETs¹⁴ overcome the μ‐n_s tradeoff by using polarization-induced doping¹⁵ to realize high free electron density without impurity doping while also increasing μ relative to a HFET for a given Al contrast. AlGaN HFETs and PolFETs rely on the spontaneous and piezoelectric polarization charge of AlGaN heterostructures to produce free electrons in lieu of impurity doping. The lack of ionized impurity scattering improves channel μ for HFETs and PolFETs relative to MESFETs with a similar Al composition in the channel. Between Al-rich AlGaN PolFETs and HFETs, the former has a distinct advantage in μ. In an AlGaN HFET, the two-dimensional electron gas forms below the abrupt heterointerface between the barrier (higher Al composition) and channel (lower Al composition) layers, so the channel μ is set by the AlGaN region with lower μ. In a PolFET, the Al composition is continuously graded so that a three-dimensional electron slab (3DES) forms a volumetric channel that includes the AlGaN regions with the highest Al composition in the heterostructure and therefore the highest available μ. Thus, in the ideal case, the average μ of a PolFET channel will be higher than an HFET with the same Al contrast. In practice, there are potential drawbacks for a PolFET compared to a HFET. One undesired aspect is that much of the charge in the PolFET channel is close to the surface, whereas the HFET channel is separated from the surface by the barrier. Thus, a PolFET could be more strongly influenced by surface states, which reduce μ and cause high frequency dispersion. Thus, surface passivation will be important for maximizing PolFET performance.

Our previous report of an Al_xGa_1-xN PolFET with Al compositional grading from 0.70 ≤ x ≤ 0.85 (hereafter termed the 70→85 PolFET) achieved both linear ohmic contacts and μ > 200 cm²/V s,¹⁶ which has yet to be shown for an AlGaN MESFET or HFET of similar Al composition. However, the maximum drain current (I_DS,max) for the 70→85 PolFET was only 24 mA/mm because the relatively low Al content limited μ and the low Al contrast limited n_s = 5.4 × 10¹² cm⁻². In this work, we examine an Al_xGa_1-xN PolFET with 0.60 ≤ x ≤ 1.0 Al compositional grade (hereafter termed the 60→100 PolFET) to demonstrate improved μ by accessing a larger Al content in the channel and increased n_s by increasing the Al contrast, resulting in improved I_ds,max = 188 mA/mm.

Figure 1 shows the epitaxial heterostructure and device geometry of the 60→100 PolFET. All group-III nitride epilayers were grown by metal-organic vapor phase epitaxy in a Veeco D-125 system at 75 Torr using conventional precursors, including trimethylgallium (TMGa), trimethylaluminium (TMAl), and ammonia. Initially, a 2.3 μm-thick AlN epilayer was grown on (0001) c-plane sapphire substrates misoriented 0.2° toward the m-plane to serve as templates for subsequent growth of AlGaN PolFETs. PolFET samples consisted of a 0.1 μm-thick AlN and a 0.25 μm-thick unintentionally doped (UID) Al_0.60Ga_0.40N buffer layer structure followed by a 75 nm-thick Al_xGa_1-xN epilayer over which x was increased linearly over a nominal range of 0.60 to 1.0. The Al composition in the buffer was verified using x-ray diffraction (XRD) to be 0.61. During the grade, both the TMAl and TMGa source flows were adjusted to maintain a constant Group-III molar flux, ensuring a constant growth rate regardless of the Al composition. The length and linearity of the Al compositional grading were verified to be very close to the intended heterostructure using secondary ion mass spectroscopy (SIMS). Figure 2 shows the depth profile of the Al atomic composition near the surface from SIMS performed by the Evans Analytical Group. Due to the large uncertainty (±20%) of the absolute Al composition measured by SIMS, the Al composition in the buffer was set to the value determined form XRD.

The average μ of the PolFET channel was examined. Contactless sheet resistance (R_sh) measurement confirmed strong electrical conductivity in the UID PolFET with R_sh = 1500 Ω/◻. Mercury probe capacitance-voltage (C-V) measurements indicated a sheet charge density (n_s) of 1.3 × 10¹³ cm⁻², implying an average μ = 320 cm²/V s in the graded AlGaN channel from the relation μ = 1/(qR_shn_s), where q is the Coulomb charge. This average μ is expected to reflect the average channel mobility when the PolFET is biased in strong accumulation. Contactless sheet resistance measures the net conductivity of the portion of the channel that is not depleted by the surface potential, which is the same portion of the channel that establishes the drain current in the access region.

Circular, long-channel PolFETs were fabricated. Devices utilized planar Zr/Al/Mo/Au (15/120/35/500 nm) source and drain contacts¹⁷ followed by rapid thermal annealing under a 1 mTorr nitrogen atmosphere for 30 s at 1000 °C and 1100 °C. Gates utilized the Ni/Au Schottky metal. No surface passivation layer was applied to the devices. Transistors had a nominal 10 μm spacing between the source and drain (L_sd), with gates centered between the source and drain. The gate width was defined by its 660 μm circumference. After transistor fabrication, the gate length (L_g) and gate-to-drain spacings (L_gd) and L_sd were measured by secondary electron microscopy to be 3.1, 2.9, and 9.6 μm, respectively.

The source/drain contacts were characterized using circular transfer length method (CTLM) patterns defined with nominal electrode spacings of 5, 10, 15, 20, 25, 30, and 35 μm. The Zr/Al/Mo/Au contacts had non-linear I-V properties with a large voltage offset, so the specific contact resistance could not be quantified but was estimated to be ∼10⁻² Ω cm². The degradation in ohmic contacts for the 60→100 PolFET compared to the previous report of the 70→85 PolFET¹⁶ likely results because it is more difficult to form ohmic contacts as the AlN mole fraction at the surface increases. Previous work has shown that the specific resistivity of contacts increases exponentially with the AlN mole fraction,¹⁷ so the 15% increase in the Al composition at the surface for a 60→100 PolFET relative to a 70→85 PolFET could cause ohmic contacts to become highly resistive. The underlying physics for this trend is not clear, but it is likely due in part to the decreasing electron affinity with the increasing AlN mole fraction.

