III-Nitride microwave power transistors

CHAPTER THREE

III-Nitride microwave power transistors Jeong-Sun Moon* HRL Laboratories, Malibu, CA, United States *Corresponding author: e-mail address: [email protected]

Contents 1. Overview of microwave power GaN HEMTs 2. Field-plated GaN HEMTs for microwave power applications 2.1 Gate-recessed field-plated GaN HEMTs 2.2 Linearized field-plated GaN HEMTs 3. Dual-gate and cascode field-plated GaN HEMTs for High-efficiency applications 4. Speed performance and scaling of field-plated GaN HEMTs 5. Future development of microwave power GaN HEMTs Acknowledgments References Further reading

115 116 118 123 127 133 137 138 138 140

1. Overview of microwave power GaN HEMTs Since the early 2000s, field-plated (FP) GaN HEMTs were studied and showed very promising high-power RF and microwave performance, as summarized in Fig. 1. Their performance benefits include 10 W/mm at 2 GHz (Ando et al., 2003a), 32 W/mm at 4 GHz (Wu et al., 2004), and >10 W/mm at 30 GHz (Moon et al., 2005, 2008) and 40 GHz (Palacios et al., 2005). The first generation of FP GaN HEMTs was made with the source-connected FP gates (Ando et al., 2003a; Wu et al., 2004), and showed excellent RF performance with high breakdown voltage mainly due to a reduction of the gate-to-drain electric field. These GaN HEMTs are very robust and commercially available for mostly wireless frequency applications. Recently, the second generation FP GaN HEMTs are made by integrating a short-gate-length gamma-gate structure for the FP gate without the source-connected FP (Moon et al., 2005, 2008; Palacios et al., 2005). Semiconductors and Semimetals, Volume 102 ISSN 0080-8784 https://doi.org/10.1016/bs.semsem.2019.08.001

#

2019 Elsevier Inc. All rights reserved.

115

116

Power Density (W/mm)

Jeong-Sun Moon

10

GaN HEMTs with T-gate

1

GaAs HEMTs

10

100

Frequency (GHz)

Fig. 1 State-of-the-art microwave power transistors are shown in terms of their power density versus frequency.

These integrated FP GaN HEMTs demonstrated high cut-off frequency (fT) of 60 GHz and maximum oscillation frequency (fMAX) of
150 GHz (Moon et al., 2011). With a breakdown voltage of
90 V, excellent progress was also made in MMIC amplifiers, with early results of 20 W and 43% PAE (Moon et al., 2006) to the latest results of 58 W and 38% PAE (Piotrowicz et al., 2008) in the X-band. Using 150-nm gate length FP GaN HEMTs, excellent class E switched-mode amplifiers were demonstrated in the X-band with 61% PAE at 3.5 W of output power (Moon et al., 2012). This chapter will describe progress made in high-power and highefficiency microwave power GaN HEMTs with the integrated field-plated gate layout with state-of-the-art performance up to Ka-band applications.

2. Field-plated GaN HEMTs for microwave power applications Continuous progress in wide bandgap GaN/AlGaN-based HEMTs pushed their frequency performance beyond the X-band into the millimeter-wave range. In particular, there is growing interest in wide bandgap semiconductors for high-power applications in the X-, K-, Ka-, and Q-band. Early performance in these bands was obtained from GaN/ AlGaN HEMTs fabricated with T-gate devices. In 2001, Moon et al. reported

117

III-Nitride microwave power transistors

a CW power density of 6.6 W/mm and a PAE of 37% at 20GHz (Moon et al., 2001). At 29GHz (Sandhu et al., 2001) and 30GHz (Kasahara et al., 2002), power density of below 2 W/mm and PAE was <30%. At 30 GHz, a 2.3 W (Kasahara et al., 2002) and 5.8W (Inoue et al., 2004) GaN HEMT cells demonstrate PAE of 40% and 43%, respectively. Also, Lee et al. reported PAE as high as 44% at 30GHz, but with a much lower power density of 3 W/mm. The decrease of PAE to 30% was observed as the power density increased to 5.68 W/mm (Saunier et al., 2003). However, the GaN HEMTs with T-gates often suffer from the surface-trap related current collapse and have not been able to deliver high power density and high efficiency operation simultaneously. Recently, Fitch et al. demonstrated 140 nm T-gate GaN HEMTs with a source-connected FP, operating in the X- and Ka-band, with PAE exceeding 50% at 10 GHz (Fitch et al., 2015). Following a demonstration of field-plated (FP) GaN-based HEMTs with dramatically improved RF power performance, the integrated FP GaN HEMTs with reduced gate capacitance were fabricated for high-frequency applications. The integrated FP GaN/AlGaN HEMTs with deep-submicron gate lengths offer excellent large-signal performance with power density of 10 W/mm at 10 GHz and 30 GHz (Moon et al., 2005, 2008). The associated PAEs were 60% at 10 GHz and 38% at 30 GHz, respectively. At 5.7 W/mm power density, the PAEs of 74% at 10 GHz and 45% at 30 GHz were demonstrated (Moon et al., 2002a). At 40 GHz, a CW power density of 10.5 W/mm was measured with PAE of 34% (Palacios et al., 2005). Fig. 2 shows a summary

A

B 10 Microwave Power performance

Microwave Power performance at 30 GHz

CW Power Density (W/mm)

80 10 GHz

PAE (%)

60 30 GHz

40

20 Integrated FP GaN HEMTs

0 0

2

4

6

8

Power Density (W/mm)

10

12

8

6

4

2

0

0

5

10

15 20 Vd (V)

25

30

35

Fig. 2 (A) Measured microwave power performance of integrated FP GaN HEMTs and T-gate GaN HEMTs at 30 GHz are shown. (B) Plot of measured power-added-efficiency versus Power density of the integrated FP GaN HEMTs at 10 GHz and 30 GHz are shown with 10 W/mm operation.

118

Jeong-Sun Moon

of microwave power performance of the integrated FP GaN HEMTs at 10 GHz and 30 GHz, illustrating state-of-the-art performance. The improvement is attributed to minimized surface-trap related DC-RF dispersion with re-distribution of the gate-to-drain breakdown field (Mishra et al., 2016). While the large-signal performance of GaN HEMTs is improved in the FP devices, their PAE, power density and linearity can further improve especially for high-frequency applications. In particular, their high 2DEG sheet resistance (
3 times higher than that of GaAs PHEMTs or InP HEMTs), non-linearity in transconductance, and saturation velocity need to be addressed properly in device design. Nidhi et al. (2006) indicated that the access resistance at the source (Rs) and drain (Rd) can impact the high-frequency performance of GaN HEMTs through a parasitic charging delay (Tasker and Hughes, 1989), (Rs + Rd) * Cgd, where Cgd is a gateto-drain feedback capacitance. This chapter will describe high-power and high-efficiency FP GaN HEMTs with both the gate-recessed process and n + GaN linearization.

2.1 Gate-recessed field-plated GaN HEMTs Since 2001, RF power testing of discrete AlGaN/GaN HEMTs was done at various microwave frequencies up to 40 GHz. Most of these results were obtained from the devices, in which the gate metals were fabricated in a planar geometry (i.e., no gate-recessed geometry). Non-planar or gate-recessed AlGaN/GaN HEMTs were demonstrated previously by various groups (Ando et al., 2003b; Chini et al., 2004; Chu et al., 2008; Moon et al., 2002b). Utilizing both gate-recess and field-plating, Ando et al. (2003b) showed 12 W/mm power density and 49% PAE at 2 GHz, with 3–7 dB improved gain over the non-planar FP devices. Chini et al. also reported a similar improvement in RF power performance at 4 GHz (Chini et al., 2004). With gate-recess and V-gate for mitigation of the high electric field at the gate-edge, Chu et al. reported an excellent performance at 10 GHz with 12.2 W/mm and 65% PAE (Chu et al., 2008). This section describes high-speed, non-planar, i.e., gate-recessed, AlGaN/GaN HEMTs, in which deep-submicron gates are placed vertically in an etched surface. At 30 GHz, with 1 dB gain compression, a power density of 5.7 W/mm with 45% PAE was observed at Vds ¼ 20 V. At Vds ¼ 28 V, the output power density was as high as 6.9 W/mm. The PAE and output power are still increasing with the input power level. Compared to planar AlGaN/GaN HEMTs, the measured output power density and PAE of

