- Email: [email protected]
Reliability of III–V devices – The defects that cause the trouble
Microelectronic Engineering 90 (2012) 3–8
Contents lists available at ScienceDirect
Microelectronic Engineering journal homepage: www.elsevier.com/locate/mee
Reliability of III–V devices – The defects that cause the trouble Sokrates T. Pantelides a,b,c,⇑, Yevgeniy Puzyrev a, Xiao Shen a, Tania Roy b, Sandeepan DasGupta b, Blair R. Tuttle a, Daniel M. Fleetwood b,a, Ronald D. Schrimpf b a
Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37235, USA Department of Electrical Engineering and Computer Science, Vanderbilt University, Nashville, TN 37235, USA c Oak Ridge National Laboratory, Oak Ridge, TN 37235, USA b
a r t i c l e
i n f o
Article history: Available online 13 April 2011 Keywords: Degradation Hot electrons
a b s t r a c t Degradation of electronic devices by hot electrons is universally attributed to the generation of defects, but the mechanisms for defect generation and the specific nature of the pertinent defects are not known for most systems. Here we describe three recent case studies in III–V high-electron-mobility transistors that illustrate the power of combining density functional calculations and experimental data to identify the pertinent defects and associated degradation mechanisms. In all cases, benign pre-existing defects are either depassivated (irreversible degradation) or transformed to a metastable state (reversible degradation). Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction Degradation of electronic devices under operating conditions remains a major issue that compromises the introduction of new technologies in niche applications, e.g., power devices, high-frequency response, etc. Hot carriers have long been known to cause degradation by generating defects that lead to the trapping of carriers and subsequent decrease in mobilities by Coulomb scattering. The atomic-scale processes of defect generation by hot carriers, however, are not generally addressed. It should be noted that hot carriers cannot in general create defects in an otherwise perfect crystal structure. For example, hot carriers have neither the energy nor momentum that is needed to create a Frenkel pair. On the other hand, they can provide sufficient energy to depassivate existing defects. A well-known example is the removal of H from passivated dangling bonds at the Si–SiO2 interface, a process that was studied extensively since the 1980’s [1–3]. In this case, hot carriers provide the energy to overcome the energy barrier for H release. Another possibility is the transformation of a pre-existing benign defect to a metastable configuration with detrimental consequences, e.g., capture of a carrier by a neutral defect into a long-lived metastable charged configuration. In this case, the hot carriers provide the energy to overcome the energy barrier for the transformation. If the reverse process is somehow impeded, the resulting degradation may be long-lived or reversible. In this paper we will summarize three recent case studies where a combination of first-principles quantum mechanical ⇑ Corresponding author at: Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37235, USA. E-mail address: [email protected] (S.T. Pantelides). 0167-9317/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.mee.2011.04.019
(density functional theory) calculations of defect properties and experimental data led to the identification of specific defects and processes that cause the degradation of III–V high-electron-mobility transistors (HEMTs) [4–6]. Calculations were used to obtain defect formation energies and defect electronic energy levels. Only point defects were considered, focusing on candidates that are likely to form during the different growth conditions of interest, plus impurities that are known to be present. In all cases, specific point defects and associated processes were identified that account for the observations, while the remaining candidates are benign under the conditions of interest. Dislocations or other extended defects were not considered because they are generically present in all cases.
2. Case study I. GaN HEMTs [4] Meneghesso et al. [7] reported degradation of GaN/AlGaN HEMTs and demonstrated that hot electrons must be responsible because degradation is worst under semi-on conditions (high electric field and moderate carrier concentration) when the concentration of hot electrons is maximum (Fig. 1). They attributed the degradation to defect generation, but no specific defects were identified. A first clue on the nature of the pertinent defect is provided by luminescence data reported later under the same operating conditions by Lin et al. [8]. These authors found that the well-known yellow luminescence increases substantially, while blue-green luminescence increases more modestly. Since the yellow luminescence has been attributed by several authors to Ga vacancies [9,10], a likely mechanism is the dehydrogenation of pre-existing hydrogenated Ga vacancies.
4
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 1. Average percent decrease of the maximum value of transconductance measured at VDS = 10 V, for ‘‘G’’ devices during 10-h ‘‘ON-state’’ tests (VDS = 20 V, VGS = 0 V, diamonds) ‘‘OFF-state’’ tests (VDS = 20 V, VGS = 7.7 V, squared symbols), ‘‘semi-ON-state’’ tests (VDS = 20 V, VGS = 5.5 V, triangles).
