ICMBE 2008, Aug. 3~8th, 2008 at University of British Columbia, BC, Canada

This year’s MBE conference focused on the following main topics:

1. Self-assembled nanostructures: Quantum Dots, Quantum Wires on compound semiconductors and Si/Ge material systems.
2. Nitrides: Growth mechanisms, rf vs. NH₃ MBE, material qualities and a little bit of electronic and photonic device results.
3. Photonic devices: QD lasers, diluted nitride long wavelength lasers, quantum cascade lasers, LEDs and photodetectors
4. III-V MOSFETs
5. Widegap oxides
6. Solar cells
7. Large production scale MBE operations
8. Growth control: RHEED with rotation and temperature control

Some topics will be discussed below.

Multi-D Nanostrucutres:

Multi-dimensional nanostructures and devices continue to attract researchers’ interests. The studies range from III-V quantum dot (QD) lasers for telecommunication applications (QD lasers were finally commercialized after being proposed ¼ century ago); to Si/Ge island formation for Si hole mobility enhancement in CMOS applications; to vertical nanowire (whisker) growths on various of semiconductor substrates for practical applications yet to be realized. All these show that crystal growers are enthusiastic about finding a way to improve the lateral pattern control down to nanometer scale.

Nitrides:

There are several interesting papers on growing GaN heterostructures (using MBE of course). In the plenary talk PL2.2 Dr. Skierbiszewski (Institute of High Pressure Physics, Polish Academy of Science, Poland) described how high quality InGaN/GaN QW lasers can be grown using plasma-assisted MBE (PAMBE) with post-growth surface/optical/electrical properties surpass MOCVD-grown counterparts. By using extreme Ga-rich conditions close to the onset of Ga-droplet formation, a two-dimensional growth can happen at relatively low temperature (700~800^oC, compared to half of GaN melting point of 1050^oC, which is used by MOCVD growths). The Ga-rich condition produces a metallic Ga adlayer within which the nitrogen adatoms can diffuse with lowered diffusion barrier energies, and therefore grow two-dimensionally. What makes Dr. Skierbiszewski’s talk even more interesting is that in their institute they can grow GaN single crystal substrates with threading dislocation density <>-2, the synthesis happens in Ga solution at 1300-1600^oC under N₂ pressure of 8-17 kbar ( 1bar ~ 750 Torr ~ 1 atm, that’s why he’s from the Institute of High Pressure Physics). The size of the single crystal GaN is only about 1cm x 1cm but can be sold for 8~10 K USD a piece! I talked to him during the recess and discussed about whether they do GaN HEMT or not. I found out that somehow they rather grow green-blue lasers than GaN HEMTs (Blue/green lasers make more money than GaN HEMT ICs?). BTW, along the same direction I found out that Sumitomo also produces and sells single crystal 2” GaN substrates, but they need NDA if I want to discuss further with them.

