Materials Science and Engineering, B4 (1989) 423-427 423

A Study of Growth Defects in Seeded and Unseeded Silicon on Insulator Layers

D. A. WILLIAMS, R. A. MCMAHON AND H. AHMED

Microelectronics Research Laborato~', Department of Physics, Cambridge Universiff, Science Park, Milton Road, Cambridge CB4 4FW (U.K.)

(Received May 30, 1989)

Abstract

Zone melt recrystallization has been performed using a dual electron beam system to form layers of silicon on insulator. Both seeded and unseeded strategies have been explored, using fast (35 cm s t) scans to ensure compatibility with the processing conditions necessary for making three-dimensional integrated circuits. The quality of the resulting film has been assessed using a variety of microscopic techniques, including scanning and transmission electron microscopy and optical microscopy. Defect structures were found in the recrystallized films which indicated planar and faceted growth in the seeded case, and a mixture of faceted, cellular and dendritic growth in the unseeded case. The structures are described and linked to possible regrowth mechanisms, and the implications for three-dimensional circuit formation are discussed briefly.

1. Introduction

Silicon on insulator layers has been prepared using seeded and unseeded zone melt recrystal- lization in a dual electron beam system. The defect morphologies in both cases have been studied using scanning and transmission electron microscopy and optical microscopy, and found to display several different growth mechanisms.

A thin silicon film on an insulating layer is the principal material for building three-dimensional integrated circuitry [1], and has many other applications in microelectronics [2]. Three- dimensional circuits are made by stacking layers of devices one above another, and are produced by fabricating one layer of devices then ;forming a film of single-crystal silicon above them. The film should be defect-free single-crystal silicon in order that reliable devices may be made repeat-

ably. Several techniques have been developed for the fabrication of silicon on insulator films, but the requirement that the formation of an upper layer does not adversely affect devices already fabricated in lower layers prohibits the use of some of these methods for three-dimensional integration.

The method which seems most promising for the formation of three-dimensional circuits is the zone melt recrystallization of polycrystalline silicon films using laser [3] or electron [4] beams. These have the advantage that, although the upper film reaches the silicon melting point, the heat pulse is of very short duration (of the order of milliseconds in the regime described in this report) and so diffusion is negligible in the layers which do not melt. It is clearly necessary to characterize the mechanisms occurring during regrowth in these strategies, and many investiga- tions have been made. Seeded and unseeded laser recrystallization [3, 5] and seeded electron beam recrystallization [4] have been studied extensively, but the experiments reported here used electron beam recrystallization for both seeded and unseeded material. The flexibility of the electron beam system allows recrystallization conditions to be explored which are similar to or very differ- ent from the laser case.

Seeded growth gives single-crystal silicon films, which are needed for device production, but the requirements of circuit layout may not in all cases allow the inclusion of seeding structures. For this reason, it is necessary to study the mechanisms occurring during unseeded regrowth in electron beam recrystallization, so that this material may also be improved and so that it could be used for circuit formation if design con- siderations dictated.

The starting material consists of a single- crystal silicon wafer of (001) orientation, with a

0921-5107/89/$3.5I) © Elsevier Sequoia/Printed in The Netherlands

424

layer of silicon dioxide 1 pm thick on the surface. This is covered with 1 #m of polycrystalline sili- con and the structure capped with 0.3 pm of undoped silicon dioxide and 0.7 pm of phosphorus-doped silicon dioxide. For the seeded structure, the underlying isolating oxide is patterned by removing stripes 2-6 pm wide and spaced by 20-100 pm before deposition of the polycrystalline silicon. This gives seeding windows which also act as heat sinks during re- crystallization. The composite capping layer of silicon dioxide prevents the molten silicon from bailing up during recrystallization, as molten silicon does not wet silicon dioxide. The upper doped layer is flexible and does not crack during processing, and the lower undoped layer provides strength and acts as a diffusion barrier to prevent unwanted doping of the silicon layer by dopant diffusion from the oxide. A stiff cap also reduces thickness variations in the silicon layer due to stress relief in the cap during recrystallization.

The recrystallization system [6] uses one electron beam which is rapidly raster scanned over the back surface of the wafer or chip to give background heating in the range 700-1000 °C. A second electron beam of about 100 pm, at full width half maximum, is scanned with a triangle waveform to produce a line which is swept over the front surface of the specimen at 35 cm s ~. As the line passes over, the layer of poly- crystalline silicon melts and then refreezes. In the seeded case the regrowth process is controlled by the heat sinking seed windows which results in a layer of single-crystal silicon [7]. The unseeded case results in a complex polycrystalline morph- ology. The duration of the heat pulse as the line beam passes over is such that if the silicon does not melt, the dopant diffusion in underlying regions is negligible.

