Fabrication and Characterization of In0.53Ga0.47As/InAs/In0.53Ga0.47As Composite Channel Metamorphic HEMTs (mHEMTs) on a GaAs Substrate

In this work, we successfully demonstrated In0.53Ga0.47As/InAs/In0.53Ga0.47As composite channel metamorphic high electron mobility transistors (mHEMTs) on a GaAs substrate. The fabricated mHEMTs with a 100 nm gate length exhibited excellent DC and logic characteristics such as VT = −0.13 V, gm,max = 949 mS/mm, subthreshold swing (SS) = 84 mV/dec, drain-induced barrier lowering (DIBL) = 89 mV/V, and Ion/Ioff ratio = 9.8 × 103 at a drain-source voltage (VDS) = 0.5 V. In addition, the device exhibited excellent high-frequency characteristics, such as fT/fmax = 261/304 GHz for the measured result and well-matched modeled fT/fmax = 258/309 GHz at VDS = 0.5 V, which is less power consumption compared to other material systems. These high-frequency characteristics are a well-balanced demonstration of fT and fmax in the mHEMT structure on a GaAs substrate.


Introduction
High electron mobility transistors (HEMTs) based on indium-rich In x Ga 1-x As channel materials on an InP substrate have demonstrated excellent high-frequency and logic characteristics. In terms of high-frequency characteristics of the InGaAs channel HEMT [2], and Northrop Grumman Corporation exhibited an f T of 610 GHz/f max of 1.5 THz by using an In 0.53 GaAs/InAs/In 0.53 GaAs composite channel with a L g of 25 nm [3]. These remarkable performances have been achieved through downscaling of device feature size, an optimized fabrication process, and optimized InGaAs channel materials for excellent transport properties. In addition, InGaAs channel MOSFETs have shown outstanding logic performance on various substrates, such as InP and flexible substrates, with extensive efforts to enhance their capability in new device structures, S/D Ohmic contacts, and optimization of the gate stack [4][5][6][7]. Meanwhile, large-size and cheaper-cost substrates will be essential for largevolume manufacturing from a mass production point of view, but an InP substrate is more expensive than a GaAs substrate, and the size to date is limited to 6 inches. To overcome these limitations of the InP substrate, many groups have demonstrated many outstanding results for mHEMTs on a GaAs substrate [8][9][10][11]. In particular, Teledyne demonstrated excellent results of a 688 GHz f T by utilizing an In 0.7 GaAs mHEMT structure with dual Si δ-doping and an InAs-rich In 0.7 Al 0.3 As spacer on a GaAs substrate in 2011 [9]. Fraunhofer showed a maximum oscillation frequency (f max ) exceeding 1000 GHz by using an In 0.8 GaAs mHEMT structure on a GaAs substrate in 2013 [11]. Among various HEMT structures, an InGaAs/InAs/InGaAs composite channel was used to enhance high-frequency characteristics in HEMT structures because of its excellent electron transport properties, such as electron velocity and mobility [2,3,12]. However, an In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel structure on a GaAs substrate has not been demonstrated yet. In this work, we fabricated an In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel HEMT on a GaAs substrate incorporating a molybdenum (Mo)-based Ohmic contact using blanket Mo deposition and investigated its electrical performance, such as DC, logic, and RF characteristics, with an L g of 100 nm.

