Formation of buried insulating layers by high dose oxygen implantation under controlled temperature conditions

Formation of buried insulating layers by high dose oxygen implantation under controlled temperature conditions

Vacuum/volume Printed in Great 35/number Britain 12lpages 589 to 593/l 0042-207X/85$3.00+ Pergamon 985 .OO Press Ltd Formation of buried insula...

2MB Sizes 5 Downloads 29 Views

Vacuum/volume Printed in Great

35/number Britain


589 to 593/l

0042-207X/85$3.00+ Pergamon


.OO Press Ltd

Formation of buried insulating layers by high dose oxygen implantation under controlled temperature conditions M Bruel,

J Margail,


J Stoemenos*,

CEN, 38047


P Martin

and C Jaussaud,



et de Technologie



received for publication 24 July 1985

A specially designed sample holder was used to form SOI structures by high dose oxygen implantation under controlled temperature conditions. This system uses a layer of molten tin to provide a good thermal contact between the silicon wafer and a resistively heated support. The layers formed under these conditions were characterized by RBS and XTEM. After annealing at 1150°C for 2 h the only remaining defects in the top silicon layer are dislocations with a density of less than 70’ cm-= and SiO, precipitates. Annealing at 7 185°C for 6 h does not change the density of dislocations but leads to the formation of a 200 nm thick silicon layer free of SiO, precipitates, suitable for VLSI processing without growing an epitaxial layer.

Introduction It is now well established that implantation of oxygen into silicon at very high doses (about 2 x 10” ions cm-‘) and medium energies (200 keV) can form high quality silicon-on-insulator (SOI) substrates for VLSI technology (for a review paper, see ref 1). Among all the implantation parameters, wafer temperature has been shown to be a key parameter for making good quality SO1 substrateszm4. Up to a recent date, most implantations were done with the wafer thermally insulated, and its temperature is thus set by the balance between the energy received from the beam and that lost by radiation. Using this method, SO1 substrates were formed at temperatures up to 600°C. Though this method has been shown to be very efficient, it suffers from some limitations: (a) Since wafer temperature is set by the beam current density, limiting this temperature leads to a decrease of the dose rate, involving longer implantation times. (b) In most cases, the implantations are done with low or medium current implanters and in order to reach the required temperature, the implanted zone must be restricted to a small area of the wafer. This results in steep temperature gradients across the sample which may cause it to break4. (c) Since the current density and the temperature of the wafer cannot be set independently their effects on the quality of the SO1 substrate cannot be separated. This might lead to erroneous conclusions if an increase in wafer temperature and an increase in current density have opposite effects on damage creation. (d) One of the most severe limitations to the quality of the layers obtained by using beam heating arises from the fact that the * Physics department,





implantation begins with the wafer at room temperature, and a lot of defects are created during this period. Heating the specimen before irradiation should avoid this problem. The first experiment using external heating reported in the literature is that of Wilson and co-workers4. They have shown that implanting with the wafer previously heated at 450°C results in a better crystallinity of the top silicon layer. In a recent paper, Holland et al 5 used a resistively heated holder to heat the sample during the implantation and obtained a better crystalline surface. They explain this result stating that when the sample is irradiated at low temperature, clustering of point defects to form linear or dislocation loops, occurs. Once nucleated, these defects are difficult to eliminate even after a high temperature treatment. Nevertheless these experiments were conducted under conditions where thermal radiation is the dominant heat loss mechanism, and during the implantation beam heating raised the sample temperature far above that at the beginning of the irradiation, to a value not accurately known. So it appears at this date that though the effects of temperature on SO1 quality have been investigated, the formation of SO1 structure by high dose oxygen implantation has never been performed under complete control of the wafer temperature, because of experimental difficulties. Moreover, the industrial development of this technique requires very high current (- 100 mA) implanters. Such machines are presently being developed. They will provide preheating of the wafer at a selected temperature and will maintain this temperature during the implant. Thus a good knowledge of the effects of wafer temperature on the crystalline quality of the silicon over-layer is highly desirable. For all the above indicated reasons, we developed a system that allows wafer temperature control during the implantation, 589