119

III-Nitride microwave power transistors

gate-recessed AlGaN/GaN HEMTs dramatically improve, demonstrating both high power density and high PAE, simultaneously at the Ka-band. The deep-submicron gate-recessed Al0.25Ga0.75N/GaN HFETs are fabricated using a 1000-A˚-thick SiNx mask defined by electron beam lithography and Cl2 plasma etching of the Schottky Al0.25Ga0.75N layer, similar to the process reported (Moon et al., 2006). The key processing step is to produce low damage gate recess etching of the Al0.25Ga0.75N layer. Gate metallization follows to fill the recessed area completely. This allows placing the gate foot plane separated from surface states vertically. Fig. 3A is a schematic diagram of the gate-recessed devices, where dimensions for the gate foot, recess-etched depth, gate-to-drain overhang, and gate-tosource overhang are denoted as Lg, Lr, Ls, and Ld, respectively. The gate ˚ . The Ls is fixed foot dimension was 0.2 μm. The recess depths are 80–100 A at 0.25 μm and the Ld varies from 0.35 μm, 0.45 μm, and 0.55 μm. The DC and small-signal RF performance is summarized and compared in Table 1. The gate-recessed devices show Idss of 0.65 A/mm to 0.75 A/mm measured ˚ , respecat Vds ¼ 10 V, depending on the etch depth of Lr of 100 A˚ and 80 A tively. Fig. 3B shows the measured extrinsic transconductance is as high as 635 mS/mm with Lr ¼ 100 A˚, which is comparable to that of GaAs PHEMTs. Fig. 4 shows the measured small signal S-parameter data of the gaterecessed devices with Ld ¼ 0.35 μm show a unity-gain cut-off frequency (fT) of 60 GHz and maximum oscillation frequency (fmax) of
100 GHz at Vdd ¼ 10 V, respectively. As the Ld increases to 0.55 μm, the fT is reduced to 54 GHz due to the increased gate-to-drain feedback capacitance. B 1500 Is lg

SiNx

ld

AIGaN lr

GaN 4H-SiC

Recessed-gate

Gm (mS/mm), ld (mA/mm)

A

1000

500

0 –4

–3

–2

–1 Vgs[V]

0

1

2

Fig. 3 (A) A schematic diagram of a non-planar gate-recessed AlGaN/GaN HEMT with device design parameters such as Ls, Ld, Lg, and Lr. (B) Measured 2  100 μm AlGaN/GaN HEMT transfer curve at Vds ¼ 10 V, yielding Imax of 1.4 A/mm.

120

Jeong-Sun Moon

Table 1 DC and RF performance of gate-recessed AlGaN/GaN HEMTs. Peak Gm Idss* Lr (A) Ld (μm) Ls (μm) (mS/mm) (mA/mm) fT (GHz)

Gain (dB) @ 30 GHz

80

0.55

0.25

560

700

54

8.2

80

0.45

0.25

545

715

58

9.3

80

0.35

0.25

550

715

60

9.7

100

0.35

0.25

635

650

63

10.7

Idss was measured at Vds ¼ 10 V. Ft was measured at Vds ¼ 10 V and peak gm bias.

50

Gain (dB)

40

30

IH21I2

MSG/MAG

20

10

0 0.1

Wg = 2x100 mm Vdd =10 V Ld = 0.35 mm

fT = 60 GHz

1

10

100

Frequency [GHz]

Fig. 4 Measured S-parameter gains, current gain and maximum stable gain (MSG)/ maximum available gain (MAG) as a function of frequency are shown.

The measured maximum-stable-gain (MSG) is compared in Table 1. The MSG is 10.7 dB at 30 GHz with Ld ¼ 0.35 μm, Lr ¼ 100 A, and Lg ¼ 0.18 μm. With Ld ¼ 0.55 μm, the field-plating effect is expected to increase, but the device gain is reduced by 2–2.5 dB compared to ones with Ld ¼ 0.35 μm. For high PAE operation at mmWave frequency, we characterized devices with Ld ¼ 0.35 μm. Table 2 compared the measured fT * Lg values from the planar devices and the gate-recessed devices. With the Ld ¼ 0.35 μm, fT * Lg is 11.3–12 GHz μm, which is comparable to that of the planar devices, implying that the gate-to-drain feedback capacitance is insignificant. The gate-recessed AlGaN/GaN HEMTs are characterized for mmWave power performance using an on-wafer Ka-band Focus Load-pull setup in a

121

III-Nitride microwave power transistors

Table 2 Speed comparison of gate-recessed versus planar AlGaN/GaN HEMTs. Device layout Lg (μm) fT (GHz) fT * Lg (GHz μm)

Planar

0.25

50–55

12.5–13.7

Planar

0.15

80

12

Gate-recessed

0.2

60

12

Gate-recessed

0.18

63

11.3

30

60

Wg = 0.15 mm

50 25 Vds = 20 V Vds = 16 V

30

15 Vds = 25 V

10

20 10

5 0 0

5

10

15

Pin (dBm)

20

0 25

12

50 PAE

10

40

8 30 6 20

PAE (%)

40

20

PAE (%)

Pout (dBm), Gain (dB)

B

35

Power Density (W/mm), Gain (dB)

A

4 PD gain (dB) MSG

2 0 5

10

15

20

25

30

10

0 35

Vds (V)

Fig. 5 (A) Measured CW power performance of 2  75 μm gate-recessed AlGaN/GaN HEMTs at f ¼ 30 GHz. The device is biased at Vds ¼ 20 V and Vgs ¼  1.9 V. (B) Measured power density, gain and PAE versus drain bias. At Vds > 20 V, the input power is limited to compress the devices.

continuous wave mode at 30 GHz. Both input and output matching are provided by mechanical tuners. The matching is tuned for optimum PAE. Fig. 5A shows CW power performance measured from a 2  75 μm device at Vds ¼ 20 V. The linear gain is 8.3 dB. With 1 dB gain compressed, power density reaches 5.7 W/mm with drain-efficiency (DE) and PAE as high as 58% and 45%. The measured high DE implies that the gate-recessed GaN HFETs would provide further higher PAE when device gain improves at the mmWave frequency. Since the PAE peaks with 2–3 dB gain compression, we expect both the peak PAE and saturated power to be higher. For instance, at Vds ¼ 16 V, PAE peaked at 47% with power density of 4.5 W/mm. Fig. 5B shows measured output power density, gain, and PAE versus drain bias, while the load impedance is fixed. The measured output power density increases almost linearly until Vds ¼ 20 V, while the PAE is almost unchanged as 47% to 45%. At Vds ¼ 28 V, the output power density at P1dB is as high as 6.9 W/mm, while the PAE is still rising above 29%. The delivered input power is limited well below the peak PAE point.

122

Jeong-Sun Moon

Interestingly, as the drain bias increases from 12 V to 28 V, the measured gain improves by almost 3 dB. This helps to operate devices at high PAE at high drain biases. In the case of planar devices, when the drain bias increases, PAE is greatly reduced such that at power density of 3.3 W/mm, PAE is measured as 32%. PAE is measured as high as 51% when the devices are biased at Vds ¼ 6.5 V. The measured power density is, limited below 1.5 W/mm, however. The observed significant reduction in PAE with increasing drain biases was observed by different groups, as well. This has been attributed to the knee voltage walkout, or high-field-induced trapping under high Vds and high RF power operation. Both planar and gate-recessed AlGaN/GaN HEMTs are characterized in terms of pulsed IV’s where 200 ns pulses are used at several different Q-biases up to Vds ¼ 40 V. In contrast to the planar devices, the knee voltage walkout was eliminated in the gate-recessed devices. Interestingly, the pulsed IV exhibits Idss enhancement of
10% with the knee voltage and threshold voltage unchanged. The improved pulsed IV behaviors of the gate-recessed devices are attributed to the fact that the gate metals are vertically separated from the surface states. Due to the physical separation, the time constant for the surface states to respond to the electrical pulses is much longer than that of the gate itself, for the first order. To understand PAE versus frequency, we measured the CW power performance at 10 GHz. The CW output power density of 11 W/mm is obtained with PAE of 50% and gain of 12 dB at Vds ¼ 30 V. The PAE is measured as 59% with 7 W/mm output power density when the devices were biased at Vds ¼ 20 V. The difference in PAE between 10 GHz and 30 GHz is about 12%, which is, in the first order, due to the reduced device gain at the Ka-band. In summary, high performance Ka-band power performance is demonstrated from deep-submicron and non-planar, i.e., gate-recessed, AlGaN/ GaN HEMTs, in which gates are placed vertically in an etched surface. In addition to the excellent pulsed IVs, a CW power density of 5.7 W/mm is observed with 45% PAE at Vds ¼ 20 V at 1 dB gain compression point. At Vds ¼ 28V, the output power density is as high as 6.9 W/mm. The PAE and output power still rise with input power. Thus, high performance at millimeter-wave frequency band is promising with well-optimized deepsubmicron gate-recessed AlGaN/GaN HEMTs, delivering both high power density and high PAE simultaneously.

