Modeling gallium arsenide heterojunction bipolar transistor ledge variations for insight into device reliability

https://doi.org/10.1016/S0026-2714(02)00065-3Get rights and content

Abstract

It is widely known that under normal bias conditions, GaAs heterojunction bipolar transistor (HBT) device degradation proceeds by a gradual buildup of defects in the base and base–emitter junction depletion regions. The buildup of these defects is associated with a solid-state phenomenon known as recombination enhanced defect reaction, which is the formation and migration of defects associated with nonradiative electron–hole recombination events. These defects are often associated with midgap traps, which serve as additional recombination centers for electron–hole pairs. The resulting increased recombination current is an additional base leakage current, which reduces current gain. By extension, a high electron–hole recombination density in a region with an initially high defect density––such as an unpassivated or poorly passivated base surface––will lead to quick device degradation.

This paper reports the modeling of the effects of various different extrinsic base passivation ledge parameters––material composition, thickness, width, and spacing from ledge to base contact––to determine the microscopic effects these parameters have on electron–hole recombination density. Through this we can qualitatively predict the effects these parameters will have on HBT reliability.

Introduction

A great deal of interest continues to be attached to GaAs-based heterojunction bipolar transistor (HBT) reliability. For example, there is an ongoing debate within the industry about the comparative reliability of AlGaAs versus InGaP emitter structures [1], [2], [3]. Additionally, there is also continuing interest in the optimum parameters for the extrinsic base passivation ledge [4]. Indeed, as early as 1991, it became clear that a passivation ledge composed of the high bandgap emitter material is essential for good reliability [5]. The reasons for this are rooted in the device physics of the HBT.

Base dopant outdiffusion is a possible degradation mechanism under extreme conditions (e.g., very high temperatures [6], [7]). However several authors [3], [7], [8], [9], [10] have demonstrated that degradation under most HBT bias conditions is due to a steady increase in base recombination current over time. We have demonstrated that the excess base leakage current is due to a serial tunneling–recombination process [8], [11]. Electrons tunnel from the emitter into midgap level traps in the base and base–emitter depletion regions. Once in the traps, the electrons then recombine with holes. The midgap states that make the leakage current possible are formed during bias stress.

A similar process is also responsible for laser diode (LD) and light emitting diode (LED) degradation [12]. The physical mechanism responsible for the formation of defect-associated midgap traps is as follows. A non-radiative recombination event may occur at a midgap site associated with a crystal defect. The energy emitted by nonradiative recombination may be localized and transformed into lattice vibrations. These vibrations may in turn result in low-temperature solid-state changes, including defect movement and propagation. Recombination enhanced defect reaction (REDR) is the term used to describe a generic class of solid state reaction that are enabled by the energy released during electron–hole recombination occuring in the presence of a defect. This energy may generate the motion of defects and dislocation climb. The REDR process may also annihilate non-radiative recombination [13], [14]. Defect motion and dislocation climb may be a positive feedback mechanism. Eventually, defects may cluster together and form larger defects and larger recombination centers that leads to degradation of the device over time under forward current bias.

As the process proceeds by electron–hole recombination, it is obviously dependent on the concentration of minority carriers injected into the base, which is the emitter current. Where Jc represents collector current density, a dependence of lifetime on Jc−2 has been reported for GaAs/AlGaAs HBTs [8], and a dependence on Jc−1/2 has been reported for GaAs/GaInP HBTs [15]. However, more work is required to understand under what conditions these relationships hold.

From the above discussion, two things should be clear. First, device reliability is critically dependent on current density; or, more correctly, recombination current density. Recombination current is easy to model using a number of commercially available device simulation software packages. Second, device reliability is also dependent on defect density. Specifically, the device must be designed to minimize electron–hole recombination at crystalline defects that may be associated with midgap traps. Clearly, since the vast majority of recombination under normal device stress will occur in the base, the quality of the base material is of paramount importance. Sources of defects in the base may include dislocations propagating from the substrate and dislocations induced during processing. Additionally, defects may be formed in the base during crystal growth due to inappropriate III/V ratios, growth temperatures, or the inclusion of a high density of impurities. These factors can all be controlled through the use of low dislocation density substrates and proper growth and processing techniques [8].