M Bruel et al: Formation of buried insulating layers by high dose oxygen implantation under controlled temperature conditions independently of the beam power. which means a system capable of heating or cooling the sample during the process. In the first part we will describe this system which is a new approach to the problem of temperature control during the implantation and in the second part some preliminary results will be given in order to show the quality of the SOI layers that can be obtained using this system. 2. Experimental 2.1. System design. Two systems are generally

used in high current machines to limit beam heating during the implantation?‘. One uses a clamping of the wafer against an elastomeric pad, the other uses gas cooling. Nevertheless. these two systems are not ver) efficient and are difficult to use at the temperature required ( - 500’ C) for oxygen implantations. the first one because the elastomer cannot withstand such temperature and the second because thermal expansion problems make it difficult to meet the required tolerances (thickness of the gas layer around 20 ilm). So we designed and built a system to insure a good thermal contact between the wafer and its holder. This system (see Figure 1) uses a film of liquid tin between the wafer and its holder. The wafer is clamped against a stainless steel holder in which a cavity has been made. It is filled with tin (melting temperature 232 ‘C) in which two thermocouples are dipped. One thermocouple is used for feedback control of the temperature of tin and the other one records this temperature during the implantation. The two thermocouples are placed behind the implanted zone, very close to the back side of the wafer in order to minimize thermal gradients between the wafer and the thermocouples. Because of the very good thermal contact between liquid tin and silicon (the effective contact surface is maximum) there is no thermal barrier at the interface between silicon and liquid tin, and they can be regarded as a continuous medium. Thus the wafer can be heated before the implantation, and during it calories can be added or subtracted, depending on the incident beam power and the chosen temperature. Tin was chosen for several reasons: it has a low melting temperature (232 C) and a high boiling point (2687 C); its vapour pressure is low enough7 (7.4 x IO-’ torr at 700°C) in the temperature range for which the system is designed: it is not electrically active in silicon*. 2.2. Sample preparation. Since the wafer is in contact with liquid tin at a high temperature for several hours, some precautions must resistive


be taken: a layer of SiOZ (about 600 nm thick) is deposited on the back side of the wafer to avoid the formation of an alloy between tin and silicon. This SiO, layer also acts as a diffusion barrier for tin: the diffusivity of tin into silicon dioxide is 10~” cm* s ’ at 600 C”, giving a diffusion length of about 1 nm for a 3 h implantation at this temperature. 2.3. Wafer temperature. At the end of the implantation the structure of the wafer is typically: top silicon layer 300 nm, buried SiOZ 500 nm, silicon 400 itm. deposited SiO, 600 nm. Typical implantation conditions are: energy 200 keV, current density SO /‘A cm ’ leading to an incident average power density of IOWcm ‘. Taking at 600 C thermal conductivities of0.6 W cm ’ K ’ for silicon. 0.023 W cm ’ K ’ (ref IO)for SiO, and 0.3 W cm ’ K ’ for liquid tin. a simple calculation of heat transfer leads to a thermal gradient of 0.6 C between the front and back sides of the wafer and 4’ C for a I mm thick layer of tin, for the implantation conditions mentioned above. Since there is no thermal gradient at the interface between silicon and tin, placing the thermocouples in the liquid tin very close to this interface allows a very good knowledge and monitoring of the temperature at the front surface of the wafer. The above considerations are valid for a uniform incident power density on the wafer. Since in fact the beam is rapidly scanned across the wafer we must estimate the instantaneous temperature raise under it. The beam size is about 0.8 x 0.3 cm’ with a slow scan speed along the smaller dimension of 560 cm s ’ and a fast perpendicular one of 4800 cm s ‘. the scanned arca being 4 x 4 cm”. This leads to a maximum instantaneous power density of 500 W cm ~’ for a 600 /tA beam current and an cncrgy of 200 keV. Solving the problem of heat conduction in a semi infinite medium with the energy deposited at the surface of the wafer ’ ‘, the temperature increase at the surface is









where P is the incident power density, I the penetration depth of the oxygen ions in the silicon target, D and K the thermal diffusivity and conductibility of silicon and