123

III-Nitride microwave power transistors

2.2 Linearized field-plated GaN HEMTs This section describes field-plated GaN HEMTs fabricated with an n + source ledge to reduce source access resistance and linearize transconductance, and their record 10 W/mm power density Ka-band performance. No n + ledge was defined on the drain side in order not to increase gate-to-drain feedback capacitance or to decrease breakdown voltage. Fig. 6A shows a schematic diagram and a scanning electron microscope photograph of GaN HEMT device with an n + source ledge. The sourcedrain spacing is 2.5 μm and the gate is centered between the source and drain. Three different lengths of n + source ledge, L_ledge as marked in Fig. 6A, were used in the experiment: 0.35 μm, 0.55 μm, and 0.75 μm. The n + source contact ledge reduces the gate-to-source spacing without placing rough ohmic metals close to the gate metal. This makes the deep-submicron gate A

L_ledge

FP gate

n+ source ledge B 1400

500 L_ledge = 0.75 µm

1200

L_ledge = 0.55 µm

400

800

300

L_ledge = 0

600

200

gm (mS/mm)

Ids (mA/mm)

1000 L_ledge = 0.35 µm

400 100 200 0 -3

-2

-1

0 0

1

2

Vgs (V)

Fig. 6 (A) A schematic diagram and SEM photograph of a field-plated GaN HEMT with an n + source ledge is shown. (B) A plot of measured transconductance and source-drain current vs. gate-to-source voltage is shown from 2  100 μm GaN HEMTs at Vds ¼ 10 V with no ledge (solid), n + ledge of 0.35 μm (long dash), n + ledge of 0.55 μm (short dash), and n + ledge of 0.75 μm (dotted), respectively.

124

Jeong-Sun Moon

process via electron beam lithography more reproducible. The GaN heterostructures were grown with an n + capping layer on top of the. AlGaN layer. After Ti/Al ohmic metallization, the source ledge was defined with a photoresist mask and Cl2-based etching of the n + GaN capping layer. The thickness of AlGaN layer was 17 nm. A 100-nm-thick PECVD SiNx layer was deposited to provide a surface passivation and to support the field-plated structure. An opening in the SiNx for the gate foot was defined by electron beam lithography with ZEP520A resist, followed by CF4 dry etching of the SiNx layer. Reference devices were fabricated without the n + source ledge. The length of the gate foot was 0.14 μm, and the gate overhang above the FP SiNx layer was 0.55 μm. Fig. 6B shows the measured transconductance (gm) and drain current (Ids) versus gate-to-source voltage (Vgs) of 2  100 μm GaN HEMT devices with different source ledge lengths. While the threshold voltage is not changed, the peak extrinsic transconductance increased by 21% from 350 mS/mm of the reference device to 425 mS/mm for the device with L_ledge ¼ 0.75 μm. The measured gm and gm/gm_max was plotted in terms of drain current (not shown here), where the peak gm occurs at Ids ¼ 400 mA/mm regardless of L_ledge. On the other hand, the gm/gm_max at a high current level such as Ids ¼ 1A/mm is reduced to 0.5 for the reference device and 0.7 for the devices with L_ledge ¼ 0.75 μm, respectively. This shows improved linear behavior over a wide source-drain current range, which is similar to the recent simulated result (Piotrowicz et al., 2008). The improved transconductance especially at high drain current should increase output power and efficiency under largesignal conditions, as shown in this work. The gate-to-drain breakdown is not affected by the n + source contact ledge and is measured to be about 100 V. Both the saturated source-drain current, Idss (Ids at Vg ¼ 0 V and Vds ¼ 10 V), and maximum source-drain current, Imax (Ids at Vg¼ 2 V and Vds ¼ 10 V), increased by
23% as the length of the n + source ledge increased from 0 μm to 0.75 μm, as shown in Table 3. A reduction of source access resistance (Rs) by almost 50% was observed from the end-point measurements. The significant reduction in Rs is attributed to the decrease in the access resistance since the contact resistance and the access resistance was
0.25 and
0.65 Ω mm, respectively, without the n + source contact ledge. Small-signal S-parameters were measured using an HP 8510B network analyzer. The extrinsic unity-gain, cut-off frequency (fT) increased from 50 GHz in the case of no ledge to 55 GHz in the case of L_ledge ¼ 0.75 μm. MSG was calculated based on S-parameters at 30 GHz. At Vds ¼ 25 V and Ids ¼30 mA, the MSG increased from 11 dB in the case of reference

125

III-Nitride microwave power transistors

Table 3 DC and RF performance of 2  100 μm GaN HEMTs with n + source ledge and a gate length of 140 nm. L_n +ledge Rs Peak Gm Idss* Imax fT Gain (dB) Lsd (A) (μm) (ohm) (mS/mm) (mA/mm) (mA/mm) (GHz) @ 30 GHz

2.5

0

5.2

350

440

970

50

11

2.5

0.35

4.0

380

465

1055

51

11.7

2.5

0.55

3.4

390

485

1105

51

11.6

2.5

0.75

2.7

425

540

1200

55

11.5

Imax was measured at Vds ¼ 10 V and Vgs ¼ 2 V. fT was measured at Vds ¼ 10 V and peak gm bias.

Table 4 Small-signal equivalent circuit model of 140-nm gate length 2  100 μm GaN HEMT with L_ledge ¼ 0.75 μm, where τ* is charging delay due to parasitics. Cgs (fF) Cgd (fF) Gm0 (mS) Ri (ohm) Rds (ohm) Rs (ohm) Rd (ohm) τ (ps)

186

20

73

1.8

595

2

7.7

0.2

sample to 11.5 dB with L_ledge ¼ 0.75 μm. For more detailed analyses, smallsignal equivalent circuits were developed for both reference devices and devices with n + source ledges at Vds ¼ 25 V and Ids ¼ 30 mA. As shown in Table 4, Rs is reduced from 0.85 Ω mm of the reference devices to 0.42 Ω mm in the case of L_ledge ¼ 0.75 μm. The Cgd and Cgs are not changed as 100 fF/mm and 930 fF/mm, respectively. As a result, the parasitic charging delay, τ* ¼ (Rs + Rd) * Cgd, is reduced by
17% from 0.24 ps to 0.2 ps. Using an on-wafer Maury load-pull setup, field-plated GaN HEMTs with and without n + source ledges were characterized in a continuous wave mode at 30 GHz. Both input and output matching were tuned for optimum PAE. The devices were biased at deep class AB with a quiescent current of 50 mA/mm. Fig. 7 shows the CW power performance of a 2  75 μm device measured at Vds ¼ 20V (circle), 30 V (square), and 40 V (diamond). The measurement gamma was fixed at 0.509 + 0.527i for source matching and 0.368 + 0.663i for load matching. This load-matching condition corresponds to an output shunt resistance of
41 Ω-mm. At Vds ¼ 20 V, the linear gain was 8.6 dB. Output power was 28.1 dBm with PAE of 52% and DE of 63%, respectively. At the peak PAE, the dynamic current was raised to 0.34 A/mm. At Vds ¼ 30 V, the linear gain improved to 10.4 dB. PAE and DE peaked at 50% and 58%, respectively, with an associated power density of 7.3 W/mm. At Vds ¼ 40 V, the device linear gain was 11.1 dB.