When the beam has passed oven and the growth changes from melt/freeze where the regrowth front moves backwards and forwards, to slow/ fast growth with the front always moving torward. facets begin to form in the regrowth melt-solid interface and then develop to a size of several micrometres. Generally, tile planar regime lasts from the seed window for a few micrometres over the isolating oxide and the final growth of the film over the isolating oxide is faceted.

The resultant film is single-crystal silicon with a subgrain boundary running down the centre of each stripe, equidistant from each seed window [9]. This arises from the strain caused by the difference in thermal expansion coefficient between silicon and the isolating silicon dioxide. Figure I is a plan view transmission electron micrograph of such a subgrain boundary. It is seen to be localized to within 1 pm laterally, because of the symmetry of the heat flow. The silicon near the seed window is found to be strained, but most of the regrown silicon over the isolating oxide is strain free,

If the silicon film contains impurities, princi- pally oxides, then branching defects are seen on either side of the central subgrain boundary, with a spacing determined by the characteristic size of faceting. Figure 2 is a plan view optical micro- graph with Nomarski contrast of a typical pattern. These appear to originate at precipitates at interior facet corners in a similar manner to the subgrain boundaries which are seen starting at interior facet corners in some types of slow regrown silicon on insulator layers [1 O, 1 I i They

2. Seeded regrowth

The seeded case has been extensively studied for rapid regrowth, and in the electron beam method used for these investigations two regrowth regimes are seen [8]. As the electron beam spot is scanned from side to side to form a line, the pool of molten silicon grows in size as the spot passes over then shrinks again. When the electron beam is over the regrowing material, the rapid oscillatory growth due to the changes in melt pool size results in planar regrowth, as any facets that form are destroyed on remelting.

Fig. 1. Planar transmission electron micrograph of a sub- grain boundary in seeded silicon on insulator.

425

Fig. 3. Optical micrograph of unseeded material showing the branching structure of "triangular defects" around a subgrain boundary.

Fig. 2. Optical micrograph of seeded silicon on insulator where branching "triangular defects" are present.

are found to be small grains of different orienta- tion to the main film, sometimes bearing a twin- like orientation relationship to the rest of the film, and also sometimes being associated with local- ized high density dislocation entrainments. The periodicity of the branching defects is usually in the range 5-10 ktm, if there are sufficient im- purities present, which is consistent with a final facet periodicity of that magnitude [8, 9].

3. Unseeded regrowth

When no seed windows are present, recrystal- lization is of random orientation and proceeds from the already grown material [12]. The film tends to have a degree of (001) texture, as the (001) plane is energetically more favourable at the silicon/silicon dioxide interface where regrowth nucleates. The defect morphologies seen are very much more complex than those seen in seeded material, and unpredictable on a local scale, but the patterns seen are remarkably similar on a scale of millimetres. Previous studies have concentrated on strategies for slow regrowth to produce as good quality single-crystal silicon as possible, but such regimes are not appropriate for multilayer regrowth for making three-dimen- sional circuits, as the thermal treatments would severely disrupt lower layers of devices. This investigation concentrates on the regrowth

mechanisms occuring at scan speeds which are known not to affect underlying devices.

In some regions, a herringbone pattern is seen to evolve from a region of single-crystal material, seen in an optical micrograph in Fig. 3. This is very similar to the patterns seen in seeded growth, and is characteristic of the melt front evolving from faceting on an atomic scale to facets of several micrometres' length. The perio- dicity of the branches is similar to that in the seeded material, which suggests that in both seeded and unseeded recrystallization the facet length has reached the maximum possible (about 5-10 ktm) for the sweep speed of 35 cm s -I. The branches are similar in structure to those seen in seeded material, being separate, often highly dis- located grains which start near the lower silicon/ silicon dioxide interface and grow upwards and out [9]. They have been called "triangular defects" [12], and a typical example is seen in a plan view transmission electron micrograph in Fig. 4 and in cross-section in Fig. 5. The region of single- crystal silicon is likely to occur when the growth is oscillatory, which is seen to suppress perturba- tions and keep the regrowth front at the surface defined by the melt isotherm, as seen in seeded material [8].

Other regions are seen to exhibit structures which are more indicative of dendritic growth, with a very intricate arrangement of grain bound- aries, as seen in an optical micrograph in Fig. 6. Some regions suggest cellular growth, but there is no conclusive evidence of this. When one mechanism has started to dominate the regrowth,

426

Fig. 4. Transmission electron micrograph showing a trian- gular defect with a dislocation tangle. One side of the grain boundary can be seen in the lower left of the micrograph.

Fig. 5. Cross-sectional transmission electron micrograph of a triangular defect.