Layer Structure and Experiments
The mHEMT heterostructures consisted of a 500 nm In 0.52 Al 0.48 As buffer, a 12 nm In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As (4/5/3 nm) channel, a 3 nm In 0.52 Al 0.48 As spacer, Si δ-doping (4.1 × 10 12 cm −2 ), an 8 nm In 0.52 Al 0.48 As barrier, a 4 nm InP etch stop layer, and a 35 nm heavily doped In 0.53 Ga 0.47 As/In 0.52 Al 0.48 As multi-layer cap from the bottom to the top as shown in Figure 1a. The energy band diagram of the epitaxial structure is shown in Figure 1b. From this structure, sheet carrier density and electron hall mobility were measured to be 2.92 × 10 12 cm −2 and 10,000 cm 2 /V·s at room temperature, respectively, with four-point probe measurement methods (Van der Pauw measurement method). Device fabrication began with a 30 nm blanket molybdenum (Mo) deposition for ohmic contact to prevent surface contamination and improve the contact resistance (R c ), then mesa isolation down to an InAlAs buffer layer by Mo dry etching and wet etching. After Ti/Au/Ni (20/150/30 nm) metallization for source and drain, dry etching in an SF 6 /Ar plasma was performed to etch Mo in the gate region using the Ni metal etch mask of the source and drain [13]. A 30 nm thick layer of SiO 2 was deposited by plasma-enhanced chemical vapor deposition (PECVD), and then the pad patterns with Ti/Au (20/300 nm) were defined for ground-signal-ground probing. After e-beam exposure, the defined e-beam resist pattern was transferred to define the T-gate by using reactive ion etching based on CF 4 plasma. Gate recessing was performed in two different step stages, followed by anisotropic reactive ion etching of the InP etch stop layer in an Ar-based plasma [14]. After InP etching, Schottky gate metallization of Ti/Pt/Au (20/30/300 nm) was deposited on top of the InAlAs layer. Finally, the mHEMT with a width of 2 × 50 µm was fabricated, and a schematic of the fabricated mHEMT is shown in Figure 1c. Figure 1d shows the SEM image of the fabricated t-gate, whose foot and head sizes are 100 nm and 470 nm, respectively.
Micromachines 2023, 14, x FOR PEER REVIEW 2 particular, Teledyne demonstrated excellent results of a 688 GHz fT by utilizing In0.7GaAs mHEMT structure with dual Si δ-doping and an InAs-rich In0.7Al0.3As space a GaAs substrate in 2011 [9]. Fraunhofer showed a maximum oscillation frequency ( exceeding 1000 GHz by using an In0.8GaAs mHEMT structure on a GaAs substrate in 2 [11]. Among various HEMT structures, an InGaAs/InAs/InGaAs composite channel used to enhance high-frequency characteristics in HEMT structures because of its ex lent electron transport properties, such as electron velocity and mobility [2,3,12]. H ever, an In0.53Ga0.47As/InAs/In0.53Ga0.47As composite channel structure on a GaAs subst has not been demonstrated yet. In this work, we fabricated In0.53Ga0.47As/InAs/In0.53Ga0.47As composite channel HEMT on a GaAs substrate inco rating a molybdenum (Mo)-based Ohmic contact using blanket Mo deposition and in tigated its electrical performance, such as DC, logic, and RF characteristics, with an L 100 nm.

Layer structure and experiments
The mHEMT heterostructures consisted of a 500 nm In0.52Al0.48As buffer, a 12 In0.53Ga0.47As/InAs/In0.53Ga0.47As (4/5/3 nm) channel, a 3 nm In0.52Al0.48As spacer, Si δ-d ing (4.1 X 10 12 cm -2 ), an 8 nm In0.52Al0.48As barrier, a 4 nm InP etch stop layer, and a 35 heavily doped In0.53Ga0.47As/In0.52Al0.48As multi-layer cap from the bottom to the to shown in Figure 1a. The energy band diagram of the epitaxial structure is shown in Fig  1b. From this structure, sheet carrier density and electron hall mobility were measure be 2.92 x 10 12 cm -2 and 10,000 cm 2 /V·s at room temperature, respectively, with four-p probe measurement methods (Van der Pauw measurement method). Device fabrica began with a 30 nm blanket molybdenum (Mo) deposition for ohmic contact to pre surface contamination and improve the contact resistance (Rc), then mesa isolation d to an InAlAs buffer layer by Mo dry etching and wet etching. After Ti/Au/Ni (20/15 nm) metallization for source and drain, dry etching in an SF6/Ar plasma was perform to etch Mo in the gate region using the Ni metal etch mask of the source and drain [13 30 nm thick layer of SiO2 was deposited by plasma-enhanced chemical vapor deposi (PECVD), and then the pad patterns with Ti/Au (20/300 nm) were defined for grou signal-ground probing. After e-beam exposure, the defined e-beam resist pattern transferred to define the T-gate by using reactive ion etching based on CF4 plasma. G recessing was performed in two different step stages, followed by anisotropic reactive etching of the InP etch stop layer in an Ar-based plasma [14]. After InP etching, Scho gate metallization of Ti/Pt/Au (20/30/300 nm) was deposited on top of the InAlAs la Finally, the mHEMT with a width of 2 × 50 µm was fabricated, and a schematic of fabricated mHEMT is shown in Figure 1c. Figure 1d shows the SEM image of the fa cated t-gate, whose foot and head sizes are 100 nm and 470 nm, respectively.