Equilibrium concentrations of defects are determined by their Gibbs free energies of formation. Entropy contributions are typically small, and one examines formation energies obtained from an electronic structure calculation that are in fact enthalpies. The formation energies of Ga vacancies in Ga-rich and N-rich conditions and hydrogenated Ga vacancies are shown in Fig. 2 as a function of the position of the Fermi energy in the band gap. Formation energies also depend on the chemical potentials of the elements that are traded between the crystal and a hypothetical reservoir in order to establish an equilibrium concentration. The sum of the chemical potentials of Ga and N is fixed by the stoichiometry of the perfect crystal, whereby the only chemical potentials that are fixed by the growth conditions are those of H and either Ga or N. N-rich conditions are defined to be when the N chemical potential is fixed by a reservoir of N2 molecules, and Ga-rich conditions are defined to correspond to a Ga chemical potential defined by the energy per atom in a Ga metal. The hydrogen chemical potential is the energy per H atom in a H2 molecule. The key result in Fig. 2 is that, during high-temperature growth under equilibrium conditions, when the Fermi energy is in the midgap region, neutral triply hydrogenated Ga vacancies, which are electrically benign, are the dominant defect under either N-rich or Ga-rich growth conditions. The formation energies of Fig. 2 show that hot carriers can provide energy to release one, two, or three hydrogen atoms, leaving behind negatively-charged defects. Conductance degradation would then arise from Coulomb scattering. The energy-level diagram of Fig. 3 accounts for the observed
Fig. 3. Plot of Kohn–Sham eigenvalues for the hydrogenated gallium vacancy, where arrows show optical transitions corresponding to the yellow and blue luminescence lines (yellow and blue circles) (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
Fig. 4. Threshold voltage shift after electrical stress in several MBE-grown devices. Ellipses show the growth conditions of the measured devices.
increase in yellow luminescence and the lesser increase in green–blue luminescence.
Fig. 2. Hydrogenated gallium vacancy formation energies suggest that the defects have higher concentrations in N-rich than in Ga-rich environments.
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
5
Fig. 6. Ga vacancy dehydrogenation leads to additional acceptors and negative threshold shifts. Triply hydrogenated Ga vacancy is in the neutral state, denoted as ‘‘0’’, and the dehydrogenated defect has the charge state ‘‘3’’. Blue spheres are N, green spheres are Ga and small red spheres are H. The small arrows indicate crystal directions (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
Fig. 5. Threshold voltage shift after electrical stress in MOCVD-grown devices. Each curve represents a different electrically-stressed device. A negative voltage shift is observed in all the devices.
3. Case study 2 – GaN/AlGaN HEMTs fabricated under different growth conditions [5] Electrical stress measurements were performed on devices fabricated by molecular beam epitaxy (MBE) under Ga-rich, N-rich, and ammonia-rich conditions and by metal-organic chemical vapor deposition (MOCVD) using ammonia. Under electrical stress at a bias determined to produce the same power dissipation in all devices, the Ga-rich and N-rich MBE devices exhibit a positive threshold voltage shift, indicating a net negative change in charge density, i.e., increase of donors or decrease of acceptors; the ammonia-grown devices exhibit a negative threshold voltage shift, indicating a net positive change in charge density, i.e., decrease of acceptor concentration or increase of donor concentration (Figs. 4– 7) (the dependence of the threshold voltage shift on device parameters and the change in charge density in the device are described in Ref. [11]). The opposite threshold voltage shifts that are observed in the two kinds of devices, namely Ga-rich and N-rich vs. ammonia-rich growth conditions, suggest that a different defect is at play in each
Fig. 7. N antisite dehydrogenation leads to removal of acceptors and positive threshold shifts. Negatively charged N antisites, denoted as ‘‘1’’, become neutral, shown as ‘‘0’’, after removal of a hydrogen. Color code of spheres as in Fig. 6 (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
case. Following up on the results of case study I, we find that dehydrogenation of hydrogenated Ga vacancies accounts for the data obtained in devices fabricated in Ga-rich and N-rich conditions in the absence of ammonia: dehydrogenation leads from neutral triply hydrogenated Ga vacancies to negatively charged doublyhydrogenated, singly-hydrogenated, and bare Ga vacancies, i.e., to a net increase in the negative charge, in accord with the data. For ammonia-rich growth environments, entire ammonia molecules may take the place of Ga atoms, yielding a non-equilibrium concentration of N antisite defects, NGa. As shown in Fig. 8, triply hydrogenated N antisites are negatively charged in n-type GaN. Energy from hot electrons can then release a hydrogen. The resulting
Fig. 8. Hydrogenated N antisite formation energies in Ga-rich and N-rich conditions. The color bars highlight the pertinent charge states under zero bias and under stress (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
6
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 9. (a) Structure of an InAs/AlSb HEMT (left) and its vertical band diagram under electrical stress conditions (right). (b) Transconductance vs gate voltage of an InAs/AlSb HEMT before stress, after 5 h of electrical stress, after 8 h of room temperature anneal, and after two days of room temperature anneal.