Fig. 6. Optical micrograph showing evidence of dendritic growth.

it will propagate for relatively large distances (hundreds of micrometres) before a change. However, on a scale of millimetres, the patterns observed are remarkably similar in different

Fig. 7. Transmission electron micrograph of a subgrain boundary in unseeded material. Many straggling dislocations and a void can be seen.

regions, suggesting that the probabilities of each type of growth predominating are constant. Faceted appears to be the predominant mechan- ism, accounting for approximately 70% of the regrown area.

The defects seen in unseeded regrowth are similar to those observed in seeded recrystal- lization, but not as well localized and not in pre- dictable positions. Figure 7 shows a subgrain boundary in unseeded material, which may be compared with Fig. 4. Th e boundary is very narrow in one region, but then widens, and is associated with a wide tangle of dislocations. Also, the silicon film strains are very different. The unseeded material is heavily strained, with no preferred direction, but the seeded material is only strained in the seed window region and is strain free elsewhere. This is again a conse- quence of the heat flow directionality imposed by the seeding strategy.

The direction of propagation of the regrowth mechanism is also variable in unseeded material. Figure 8 shows a plan view optical micrograph of a region where several growth mechanisms are evident. Th e sweep direction was from left to right, and the defect pattern shows a general trend in this direction, but with much variation.

4. Discuss ion

It has been shown that seeded fast regrowth of silicon on insulator layers can produce single- crystal films, and that the regrowth mechanisms in the electron beam system used are planar and evolving faceted regrowth. When seeding is not used, for the same beam processing conditions, the regrowth mechanisms seen are planar,

427

[13]. It is apparent that perturbations such as facets, dendrites and growth cells allow the aggre- gation of impurities and the build up of stresses which lead to breakdown of the regrowth front, and so any mechanism reducing perturbations is likely to be beneficial.

Acknowledgments

D.A.W. acknowledges the support of a post- doctoral research fellowship from the U.K. Science and Engineering Research Council.

. . . . . . . . = ili~i~!!i ¸¸ j iii ~ iiilii ̧i!̧~!~̧ ~̧ ¸̧

S

Fig. 8. Optical micrograph of a region where many mechan- isms are evident.

faceted, dendritic and possibly cellular. The pre- dominant mechanism is faceted growth, but once one type of regrowth has started, it will propagate for many hundreds of micrometres. The resultant material is highly defective in most regions, but large grains can be produced. It would appear that unseeded growth of this form may have applications in some three-dimensional integra- tion systems, but that control of the heat flow during recrystallization is usually desirable. Heat sinking structures without seeding have been used extensively for laser recrystallization of single layers, and these may be useful in electron beam recrystallization, particularly if used in con- junction with regrowth which is predominantly oscillatory, with the subsequent reduction in per- turbations. Some early studies in multiple layer recrystallization have shown promising results

References

1 D.A. Williams, R. A. McMahon, H. Ahmed, K. Barfoot, D. J. Godfrey, B. Dunne and A. Mathewson, Mater. Res. Soc. Symp. l'roc., 107 (1988) 235.

2 G. Furakawa (ed.), SOl-Technology and Applications, KTK Toyko and Reidel, 1985.

3 J. R Colinge, E. Demoulin, D. Bensahel and G. Auvert, Appl. Phys. Lett., 41 (4)(1982) 346.

4 R.A. McMahon, Microcircuit Eng., 8(1988) 255. 5 J: P. Colinge, D. Bensahel, M. Alamome, M. Haond and

C. Leguet, Mater. Res. Soc. Symp. Proc., 23 (1984) 597. 6 R. A. McMahon, J. R. Davis and H. Ahmed, Mater. Res.

Soc. Symp. f'roc., 4 (1982) 783. 7, J. R. Davis, R. A. McMahon and H. Ahmed, J.

Eleetrochem. Sot'., 132 (8) ( 1985 ) 1919. 8 D. A. Williams, R. A. McMahon and H. Ahmed, l'hys.

Rev. B, 3g(14)(1989) 10467. 9 D. A. Williams, R. A. McMahon, H. Ahmed and W. M.

Stobbs, Inst. Phys. Con]~ Ser., 87 ( 1987 ) 415. 10 L. Pfeiffer, S. Paine, G. H. Gilmer, W. vanSarloos and K.

W. West, Phys. Rev. Lett. 54 (17) (1985) 1944. 11 J. C. C. Fan, B-Y. Tsaur, C. K. Chen, J. R. Dick and L. L.

Kazmerski, Appl. l'hys. Lett., 44 ( 1 l ) (1984) 1(186. 12 M.W. Gels, H. I. Smith and C. K. Chen, J. Appl. t'hys., 00

(,3) (1986) 1152. 13 D. A. Williams, R. A. McMahon, H. Ahmed, K. Barfoot,

D. J. Godfrey, B. Dunne and A. Mathewson, Mater. Res. Soc. Symp. Proc., 107 (1988) 235.