NH₃ MBE for GaN attracts more and more attention, too. In my opinion (and maybe many other crystal growers’) energetic particles (in this case N₂^* radicals) coming out of rf-plasma source are not ideal nitrogen source. CNRS/Picogiga group presented several papers (WB1.6, THB2.1 invited, and THB2.2) on GaN layers grown by NH₃ MBE on Si(111). Nitrogen source is provided by injecting 100-200 sccm (as opposed to less than 10 sccm of N₂ for normal PAMBE with rf-plasma source) of NH₃ through a leak valve and thermally cracked on substrate. Therefore the growth temperature is ~ 100^oC higher than PAMBE. Besides the source of nitrogen, the development of III-nitride devices on Si substrate mostly relies on the management of the dislocation density and the residual strain. The CNRS/Picogiga group managed to grow high quality AlGaN/GaN HEMT structures with n_s ~ 10¹³ cm^-2 and μ_n ~ 2000 cm²/V-s (the number is comparable to, if not better than, MOCVD grown ones. As they said, mobility is not an issue for MBE-grown HEMT anymore). The tricks include first nitridizing the Si substrate surface by impinging NH₃ on cleaned hot Si(111) surface to form a very thin layer of single crystal Si₃N₄, this will produce a sharp bottom interface; then using an AlN/GaN/AlN 40nm/250nm/250nm stress mitigating layer to reduce/bend the dislocations resulted from the large lattice mismatch between Si(111) and AlN/GaN; finally grow a thick ( >1.5μm ) GaN buffer layer at an optimized temperature at which the dislocations incline and cancel each other (although the dislocation density of grown epi is still in ~ 10⁹ cm^-2 range). The AlGaN/GaN active layers are then grown on top of the buffer layer. They also monitored in-situ the curvature of the epi and established a relation between wafer bowing vs. epilayer dislocation density. The summary: convex bowing (means the GaN grows compressively strained on the stress mitigating layer) reduces the dislocation density. But there are more subtle interaction between dislocation density, leakage current of the HEMTs, and the donor density of the 2DEG. When dislocation density drops, the 2DEG donor density increases and the HEMT leakage current also increases.

Dislocation reduction and strain engineering are only two of many obstacles that need to be overcome to produce high quality and reliable GaN HEMTs, no matter what crystal growth technique chosen. Crack formation in grown epilayer resulted from thermal expansion coefficient mismatch between Si or SiC substrates and GaN (or even between AlN and GaN) is another issue, for example. Growing HEMTs on single crystal GaN substrates (if they get large enough and cheap enough) could theoretically solve most of the problems, but poor thermal conductivity of GaN (656 mW/cm^-1K^-1) compared to 6H-SiC (4900 mW/cm^-1K^-1) could potentially hamper its practical use.

MBE Large-Scale Production and Growth Control:

Besides nitrides there are also several papers on MBE growth control and large scale MBE production operation. Tom Rogers of RFMD shared their operation details (TUA2.1), which can be leveraged by us. Details include:

1. Minimal number of device structures
2. Several items can be used as production metrics:
1. System uptime, throughput, and yield
2. Manpower efficiency
3. Reproducibility inter MBE systems
4. Wafer uniformity and run-to-run reproducibility
3. Documentation, training and uniform methods are critical for multi-shift pass-downs
4. Using of band-edge thermometry (from kSA) to reduce temperature variation of wafers across a huge platen (7 x 6”, Veeco Gen2000) from 5.78^oC to 1.5^oC by optimizing the platen structure (wafer ledge and backing ring wideth), proper coating of platens with semiconductors, and fine-tuning the power ratio and distance between the heater and the platen (TUA2.3).
5. Precision MBE system downtime forecasting
6. 24 hr HBT QTA turn-around time
7. Pareto analysis of reject causes
8. SEMI-standard definitions and tool states

For growth control I found out that RHEED oscillation can be done on rotating substrates. It is simple in theory: synchronize the RHEED image capturing with substrate rotation. A rotation speed detection mechanism will be added by the CAR outside the vacuum and triggers the capture of RHEED intensity. In practice there are few concerns such as less available data points and therefore worse fitting; and the annoying wafer skidding. But it proves that RHEED works better than just monitoring substrate oxide desorption.

MBE Native Oxide Desorption

Native oxide desorption process depends on various conditions such as the following:

1 Temperature monitor calibration: including pyrometer and desorption mass spectroscopy (DMS); one has to take the viewport deposition, reflectance from heated effusion cells, etc. into consideration when measuring temperature using pyrometer, while the calibration of DMS is not easy, either.

2 The gas species used for overpressure: Basically, using the same group V species as the group V component of the substrate will results in lower desorption temperature [1-3].