I’ erfc .Y = ’

4 L

Expanding leads to

(I +2s’)erfc

.\- -

z L rr

(2) in series and considering


.~e~-” I

only the tirst order term

I thermocouples & clamping_ ring silicon wafer --\ ion beam -

Figure 1. Schematic of wafer holder





(3) From beam geometry considerations we find t = I70 ,us leading to a local temperature increase of 5 C. After passage of the beam the temperature falls very rapidly to the average one (90% of the extra 5 ‘C are lost in a few milliseconds). In conclusion this system allows a very good control of the temperature of the wafer during the implantation, even with very high (50 PA cm ‘) current densities. The layers of oxide (one formed by the implantation, the other one deposited at the back side of the wafer) do not affect the heat transfer from the surface of the wafer to the layer of tin, and local heating under the ion beam IS negligible.

M Bruel er a/: Formation of buried insulating layers by high dose oxygen implantation under controlled temperature conditions

3. Results

depth in Si (x lO$

(lOO), n type, 10 R cm resistivity silicon wafers were implanted at an energy of 200 keV and a dose of 1.7 x 10’s (160+ ions) cm-‘. The dose was measured after the implantation by nuclear reaction t2. The implantation was done with the wafer not tilted in order to minimize radiation damage, with a current density of 50pAcm-‘, at a temperature of 600 + 5°C. The beam scanning was reduced to a 4 x 4 cm’ area in order to achieve this high current density and thus reduce the implantation time. The implantation was done through a 2 x 2 cm2 silicon aperture in order to avoid contamination of the sample by sputtering13. After implantation, one part of the sample (referred to as B) was annealed at 1150°C for 2 h, and another part (referred to as C) was annealed at 1185°C for 6 h. Both anneals were done under nitrogen, and with a 600 nm thick plasma-deposited SiO, capping. The as implanted sample will be referred to as A. All these samples were characterized by Rutherford backscattering spectroscopy (RBS) and a cross-sectional transmission electron microscopy (XTEM). 3.1. As-implanted (sample A). Figure 2 shows the (100) aligned and random backscattering spectra from samples labelled A, B and C. In this figure only the highest energy portion of the experimental spectra corresponding to scattering from Si atoms in the top layer and in the beginning of the buried SiO, layer are shown. From the random spectrum, the top surface Si layer is 400 nm thick and the buried oxide layer is 200 nm thick from the part not shown of the spectrum. These thickness calculations are based on tabulated stopping cross-sectionst4 for the He+ ions and Bragg’s additivity rules in SiO,. Now, considering the aligned spectrum of the as-implanted sample, the minimum yield xmi, defined as the ratio of aligned to random yield just behind the surface peak is 5% compared to 2.5% in non-implanted single crystal silicon. Figure 3 shows the dechannelled fraction defined as the ratio of channelled to random yield calculated at each channel. A depth scale is also plotted, assuming that the energy loss of the He+ ions is the same in the (100) aligned and random directions. For this sample the dechannelled fraction is constant (about 8%) only in the first 60 nm behind the surface peak and it progressively degrades towards the Si-SiO, interface. An important hump,

4He+1500 keV 91.b=165°







I xI

I x I x lx

: as-implanted . B : annealed 1150



+ A

OC, 2h

. C :annealed 1185 OC, 2h

channel number (2.16 ksV/channel)

Figure 2. Random and (100) channelling after annealing

RBS spectra. A: as implanted. B: at 1150°C for 2 h. C: after annealing at 1185°C for 6 h.













.45 .40 g .35 = $j .30 .k v P) .25i

+ A B

: as-implanted : annealed 1150






II * *

OC, 2 h

1 c :annealed 1165 OC, 6 h

channel number

Figure 3. Ratio of aligned and random yield as a function of the depth using data from Figure 2. A: as implanted. B: after annealing at 1150°C for 2 h. C: after annealing at 1185°C for 6 h.