126

70

35 Wg = 0.15 mm

60

Pout

Vds = 30 V

20 s= Vd

15

V 20

40

PAE

Vds = 40 V

30

Gain

10

20

5

10

0 0

5

10

15

Pin (dBm)

20

0 25

12

70

10

60 50

8

40 6 30 4 2 0 15

20 PAE DE

Power Density Linear Gain

25

35

45

PAE (%), DE (%)

50

25

PAE (%)

Pout (dBm), Gain (dB)

30

Power Density (W/mm), Gain (dB)

Jeong-Sun Moon

10 0

Vds (V)

Fig. 7 (A) CW power performance of a 2  75 μm field-plated GaN HEMT with n + source ledge of 0.55 μm at f ¼ 30 GHz with Vds ¼ 20 V (circle), 30 V (square), and 40 V (diamond). The power performance was tuned for PAE in a deep class AB bias. (B) Measured power density, linear gain, PAE, and DE versus source-drain bias at f ¼ 30 GHz from 2  75 μm field-plated GaN HEMTs with n + ledge of 0.55 μm.

The output power was 31.6 dBm, yielding a power density of 9.7 W/mm at PAE of 43% and DE of 50%, respectively. As shown in Fig. 6, the power density is as high as 10.3 W/mm at Vds ¼ 45 V. The improvement in the linear gain from 8.6 dB to 11.3 dB is associated with a reduced gate-to-drain feedback capacitance (Cgd). The measured output power density increased almost linearly to Vds ¼ 37 V, reaching a power density of 9.3 W/mm. Above Vds >35 V, there is a deviation from linear behavior in the measured power density. Accordingly, there is a reduction in PAE. These characteristics at higher drain biases can be attributed to several factors such as nonoptimum matching condition due to a limited load Gamma available, rise of junction temperature, or reduced field-plating effect. The CW power performances of 2  75 μm devices with L_ledge of 0.55 μm and of 2  75 μm reference sample were measured and compared at 30 GHz and at Vds ¼ 30 V. The devices with n + source ledge showed both improved linear gain and output power by 0.6 dB and PAE by
5% over the reference samples. Fig. 8 summarizes measured PAE versus power density of 2  75 μm GaN HEMTs with and without n + source ledge at 30, 33, and 36 GHz. At 30 GHz, the devices with different length of n + source ledges (filled circles) showed improvement in PAE by 5–6% for the same power density, compared to those (open squares) without n + contact ledge. Similarly, these devices with n + contact ledge exhibited very high PAE of 55% at 33 (open diamonds) and 36 GHz (open triangles). At 33 GHz, a peak PAE of 55% was observed with power density of 5 W/mm. The load-pull testing was limited to reduced drain bias due to the available RF input power at 33 GHz and 36 GHz. The improvement can be attributed

127

III-Nitride microwave power transistors

36 GHz

33 GHz

60 30 GHz 50

PAE (%)

40 Without n+ contact ledge at 30 GHz

30 20 10 0 0.0

2.0

4.0 6.0 8.0 Power density (W/mm)

10.0

Fig. 8 A plot of measured PAE versus power density from 2  75 μm field-plated GaN HEMT devices with different length of n + source ledges (filled circle) and no n + source ledge (open square) at 30 GHz, respectively. The gate length is 140 nm. Measured PAEs and power densities at 33 GHz (open diamond) and 36 GHz (open triangle) are also plotted.

to an increase in Imax and gm, linearized gm over the large Ids, and reduced on-resistance. Improvement in PAE at power densities above 8 W/mm is not observed partially due to the non-optimum measurements associated with limited load-pull gamma. In summary, GaN HEMT devices were designed with n + source ledge without impacting their gate-to-drain capacitance and breakdown voltages. Improvement in both small-signal and large-signal performances was observed, yielding state-of-the-art performance at the Ka-band. At 30GHz, CW output power density of 10 W/mm was obtained with PAE of 40% and gain of 11dB at Vds ¼ 42 V. The peak PAE was measured as high as 55% at 5 W/mm output power density at 33 and 36 GHz. Thus, deeply-scaled FP GaN HEMTs can offer an excellent MMW power performance with very high power density and total output power.

3. Dual-gate and cascode field-plated GaN HEMTs for High-efficiency applications The short-gate-length integrated FP devices with gamma-gate structures, without the source-connected FP, demonstrated high cut-off frequency (fT) of 60 GHz and maximum oscillation frequency (fMAX) of
150 GHz (Moon et al., 2006), with a breakdown voltage of
90 V. Excellent progress

128

Jeong-Sun Moon

also has been made in MMIC amplifiers, with early results of 20 W and 43% PAE (Moon et al., 2006) to the latest results of 58 W and 38% PAE (Piotrowicz et al., 2008) in the X-band. Using 150-nm gate length FP GaN HEMTs, excellent class E switched-mode amplifiers were demonstrated in the X-band with 61% PAE at 3.5 W of output power (Moon et al., 2012). Also, excellent X-band GaN power amplifiers were demonstrated in class-F design with 61% PAE at 9.2 W of output power (Kang and Moon, 2017). On the other hand, the short-gate-length FP structures (Moon et al., 2005, 2008; Palacios et al., 2005) increase gate-to-drain capacitance and reduce gain. Furthermore, short-gate-length GaN HEMTs, either in the field-plated gate or T-gate, suffer from short-channel effect, which results in reduced output resistance and poor pinch-off behaviors at high drain voltage. This often limits the output power and PAE scaling, which was the focus of this study. While dual-gate or cascode devices provide a way to improve small-signal gain due to the cancellation of gate-to-drain capacitance and higher output resistance (Chen et al., 1988), only a few groups have reported dual-gate or cascode GaN HEMTs with large-signal performances. Fig. 9A shows dual-gate FETs consisting of an RF gate (Vg1) and DC gate (Vg2) defined within a source-drain spacing. This is electrically equivalent to a cascode device where a common-gate (CG) FET and a common-source (CS) FET are connected in series. Green et al. (2000) reported the first cascodeconnected 0.25-μm gate length AlGaN/GaN HEMTs with improved linear gain by 7 dB and associated gain by 3 dB over common-source AlGaN/GaN HEMTs. Earlier work demonstrated PAE from 30% in CS HEMTs to 36% in cascode-connected AlGaN/GaN HEMTs at 4 GHz. Chen et al. (2000) reported dual-gate AlGaN/GaN HEMTs with 160-nm RF gate and 350-nm DC gate, which delivered power density of 3.5 W/mm, PAE of 45%, and associated gain of 12 dB at 8.2 GHz. Later, dual-gate GaN transistors were utilized for high-voltage applications (Lin et al., 2010) and low-noise amplifiers (Aust et al., 2006). This section describes state-of-the-art microwave performance of both dual-gate field-plated GaN HEMTs and cascode field-plated GaN HEMTs with a short gate length of 150 nm, where peak PAE of 71% to 74% was demonstrated up to continuous-wave output power of 2.5 W, without harmonic tuning at 10 GHz. The technology offers record performance for combined PAE, output power, and gain among reported GaN HEMTs in the X-band. Fig. 9A shows a schematic of a dual-gate FP GaN HEMT on a semiinsulating SiC substrate with RF gate (Vg1) and DC gate (Vg2) defined

129

III-Nitride microwave power transistors

A

B Dual-gate GaN FETs Vg1 RF gate

Vg2 DC gate

source

Drain N+ GaN D

1.2 1

Ids (A/mm)

0.8

SG: Black lines DG: Blue lines

0.6 0.4 0.2

600

Vdd = 10 V, Vg2 = +1 V, 0 V, -1 V

SG and DG GaN HEMTs: Gate length = 150 nm Vg1 = +1V to -3V, step = 1V Dual-gate: Vg2 fixed at +1V

Ids (mA/mm), gm (mS/mm)

C

DG: Vg1 = -2V

500

0

10

20

30

Vds (V)

40

50

60

gm curve Vg2 = +1 V

300 200 100 0

0

Ids curve Vg2 = +1 V

400

−4

−3.5

−3

−2.5

−2

−1.5

−1

Vg1 gate voltage (V)