doubly-hydrogenated N antisite then loses an electron to the conduction band and becomes neutral. The net result is the conversion of negatively charged defects to neutral defects, which accounts for the observed negative threshold voltage shift. We propose, therefore, that in ammonia-rich growth a non-equilibrium concentration of hydrogenated N antisites dominates over hydrogenated Ga vacancies and is responsible for the observed degradation. 4. Case study 3. Recoverable degradation of InAs/AlSb HEMTs [6] Electrical stress measurements were performed on InAs/AlSb HEMTs [12], shown schematically in Fig. 9a. The data are shown in Fig. 9b, indicating a negative threshold voltage shift, i.e., stress-induced generation of positive charge. Unlike the previous cases of GaN/AlGaN HEMTs, however, the degradation of InAs/AlSb HEMTs is recoverable, which means that the underlying process is not release of hydrogen from hydrogenated defects. Instead, in order to explain the data we must identify a pre-existing defect with a metastable configuration. More specifically, since the electron mobility in the InAs channel does not degrade, we must look for defects in the top AlSb layer. Furthermore, since the applied stress results in holes being injected from the InAs channel into the top AlSb layer (the holes are generated by impact ionization of hot electrons in the channel), we must look for a defect that undergoes a transformation to a metastable configuration upon capture of one or more holes. The obvious first suspects are native defects, e.g., vacancies, antisites, and self-interstitials. The lowest-energy such defects had already been identified in the literature as the aluminum interstitial and the SbAl antisite defect (Sb sitting at the Al site) [13,14]. The former can be ruled out because its calculated migration energy is only 1.3 eV, so it can easily anneal out during growth. The
Fig. 10. (Left) Formation energies of SbAl. The shaded region illustrates the estimated range of the Fermi level during stress. (Right) Structures of SbAl with Td symmetry, [SbAl (Td)], and SbAl with C3V symmetry, [SbAl (C3V)].
SbAl antisite defect can be ruled out for a different reason. As shown in Fig. 10 and as discussed further below, for the range of Fermi levels of interest here, the SbAl antisite defect is positively charged. Though a metastable configuration exists, it can be reached only if the defect captures an electron to become neutral. The other possibility is impurities, either contaminants or dopants, in this case Te. Tellurium is an n-type dopant and exists in AlSb as TeSb (a Te impurity occupying a Sb site). It can be ruled out as the responsible defect because it is positively charged at the Fermi energies of interest (as would be expected for the n-type dopant of the adjacent InAs layer), and its metastable configurations can be reached only by capturing electrons (Fig. 11, left). The known contaminants are carbon and oxygen. Substitutional carbon CSb (a carbon impurity occupying a Sb site) can be ruled
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
7
Fig. 11. (Left) Formation energies of stable configurations of TeSb. Td stands for a structure with Td symmetry. BB-DX stands for a DX structure that has a broken-bond. a/bCCBDX stands for two DX structures with cation–cation bond. (Right) Formation energies of stable configurations of CSb.
Fig. 12. (Left) Formation energies of OSb; the arrow shows changes in defect charge states and configuration that are consistent with the observed hot-carrier degradation. C3V stands for a structure with C3V symmetry. OBBDX stands for a DX structure that has orthorhombic symmetry and a broken-bond. (Right) One structure of Oþ Sb and one structure of OSb .
Fig. 13. (Left) Formation energies of Oi; the arrow shows changes in defect charge states and configuration that are consistent with the observed hot-carrier degradation. Td and C3V stand for structures with Td and C3V symmetry, respectively. The symbol bb stands for a bond bridge structure, which means the interstitial is situated as a bridge between a cation and an anion. (Sb–O)sp stands for a split interstitial structure at which the O and Sb form a pair and share a Sb site. (Right) A structure of O0i and a structure of O2 i .