The electrical properties of the PolFET channel Schottky gate were characterized by C-V measurements. Figure 3 shows the polarization-induced doping profile (ρ_π) of the channel, which was calculated from the C-V measurement of a Ni/Au Schottky contact surrounded by an annular ohmic contact. The measured ρ_π value is close to the expected value for linear compositional grading ρ_π = 5 × 10¹³ × (Δx)/d = 2.7 × 10¹⁸ cm⁻³, where the leading term is the polarization charge associated with a GaN/AlN heterointerface, Δx is difference in the initial and final Al compositions, and d is the length of the grade, i.e., the channel thickness.¹⁸ The increase in polarization doping near 75 nm is less abrupt than expected. This could result from backside depletion of the graded channel by the large negative polarization charge associated with the underlying Al_0.60Ga_0.40N/AlN heterointerface. It is noted that SIMS data verified that unintentional doping with oxygen or silicon cannot be the source of the channel conductivity. From SIMS (data not shown), the oxygen concentration in the channel was 1 × 10¹⁷ cm⁻³ and the Si concentration was at or below the detection limit of 2 × 10¹⁶ cm⁻³, both of which are well below the measured ρ_π.

Figure 4 presents the DC I-V characteristics for a 60→100 PolFET, and Fig. 5 shows the transfer characteristics and mutual transconductance (g_m). The drain current density reached I_DS,max = 188 mA/mm; however, there was significant offset voltage in the triode region due to highly resistive ohmic source and drain contacts. Nonetheless, the PolFET achieved R_on = 85 Ω mm. The Schottky gate contact to the AlN surface proved robust enough that PolFET operation was stable for V_GS = +10 V. Operating at V_GS > +4 V improved I_DS,max and R_on significantly, although I_DS was negative at low V_DS due to forward current from the Schottky gate diode. Such operation is functionally similar to that of an insulated gate bipolar transistor, where a large V_DS offset is required to achieve high I_DS (although the underlying device physics of the two types of devices is fundamentally different). These I_DS,max and R_on values are greatly improved over the previous report for a UID 70→85 AlGaN PolFET with 24 mA/mm and 195 Ω mm,¹⁶ respectively. The higher maximum Al composition and greater Al contrast for the 60→100 PolFET improved μ > 60% and n_s > 100%, respectively, relative to the 70→85 PolFET. The PolFET transfer characteristics exhibited pinch off at V_th = −15 V but with an on/off ratio >10⁴ that was somewhat low, possibly due to a buried extrinsic channel in the buffer. The mutual transconductance reached 9.5 mS/mm for V_DS = 10 V, and a plateau was evident in g_m near −13 V, again likely due to an unintentional channel forming in the buffer.

Figure 6 displays breakdown measurements performed on a device with a realized L_gd = 2.7 μm with devices immersed in fluorinert. The gate voltage was held at −20 V V_gs, and breakdown did not occur up to the instrument limit of V_ds = 600 V, i.e., V_gd = 620 V, indicating E_crit > 210 V/μm. The demonstrated average E_crit for the 60→100 PolFET is significantly less than E_crit = 810 V/μm reported for vertical AlGaN diodes for x = 0.58¹⁹ and E_crit = 360 V/μm reported for Al_0.70Ga_0.30N MESFETs.¹⁰ Lower E_crit for lateral FETs compared to vertical diodes is expected due to more challenging electric field management for the former due to near-surface effects. Lower E_crit for the 60→100 PolFET compared to the Al_0.70Ga_0.30N MESFET reported in Ref. 10 is likely due to the lack of a field plate for the former device. Nonetheless, E_crit for the 60→100 PolFET is substantially better than ∼100 V/μm that is typically achieved in GaN HEMTs, despite no effort to optimize the PolFET device geometry with a field plate design or with surface passivation. It is worthwhile to note that E_crit for the 60→100 PolFET exceeds that of our previously reported 70→85 PolFET while also enjoying 2× greater n_s. Simultaneous improvement of n_s and E_crit for increasing Al contrast and Al composition demonstrates that Al-rich PolFETs realize the expected benefits of accessing Al-rich AlGaN for high voltage transistors.

In summary, UID Al-rich PolFETs were realized with a channel AlN mole fraction compositionally graded from 0.60 to 1.0. A high Al composition provided high average channel μ = 340 cm²/V s and E_crit > 210 V/mm to achieve R_on = 85 mΩ mm and V_br > 620 V. The PolFET μ exceeded reported values for AlGaN MESFETs and HEMTs of similar Al composition. The expected benefits of increasing the PolFET maximum Al composition and Al contrast were realized with a simultaneous increase in channel μ, n_s, and E_crit relative to previous UID Al-rich PolFETs graded from AlN mole fractions of 0.70 to 0.85. This demonstrates PolFETs as an attractive alternative to MESFETs and HEMTs for Al-rich rf and power switching transistors.

The authors thank Dr. Sanyam Bajaj and Professor Siddharth Rajan for their helpful discussion. This work was supported by the Laboratory Directed Research and Development program at Sandia National Laboratories. Sandia National Laboratories is a multi-program laboratory managed and operated by the National Technology and Engineering Solutions of Sandia, LLC (NTESS), a wholly owned subsidiary of Honeywell Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under Contract No. DE-NA0003525. The views expressed in the article do not necessarily represent the views of the U.S. Department of Energy or the U.S. Government.