ATOMIC STRUCTURE OF THE INTERFACES BETWEEN SILICON DIRECTLY BONDED WAFERS M Benamaral, A Rocher1, A Laporte2,3, G Sarrabayrouse2, L Lescouzfcres3, A PeyreLavigne3, M Fnaiech4 and A Claverie1. 1

CEMES-LOE/CNRS, 29 rue Jeanne Marvig, BP 4347,31055 Toulouse Cedex (France). LAAS/CNRS, 7 av. du Colonel Roche, 31077 Toulouse Cedex (France). 3 Motorola Semiconducteurs S. A., Av. General Eisenhower, 31023 Toulouse Cedex (France). 4 Facultes des Sciences de Monastir, 5000 Monastir (Tunisie). 2

ABSTRACT The so-called Direct Wafer Bonding (DWB) technique opens new possibilities for the electronic industry but still suffers from the poor knowledge we have of the microstructure of these interfaces and hence of their electrical activity. In this work, we have extensively used Transmission Electron Microscopy techniques in plan-view and cross-section to identify the structure of the interfaces found between two bonded silicon wafers. The general structure of these interfaces is that of a perfect grain boundary and evidently depends on the misorientation between the two bonded wafers. A twist component in the range 0>θ >13° creates a square network of pure screw dislocation whereas an unavoidable tilt component (<0.5°) is compensated by a periodic array of 60° dislocation lines perpendicular to the tilt direction. Therefore, the regularity of these networks can be disrupted by the presence of steps (of up to several nanometers) in the interface plane. Silicon oxide precipitates are seen heterogeneously distributed on the interface with preferential nucleation sites on the dislocations. INTRODUCTION The direct wafer bonding (DWB) technology where two materials of different compositions or doping levels are bonded together at the atomic level opens new possibilities for the realization of electronic devices1. With regard to bipolar devices and high voltage integrated circuit applications, it is investigated as an attractive technique since it has the potential for inexpensively producing thick and doped epilayers including n-n, p-p and p-n type junctions. For such applications, the electrical properties of the bonding interface are of crucial importance and were shown, contrary to what it is required, to be not free of anomalies most probably related to residual impurities or cristallographic defects. It has been shown that the electronic transport across 2such interfaces is restricted by a potential barrier resulting from charged interfacial states -3. However, the origin of these interfacial states remains still unknown because of the poor knowledge we have of the microstructure of these interfaces. It is the purpose of this work to carry out a complete transmission electron microscopy (TEM) study of these interfacial structures in an effort to improve our understanding of their electrical properties. From a fundamental point of view, this study consists in a considerable outgrowth of previous investigations of (001) twist boundaries in silicon4". EXPERIMENTAL Polished 4-inch, (001) oriented, 525 (im-thick, CZ-grown type, and lightly boron doped (Ifll5 cm~3) silicon wafers were cleaned in an automatic equipment aimed at rendering their surfaces hydrophobic. Then, pairs of wafers were put in contact by manual assembling in a class 10 clean room with the wafer flats intentionally misoriented from aligned to about 15°. Afterwards, the bonded wafers pairs were annealed at 1200°C for 50 mn in an N2 ambiant The TEM studies were performed with a JEOL 200 CX and a Philips CM20 microscopes operating at 200 kV, using conventional images and diffraction patterns, on both cross-section and planview samples. 863 Mat. Res. Soc. Symp, Proc. Vol. 378 * 1995 Materials Research Society

RESULTS 1) GENERAL OBSERVATIONS The figure l.a is a bright field image showing the general view of a bonding interface when the electron beam is closed to the [001] direction. Large domains are periodically separated by dark lines. Amorphous precipitates of silicon oxide are often seen located at the interface. The largest of them are connected to the black lines. The overlapping of the two misoriented crystals gives rise to double-diffraction effects (fig l.c). Moire fringes are resulting from this doublediffraction effect and arise owing to the interference essentially between the (220) reflections of the two crystals. They appear in the imaging mode as it is shown in the fig l.b. These fringes give no direct information about the boundary structure but they show that the two crystals are misoriented around their common <001> axis. In this case, the moire spacing is measured to be equal to d=23 + 1 A. From this value, one can deduce the corresponding twist angle'θ =4.8 ± 0.2°.

figure 1: a) BF plan-view micrograph taken under [001] zone axis. Note the formation of terraces (A) resulting from a small tilt misorientation and oxide precipitates (P) mostly connected to the 60° dislocation lines, b) the moire fringes, seen within terraces and parallel to <110> directions, c) diffraction pattern along the [001] zone axis. The sharp extra-satellite spots that are pointed by the arrows in fig 1 .d attest of the periodicity of the dislocation network on a large scale.