out because, though it is initially negatively charged, it requires electron capture to undergo a transformation (Fig. 11, right). Finally, we find that oxygen works either as a substitutional impurity OSb (oxygen occupying an Sb site) or an interstitial
þ Fig. 14. (Left) Energy diagram for the transition from O Sb to OSb . The approximate position of the Fermi level when device is under zero gate bias is from device simulation. (Right) Energy diagram for the transition from O2 to O0i . i
impurity. The pertinent formation energies of the various configurations of both substitutional and interstitial oxygen are shown in Figs. 12 and 13. In the case of OSb, the initial configuration is negatively charged. Upon capture of two holes, not necessarily simultaneously, the defect undergoes a transformation to a different configuration that is now positively charged and can, therefore, account for the observed negative threshold voltage shift. Similarly, interstitial oxygen starts out negatively charged and converts to a neutral but different configuration upon capture of a hole. Again, the observed negative threshold voltage shift is explained. The final task is to explain the recovery that is observed at room temperature. Substitutional oxygen is an example of the well-known DX centers that have been studied extensively in the context of donors in bulk GaAs and other III–V semiconductors. DX centers typically exhibit an energy barrier for the capture of an electron in the metastable configuration, which would take the defect back to its initial configuration. In the case of the HEMT of interest here, an altogether different mechanism is delaying the return to the original configuration: the Fermi energy in the AlSb layer is controlled by the InAs layer and is generally in the lower part of the AlSb energy gap (Fig. 14). In the initial configuration, the energy level of O Sb is below the Fermi energy and contains two electrons. Upon capture of two holes, the empty level rises far above the Fermi energy by lattice relaxations. The lifetime of the metastable configuration, therefore, is controlled by the scarcity of electrons that can be captured: the most likely source of electrons is the InAs channel, from which electrons must tunnel into the empty defect level. An estimate of the lifetime of the metastable configurations was given in Ref. [6], in accord with the observed time for recovery. Interstitial oxygen behaves in a similar way.
8
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 15. Transconductance vs. Vgs plots for devices before and after stress, and after 6 h of annealing at zero drain bias, for (a) Vgs = 1 V and (b) Vgs = 0.7 V. (c) Fractional recovery under four different gate biases, with zero drain bias in all cases.
A test of the above mechanism for degradation and recovery was performed as follows. By applying a larger and larger negative gate voltage, the InAs bands are bent up near the InAs/AlSb interface, pushing electrons away from the interface, which should extend the lifetime of the metastable configuration. The data of Fig. 15 indeed show a slower recovery with increased negative bias. 5. Summary We have shown that the defects responsible for hot-electroninduced degradation in HEMTs can be successfully identified by combining results of density functional calculations of formation energies and energy levels of candidate defects with experimental data. Permanent degradation can occur if hot electrons remove hydrogen atoms from pre-existing hydrogenated defects, while recoverable degradation arises from the transformation of preexisting defects into metastable configurations. Acknowledgments The authors would like to thank D.F. Brown, J. Speck, and U. Mishra for providing the GaN/AlGaN devices and J. Bergman and
B. Brar for providing the InAs/AlSb devices. The work reported here was supported in part by ONR MURI grant No. N-00014-08-100655 and by the McMinn Endowment at Vanderbilt University. References [1] D.J. DiMaria, J.W. Stasiak, J. Appl. Phys. 65 (1989) 2342. [2] D.J. DiMaria, J. Appl. Phys. 87 (2000) 8707. [3] S.N. Rashkeev, D.M. Fleetwood, R.D. Schrimpf, S.T. Pantelides, Phys. Rev. Lett. 87 (2001) 165506. [4] Y.S. Puzyrev, B.R. Tuttle, R.D. Schrimpf, D.M. Fleetwood, S.T. Pantelides, Appl. Phys. Lett. 96 (2010) 053505. [5] T. Roy, Y.S. Puzyrev, B.R. Tuttle, D.M. Fleetwood, R.D. Schrimpf, D.F. Brown, U.K. Mishra, S.T. Pantelides, Appl. Phys. Lett. 96 (2010) 133503. [6] X. Shen, S. DasGupta, R.A. Reed, R.D. Schrimpf, D.M. Fleetwood, S.T. Pantelides, J. Appl. Phys. 108 (2010) 114505. [7] G.V.G. Meneghesso, F. Danesin, F. Rampazzo, A. Tazzoli, M. Meneghini, E. Zanoni, IEEE Trans. Dev. Mater. Reliab. 8 (2008) 332. [8] C.-H. Lin, T.A. Merz, D.R. Doutt, M.J. Hetzer, J. Joh, J.A. Del Alamo, U.K. Mishra, L.J. Brillson, Appl. Phys. Lett. 95 (2009) 033510. [9] C.G. Van de Walle, Phys. Rev. B 56 (1997) R10 020. [10] C.G. Van de Walle, J. Neugebauer, J. Appl. Phys. 95 (2004) 3851. [11] D. Delagebeaudeuf, N.T. Linh, IEEE Trans. Electron Devices 28 (1981) 790. [12] S. DasGupta, R.A. Reed, R.D. Schrimpf, D.M. Fleetwood, X. Shen, S.T. Pantelides, J. Bergman, B. Brar, Proc. IEEE IRPS 813 (2010). [13] D. Aberg, E. Paul, A.J. Williamson, V. Lordi, Phys. Rev. B 77 (2008) 165206. [14] M.-H. Du, Phys. Rev. B 79 (2009) 045207.