Results and discussion
The transfer characteristic and output characteristic of the mHEMTs are show Figure 2. The maximum transconductance (gm,max) and maximum drain current den (ID,max) were 949 mS/mm and 413 mA/mm at VDS = 0.5 V, respectively. The output cha teristics presented in Figure 2b show good pinch-off characteristics, but the measured was 733 Ω-um, which is a higher value than the lift-off Mo/Ti/Mo/Au metal scheme w an InAs rich InAlAs barrier spacer [15]. The ohmic contact resistance (Rc) and shee sistance (Rsh) measured by the transmission line method (TLM) was as low as 0.01 Ωand 75.5 Ω /sq as shown in Figure 2c. When compared to the lift-off method usin Mo/Ti//Mo/Au scheme (30/20/20/150 nm) on the same multi-cap layer, the blanket method shows a higher Rsh value of 75.5 Ω /sq than the lift-off method of 69.2 Ω /sq cause of SF6/Ar plasma damage in the active region during Mo etching to define the ac gate region. However, the blanket Mo method shows a Rc of 0.011 Ohm-mm, whic lower than that of 0.026 Ohm-mm with the lift-off method because it is beneficial to pro the surface underneath the metal contact region from contaminants during the device cess. Due to the lower Rc of the Mo blanket method, a lower Ron value could be achie if the S/D distance was reduced, as in the self-aligned gate scheme.

Results and Discussion
The transfer characteristic and output characteristic of the mHEMTs are shown in Figure 2. The maximum transconductance (g m,max ) and maximum drain current density (I D,max ) were 949 mS/mm and 413 mA/mm at V DS = 0.5 V, respectively. The output characteristics presented in Figure 2b show good pinch-off characteristics, but the measured R on was 733 Ω-µm, which is a higher value than the lift-off Mo/Ti/Mo/Au metal scheme with an InAs rich InAlAs barrier spacer [15]. The ohmic contact resistance (R c ) and sheet resistance (R sh ) measured by the transmission line method (TLM) was as low as 0.01 Ω-mm and 75.5 Ω/sq as shown in Figure 2c. When compared to the lift-off method using a Mo/Ti//Mo/Au scheme (30/20/20/150 nm) on the same multi-cap layer, the blanket Mo method shows a higher R sh value of 75.5 Ω/sq than the lift-off method of 69.2 Ω/sq because of SF 6 /Ar plasma damage in the active region during Mo etching to define the active gate region. However, the blanket Mo method shows a Rc of 0.011 Ohm-mm, which is lower than that of 0.026 Ohm-mm with the lift-off method because it is beneficial to protect the surface underneath the metal contact region from contaminants during the device process. Due to the lower R c of the Mo blanket method, a lower R on value could be achieved if the S/D distance was reduced, as in the self-aligned gate scheme.

