indium gallinum arsenide metamorphic high electron mobility transistors with various work function-gate metals

Materials Science in Semiconductor Processing 30 (2015) 41–47

Contents lists available at ScienceDirect

Materials Science in Semiconductor Processing journal homepage: www.elsevier.com/locate/mssp

Investigation of impact ionization and flicker noise properties in indium aluminum arsenide/indium gallinum arsenide metamorphic high electron mobility transistors with various work function-gate metals Hsien-Chin Chiu a,n, Chia-Hsuan Wu a, Che-Kai Lin a, Feng-Tso Chien b a b

Department of Electronics Engineering, Chang Gung University, Taoyuan, Taiwan, ROC Department of Electronics Engineering, Feng-Chia University, Taichung, Taiwan, ROC

a r t i c l e i n f o

Keywords: MHEMT Flicker noise Impact ionization Kink effect

abstract This study investigates the effect of impact ionization using Ir, Pt, Pd, Ti gate metals and the direct correlation between these high work function metals and low frequency noise (LFN) on an In0.4Al0.6As/In0.4Ga0.6As metamorphic high electron mobility transistor (MHEMT). The effect of impact ionization on DC, RF, and cryogenic LFN is systematically studied and discussed. Gate metals with high work functions are used to suppress the kink effect and gate leakage current. Experimental results suggest that the Ir gate MHEMT exhibits superior thermal stable properties in a strong electrical field at various temperatures, associated with high gain, high current, and excellent low-frequency noise performance. & 2014 Elsevier Ltd. All rights reserved.

1. Introduction As the scaling down of silicon based devices reaches its limit, III–V compound semiconductor field effect transistors have been identified as some of the most attractive nanoelectronic devices [1–3]. In particular, InP-based high electron mobility transistors (HEMT) exhibit high power, high operating frequency, low distortion, and low noise characteristics [4–7]. These are favorable for millimeter wave applications. A high indium content in a channel layer generally corresponds to high electron mobility and velocity, making InAs channel hetero-structured FETs (HFETs) highly suitable for low-power and high-speed logic applications, since they have an extremely high electron mobility of more than 30,000 cm2/V s [8,9]. However, the narrow bandgap of an InGaAs channel

n

Corresponding author. Tel.: þ886 3 2118800; fax: þ 886 3 2118507. E-mail address: [email protected] (H.-C. Chiu).

http://dx.doi.org/10.1016/j.mssp.2014.09.035 1369-8001/& 2014 Elsevier Ltd. All rights reserved.

with high indium content clearly has a negative effect on impact ionization. In a strong electric field beneath the gate depletion region, the “highly energetic” electrons and holes in the conduction or valence band can collide with each other, creating electron–hole pairs, which are excited to the conduction band. Impact ionization further induces the kink effect, which is a sudden rise in the drain current at a certain drain-to-source voltage that causes high drain conductance and transconductance (Gm) compression, leading to reduced voltage gain and yielding linearity [1,10]. The kink effect corresponds to Schottky characteristics. Gate metal stacks strongly dominate the device performance, including threshold voltage, Gm, gain, and reliability. Most importantly, the use of gate metals with a high work function can limit the kink effect and the gate leakage current. The metals that have previously been studied as gate metals include Ti [11–14], Pt [15–17] Pd [11,12,18], and Ir [19–21]. Pt exhibits the highest SBH (Schottky barrier height) Z0.8 eV on InAlAs, and has been widely used for the

42

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

undoped In0.4Al0.6As spacer layers. The 150 Å thick In0.4Al0.4As Schottky layer sits on top of the upper Si δ-doped layer to increase the height of the Schottky barrier, and a 100 Å-thick un-doped In0.52Ga0.48As cap layer was grown to enhance ohmic contact resistivity. The designed structure had a sheet-charge density of 3.5  1012 cm  2 and a Hall mobility of 8500 cm2/V s at 300 K after the un-doped In0.52Ga0.48As cap layer was removed. For device fabrication, low-resistivity ohmic contacts of Ni/Ge/Au alloy metals were deposited by thermal evaporation and patterned using a conventional liftoff process, followed by annealing at 320 1C in an N2-rich chamber for 20 s. A chemical etching method was utilized for the mesa isolation. The drain-to-source spacing was set to 3 μm to minimize series resistance. Following the highly selective chemical gate recess process, 2  50 μm2 Ir/Ti/Au (10 nm/20 nm/200 nm), Pd/Ti/Au (10 nm/20 nm/200 nm) and Pt/Ti/Au (10 nm/20 nm/200 nm) gate stacks were deposited and lifted off. After evaporation of the Ti/Au interconnection, a 20 nm-thick layer of SiO2 was deposited for passivation. For comparison, a conventional Ti/Au gate was also fabricated.