Fig. 9 (A, B) A schematic diagram of a dual-gate field-plated GaN HEMT with a 0.5 μm long n + source ledge is shown with a SEM photograph. The source-drain spacing was 2.5 μm. (C) Measured pulsed I-V curves of SG and DG devices are shown with 250 ns pulses, where output conductance of the DG device is greatly reduced, and (D) Transfer curves of a 4  37.5 μm dual-gate field-plated GaN HEMT at several different Vg2 of 1 V, 0 V and 1 V. The Vdd was fixed at 10 V.

within 3-μm source-drain spacing. The gate length of the RF gate was 150 nm, and the gate length of the DC gate was 250 nm. The gates were fabricated with a double heterojunction Al0.3Ga0.7N/GaN/Al0.04Ga0.96N structure. The Ti/Al ohmic metallization was done by liftoff processing. The 0.5-μm n + GaN ledge with Si doping density of 7  1019/cm3 was defined at the source side only, which reduced access resistance without impacting the gate-to-drain breakdown voltage (Moon et al., 2005). At the same time, the n + GaN capping layer was recessed by a chlorine-based dry etching process, and the gate metal was defined. A transmission-linemethod (TLM) measurement showed ohmic contact resistance of 0.4 Ω mm. The gate-to-channel separation was
15 nm after recessing the AlGaN layer. A 75-nm-thick SiNx layer was deposited to provide surface passivation and to support the field-plated structure. The Pt/Au gate metallization was done by liftoff processing, and the gate overhang above the FP

130

Jeong-Sun Moon

SiNx layer was 550 nm for the 150-nm RF gate. For comparison, we fabricated three different device types on the same wafer; single-gate, dual-gate, and cascode GaN HEMTs. The on-state resistance (Ron) was 1.3 Ω mm for the single-gate GaN HEMTs with a source-drain spacing of 2.5 μm. Fig. 9B shows an SEM photograph of a fabricated dual-gate FP GaN HEMT. Fig. 9C-D shows measured on-wafer common source DC I-V and transconductance characteristics of a 0.15-mm (4  37.5 μm) dual-gate GaN HEMT. Ron was 1.4 Ω mm at Vg2 ¼ +1 V. Knee voltage was 2 V. The device had an Idss of 800 mA/mm at Vg2 ¼ +1 V. The I-V curves of the dualgate GaN HEMT show excellent saturation and high output resistance with measured pinch-off characteristics up to 60 V, compared to the single-gate GaN HEMT. The Imax was reduced in DG devices due to an additional gate-to-drain parasitic resistance. As shown in Fig. 9D, peak gm was 370 mS/mm at Vg2 ¼ +1 V. Fig. 10A-B shows the measured small-signal S-parameters of single-gate and dual-gate GaN HEMTs. The high output resistance, as shown in the I-V curves, led to a shift of S22 toward OPEN impedance. The S12 magnitude of dual-gate GaN HEMTs is greatly reduced below 40 dB compared to
 20 dB for single-gate GaN HEMTs, which is due to reduced gateto-drain capacitance of 0.025 pF/mm for dual-gate devices compared to 0.22 pF/mm for single-gate devices. The S21 magnitude of dual-gate GaN HEMTs is higher due to high output resistance. The extra phase shift in S21 is due to the delay from the DC control gate capacitance (not shown here). Fig. 9C compares short-circuit current gain j h21 j2, and maximum stable/available gain (MSG/MAG) versus frequency at Vds ¼ 10 V and peak gm gate bias. The measured fT values were 53 GHz for single-gate GaN HEMTs and 38 GHz for dual-gate GaN HEMTs. The MSG values at 10 GHz are 14 dB for single-gate GaN HEMTs and 25 dB for dual-gate GaN HEMTs, respectively. The dual-gate GaN HEMTs provide almost 11 dB improvement in MSG. The non-smoothness in MSG/MAG of dual-gate GaN HEMTs are due to very small S12, not measurement error. Table 5 shows extracted output conductance of both SG and DG devices. Fig. 11A shows measured CW RF power performance of 2  110-μm dual-gate FP GaN HEMTs (filled circles with Vg2 ¼ +1 V, open circles with Vg2 ¼ 0 V) and 2  110-μm single-gate FP GaN HEMTs (open squares) at 10 GHz using Maury load-pull without the harmonic tuning. The devices were biased at a quiescent current of 50 mA/mm at a fixed Vdd¼ 20 V. The load impedances for the load-pull measurements were ZL ¼ 47.4 Ω + j130 Ω for single-gate.

131

III-Nitride microwave power transistors

A

B 20 Wg = 2x110 µm Vdd =10 V SG: Vg at peak gm bias DG: Vg1 at peak gm bias, Vg2 = +1 V

S21_DG S21_SG

0

S12_SG S22_DG

S11_DG S11_SG

-20

S22_SG -40

S12_DG -60

-80 0

5

10

15

20

25

30

Frequency (GHz)

C 50

S-par Gains (dB)

40

MSG/MAG_DG Ih21I2 _DG

30

Ih21I2 _SG

20

MSG/MAG_SG

10

0 0.1

Wg = 2x110 µm Vdd =10 V DG Vg2 = +1 V

1

10

100

Frequency (GHz)

Fig. 10 (A, B) Measured small-signal S-parameter gains of 2  110 μm single-gate (SG) and dual-gate (DG) FP GaN HEMT are shown at Vds ¼ 10 V and Vgs bias for peak gm operation. For the DG device, the Vg2 was biased at +1 V. (c) Measured S-parameter gains, current gain and maximum stable gain (MSG)/maximum available gain (MAG), as a function of frequency for both SG and DG FP GaN HEMTs, are shown. Table 5 Key equivalent circuit parameters of single-gate and dual-gate GaN HEMTs. Single-gate Dual-gate (Vg2 5 0 V) Dual-gate (Vg2 5 0 V)

Cgs (pF/mm)

1.16

1.45

1.41

Cgd (pF/mm)

0.21

0.024

0.024

Gds (mS/mm)

22.8

4.5

4.4

HEMTs, and ZL ¼ 36.3 Ω + j117 Ω for dual-gate HEMTs. Both devices show similar output power of 27.4 dBm. But power gain and PAE were higher in the dual-gate FP GaN HEMTs, which showed power gain of 13.7 dB at peak PAE of 71%. Linear gain was 17 dB. Single-gate FP GaN

132

Jeong-Sun Moon

70 PAE (%), Gain (dB)

B

80

60

Dual-gate GaN HEMT PAE = 71 %

Wg = 2x110 µm Vd = 20 V Iq = 11 mA F = 10 GHz

50 PAE of single-gate GaN HEMT

40

80 Cascode GaN HEMT Wg = 0.6 mm Vdd = 28 V Iq = 1 mA (filled circle) Iq = 6 mA (empty square)

70

30 Gain of dual-gate GaN HEMT

PAE (%), Gain (dB)

A

60 50

PAE

40 30 Gain

20

20

10 10 16

Gain of single-gate GaN HEMT

18

20

22

24

Pout (dBm)

26

0 28

30

20

25

30

35

Pout (dBm)

Fig. 11 Measured power data at 10 GHz are shown from (A) single-gate and dual-gate FP GaN HEMTs (filled circles with Vgs2 ¼ +1 V, open circles with Vgs2 ¼ 0 V) and (B) cascode FP GaN HEMTs without harmonic tuning.

HEMTs showed power gain of 13 dB at peak PAE of 67%, and with linear gain of 16 dB. While the dual-gate GaN HEMTs provide almost 11 dB improvement in small-signal gain compared to the single-gate GaN HEMTs, the large-signal power gain improvement is about 1 dB. This is due to the fact that the load-pull measurements were done at class AB bias with a quiescent current of 11 mA (50 mA/mm), and both load and source impedance were tuned to maximize PAE, rather than associated power gain. Fig. 11B shows measured CW RF power performance of 0.6 mm CG and 1 mm CG cascode devices at 10 GHz. The devices were biased at different quiescent currents of 1 mA and 6 mA at a fixed Vdd ¼ 28 V. The peak output power ranged from 2.3 to 2.5 W with associated power gain of
13.4 dB at the peak PAE point. Peak PAE was 71–74%, which is the highest PAE reported in GaN HEMTs at 10 GHz with a few Watt level. Fig. 12 summarizes load-pull data taken from single-gate, dual-gate, and cascode FP GaN HEMTs at 10 GHz. Both dual-gate and cascode FP GaN HEMTs showed improved PAE given CW output power. PAE improvement was >10% at a 2 W output power level, which is attributed to improved output resistance, gain, and pinch-off characteristics. We note that we also characterized the dual-gate GaN HEMTs in terms of harmonic loadpull at a fundamental frequency of 2 GHz. With a second harmonic (4 GHz) impedance of an open circuit presented to the devices, peak PAE was improved by
10%. In summary, 150-nm gate length dual-gate or cascode field-plated GaN HEMTs offer state-of-the-art microwave power performance with a record combined PAE and power density at 10 GHz.