Results and discussion
The transfer characteristic and output characteristic of the mHEMTs are shown in Figure 2. The maximum transconductance (gm,max) and maximum drain current density (ID,max) were 949 mS/mm and 413 mA/mm at VDS = 0.5 V, respectively. The output characteristics presented in Figure 2b show good pinch-off characteristics, but the measured Ron was 733 Ω-um, which is a higher value than the lift-off Mo/Ti/Mo/Au metal scheme with an InAs rich InAlAs barrier spacer [15]. The ohmic contact resistance (Rc) and sheet resistance (Rsh) measured by the transmission line method (TLM) was as low as 0.01 Ω-mm and 75.5 Ω /sq as shown in Figure 2c. When compared to the lift-off method using a Mo/Ti//Mo/Au scheme (30/20/20/150 nm) on the same multi-cap layer, the blanket Mo method shows a higher Rsh value of 75.5 Ω /sq than the lift-off method of 69.2 Ω /sq because of SF6/Ar plasma damage in the active region during Mo etching to define the active gate region. However, the blanket Mo method shows a Rc of 0.011 Ohm-mm, which is lower than that of 0.026 Ohm-mm with the lift-off method because it is beneficial to protect the surface underneath the metal contact region from contaminants during the device process. Due to the lower Rc of the Mo blanket method, a lower Ron value could be achieved if the S/D distance was reduced, as in the self-aligned gate scheme.  Figure 3 shows the subthreshold characteristics at VDS = 0.5 V and 0.05 V, respectively At VDS = 0.5 V, the threshold voltage (VT) is -0.13 V, defined as the value of VGS that yields at ID = 1 mA/mm, and a VT of -0.13 V indicates that the fabricated mHEMT operated in depletion mode (D-mode). The fabricated device shows excellent electrostatic integrity such as the subthreshold swing (SS) at 84 mV/dec, the drain-induced barrier lowering (DIBL) at 89 mV/V, and the Ion/Ioff ratio of 9.8 × 10 3 , respectively. Additionally, the gate leakage current of the fabricated mHEMT was measured at VDS = 0.5 V and shows that the gate Schottky metallization is in good contact with the In0.52Al0.48As barrier layer. These outstanding logic performances are due to the well-designed heterostructure and opti mized fabrication process on the GaAs substrate.  Figure 3 shows the subthreshold characteristics at V DS = 0.5 V and 0.05 V, respectively. At V DS = 0.5 V, the threshold voltage (V T ) is −0.13 V, defined as the value of V GS that yields at I D = 1 mA/mm, and a V T of −0.13 V indicates that the fabricated mHEMT operated in depletion mode (D-mode). The fabricated device shows excellent electrostatic integrity, such as the subthreshold swing (SS) of 84 mV/dec, the drain-induced barrier lowering (DIBL) of 89 mV/V, and the I on /I off ratio of 9.8 × 10 3 , respectively. Additionally, the gate leakage current of the fabricated mHEMT was measured at V DS = 0.5 V and shows that the gate Schottky metallization is in good contact with the In 0.52 Al 0.48 As barrier layer. These outstanding logic performances are due to the well-designed heterostructure and optimized fabrication process on the GaAs substrate.
To verify the high-frequency RF characteristics of the mHEMT, S-parameters were measured from 0.5 to 40 GHz using a vector network analyzer (VNA). In addition, smallsignal modeling was performed by using a small-signal equivalent circuit [16], and we found that small-signal modeling and measured S-parameters are well matched, as shown in Figure 4a. Figure 4b shows the unity current gain cutoff frequency (f T ), maximum oscillation frequency (f max ), and maximum stable gain (MSG)/maximum available gain (MAG) against frequency for the measured results (symbols) and modeled results (solid lines) at V DS = 0.5 V and V GS = 0.2 V with a L g of 100 nm mHEMT device. The de-embedding method was done by using open and short patterns to extract parasitic pad capacitance and inductance. We obtained 261 GHz/304 GHz for f T /f max by extrapolation (dashed lines) and 258 GHz/309 GHz for f T /f max by small-signal modeling, respectively. This excellent high-frequency response is due to the high value of the intrinsic transconductance (g mi ) of 2.0 mS/µm. The extracted intrinsic parameters of the mHEMT are summarized in Table 1 and are well-matched to the measured results. The difference between g m,ext (0.95 mS/µm) and g mi (2.0 mS/µm) is due to the R s and g o values according to equation (1) [17]. To verify the high-frequency RF characteristics of the mHEMT, S-parameters were measured from 0.5 to 40 GHz using a vector network analyzer (VNA). In addition, smallsignal modeling was performed by using a small-signal equivalent circuit [16], and we found that small-signal modeling and measured S-parameters are well matched, as shown in Figure 4a. Figure 4b shows the unity current gain cutoff frequency (fT), maximum oscillation frequency (fmax), and maximum stable gain (MSG)/maximum available gain (MAG) against frequency for the measured results (symbols) and modeled results (solid lines) at VDS = 0.5 V and VGS = 0.2 V with a Lg of 100 nm mHEMT device. The de-embedding method was done by using open and short patterns to extract parasitic pad capacitance and inductance. We obtained 261 GHz/304 GHz for fT/fmax by extrapolation (dashed lines) and 258 GHz/309 GHz for fT/fmax by small-signal modeling, respectively. This excellent highfrequency response is due to the high value of the intrinsic transconductance (gmi) of 2.0 mS/µm. The extracted intrinsic parameters of the mHEMT are summarized in Table 1 and are well-matched to the measured results. The difference between gm,ext (0.95 mS/µm) and gmi (2.0 mS/µm) is due to the Rs and go values according to equation (1) [17].  To verify the high-frequency RF characteristics of the mHEMT, S-parameters were measured from 0.5 to 40 GHz using a vector network analyzer (VNA). In addition, smallsignal modeling was performed by using a small-signal equivalent circuit [16], and we found that small-signal modeling and measured S-parameters are well matched, as shown in Figure 4a. Figure 4b shows the unity current gain cutoff frequency (fT), maximum oscillation frequency (fmax), and maximum stable gain (MSG)/maximum available gain (MAG) against frequency for the measured results (symbols) and modeled results (solid lines) at VDS = 0.5 V and VGS = 0.2 V with a Lg of 100 nm mHEMT device. The de-embedding method was done by using open and short patterns to extract parasitic pad capacitance and inductance. We obtained 261 GHz/304 GHz for fT/fmax by extrapolation (dashed lines) and 258 GHz/309 GHz for fT/fmax by small-signal modeling, respectively. This excellent highfrequency response is due to the high value of the intrinsic transconductance (gmi) of 2.0 mS/µm. The extracted intrinsic parameters of the mHEMT are summarized in Table 1 and are well-matched to the measured results. The difference between gm,ext (0.95 mS/µm) and gmi (2.0 mS/µm) is due to the Rs and go values according to equation (1) [17].