Contents lists available at ScienceDirect
Microelectronic Engineering journal homepage: www.elsevier.com/locate/mee
Reliability of III–V devices – The defects that cause the trouble Sokrates T. Pantelides a,b,c,⇑, Yevgeniy Puzyrev a, Xiao Shen a, Tania Roy b, Sandeepan DasGupta b, Blair R. Tuttle a, Daniel M. Fleetwood b,a, Ronald D. Schrimpf b a
Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37235, USA Department of Electrical Engineering and Computer Science, Vanderbilt University, Nashville, TN 37235, USA c Oak Ridge National Laboratory, Oak Ridge, TN 37235, USA b
a r t i c l e
i n f o
Article history: Available online 13 April 2011 Keywords: Degradation Hot electrons
a b s t r a c t Degradation of electronic devices by hot electrons is universally attributed to the generation of defects, but the mechanisms for defect generation and the specific nature of the pertinent defects are not known for most systems. Here we describe three recent case studies in III–V high-electron-mobility transistors that illustrate the power of combining density functional calculations and experimental data to identify the pertinent defects and associated degradation mechanisms. In all cases, benign pre-existing defects are either depassivated (irreversible degradation) or transformed to a metastable state (reversible degradation). Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction Degradation of electronic devices under operating conditions remains a major issue that compromises the introduction of new technologies in niche applications, e.g., power devices, high-frequency response, etc. Hot carriers have long been known to cause degradation by generating defects that lead to the trapping of carriers and subsequent decrease in mobilities by Coulomb scattering. The atomic-scale processes of defect generation by hot carriers, however, are not generally addressed. It should be noted that hot carriers cannot in general create defects in an otherwise perfect crystal structure. For example, hot carriers have neither the energy nor momentum that is needed to create a Frenkel pair. On the other hand, they can provide sufficient energy to depassivate existing defects. A well-known example is the removal of H from passivated dangling bonds at the Si–SiO2 interface, a process that was studied extensively since the 1980’s [1–3]. In this case, hot carriers provide the energy to overcome the energy barrier for H release. Another possibility is the transformation of a pre-existing benign defect to a metastable configuration with detrimental consequences, e.g., capture of a carrier by a neutral defect into a long-lived metastable charged configuration. In this case, the hot carriers provide the energy to overcome the energy barrier for the transformation. If the reverse process is somehow impeded, the resulting degradation may be long-lived or reversible. In this paper we will summarize three recent case studies where a combination of first-principles quantum mechanical ⇑ Corresponding author at: Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37235, USA. E-mail address: [email protected] (S.T. Pantelides). 0167-9317/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.mee.2011.04.019
(density functional theory) calculations of defect properties and experimental data led to the identification of specific defects and processes that cause the degradation of III–V high-electron-mobility transistors (HEMTs) [4–6]. Calculations were used to obtain defect formation energies and defect electronic energy levels. Only point defects were considered, focusing on candidates that are likely to form during the different growth conditions of interest, plus impurities that are known to be present. In all cases, specific point defects and associated processes were identified that account for the observations, while the remaining candidates are benign under the conditions of interest. Dislocations or other extended defects were not considered because they are generically present in all cases.
2. Case study I. GaN HEMTs [4] Meneghesso et al. [7] reported degradation of GaN/AlGaN HEMTs and demonstrated that hot electrons must be responsible because degradation is worst under semi-on conditions (high electric field and moderate carrier concentration) when the concentration of hot electrons is maximum (Fig. 1). They attributed the degradation to defect generation, but no specific defects were identified. A first clue on the nature of the pertinent defect is provided by luminescence data reported later under the same operating conditions by Lin et al. [8]. These authors found that the well-known yellow luminescence increases substantially, while blue-green luminescence increases more modestly. Since the yellow luminescence has been attributed by several authors to Ga vacancies [9,10], a likely mechanism is the dehydrogenation of pre-existing hydrogenated Ga vacancies.
4
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 1. Average percent decrease of the maximum value of transconductance measured at VDS = 10 V, for ‘‘G’’ devices during 10-h ‘‘ON-state’’ tests (VDS = 20 V, VGS = 0 V, diamonds) ‘‘OFF-state’’ tests (VDS = 20 V, VGS = 7.7 V, squared symbols), ‘‘semi-ON-state’’ tests (VDS = 20 V, VGS = 5.5 V, triangles).