enhanced mode HEMT process. However, during the production or thermal treatment of process baking, gates with incorporated Pt suffer from severe diffusion problems, further influencing the characteristics and reliability of the devices [16]. A thermally stable gate metal is urgently required. Recent investigations revealed that iridium is a promising gate metal. Ir gates have a similar SBH to those of Pt gates on InAlAs epi-layers and exhibit a lower diffusion rate and greater thermal stability than Pt on InAlAs [22]. This investigation examines the kink effect associated with gate metals with high work functions and it has direct correlation with low frequency noise (LFN). The total effects of impact ionization on DC, RF, and cryogenic LFN will be systematically studied and discussed. 2. Device fabrication and structure Fig. 1 presents a cross-sectional photograph of the MHEMT with an epitaxial structure, grown on GaAs substrate. The grown wafers comprised of a 1 mm thick InxAl1 xAs metamorphic graded buffer layer with an indium content grading of x¼0–40%. Two two-dimensional electron gases formed in this In0.4Ga0.6As quantum well and electrons were transferred from both upper and lower silicon δ-doped layers through

3. Measurement results and discussion Fig. 2 plots the DC characteristics of devices with a gate length of 1 μm and Ir, Ti, Pd, Pt gate MHEMTs at room temperature. Notably, the drain currents of the Ir gate MHEMT show a good output I–V characteristics. In contrary, Pd and Pt gate MHEMTs increase at a constant rate with drain-to-source voltage (VDS) 41 V. Ti gate MHEMT has the worst drain current. The humping curves are direct evidence of impact ionization. Fig. 3 plots the dc transconductance (Gm) versus gate-to-source voltage (VGS) at various VDS from 0.5 to 3 V. At VDS ¼1 V, the Gmmax values of Ir, Ti, Pt, and Pd gates are 335, 372, 367, and 370 mS/mm, respectively. However, as VDS increases, the impact ionization that is caused by the increasingly strong electric field becomes severe. The generation of electrons in the channel enhances both IDS and Gm. Generated holes are collected at the gate terminal, causing a negative shift of VGS. From Fig. 3, Ti and Pd gate MHEMTs suffered greater impact ionization than did Ir and Pt gate MHEMTs. Fig. 4 plots the

D r a in -to - S o u r c e C u r r e n t , IDS (mA/mm)

Fig. 1. Cross-sectional structure of the fabricated MHEMT.

600

600

Ir 500

Pt

Ti

Pd 500

VGS =1V to -1.5V, step= -0.5V

400

400

300

300

200

200

100

100

0

0 0

1

2

3

0

1

2

3

0

1

2

3

Drain-to-Source Voltage, VDS (V) Fig. 2. The IDS–VDS characteristics of various devices.

0

1

2

3

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

VDS=1V

400

IDS(mA/mm) and Gm (mS/mm)

increasing V

500

increasing V

VDS=2V VDS=3V

Ir Ti Pt Pd

500

300

300

200

200

100 0 500

550

400

100

Ir

Ti increasing V

0 500

increasing V

400

400

300

300

200

200

Gmmax (mS/mm)

500

43

450 400 350 300 250

0.5

1.0

1.5

2.0

2.5

3.0

VDS (V) 100

0 -2

Pt

Pd -1

0 1 -2 -1 Gate-to-Source Voltage, VGS (V)

0

1

0

Fig. 3. The Gm and IDS versus VGS characteristics of various devices.