133

III-Nitride microwave power transistors

80 Dual-gate or Cascode GaN HFETs

70 60

PAE (%)

50 Vg2

40

Vdd

Single-gate GaN HFETs

MIM Cap

30 20 10 0 25

RF in, Vg1

Dual gate, Cascode GaN HEMTs Vg2 = +1V

30 35 Output Power (dBm)

40

Fig. 12 Measured PAE versus output power of GaN HEMTs at 10 GHz without harmonic tuning is shown. Both dual-gate and cascode devices were biased with Vg2 ¼ +1 V.

4. Speed performance and scaling of field-plated GaN HEMTs Field-plated (FP) GaN HEMTs show very promising RF performance with both high power density and high PAE up to 40 GHz frequency operation. These short-gate-length FP devices were fabricated with a gammagate structure and demonstrated high cut-off frequency (fT) of 60GHz and maximum oscillation frequency (fMAX) of
150 GHz at the gate length of 120 nm (Moon et al., 2008). Also, the fT and fMAX of
70 GHz and
100 GHz were demonstrated with the gate length of 160 nm (Palacios et al., 2005). The off-state breakdown voltage was
80V (Palacios et al., 2005) and
90V (Moon et al., 2002a). Recently, the short-gate-length FP devices also demonstrated state-of-the-art microwave noise performance with a minimum noise figure of 0.89 dB and an associated gain of 11 dB at 10 GHz (Moon et al., 2006). While there are growing demands for high fT and fMAX of FP GaN HEMTs, the fT and fMAX of FP GaN HEMTs has been limited primarily due to increased gate (i.e., gate-to-source and gate-to-drain) capacitance due to the FP gate layout. In particular, their high 2DEG sheet resistance (
3 times higher than that of GaAs PHEMTs or InP HEMTs), non-linearity in transconductance, and saturation velocity (
1.5  107 cm/s)

134

Jeong-Sun Moon

need to be addressed properly in device design. Nidhi et al. (2006) indicated that the access resistance at the source (Rs) and drain (Rd) can impact the high-frequency performance of GaN HEMTs through a parasitic charging delay (Tasker and Hughes, 1989), (Rs + Rd) * Cgd, where Cgd is a gateto-drain feedback capacitance. This section describes a scaling of the FP GaN HEMTs to the gate length of 90 nm, which yielded a record fT and fMAX of 95 GHz and 200 GHz, respectively. The source-drain spacing was scaled to 1 μm, along with n+-GaN ohmic contact, to minimize the parasitics. Fig. 13A shows the scaled field-plated GaN HEMTs fabricated with both gate lengths of 120 nm and 90 nm. The SEM image in Fig. 13B shows the 90-nm FP gate design, where the field-plated gate overhang is
500 nm and A

90 nm FP gate

120 nm FP gate

Lg = 120 nm

Lg = 90 nm ~0.5 or 0.7 m

~0.7 m SiNx

Regrown Source

AlGaN/GaN

SiNx Regrown Drain

Regrown Source

AlGaN/GaN

Regrown Drain

1 m

2 m SiC substrate B

SiC substrate C

E-beam resist opening FP2 gate foot = 89 nm

0.5 µm

Gate metal Sourcedrain spacing = 1 µm

FP2 gate overhang ~500 nm

Vd = 5 V Wg = 2x37.5 um

600

400 Gm (mS/mm))

200

0 –3

1 µm

Ids (mA/mm)

800

Ids (mA), Gm (mS)

FP2 gate overhang = 354 nm

1000

Black: FP1 GaN (Lg = 120 nm) Blue: FP2 GaN (Lg = 90nm)

–2.5

–2

–1.5

–1

–0.5

0

Vgs (V)

Fig. 13 (A) Schematics of scaled FP GaN HEMT with gate length of 120 nm and 90 nm. (B) A scanning electron microscope (SEM) image of the fabricated FP GaN HEMT with 90-nm gate length. (C) Measured transfer curves and gm curves of the scaled FP GaN HEMTs.

135

III-Nitride microwave power transistors

the gate foot is 90 nm. Also the source-drain spacing was reduced to 1.0μm from 2.0 μm. To minimize the ohmic contact resistance, the FP GaN HEMTs were processed with n + GaN ohmic regrowth. As a result, the on-state resistance was 1.0 Ω mm in comparison to 1.7 Ω mm without the n+-GaN ohmic regrowth. The gate-to-channel depth was 170 nm. As shown in Fig. 13C, the source-drain current at zero-gate voltage (Idss) was 980 mA/mm for 90nm FP2 devices, compared to
800 mA/mm for 120 nm FP devices at Vds ¼ 5 V. The measured peak gm is as high as 530 mS/mm for 90-nm FP devices, compared to
420 mS/mm for 120-nm FP devices. The small-signal S-parameters of the scaled 2  37.5-μm FP GaN HEMTs were measured at peak gm gate bias and Vd ¼ 5 V. Fig. 14A shows A 50 FP2 GaN HEMT Lg = 90 nm FP overhang ~500 nm Vd = 5 V

Gain (dB)

40

30

|H |

U (dB)

20

fMAX = 200 GHz

MSG (dB)

10

fT = 95 GHz 0

1

10

100

Frequency (GHz)

0.8

1.0

B Max SwpSwp Max 67GHz 67 GHz

2. 0

6 0.

Lg = 90 nm Wg = 2x37.5 µm Vdd = 5 V

0 3.

0 4.

0. 4

5.0

0.2

10.0

4.0 5.0

3.0

2.0

1.0

0.8

0.6

0.4

S22

-0.2

-10.0

-4 .0 -5.0

-1.0

-0.8

-0 .6

.0 -2

S11

.4 -0

-3 .0

0

0.2

10.0

S12

Swp Min Swp Min 0.5 0.5GHz GHz

Fig. 14 (A) Measured RF gains versus frequency for 90-nm gate length FP GaN HEMTs with equivalent model parameters. (B) Measured and simulated S-parameters are compared at Vd ¼ 5 V with Table 6 showing equivalent circuit parameters.

136

Jeong-Sun Moon

that the 90-nm FP GaN HEMTs have an extrinsic fT of 95 GHz and fMAX of 200 GHz, respectively. For the 120-nm FP GaN HEMTs, the extrinsic fT and fMAX are reduced to 70 GHz and 175 GHz, respectively. To see the impact of the FP gate overhang on fT and fMAX of 90-nm gate length FP GaN HEMTs, we have also fabricated FP GaN HEMTs with different overhang length. As the FP gate overhang increases to 0.7 μm, the extrinsic fT and fMAX are reduced to 80 GHz and 200 GHz, respectively. This is due to the increased gate-to-drain feedback capacitance (Cgd), since the extrinsic fT depends on Cgd in fT ¼ gm/[(Cgs + Cgd) * 2π] without Miller effect correction. Fig. 14B shows measured S-parameter data overlapped with modeled S-parameters with an excellent fit up to 67 GHz. Table 6 summarizes key device parameters such as gm, Rds, Cgs, and Cgd and the modeled extrinsic fT of 94 GHz and fMAX of 204 GHz are very close to the measured values. Fig. 15A shows fT and fMAX of a 90-nm gate length FP GaN HEMT device as a function of gate voltage at Vd ¼ 5 V and 10 V. The fT and fMAX are very similar for both Vd ¼ 5 V and 10V, with a small reduction in fT at Vd ¼ 10 V. This is attributed to the fact that the extracted RF gm is reduced from 39 mS at Vd ¼ 5 V to 36 mS at Vd ¼ 10 V, along with the Cgs + Cgd value changed from 66fF at Vd¼ 5 V to 69 fF at Vd ¼ 10 V. Fig. 15B compares measured fT and fMAX values of the scaled 90-nm FP GaN HEMTs with previously reported FP GaN HEMTs (Ando et al., 2003a; Kumar et al., 2006; Moon et al., 2005, 2008; Palacios et al., 2005; Wu et al., 2004). As shown in the blue symbols, fT and fMAX of the scaled FP GaN HEMTs show a record values of
100 GHz and 200 GHz, respectively. Interestingly, fT and fMAX of

A

B

250

fMAX

Black: Vd = 5 V Blue: Vd = 10 V

150

(GHz)

100

f ,f

150

MAX T

T

f ,f

fMAX

200

MAX

(GHz)

200

250

fT

50

100

fT

50

This work: Blue symbols 0 –3

–2

–1

Vgs (V)

0

1

0 60

80

100

300

Gate Length (nm)

Fig. 15 (A) Measured fT and fMAX versus gate bias for 90 nm gate length FP GaN HEMTs. (B) Measured fT and fMAX of 90 nm gate length FP GaN HEMTs (in filled symbols) are plotted in terms of gate length, in comparison to the previously reported FP GaN HEMTs.