In addition to good growth and processing techniques, proper device design also plays a role in device reliability. As noted above, the use of an emitter ledge that passivates the extrinsic base is critical to good device reliability. Indeed, the absence of such a ledge will allow for a very high recombination density at the unpassivated surface. Of course, the unpassivated surface is an infinite source of crystalline defects in GaAs. Defects that may move into the active base from the surface will be replaced by surface reconstruction.

Design of this emitter ledge passivation is obviously crucial; if it is too thick or heavily doped, current will flow through the undepleted ledge and effectively serve to increase the emitter size. In addition, significant surface recombination would occur at the end of the ledge. By contrast, if the ledge is too thin, it may not effectively passivate the surface [16]. Additionally, recombination within the high bandgap ledge needs to be minimized; the phonon energy released by non-radiative recombination in wide-bandgap AlGaAs or InGaP is higher than in GaAs, and presumably is more likely to result in an REDR event that may be associated with device degradation. In contrast, an REDR event in a low bandgap material such as InGaAs would be less likely to lead to REDR. Clearly, understanding the limitations of emitter ledge design is critical to HBT device reliability.

Section snippets

Model and assumptions

Using commercially available two-dimensional HBT simulation software (G-PISCES-2BTM) provided by Gateway Modeling, we have investigated the recombination rate in and near the base and emitter passivation ledge as a function of various different ledge parameters. Values for key material parameters used are shown in Table 1. These values were selected not only to be consistent with published values, but also so the model generated would provide a good fit to our own experimentally measured device

Ledge versus no ledge

As previously noted, the advantages of a high bandgap passivation ledge have been made clear in a number of publications. First, the ledge improves current gain in the device [16]. Second, as previously noted, a ledge has been shown essential to good device reliability [5]. This point is illustrated in the simulations of recombination current of otherwise identical devices with a 1 μm spacing between the emitter mesa and the base contact. Fig. 2 shows modeled electron–hole recombination current

Recombination profile versus temperature

The next step in this study was to determine recombination rate within the active region of the device as a function of junction temperature. This is a valuable benchmark, especially since device operation may normally be in a range near room temperature, whereas most bias stress tests are performed at elevated temperatures, typically above 200 or even 300 °C [9]. However, we note here an important qualifier: as previously noted, the model assumes a uniform temperature throughout the device,

Recombination profile versus current density

Fig. 4 shows otherwise identical HBTs with the ledge as described in Fig. 2(b), operated at different current densities but otherwise identical collector voltage bias and temperature. Fig. 4(a) shows a device with Jc=6 kA/cm2, Vce=3 V, and Tj=230°C. Peak recombination currents in the base, InGaAs cap layer, and the surface of the InGaP emitter are 1×107, 1×108 and 3×107 A/cm3, respectively. Fig. 4(b) shows the same device, except Jc=120 kA/cm2, or increased by a factor of 20. The increases in

Recombination profile versus emitter composition

Recently, there has been some controversy over the relative effectiveness of InGaP and AlGaAs ledges [1], [2], [3], with experimental evidence indicating that an HBT with an InGaP emitter is much more reliable than one with an AlGaAs emitter. In order to get better insight into the matter, we have modeled HBTs with two different thicknesses of Al0.25Ga0.75As emitters. Both structures are doped 5×1017 cm−3 in the AlGaAs. Each has an additional 300 Å of bandgap graded AlGaAs inserted between the

Recombination profile versus ledge doping

The effect of ledge doping has been modeled in an InGaP emitter device, again at Tj=230°C, Jc=20 kA/cm2, and Vce=3.0 V. Fig. 6 shows the results for two otherwise identical structures with InGaP emitter doping levels of 1×1017 cm−3 shown in Fig. 6(a) and 5×1017 cm−3 shown in Fig. 6(b). The former was chosen as a reasonable bottom limit in doping; significantly lower doping levels than that can lead to a non-negligible resistance in the emitter. As shown, there is very little difference in