2.1 For InP substrate, according to [1] under P₂ overpressure the oxide (thermal oxide) desorbs within 5 min at 490^oC (actually at 473±6^oC the (2×4) surface is observed and the oxygen is below detection limit for the XPS measurement) and a (2×4) reconstruction is observed. This report suggests that under P₂ overpressure, the appearance of the (2×4) reconstruction can be regarded as a sign for the complete desorption of the oxide. On the other hand, heating in a flux of As₄ molecules results in the formation of an InAs layer at the top of the oxide, which strongly reduces the oxide desorption rate. The complete removal of the oxygen requires at least 520^oC and typically leaves a 1-nm-thick overlayer of InAs. In this case RHEED gives no indication whether the surface is oxide free.

2.2 Authors in [2] suggest from In-P-O phase diagram that the UV-ozone oxide prepared on InP substrate should have a composition of InPO₄. The (2×4) reconstruction emerges at the temperature of 528^oC under P₂ overpressure. They claim the removal of thin oxides from InP surface is probably not a matter of simple decomposition of the oxide but rather involves a reaction with the overpressure species or the substrate. The results suggest that the most probable reaction is with atomic phosphorus to form volatile phosphorus oxide, of which the most prevalent one would be P₂O₃, and atomic In as following formula:

3InPO₄ + 5P → 4P₂O₃ + 3In

The phosphorus from the substrate is the dominant driving force for the reaction and the reaction is both temperature and time dependent. During the desorption process there is a coexistence of areas covered by oxide and of some oxide-free regions which gradually increase in size until the oxide is desorbed. The phosphorus overpressure prevents decomposition of the oxide-free, exposed InP substrate surface. The free In atom from the reaction will evaporate rather than staying on the InP substrate. However, to the extent that a P₂ overpressure does not entirely replace all the atomic P lost from the substrate, which is due to thermo-etching at overheated temperature, even desorption under a P₂ overpressure would not result in a perfect interface between substrate and subsequent epilayer, namely In-ball will form.

2.3 In [3] As₄ and/or Sb₄ are used to protect the surface during InP desorption. The authors show that even if the deoxidation is feasible using one of the different beam pressures, the less critical temperature control leading to a good surface quality is obtained under (As₄+Sb₄) combined pressures. However, in my opinion, using different group V species when desorbing native oxides is not recommended and should be avoided.

3 Substrate temperature ramp rate: Some studies use very slow ramp rate (1^oC/min) [2] while some studies use fast ramp rate (~50^oC/min) [3]. The ramp rate should be established by oneself according to each system.

4 The composition and the thickness of the native oxide: Thermal oxide will have a lower desorption temperature than ozone plasma oxide [4], the thermal oxide grown on GaAs substrate has a desorption temperature of 582^oC (in this report the temperatures were measured using a Pt/Pt13%Rh thermocouple and the thermocouple was calibrated at the melting point of aluminum: 660^oC). Whereas ozone oxide has a desorption temperature of 638^oC. The authors attribute the 56^oC difference to the composition difference of the two oxides. And desorption temperature can be used as an accurate thermal reference point in the MBE growth of GaAs. As for thickness, the thinner, the better, in order to reduce the desorption time [2].

Reference

[1] R. Averbeck et al., “Oxide desorption from InP under stabilizing pressures of P₂ and As₄,” Appl. Phys. Lett., vol. 59, pp. 1732-1734, 1991.

[2] P. G. Hofstra et al., “Desorption of ultraviolet-ozone oxides from InP under phosphorus and arsenic overpressures,” J. Appl. Phys., vol 77, pp. 5167-5172, 1995.

[3] A. Godefroy et al., “X-ray and UV photoelectron spectroscopy of oxide desorption from InP under As₄ and/or Sb₄ overpressures: exchange reaction AsóSb on InP surfaces, “ J. Crystal Growth, vol. 179, pp. 349-355, 1997.

[4] A. J. SpringThorpe et al., “Measurement of GaAs surface oxide desorption temperatures,” Appl. Phys. Lett., vol. 50, pp. 77-79, 1987.