III-Nitride microwave power transistors

137

the scaled FP GaN HEMTs has not saturated due to parasitics, so further optimization of short gate length FP GaN HEMTs can further improve their small-signal RF performance. In brief, high-speed field-plated GaN HEMTs were demonstrated with a record fT of 95 GHz and fMAX of 200 GHz among the field-plated GaN HEMT devices, with aggressive scaling of both lateral scaling of source-to-drain distance to 1 μm and the gate length scaling to 90nm.

5. Future development of microwave power GaN HEMTs RF communications with spectral efficiency utilizes complex modulation schemes that require amplifier linearity. GaN HEMTs have a high breakdown voltage that offers high output impedance and power density per input capacitance over GaAs PHEMTs. Over the last decade, various wideband GaN HEMT MMIC power amplifiers and low-noise amplifiers (Ellis et al., 2004; Koayashi, 2012) have demonstrated high dynamic range with an excellent output third-order intercept point (OIP3) of 40–52 dBm. However, for the high OIP3, these wideband GaN amplifiers exhibit an OIP3/Pdc ratio <5, and consume large amounts of DC power. RF communications with spectral efficiency utilizes complex modulation schemes that require amplifier linearity. As for wideband amplifiers with high dynamic range, the highest value of the OIP3/Pdc was 16 (Moon et al., 2016, 2017). The PAE was limited to 40%—20% over 100 MHz to 8 GHz bands at Psat of
35 dBm. Challenges to improving PAE versus linearity in GaN amplifiers remain. Current linearity/PAE/output power trade-offs impose a significant increase in an RF system’s size, weight, and power (SWaP). As amplifier-operating frequency moves into the millimeter wave (mmwave) range; the PAE also becomes important to save the system prime power. As for 5G applications, high efficiency and linearity of amplifiers are required to support complex waveforms with high peak-to-average ratio (PAPR) and large instantaneous bandwidth. Recently, CMOS-based 28-GHz power amplifiers were demonstrated with 40% peak PAE at 18.9 dBm saturated output power using 45-nm n-type CMOS SOI technology (Rostomyan et al., 2018). With a 64-QAM OFDM signal with 800 MHz bandwidth, the nMOS PA demonstrated an average output power of 9.8 dBm with PAE of 14.8%. GaN HEMTs have a high breakdown voltage that offers high output impedance and power density per input capacitance over GaAs PHEMTs

138

Jeong-Sun Moon

or CMOS. Over the last decade, various GaN HEMT MMIC power amplifiers have been demonstrated, which include high-power (
10 W) X-band GaN PAs with PAE on the order of 50% (Kang and Moon, 2017), and 4–9 W Ka-band GaN PAs with PAE of around 30% (Campbell et al., 2012; Ng et al., 2014). Also, GaN amplifiers were designed for satellite communications and showed 29.5% PAE at C/IM3 of 25 dBc and with 5 W output power at 30 GHz (Noh and Eom, 2016). Challenges to improving PAE versus output power versus linearity in GaN amplifiers remain. Current linearity/PAE/output power trade-offs impose a significant increase in an RF system’s size, weight, and power (SWaP).

Acknowledgments This material is based upon work supported by the Office of Naval Research, which was monitored by Dr. Paul Maki. The views, opinions, and/or findings contained in this article/presentation are those of the author/presenter and should not be interpreted as representing the official views or policies, either expressed or implied, of the ONR.

References Ando, Y., Okamoto, Y., Miyamoto, H., Nakayama, T., Inoue, T., Kuzuhara, M., 2003a. 10-W/mm AlGaN-GaN HFET with a field modulating plate. IEEE Electron Device Lett. 24, 289–291. https://doi.org/10.1109/LED.2003.812532. Y. Ando, Y. Okamoto, K. Hataya, T. Nakayama, H. Miyamoto, T. Inoue, and M. Kuzuhara, “12 W/mm recessed-gate AlGaN/GaM Heterojunction Field-plate FET” in IEDM Technical Digest, Washington, DC, Dec 2003b, pp 2311-2314. Aust, M.V., Sharma, A.K., Chen, Y.-C., Wojtowicz, M., 2006. Wideband dual-gate GaN HEMT low noise amplifier for front-end receiver electronics. In: Proceedings of Compound Semiconductor IC Symposium (CSICS), pp. 89–92. https://doi.org/10.1109/ CSICS.2006.319921. Campbell, C.F., et al., 2012. High efficiency Ka-band power amplifier MMICs fabricated with a 0.15 μm GaN on SiC HEMT process. In: IEEE MTT-S International Microwave Symposium Digest. Chen, Y.K., Wang, G.W., Radulescu, D.C., Eastman, L.F., 1988. Comparison of microwave performance between single-gate and dual-gate MODFETs. IEEE Electron Device Lett. 9, 59–61. https://doi.org/10.1109/55.2040. Chen, C.-H., Coffie, R., Krishnamurthy, K., Keller, S., Rodwell, M., Mishra, U.K., 2000. Dual-gate AlGaN/GaN modulation-doped field-effect transistors with cut-off frequencies fT > 60 GHz. IEEE Electron Device Lett. 21, 549–555. https://doi.org/ 10.1109/55.887461. Chini, A., Buttari, D., Coffie, R., Shen, L., Heikman, S., Chakraborty, A., Keller, S., Mishra, U.K., 2004. Power and linearity characteristics of field-plated recessed-gate AlGaN-GaN HEMTs. IEEE Electron Device Lett. 25, 229–231. Chu, R., Shen, L., Fichtenbaum, N., Brown, D., Chen, Z., Keller, S., Denbaars, S.P., Mishra, U.K., 2008. V-ggate GaN HEMTs for X-band power applications. IEEE Electron Device Lett. 29, 974–976. Ellis, G.A., Moon, J.S., Wong, D., Micovic, M., Kurdoghlian, A., Hashimoto, P., Hu, M., 2004. Wideband AlGaN/GaN HEMT MMIC low noise amplifier. In: IEEE MTT-S Digest, p. 153.