Equilibrium concentrations of defects are determined by their Gibbs free energies of formation. Entropy contributions are typically small, and one examines formation energies obtained from an electronic structure calculation that are in fact enthalpies. The formation energies of Ga vacancies in Ga-rich and N-rich conditions and hydrogenated Ga vacancies are shown in Fig. 2 as a function of the position of the Fermi energy in the band gap. Formation energies also depend on the chemical potentials of the elements that are traded between the crystal and a hypothetical reservoir in order to establish an equilibrium concentration. The sum of the chemical potentials of Ga and N is fixed by the stoichiometry of the perfect crystal, whereby the only chemical potentials that are fixed by the growth conditions are those of H and either Ga or N. N-rich conditions are defined to be when the N chemical potential is fixed by a reservoir of N2 molecules, and Ga-rich conditions are defined to correspond to a Ga chemical potential defined by the energy per atom in a Ga metal. The hydrogen chemical potential is the energy per H atom in a H2 molecule. The key result in Fig. 2 is that, during high-temperature growth under equilibrium conditions, when the Fermi energy is in the midgap region, neutral triply hydrogenated Ga vacancies, which are electrically benign, are the dominant defect under either N-rich or Ga-rich growth conditions. The formation energies of Fig. 2 show that hot carriers can provide energy to release one, two, or three hydrogen atoms, leaving behind negatively-charged defects. Conductance degradation would then arise from Coulomb scattering. The energy-level diagram of Fig. 3 accounts for the observed
Fig. 3. Plot of Kohn–Sham eigenvalues for the hydrogenated gallium vacancy, where arrows show optical transitions corresponding to the yellow and blue luminescence lines (yellow and blue circles) (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
Fig. 4. Threshold voltage shift after electrical stress in several MBE-grown devices. Ellipses show the growth conditions of the measured devices.
increase in yellow luminescence and the lesser increase in green–blue luminescence.
Fig. 2. Hydrogenated gallium vacancy formation energies suggest that the defects have higher concentrations in N-rich than in Ga-rich environments.
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
5
Fig. 6. Ga vacancy dehydrogenation leads to additional acceptors and negative threshold shifts. Triply hydrogenated Ga vacancy is in the neutral state, denoted as ‘‘0’’, and the dehydrogenated defect has the charge state ‘‘3’’. Blue spheres are N, green spheres are Ga and small red spheres are H. The small arrows indicate crystal directions (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
Fig. 5. Threshold voltage shift after electrical stress in MOCVD-grown devices. Each curve represents a different electrically-stressed device. A negative voltage shift is observed in all the devices.
3. Case study 2 – GaN/AlGaN HEMTs fabricated under different growth conditions [5] Electrical stress measurements were performed on devices fabricated by molecular beam epitaxy (MBE) under Ga-rich, N-rich, and ammonia-rich conditions and by metal-organic chemical vapor deposition (MOCVD) using ammonia. Under electrical stress at a bias determined to produce the same power dissipation in all devices, the Ga-rich and N-rich MBE devices exhibit a positive threshold voltage shift, indicating a net negative change in charge density, i.e., increase of donors or decrease of acceptors; the ammonia-grown devices exhibit a negative threshold voltage shift, indicating a net positive change in charge density, i.e., decrease of acceptor concentration or increase of donor concentration (Figs. 4– 7) (the dependence of the threshold voltage shift on device parameters and the change in charge density in the device are described in Ref. [11]). The opposite threshold voltage shifts that are observed in the two kinds of devices, namely Ga-rich and N-rich vs. ammonia-rich growth conditions, suggest that a different defect is at play in each
Fig. 7. N antisite dehydrogenation leads to removal of acceptors and positive threshold shifts. Negatively charged N antisites, denoted as ‘‘1’’, become neutral, shown as ‘‘0’’, after removal of a hydrogen. Color code of spheres as in Fig. 6 (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
case. Following up on the results of case study I, we find that dehydrogenation of hydrogenated Ga vacancies accounts for the data obtained in devices fabricated in Ga-rich and N-rich conditions in the absence of ammonia: dehydrogenation leads from neutral triply hydrogenated Ga vacancies to negatively charged doublyhydrogenated, singly-hydrogenated, and bare Ga vacancies, i.e., to a net increase in the negative charge, in accord with the data. For ammonia-rich growth environments, entire ammonia molecules may take the place of Ga atoms, yielding a non-equilibrium concentration of N antisite defects, NGa. As shown in Fig. 8, triply hydrogenated N antisites are negatively charged in n-type GaN. Energy from hot electrons can then release a hydrogen. The resulting
Fig. 8. Hydrogenated N antisite formation energies in Ga-rich and N-rich conditions. The color bars highlight the pertinent charge states under zero bias and under stress (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.).