Recombination profile versus ledge width

Another consideration with regard to ledge design is the width of the ledge itself. Assuming the ledge is completely self-aligned to the base contact, we have not found a significant difference in recombination profile, at least within the range studied. Fig. 7 shows simulated recombination current density for an HBT with a 0.5 μm wide, 300 Å thick GaInP ledge doped 3×1017 cm−3. Tj=230°C, Jc=20 kA/cm2, and Vce=3.0 V. This should be compared with the devices in Fig. 2(b), with a 1 μm ledge. As

Recombination profile versus ledge to base contact spacing

So far, we have assumed a self-aligned ledge to base contact [2], [17]. Of course, this is not always the case. Device processing may result in a finite distance between the base and emitter ledge, which would result in a partially unpassivated base surface near the base contact. Fig. 8 shows modeled recombination current density for an HBT with a 0.5 μm base contact to emitter ledge spacing, where the ledge is 0.5 μm wide. The device and bias conditions are otherwise identical to those shown

Experimental verification

Our simulations have shown that an HBT with a 300 Å InGaP emitter doped 3×1017 cm−3 exhibits relatively good recombination current characteristics when fabricated with 1 μm InGaP ledge self-aligned to the base contact. Such a device should be very reliable. We have confirmed that this is the case. Power amplifier unit cells, each consisting of twelve 3×45μm2 emitter fingers (unit cell emitter area of 1620 μm2) have been bias stressed at Tj=208, 243, 269 °C and Jc=55 kA/cm2. This current density

Conclusion

We have reported simulations on surface and emitter recombination for a variety of GaAs-based HBTs with different ledge parameters. The advantages of a ledge for reducing surface recombination are clear. Also as expected, higher temperatures and current densities lead to different recombination profiles than seen under nominal bias conditions in the device. The advantages of InGaP over AlGaAs as a hole barrier have also been shown. Additionally, the spacing between the ledge and the base

Acknowledgements

The author gratefully acknowledges Dr. R. Anholt of Gateway Modeling for useful discussions and for making available a copy of his device modeling software.

References (21)

  • T. Henderson

    Physics of degradation in GaAs-based heterojunction bipolar transistors

    Microelectron. Reliab.

    (1999)
  • S. Mohammadi et al.

    Photo-luminescence and transmission electron microscope studies of low- and high-reliability AlGaAs/GaAs HBTs

    Solid-State Electron.

    (2000)
  • Takahashi T, Sasa S, Kawano A, Iwai T, Fujii T. High-reliability InGaP/GaAs HBTs fabricated by self-aligned process....
  • Beaulieu C, Beggs B, Bennett J, Cook JPD, Hobbs L, Lester T, et al. Degradation modes of GaAs heterojuncton bipolar...
  • T.S. Low et al.

    Migration from an AlGaAs to an InGaP emitter HBT IC process for improved reliability

    GaAs IC Symp. Tech. Digest

    (1998)
  • P. Ma et al.

    Source identification and control of 1/f noise in AlGaAs/GaAs HBTs by using an on-ledge Schottky diode

    IEEE Electron Dev. Lett.

    (2000)
  • Camnitz L. Improved current gain stability in C-doped HBTs with an AlGaAs surface passivation layer. Proc 1991 GaAs REL...
  • D.C. Streit et al.

    High-reliability GaAs–AlGaAs HBTs by MBE with BE base doping and InGaAs emitter contacts

    IEEE Electron Dev. Lett.

    (1991)
  • D. Hill et al.

    Interaction of degradation mechanisms in Be-doped GaAs HBTs

    GaAs IC Symp. Tech. Digest

    (2000)
  • Henderson TS, Ikalainen P. Reliability of self-aligned, ledge passivated 7.5 GHz GaAs/AlGaAs HBT power amplifiers under...
There are more references available in the full text version of this article.

Cited by (0)

View full text
Copyright © 2002 Elsevier Science Ltd. All rights reserved.