III-Nitride microwave power transistors

139

Fitch, R.C., Walker Jr., D.E., Green, A.J., Tetlak, S.E., Gillespie, J.K., Gilbert, R.D., Sutherlin, K.A., Gouty, W.D., Theimer, J.P., Via, G.D., Chabak, K.D., Jessen, G.H., 2015. Implementation of high-power-density X-band AlGaN/GaN high electron mobility transistors in a millimeter-wave monolithic microwave integrated circuit process. IEEE Electron Device Lett. 36, 1004–1007. https://doi.org/10.1109/ LED.2015.2474265. Green, B.M., Chu, K.K., Smart, J.A., Tilak, V., Kim, H.-T., Shealy, J.R., Eastman, L.F., 2000. Cascode-connected AlGaN/GaN HEMTs on SiC substrates. IEEE Microwave Guid. Wave Lett. 10, 316–318. https://doi.org/10.1109/75.862226. Inoue, T., et al., 2004. 30 GHz-band 5.8 W High-power GaN heterojunction FET. In: 2004 IEEE MTT-S International Microwave Symposium Digest, pp. 1649–1652. Kang, J., Moon, J.S., 2017. Highly efficient wideband X-band MMIC class-F power amplifier with cascode FP GaN. Electron. Lett. 53, 1207–1290. Kasahara, K., Miyamoto, H., Ando, Y., Okamoto, Y., Nakayama, T., Kuzuhara, M., 2002. Ka-band 2.3 W power GaN heterojunction FET. In: IEDM Technical Digest, San Francisco, CA, pp. 673–676. Khan, M.A., Bhattarai, A., Kuznia, J.N., Olson, D.T., 1993. High electron mobility transistor based on a GaN-AlxGa1-xN heterojunction. Appl. Phys. Lett. 63, 1214–1415. Koayashi, K.W., 2012. An 8-W 250 MHz to 3 GHz decade bandwidth low-noise GaN MMIC feedback amplifier with >+51 dBm OIP3. IEEE J. Solid State Circuits 47, 2316. Kumar, V., Chen, G., Guo, S., Adesida, I., 2006. Field-plated 0.25 μm gate-length AlGaN/GaN HEMTs with varying field-plate length. IEEE Trans. Electron Devices 53, 1477–1480. Lee, C., Saunier, P., Yang, J., Khan, M.A., 2003. AlGaN-GaN HEMTs on SiC with CW power performance of >4 W/mm and 23% PAE at 35 GHz. IEEE Electron Device Lett. 24, 616–618. Lin, B., Saadat, O.I., Palacios, T., 2010. High-performance integrated dual-gate AlGaN/GaN enhancement-mode transistor. IEEE Electron Device Lett. 31, 990–992. https://doi.org/10.1109/LED.2010.2055825. Mishra, U.K., Parikh, P., Wu, Y.-F., 2016. AlGaN/GaN HEMTs—an overview of device operation and applications. Proc. IEEE 90, 272–275. https://doi.org/10.1109/ JPROC.2002.1021567. Moon, J.S., Micovic, M., Janke, P., Hashimoto, P., Wong, W.-S., Widman, R.D., McCray, L.M., Kurdoghlian, A., Nguyen, C., 2001. GaN HEMTs operating at 20 GHz with continuous-wave power density >6 W/mm. Electron. Lett. 37, 528–530. Moon, J.S., Graber, R., et al., 2002a. >70% power-added-efficiency dual-gate, cascode GaN HEMTs without harmonic tuning. IEEE Electron Device Lett. 37, 1022–1031. https:// doi.org/10.1109/LED.2016.2520488. Moon, J.S., et al., 2002b. Submicron enhancement-mode AlGaN/GaN HFETs. In: Device Research Conference Digest, pp. 23–24. Moon, J.S., Wu, S., Wong, D., Milosavljevic, I., Conway, A., Hashimoto, P., Hu, M., Antcliffe, M., Micovic, M., 2005. Gate-recessed AlGaN-GaN HEMTs for highperformance millimeter-wave applications. IEEE Electron Device Lett. 26, 348–350. https://doi.org/10.1109/LED.2005.848107. Moon, J.S., Wong, D., Antcliffe, M., Hashimoto, P., Hu, M., Willadsen, P., Micovic, M., Moyer, H.P., Kurdoghlian, A., Mac Donald, P., Wetzel, M., Bowen, R., 2006. High PAE 1 mm AlGaN/GaN HEMTs for 20 W and 43% PAE X-band MMIC amplifiers. In: International Electron Device Meeting, pp. 1–2. https://doi.org/10.1109/ IEDM.2006.346801. Moon, J.S., Wong, D., Hu, M., Hashimoto, P., Antcliffe, M., McGuire, C., Micovic, M., Willadsen, P., 2008. 55% PAE and high power Ka-band GaN HEMTs with linearized transconductance via n+ GaN source contact ledge. IEEE Electron Device Lett. 29, 834–837. https://doi.org/10.1109/LED.2008.2000792.

140

Jeong-Sun Moon

Moon, J.S., Wong, D., Hashimoto, P., Hu, M., Milosavljevic, I., Willadsen, P., McGuire, C., Burnham, S., Micovic, M., Wetzel, M., Chow, D., 2011. Sub-1dB noise figure performance of high-power field-plated GaN HEMTs. IEEE Electron Device Lett. 32, 297–299. https://doi.org/10.1109/LED.2010.2095408. Moon, J.S., Moyer, H., Donald, P.M., Wong, D., Antcliffe, M., Hu, M., Willadsen, P., Hashimoto, P., McGuire, C., Micovic, M., Wetzel, M., Chow, D., 2012. High efficiency X-band class-E GaN MMIC high-power amplifiers. In: IEEE Topical Conference on Power Amplifiers for Wireless and Radio Applications, pp. 9–12. https://doi.org/10.1109/PAWR.2012.6174909. Moon, J.S., et al., 2016. Wideband linear distributed GaN HEMT MMIC power amplifier with a record OIP3/Pdc. In: IEEE Conference on Power Amplifier for Wireless and Radio Applications (PAWR), p. 5. Moon, J.S., et al., 2017. 100 MHz–8 GHz linear distributed GaN MMIC power amplifier with improve power-added efficiency. In: IEEE Conference on Power Amplifier for Wireless and Radio Applications (PAWR)5. Ng, C.Y., et al., 2014. A 20-Watt Ka-band GaN high power amplifier MMIC. In: Proceedings of the 44th European Microwave Conference. Nidhi, Palacios, T., Chakraborty, A., Keller, S., Mishra, U.K., 2006. Study of impact of access resistance on high-frequency performance of GaN HEMTs by measurements at low temperature. IEEE Electron Device Lett. 27, 877–879. Noh, Y.S., Eom, Y.B., 2016. A linear GaN high power amplifier MMIC for Ka-band satellite communications. IEEE Microwave Wireless Compon. Lett. 26, 619–621. Palacios, T., Charkraborty, A., Rajan, S., Poblenz, C., Keller, S., Den Baars, S.P., Speck, J.S., Mishra, U.K., 2005. High-power AlGaN/GaN HEMTs for Ka-band applications. IEEE Electron Device Lett. 26, 781–783. https://doi.org/10.1109/LED.2005.857701. Piotrowicz, S., Morvan, E., Aubry, R., Bansropun, S., Bouvet, T., Chartier, E., Dean, T., Drisse, O., Dua, C., Floriot, D., di Forte-Poisson, M.A., Gourdel, Y., Hydes, A.J., Jacquet, J.C., Jardel, O., Lancereau, D., Mc Lean, J.O., Lecoustre, G., Martin, A., Ouarch, Z., Reveyrand, T., Richard, M., Sarazin, N., Thenot, D., Delage, S.L., 2008. State of the art 58 W, 38% PAE X-band AlGaN/GaN HEMTs microstrip MMIC amplifiers. In: IEEE Compound Semiconductor IC Symposium, pp. 1–4. https://doi. org/10.1109/CSICS.2008.39. Rostomyan, N., et al., 2018. Comparison of pMOS and nMOS 28 GHz high efficiency linear power amplifiers in 45 nm CMOS SOI. In: IEEE Conference on RF/Microwave Power Amplifiers for Radio and Wireless Applications (PAWR), pp. 26–28. Sandhu, R., et al., 2001. 1.6 W/m, 26% PAE AlGaN-GaN HEMT operation at 29 GHz. In: IEDM Technical Digest, Washington, DC, pp. 940–942. Tasker, P.J., Hughes, B., 1989. Importance of source and drain resistance to the maximum ft of millimeter-wave MODFETs. IEEE Electron Device Lett. 10, 291–293. Bolognesi, C.R., Kwan, A.C., DiSanto, D.W., 2002. Transistor delay analysis and effective channel velocity extraction in GaN HFETs. In: IEDM Tech. Dig., pp. 685–688. Wu, Y.-F., Saxler, A., Moore, M., Smith, R.P., Sheppard, S., Chavarkar, P.M., Wisleder, T., Mishra, U.K., Parikh, P., 2004. 30 W/mm GaN HEMTs by field plate optimization. IEEE Electron Device Lett. 25, 117–119. https://doi.org/10.1109/ LED.2003.822667.

Further reading Eastman, L.F., 2002. Experimental power-frequency limits of AlGaN/GaN HEMTs. In: 2002 IEEE MTT-S Digest, p. 2273.