6
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 9. (a) Structure of an InAs/AlSb HEMT (left) and its vertical band diagram under electrical stress conditions (right). (b) Transconductance vs gate voltage of an InAs/AlSb HEMT before stress, after 5 h of electrical stress, after 8 h of room temperature anneal, and after two days of room temperature anneal.
doubly-hydrogenated N antisite then loses an electron to the conduction band and becomes neutral. The net result is the conversion of negatively charged defects to neutral defects, which accounts for the observed negative threshold voltage shift. We propose, therefore, that in ammonia-rich growth a non-equilibrium concentration of hydrogenated N antisites dominates over hydrogenated Ga vacancies and is responsible for the observed degradation. 4. Case study 3. Recoverable degradation of InAs/AlSb HEMTs [6] Electrical stress measurements were performed on InAs/AlSb HEMTs [12], shown schematically in Fig. 9a. The data are shown in Fig. 9b, indicating a negative threshold voltage shift, i.e., stress-induced generation of positive charge. Unlike the previous cases of GaN/AlGaN HEMTs, however, the degradation of InAs/AlSb HEMTs is recoverable, which means that the underlying process is not release of hydrogen from hydrogenated defects. Instead, in order to explain the data we must identify a pre-existing defect with a metastable configuration. More specifically, since the electron mobility in the InAs channel does not degrade, we must look for defects in the top AlSb layer. Furthermore, since the applied stress results in holes being injected from the InAs channel into the top AlSb layer (the holes are generated by impact ionization of hot electrons in the channel), we must look for a defect that undergoes a transformation to a metastable configuration upon capture of one or more holes. The obvious first suspects are native defects, e.g., vacancies, antisites, and self-interstitials. The lowest-energy such defects had already been identified in the literature as the aluminum interstitial and the SbAl antisite defect (Sb sitting at the Al site) [13,14]. The former can be ruled out because its calculated migration energy is only 1.3 eV, so it can easily anneal out during growth. The
Fig. 10. (Left) Formation energies of SbAl. The shaded region illustrates the estimated range of the Fermi level during stress. (Right) Structures of SbAl with Td symmetry, [SbAl (Td)], and SbAl with C3V symmetry, [SbAl (C3V)].
SbAl antisite defect can be ruled out for a different reason. As shown in Fig. 10 and as discussed further below, for the range of Fermi levels of interest here, the SbAl antisite defect is positively charged. Though a metastable configuration exists, it can be reached only if the defect captures an electron to become neutral. The other possibility is impurities, either contaminants or dopants, in this case Te. Tellurium is an n-type dopant and exists in AlSb as TeSb (a Te impurity occupying a Sb site). It can be ruled out as the responsible defect because it is positively charged at the Fermi energies of interest (as would be expected for the n-type dopant of the adjacent InAs layer), and its metastable configurations can be reached only by capturing electrons (Fig. 11, left). The known contaminants are carbon and oxygen. Substitutional carbon CSb (a carbon impurity occupying a Sb site) can be ruled
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
7
Fig. 11. (Left) Formation energies of stable configurations of TeSb. Td stands for a structure with Td symmetry. BB-DX stands for a DX structure that has a broken-bond. a/bCCBDX stands for two DX structures with cation–cation bond. (Right) Formation energies of stable configurations of CSb.
Fig. 12. (Left) Formation energies of OSb; the arrow shows changes in defect charge states and configuration that are consistent with the observed hot-carrier degradation. C3V stands for a structure with C3V symmetry. OBBDX stands for a DX structure that has orthorhombic symmetry and a broken-bond. (Right) One structure of Oþ Sb and one structure of OSb .
Fig. 13. (Left) Formation energies of Oi; the arrow shows changes in defect charge states and configuration that are consistent with the observed hot-carrier degradation. Td and C3V stand for structures with Td and C3V symmetry, respectively. The symbol bb stands for a bond bridge structure, which means the interstitial is situated as a bridge between a cation and an anion. (Sb–O)sp stands for a split interstitial structure at which the O and Sb form a pair and share a Sb site. (Right) A structure of O0i and a structure of O2 i .