2) GBD accomodating low-angle twist component As expected from crystallography of low-angle twist boundaries8, the structure of the interface consists of two orthogonal sets of screw dislocations. The Burgers vectors parallel to the dislocation lines are bl=a/2[l 10] and b2=a/2[-1lO]. The dislocation spacing is given by the relation, Dd=b/2.sin(θ /2) (1), where θ is the twist misorientation angle. a) θ <1° Figure 2 is a typical multibeam plan-view image from one of these low-angle boundaries. For such angles, the dislocation spacing is still large enough (Dd>220A for θ <1°) so that the entire network can be easily imaged in strong beam conditions, as expected from the diffraction contrast theory9. The corresponding twist angle was found from the moire spacing

864

(see §3.1) to be 0.58° ± 0.03°. The separation distance between dislocations is directly measured on this image to be Dd=380A. No distance correction needs to be carried out since dislocations of the same set possess the same Burgers vector. b) 1°< θ < 7°. As the misorientation is increased, the grain boundary dislocation spacing becomes smaller and the network becomes more regular because of the attractive forces tending to hold them

figure 2: Typical BF multibeam plan-view image from the interface of two lightly misoriented wafers. The multiple diffraction condition is satisfied by using simultaneously the 220, -220 and 040 reflections. The square array of screw dislocations is disrupted by the 60° dislocation lines. Note the perfect grating of such spaced dislocations.

together in an uniform array. In this case, the structure cannot be observed in multibeam nor in two beams conditions because of the strong contrast arising from the moire patterns. This moir6 contrast may also induce confusions as its periodicity is half the separation distance between dislocations. Thus, the best way to determine the structure is to image it using the weak-beam technique because moire contrasts rapidly- fade when the deviation from the exact Bragg condition increases. Figures 3.a et b are "weak-beam" images of the sample presented in fig. 1: each of them shows one set of the screw dislocation, the other one being out-of-contrast. The dislocations appear in the images as continous and straight white lines. Their spacing is measured to be equal to 46 A. This is in good agreement with the value deduced from the relation (1) with the twist misorientation angle 4.8°. The interfacial stress is therefore minimized by this two sets of screw dislocations.

figure 3: Weak-Beam plan-view TEM images showing each of them one set of screw dislocations.

It is known that planar periodic grain boundary dislocation network examined by electron diffraction gives rise to s reflections. These GBD reflections are arranged in a network which characteristics are directly connected to the GBD network. They must be carefully analyzed to avoid confusion with spots resulting from double-diffraction effect7. For a twist boundary consisting of a square grid of screw dislocation with spacing Dd examined under normal incidence, the diffraction pattern consists of a square array of spots spaced 1/Dd centered around the two crystal reflections and in the same orientation as the square GBD's grid10. This square grid of extra spots is shown in fig l.d c) θ > 7°. The analysis of the diffraction pattern becomes a precious tool when the GBD structure cannot be determined from the image. It usually happens for dislocations rendered invisible because too closely spaced; when the corresponding misorientation angle is above 7°. In this case, it is easier to detect the periodic network through the presence of the array of the screw dislocation reflections though iheir intensity is known to decrease when decreasing Dd. Thus, a misorientation angle of 13.5° leads to the formation of a screw dislocation array detectable in the diffraction pattern [11]. Note that this sample corresponds to the maximum misorientation angle used in this work and we still do not know whether larger angles can also produce the same type of GBD network. 3) GBD accomodating low-angle tilt component The interfacial structure always exhibited an uncontrolled small tilt component. This tilt component (<0.5°) corresponds to a small rotation around random axes of the boundary plane. As a result, the interface consists of flat (001) terraces separated by steps lying along the direction perpendicular to the tilt direction. These steps are formed by the addition of families of 60° dislocation lines periodically located at the interface and that all have the same component perpendicular to the interface, a/2=2,7 A. They are clearly observed in all samples even when the screw dislocation network cannot be detected when too closely spaced. Their spacing does not depend on the screw dislocation network. In the sample seen in fig 1, they appears as the dark lines 400 nm wide spaced. This allows us 10 precisely determine the tilt angle from which they originate to be 0.04". The characteristics of this tilt angle (direction and magnitude) are in good agreement with those deduced from the shift of the Kikuchi lines in the diffraction pattern. When one of these lines is examined under different diffraction conditions as in fig 4.a, it can be seen that the separation distance between the screw dislocation on either side of this dislocation is increased by half of the usual spacing. This change in spacing is because the screw component of this dislocation is half of the other dislocation. Another remarquable feature is that the other set of screw dislocation is systematically offset by half of its periodicity when it crosses this line (fig. 4.b). The reason of this offset is that the four nodes at which this 60° dislocation meets screw dislocations split into three fold nodes. If the twist angle is below 1°, all dislocations segments are distinguable in the image. In such cases, it is possible to carried out an analysis of the resulting configuration following the Frank's procedure I2-<3. Fig. 5 shows, for instance, how a 60° dislocation lying along [-11O] with Burgers vector b=a/2[1oi] intersecting the screw dislocation with Burgers vector bl will appear in the network. At the meeting point, these two dislocations react to form a third dislocation with b3=b1+b=a/2[011] provided b2+bi2 >b32 and b, bi make an angle larger than 90°. Before the reaction takes place, the 60° dislocation b had an edge component in the plane of the interface. After the reactions, the two 60° dislocations segments with b and b3 have an equal but opposite edge component. The total edge component vanished and is essentially