-0.8

VTH (V)

-1.0

500

IDS, max (mA/mm)

100

450 400 350

Ir Ti Pt Pd

300

-1.2

250

fitting line

-1.6

0.5

Ir (-9.86mV/V ) Ti (-81.43mV/V) Pt (-7.29mV/V) Pd (-19.69mV/V)

-1.4

0.5

1.0

1.5

1.0

1.5

2.0

2.5

3.0

VDS (V) Fig.5. (a) Gmmax–VDS and (b) IDS,max–VDS characteristics of various devices.

2.0

2.5

3.0

VDS (V) Fig. 4. Threshold voltage shifts versus VDS of various devices.

variations of threshold voltage of Ir, Ti, Pt, and Pd gate samples with VDS. Clearly, the Ir gate sample exhibits a smaller variation of threshold shifts at high VDS. Threshold voltage (VTH) shifts that are caused by impact ionizationgenerated hole injection are expressed as V TH ¼ V T0  S  V DS where VT0 is threshold voltage at zero substrate bias, and S is the fitted slope in Fig. 4. A linear relationship between VTH and VDS can indicate hole injection. A larger S value represents a stronger dependence on VDS, and suggests enhanced injection of holes. This constant of this linear dependence is 0.00986 for the Ir gate and 0.00729 for the Pt gate. These two values are much smaller than those for the Pd (0.01969) and Ti (0.08143) gate samples. Metals with a high work function indeed limit the injection of holes. Fig. 5(a) and (b) plots Gmmax and IDS,max as a function of VDS for each sample. The sudden rise in IDS and Gm is direct evidence of the kink effect. Impact ionization-generated electrons flow in the channel, therefore, increasing the

channel current and the transconductance. However, if band alignment does not confine the holes thus formed, these holes will leave the channel and be collected at the gate terminal. Hence, VGS is shifted negatively. Fig. 6 plots the typical gate current characteristics of the devices of interest at room temperature. The forward turn-on voltages, VON, of the Ir, Ti, Pt, and Pd gates are 0.7, 0.46, 1 and 0.55 V, respectively. The corresponding breakdown voltages, VBR, are  14.8, 8.8,  13.8, and  14.8 V. Fig. 7 plots the gate current to drain current ratio |IG|/IDS. For VGS o  1.6, the reverse current of the gate–drain Schottky junction dominates; the device is pinched-off and the drain current decreases as VGS is pushed toward negative values. This fact explains the increase in |IG|/IDS from 1.6 V to  2 V. For 1.6 VoVGS o  0.6 V, impact-ionization dominates, and the |IG|/IDS ratios in Ti and Pd gate samples are humping. In contrast, the Ir and Pt gate samples yield smooth curves. The results reveal that a gate with a high ΦB metal efficiently suppresses impact ionization. For VGS 4  0.6 V, the decrease in the gate–drain electric field together with the possible transfer of electrons to the InAlAs layer reduces the impact ionization rate. Fig. 8(a) plots the VGS position of Gm peak value against VDS from 0.5 to 3 V. As VDS increases, electron–hole pairs are generated by impact ionization that will increase Gmmax.

44

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

1.0 -14.8V

Ti

0.46V

-8.8V

Pd

0.55V

-14.8V

Pt

1V

-13.8V

-0.5

-1.0 -16

-14

-12

-10

-8

-6

-4

-2

0

2

VGS (V) Fig. 6. IG–VGS characteristics of various devices.

0.0

-0.5

-0.5

-1.0

-1.0

E

D

0.0

0.0

-0.5

-0.5

-1.0

-1.0

max

Ir Ti Pt Pd

Apperaance V

GS

0.0

C

B

0.0

Value (V)

VBR

0.7V

Gm

IG (mA/mm)

0.5

VON Ir

0

1

2

3

0

1

2

3

VDS ( V ) 100 80 60

Frequency (GHz)

40 20

maximum oscillation frequency, f max

0

Ir Ti Pt Pd

50 40 30 20 10

Fig. 7. |IG|/IDS ratio versus VDS voltages for various devices.