out because, though it is initially negatively charged, it requires electron capture to undergo a transformation (Fig. 11, right). Finally, we find that oxygen works either as a substitutional impurity OSb (oxygen occupying an Sb site) or an interstitial
þ Fig. 14. (Left) Energy diagram for the transition from O Sb to OSb . The approximate position of the Fermi level when device is under zero gate bias is from device simulation. (Right) Energy diagram for the transition from O2 to O0i . i
impurity. The pertinent formation energies of the various configurations of both substitutional and interstitial oxygen are shown in Figs. 12 and 13. In the case of OSb, the initial configuration is negatively charged. Upon capture of two holes, not necessarily simultaneously, the defect undergoes a transformation to a different configuration that is now positively charged and can, therefore, account for the observed negative threshold voltage shift. Similarly, interstitial oxygen starts out negatively charged and converts to a neutral but different configuration upon capture of a hole. Again, the observed negative threshold voltage shift is explained. The final task is to explain the recovery that is observed at room temperature. Substitutional oxygen is an example of the well-known DX centers that have been studied extensively in the context of donors in bulk GaAs and other III–V semiconductors. DX centers typically exhibit an energy barrier for the capture of an electron in the metastable configuration, which would take the defect back to its initial configuration. In the case of the HEMT of interest here, an altogether different mechanism is delaying the return to the original configuration: the Fermi energy in the AlSb layer is controlled by the InAs layer and is generally in the lower part of the AlSb energy gap (Fig. 14). In the initial configuration, the energy level of O Sb is below the Fermi energy and contains two electrons. Upon capture of two holes, the empty level rises far above the Fermi energy by lattice relaxations. The lifetime of the metastable configuration, therefore, is controlled by the scarcity of electrons that can be captured: the most likely source of electrons is the InAs channel, from which electrons must tunnel into the empty defect level. An estimate of the lifetime of the metastable configurations was given in Ref. [6], in accord with the observed time for recovery. Interstitial oxygen behaves in a similar way.
8
S.T. Pantelides et al. / Microelectronic Engineering 90 (2012) 3–8
Fig. 15. Transconductance vs. Vgs plots for devices before and after stress, and after 6 h of annealing at zero drain bias, for (a) Vgs = 1 V and (b) Vgs = 0.7 V. (c) Fractional recovery under four different gate biases, with zero drain bias in all cases.
A test of the above mechanism for degradation and recovery was performed as follows. By applying a larger and larger negative gate voltage, the InAs bands are bent up near the InAs/AlSb interface, pushing electrons away from the interface, which should extend the lifetime of the metastable configuration. The data of Fig. 15 indeed show a slower recovery with increased negative bias. 5. Summary We have shown that the defects responsible for hot-electroninduced degradation in HEMTs can be successfully identified by combining results of density functional calculations of formation energies and energy levels of candidate defects with experimental data. Permanent degradation can occur if hot electrons remove hydrogen atoms from pre-existing hydrogenated defects, while recoverable degradation arises from the transformation of preexisting defects into metastable configurations. Acknowledgments The authors would like to thank D.F. Brown, J. Speck, and U. Mishra for providing the GaN/AlGaN devices and J. Bergman and
B. Brar for providing the InAs/AlSb devices. The work reported here was supported in part by ONR MURI grant No. N-00014-08-100655 and by the McMinn Endowment at Vanderbilt University. References [1] D.J. DiMaria, J.W. Stasiak, J. Appl. Phys. 65 (1989) 2342. [2] D.J. DiMaria, J. Appl. Phys. 87 (2000) 8707. [3] S.N. Rashkeev, D.M. Fleetwood, R.D. Schrimpf, S.T. Pantelides, Phys. Rev. Lett. 87 (2001) 165506. [4] Y.S. Puzyrev, B.R. Tuttle, R.D. Schrimpf, D.M. Fleetwood, S.T. Pantelides, Appl. Phys. Lett. 96 (2010) 053505. [5] T. Roy, Y.S. Puzyrev, B.R. Tuttle, D.M. Fleetwood, R.D. Schrimpf, D.F. Brown, U.K. Mishra, S.T. Pantelides, Appl. Phys. Lett. 96 (2010) 133503. [6] X. Shen, S. DasGupta, R.A. Reed, R.D. Schrimpf, D.M. Fleetwood, S.T. Pantelides, J. Appl. Phys. 108 (2010) 114505. [7] G.V.G. Meneghesso, F. Danesin, F. Rampazzo, A. Tazzoli, M. Meneghini, E. Zanoni, IEEE Trans. Dev. Mater. Reliab. 8 (2008) 332. [8] C.-H. Lin, T.A. Merz, D.R. Doutt, M.J. Hetzer, J. Joh, J.A. Del Alamo, U.K. Mishra, L.J. Brillson, Appl. Phys. Lett. 95 (2009) 033510. [9] C.G. Van de Walle, Phys. Rev. B 56 (1997) R10 020. [10] C.G. Van de Walle, J. Neugebauer, J. Appl. Phys. 95 (2004) 3851. [11] D. Delagebeaudeuf, N.T. Linh, IEEE Trans. Electron Devices 28 (1981) 790. [12] S. DasGupta, R.A. Reed, R.D. Schrimpf, D.M. Fleetwood, X. Shen, S.T. Pantelides, J. Bergman, B. Brar, Proc. IEEE IRPS 813 (2010). [13] D. Aberg, E. Paul, A.J. Williamson, V. Lordi, Phys. Rev. B 77 (2008) 165206. [14] M.-H. Du, Phys. Rev. B 79 (2009) 045207.