figure 4: Weak-Beam images of a 60° dislocation lying along AB a) g=220, the dislocation appears as a straight and white line which direction is the same of the screw dislocation one. b) g=220, note the shift of the other accommodated by the offset of the screw dislocation. As seen in fig. 6, the appearance of these two 60° dislocations at the interface increases the screw dislocation spacing by 0.5 Dd since their efficient component in compensating the twist misorientation is a/4[110|. The low-energy configuration is achieved when angles between dislocations change to 120°.

figure 5: Interaction between a 60° dislocation lying along [110] with Burgers vector b and screw dislocations with Burgers vector 4) Large steps The regularity of the network is usually not total but disrupted by the addition of 1111 ( steps of up to several nanometers in height which do not act in accomodating the crystal misorientation. These steps are seen in only some of the investigated samples and their origin is believed to be an neighboring unbonded area. They are not arranged with a regular periodicity as the separation distance vary from up to a few tens of micrometers. Fig. 6 shows some of these steps when the electron beam is closed to the <001> direction. These steps look tike broad ribbons lying along the two <110> directions. The corresponding schematic diagram of a perspective view of the actual structure deduced from tilting experiments is shown below. SUMMARY AND CONCLUSION Although wafers were Cz grown type, no oxide layer was encapsulated at the bonding interface whatever the twist misorientation angle. This is not surprising since it is well-known that the HF cleaning procedure prior to bonding is the best way to reduce the amount of oxide at the interface. A grain boundary is formed with all dislocations presents as part of the equilibnim boundary structure. All these dislocations are located at the interface and allow the structure to be fully relaxed with no extended volume defects.

figure 6: a) BF image along [001] zone axis showing two large (111) steps which appear as two ribbons lying along the <110> directions, b) Schematic diagram of the actual structure. The general structure of these interfaces is that of a perfect grain boundary and evidently depends on the misorientation between the two bonded wafers. A twist component creates a square network of pure screw dislocation whereas an uncontrolled tilt component (θ <0.5°) is compensated by a periodic array of 60° dislocation lines perpendicular to the tilt direction. Therefore, these low-angle equilibrium boundaries possessing perfect structure are sometimes disrupted by the presence of steps (of up to several nanometers) in the interface plane. We must emphasize that since the reproducibility of the electrical characteristics of these interfaces is of prime importance, significant efforts have to be spent to master the relative misorientation between wafers. Moreover, the 60° dislocations are seen to be preferential nucleation sites for oxide precipitates and thus the tilt component has also to be mastered. Finally, TEM experiments are well suited for the study and optimization of the process conditions needed for the routine production of directly bonded semiconductor structures. 1. Bengtsson S., (1992) J. of Electronic Materials, 21 (8), 841-862. 2. Laporte A., Sarrabayrouse G., Lescouzeres L., PeyreLavigne A., Benamara M., Rocher A. and Claverie A., 1994, Proc. 6th Int. Symp. on Power Semi. Dev. & ICs, 293-296. 3. Engstrom O., Bengtsson S., Andersson G. I., Andersson M. O. and Jauhiainen A., (1992) J. Electrochem. Soc., Vol. 139, No 12, 3638-3644. 4. Foil H. and Ast D., 1979, Phil. Mag. A 40 (5), 589-610. 5. Benamara M., Rocher A., Laanab L., Claverie A., Laporte A., Sarrabayrouse G., Lescouzeres L. and PeyreLavigne A., 1994, CRAS Paris 318 (2) 1459-1464. 6. Gafiteanu R., Chevacharoenkul S., Gosele U. M., Tan T. Y., 1993, Microscopy of Semiconducting Materials 134, 87. 7. Carter C. B., Foil H., Ast D. G. and Sass S. L., 1981, Phil. Mag. A 43 (5), 441. 8. Bollmann W., 1970, Crystal Defects and Crystalline Interfaces (Editors: Springer-Verlag). 9. Kang J.M., 1993, Ph. D. Thesis, University of Toulouse III. 10. Guan D. Y. and Sass S. L., 1973, Phil. Mag. A 27, 1211. 11 Benamara M., Rocher A., Laporte A., Sarrabayrouse G., Lescouzeres L. PeyreLavigne A. and Claverie A., 1995, Microscopy of Semiconducting Materials, in press. 12 Frank F. C., 1954, Defects in Cristalline Solid, London, Physical Society. 13 Zhu and C.B. Carter, 1990, Phil. Mag. A 62 (3), 319.