Frequency [Hz]
Figure 4. (a) Comparison of small-signal modeling and measured S-parameters at VDS = 0.5 V and VGS = 0.2 V. A Smith chart of S11, S12, and S22 (left) and a polar chart of S12 (right). (b) Measured (symbols) and modeled (solid lines) of RF gains-Maximum oscillation frequency (fmax), maximum stable gain (MSG)/maximum available gain (MAG), and unity current gain cutoff frequency (fT) of the mHEMTs at VDS = 0.5 V and VGS = 0.2 V.    Table 2 shows the benchmark high-frequency characteristics of the published stateof-the-art pHEMT and mHEMT results with an L g of 100 nm. Among various HEMT structures, the In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel HEMT on an InP substrate shows excellent high-frequency characteristics such as an f T of 421 GHz and an f max of 620 GHz because of the well-optimized fabrication process and improved carrier transport properties of the In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel [18]. Our fabricated mHEMT exhibits an excellent L g f T of 26.1 GHz-µm, which is related to carrier transport properties [19], and an outstanding f T /f max of 261/304 GHz with a L g of 100 nm at a V DS = 0.5 V. Although the performance of the fabricated mHEMT is not comparable to that of the In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel HEMT on an InP substrate, our fabricated mHEMT shows outstanding high-frequency characteristics compared to a single InGaAs channel HEMT on an InP substrate and other mHEMT structures because of the excellent transport properties of the composite channel on a GaAs substrate. Additionally, our fabricated device is operated at a V DS = 0.5 V, which has a lower power consumption than other group devices' operational voltage. These excellent performances are mainly attributed to the well-grown In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel structure by using an In 0.52 AlAs buffer layer on a GaAs substrate, and a fabricated mHEMT would be a good candidate for the high-frequency device in both 5G and 6G communications through further scaling-down of device feature size.

Conclusions
The 100 nm In 0.53 Ga 0.47 As/InAs/In 0.53 Ga 0.47 As composite channel metamorphic high electron mobility transistors (mHEMTs) on a GaAs substrate exhibited excellent logic characteristics as well as high-frequency RF performances. These outstanding performances are due to the excellent carrier transport properties of the well-grown In 0.53 Ga 0.47 As/InAs/ In 0.53 Ga 0.47 As composite channel mHEMT structure on a GaAs substrate and an optimized fabrication process. The proposed mHEMT structure on a GaAs substrate, together with optimized source/drain and gate technologies, will potentially improve logic and highfrequency characteristics. Furthermore, the proposed mHEMT structure grown on a large-size GaAs substrate could be indispensable for large-volume manufacturing.