However, holes-induced threshold voltage shifts cause the position of the Gm peak to vibrate. Since it has a low work function, the Gm peak position of the Ti metal gate sample exhibits large variation with ranges of approximately 0.4 V, corresponding to VDS ¼3 V. In contrast, Ir, Pt, Pt samples have a well confined threshold voltage. The ranges are all around 0.15 V. To determine the effect of impact ionization on RF performance, the S-parameter of the 2  50 μm devices was measured by using an Agilent 8364C network analyzer from 0.1 to 50 GHz. Fig. 8(b) plots the measured fT and fmax as functions of VDS, where VGS is biased at the Gm peak position. Clearly, fT of the Ti metal gate sample suddenly drops from 30.5 to 26.6 when the applied VDS exceeds to 1.5 V. Generally, the following well-known equation is shown as follows [23]: fT ¼

gm0 C gs þC gd

where gm0 is the intrinsic transconductance, Cgs is the gateto-source capacitance, and Cgd is the gate-to-drain capacitance. Increasing Gm accompanies increasing fT and fmax.

0

current gain cut off frequency, f 0.5

1.0 1.5 2.0 Drain-to-Source Voltage, V

DS

2.5 (V)

T

3.0

Fig. 8. (a) Gmmax appearance VGS value and (b) fT and fmax versus VDS of various devices.

However, when impact ionization occurs, the gate current becomes large. The increase in Cgs is responsible for most of the degradation of fT [1]. However, fT and fmax of the Ir, Pt and Pd samples decrease slightly with rising VDS until VDS exceeds 2 V. These results reveal that a metal with a high work function such as Ir, Pt and Pd can efficiently suppresses impact ionization-induced problems detrimental effects on DC and RF characteristics. Low-temperature flicker noise measurements were performed using a Janis cryogenic temperature-probing system and an Agilent LFN system. The Janis probe station provides a vacuum pressure of 5  10  6 Torr. A lownoise pre-amplifier was adopted to amplify the device noise signal. The Agilent resistance unit automatically matched load resistance (RL) depending on the measured IDS. The DC characteristics of the device were determined using an HP4142B. An Agilent 4446A spectrum analyzer

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

was used to monitor the pre-amplified noise voltage. Measurements in the frequency domain and averages were taken by using the spectrum analyzer. The average times for LFN in the frequency ranges 10–100 Hz, 100 Hz– 1 kHz, 1–10 kHz and 10–100 kHz were 36, 72, 144 and 144, respectively. To accurately evaluate how impact ionization influences flicker noise, bias points at which the IG bellshaped behavior was the most serious were selected. Fig. 9 plots the bell-shaped curve of IG as a function of temperature under fixed VDS ¼3 V. Generally, at lower temperature, electron mobility is higher, so impact ionization should generate more electron–hole pairs. The high IG is presumably caused by the collection of holes when impact ionization in the device is strong. Obviously, the samples with an Ir or Pt gate, which have high work functions, have smaller IG values than those with a Pd or Ti gate. Measurements of the bell-shaped curves reveal the bias conditions of the flicker noise. Fig. 9 also depicts the Schottky leakage current and the characteristic hump associated with hole current owing to impact ionization. These humped curves appeared because the impact ionization requires a significant electron concentration and a high ionization coefficient. When the large negative VGS was applied, the channel contains few free electrons. When approaches zero volts, the carrier concentration is high, but the relatively lower electric field mitigates the impact ionization coefficient. In intermediate region of VGS biases, the carrier concentration and the ionization coefficient are significant to induce bell-shape curves. Electron– hole pairs are generated in the high field region between drain and gate terminals and a portion of the holes is collected by the negatively VGS. The gate current due to impact ionization by subtracting the Schottky current was observed which was a function of both VGS and VDS voltages. The extracted impact ionization current can be correlated with the experimental drain current once the ionization constant and the hole transmission probability are known [24]. Such a rise in gate leakage current is presumably due to hole collection as impact ionization becomes significant in the device. In addition, corresponds to Fig. 7, the kink-effect of IDS appears with ΔIGS indicating that impact ionization is significant with respect to the background generation rate [10]. Therefore, for the MHEMT device with high work function gate metal, the gate leakage current due to thermionic-field phenomena can be hence suppressed which is beneficial for reducing impact ionization phenomenon. In addition, device operating temperature was also an important parameter of ionization constant. Therefore, based on Fig. 9, the Ir-gate device achieved a lower impact ionization constant at various temperatures which is beneficial for device temperature stability. Fig. 10(a) plots the I–V characteristics of all samples at the gate bias points that were selected based on the measured IG. At 77 K, the increase in electron mobility increases IDS, and the kink effect is more serious than at 300 K. The fact that the IV curves yield approximately the same of around 50 mA/mm before VDS exceeding to 1 V for all samples at the chosen bias points is coincidence. Although the VGS values that for maximizing the IG values

45

of the samples vary, their IDS values are equal, so that impact ionization occurs only under specific bias conditions for a given epi-layer. Impact ionization occurs only

Fig. 9. Temperature dependent IG–VGS Bell shape measurement at VDS ¼3 V of various devices.

Fig. 10. (a) 77 K and 330 environment IDSS–VDS characteristics and (b) 77 K output conductance (G0) of various devices.

46

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

when the depth of the depletion region reaches threshold. In Fig. 10(a), the Ir gate sample exhibits a negligible increase in drain current with decreasing temperature and the best properties at such different temperatures as 77 K and 300 K. This result is direct and strong evidence that Ir metal is thermally stable. The output conductance Go displayed in Fig. 10(b) was determined from IDS at 77 K. The graphs display a characteristic that strongly links the kink with the impact ionization. The value of Gds was maximal at approximately VDS ¼1, 1.35, and 1.45 for Ti, Pd, and Pt metal gate samples, respectively. However, the Ir gate sample yielded not significant peak. These results suggest that the kink is a function of the gate-to-drain field, as would be expected for an impact ionizationrelated mechanism. As the temperature declines, the gate current and the multiplication ratio increase. The phonon scattering is reduced, which leads to further improving of the transport properties and increase in the energy of the electrons. This phenomenon increases IDS and the 1/f noise. Fig. 11(a) presents measurements of 1/f noise at 77 K for all samples. The frequency is 20 Hz; bias points were at VDS ¼ 0.5–3 V, and VGS was selected of which IG is maximal. The spectral density Sid of the noise current of

the Ir gate sample as VDS increases is stable in Fig. 11(a). Those of the Pt, Pd, and Ti gate samples are not. The 2 normalized Sid/IDS values are in good agreement to evalu2 ate the 1/f noise intensity. Fig. 11(b) shows Sid/IDS at 77 K and 300 K. Because the output conductance of the Ir gate sample is stable, the normalized Sid/I2 shows a negligible variation at each temperature. The Pd gate yields the worst results. The 1/f noise measurements demonstrate that the kink effect dominates the flicker noise at low temperature. Impact ionization not only severely affects device performance, in terms of Gm, VTH, IG, IDS, fT, and fmax, but also affects Sid. These experimental results suggest that the 1/f noise is also related to the physical properties of the gate metal, such as its thermal expansion coefficient (Ir ¼6.4, Pt ¼8.8, Ti¼8.6, Pd ¼11.8 mm/m K). This fact explains why, at 77 K and 300 K, although Pd has a higher work function than Ti, it exhibits poor 1/f noise. The same holds for Ir and Pt. 4. Conclusion In summary, the DC, RF, and 1/f noise of In0.4Al0.6As/ In0.4Ga0.6As MHEMTs with various Schottky contact metals were systematically evaluated. The relationship between the Schottky contact metals and impact ionization was clarified. All experimental results show that the Ir gate MHEMT has the most stable thermal properties under strong electrical field and or low temperature, which are associated with high gain, high current, and low frequency noise. Ir gate technology has potential for highly reliable microwave power device applications.

Acknowledgments This work is financially supported by the Ministry of Science and Technology, ROC (NSC-101-2221-E-182-043MY3) and facility supports of High Speed Intelligent Communication (HSIC) Research Center of Chang Gung University, Taoyuan, Taiwan. References

Fig. 11. (a) 1/f noise and (b) normalized noise current at 77 K and 300 K of various devices.

[1] C.Y. Chang, H.T. Hsu, E.Y. Chang, C.I. Kuo, S. Datta, M. Radosavljevic, Y. Miyamoto, G.W. Huang, IEEE Electron. Device Lett. 28 (2007) 856–858. [2] L.D. Nguyen, A.S. Brown, M.A. Thompson, L.M. Jelloian, IEEE Trans. Electron Devices 39 (1992) 2007–2014. [3] T. Suemitsu, H. Yokoyama, T. Ishii, T. Enoki, G. Meneghesso, E. Zanoni, IEEE Trans. Electron Devices 49 (2002) 1694–1700. [4] P.C. Chao, A. Tessmer, K.H.G. Duh, M.Y. Kao, P. Ho, P.M. Smith, J.M. Ballingall, S.M. Liu, A.A. Jabra, IEEE Electron Device Lett. 11 (1990) 59–62. [5] Y.C. Chen, D.L. Ingram, R. Lai, M. Barsky, R. Grunbacher, T. Block, H.C. Yen, D.C. Streit, IEEE Microw. Guided Wave Lett. 8 (1998) 399–401. [6] A. Mahajan, M. Arafa, P. Fey, C. Caneau, I. Adesida, IEEE Electron Device Lett. 18 (1997) 284–286. [7] J.B. Shealy, M. Matloubian, T.Y. Liu, W. Lam, C. Ngo, in: Proc. Int. Conf. InP Relat. Mater., 1997, pp. 20–23. [8] C.R. Bolognesi, in: Proc. 14th Int. Conf. IPRM, 2002, pp. 55–58. [9] G. Tuttle, H. Kroememr, IEEE Trans. Electron Devices 34 (1987) 2358. [10] M.H. Sornerville, J.A. del Alamo, W. Hoke, IEEE Electron Device Lett. 17 (1996) 473–475. [11] N. Harada, S. Kuroda, T. Katakami, K. Hikosaka, T. Mimura, M. Abe, in: Int. Conf. InP Relat. Mater., 1991, pp. 377–380.

H.-C. Chiu et al. / Materials Science in Semiconductor Processing 30 (2015) 41–47

[12] A. Mahajan, M. Arafa, P. Fay, C. Caneau, I. Adesida, IEEE Trans. Electron Devices 45 (1998) 338–340. [13] L.P. Sadwick, C.W. Kim, K.L. Tan, D.C. Streit, IEEE Electron Device Lett. 12 (1991) 626–628. [14] A. Fricke, G. Stareev, T. Kummetz, D. Sowada, J. Mahnss, W. Kowalsky, K.J. Ebeling, Appl. Phys. Lett. 65 (1994) 755. [15] S. Kim, I. Adesida, H. Hwang, Appl. Phys. Lett. 87 (2005) 232102. [16] M. Dammann, A. Leuther, R. Quay, M. Meng, H. Konstanzer, W. Jantz, M. Mikulla, Microelectron. Reliab. 44 (2004) 939–943. [17] K.J. Chen, T. Enoki, K. Maezawa, K. Arai, M. Yamamoto, IEEE Trans. Electron Devices 43 (1996) 252–257. [18] H.F. Chuang, C.P. Lee, C.M. Tsai, D.C. Liu, J.S. Tsang, J.C. Fan, J. Appl. Phys. 83 (1998) 366–371.

[19] [20] [21] [22] [23]

47

S. Kim, G. Cueva, I. Adesida, Electron. Lett. 41 (2005) 665–667. S. Kim, J.-H. Jang, I. Adesida, Electron. Lett. 41 (2005) 871–872. S. Kim, I. Adesida, Electron. Lett. 27 (2006) 873–876. L. Wang, W. Zhao, I. Adesida, Appl. Phys. Lett. 89 (2006) 211910. H. Ohta, T. Tanimoto, I. Ohbu, K. Higuchi, S. Takatani, N. Kurita, K. Kamozaki, H. Kondoh, Jpn. J. Appl. Phys. 37 (1998) 1373–1376. [24] R.T. Webster, S. Wu, A.F.M. Anwar, IEEE Electron Device Lett. 21 (2000) 193–195.