corresponding to the damage peak near this interface is also present (see Figure 2). Figure 4a is a bright field micrograph from XTEM. In agreement with the RBS spectra, it shows a 200 nm thick buried oxide layer with a sharp front interface, and a tendency to banding at the lower interface, as already described”. The top 400 nm silicon layer is very defective displaying a rippled aspect, as observed by Holland et al 5. Figure 5 is an enlargement of the same region in the top layer. The first 70 nm under the surface are nearly free of defects, except for some precipitates, probably SiO,. Going deeper into the layer a larger density of precipitates appears, giving the micrograph a granular aspect. These precipitates grow in size and decrease in density when they come closer to the interface with the SiO, layer. Some dislocations can also be seen near the interface. 3.2. Annealed at 1150°C for 2 h (sample B). The RBS aligned spectrum (Figure 2) for sample B shows an improvement in the crystallinity of the top silicon layer. xmin is now 3% and the dechannelled fraction (Figure 3) is lower than that for sample A on a larger thickness (about 120 nm). The damage peak at the interface is no longer present. The XTEM micrograph of sample B (Figure 4b) displays five distinct regions: (1) a region about 30 nm thick with very small (diameter N 100 A) precipitates just under the surface; (2) a 75 nm thick region free of any defect; (3) a 100 nm thick region with polyhedral precipitatesI having a diameter of between 5 and 25 nm; (4) another defect free layer about 15 nm thick; (5) a 200 nm thick, highly defective region with a very large number of SiO, precipitates, these precipitates are larger (diameters up to 50 nm) and more numerous. Comparing the channelled spectrum (Figure 2) and XTEM, we note that even though RBS is sensitive to precipitates in region 3, the dechannelling that they create is not very important. This is to be related to the fact that channelling occurs even behind the 200 nm thick layer of oxide. A previous study16 showed that there is no strain around these precipitates. These two observations 591

M Bruel

et a/



the IOU. density







of burted


of precipitates


to bulk





layers explain


is achieved

by htgh dose oxygen why


at the







i\ not


;t mixture



Tiltcon /one





B. but



.3’!,, M hich (‘(tigure


free silicon

I X0 nm. Tilting






top silicon Ie\s

reveals and

numerous and

Figure 3. Coos\-secttonal mtcrographs 01‘ sample tmplanted at 600 (‘. Energy: 200 kcV. Dose: t .7 x IO Ix ions cm ‘. a: as implanted. h: annealed at 1150C‘ for 2 h. C: annealed at 1185<’ for 6 h. 592





is still larger.

the thickness








are largctFigure


4c layt



precipitates have


and the

the top Glicon


of the dislocations

of the defective


fA)O ntn fat-

but they insert

At the hack

and the SiO,


75 nm to

on ;I length

the buried

between layer.

ha\c I

of dislocattons

B. The

is vq



;t density


Si. The

the ‘\ I IO,, ;IW\





of dislocalion


of the interface

of virgin

\+ idcncd






layer for

the precipitates


;I thickness



to that

7) ha\



the silicon








stilt contains


are the sanic









1 cm -. The


penetrating banding







4~) shw\

us (0 mcasut-c


X /tm -









ib vcq

.> ’ and

in region

the ~mplc



at about

<‘). In the

71 the general

the good




<‘ for 6 h (sample




C‘ (P-igure




at 1185

;I\ for




of sample

700 nm.


islatid~ clotig;ttcd iti the 100I


of dislocations


arc Ic\;\



IXwussion The




was kept by

at 600


c‘ throughout and



the trradtaticln results



tn good

M Bruel et al: Formation

of buried


layers by high dose oxygen

agreement with Wilson4 and Holland5 which show that starting the implantation with the wafer already at high temperature improves the quality of the silicon layer and particularly produces a low density of dislocations which we measured to be less than 10’ cm _ ’ after annealing. These results have to be compared to those of a similar specimen”: 1.8 x 1018 Of cm-*, 200 keV, Ti = 550°C but only heated by the incident beam power, where the density of threading dislocations was about lo9 cm-‘. The different annealing steps did not change significantly the density of dislocations, and in consequence it is the oxygen behaviour, mainly present as precipitates of SiO, (ref 16) which determines the top silicon layer quality improvement after annealing. In sample A a high density of these SiO, precipitates can be observed, they are rather small although their size increases with depth. The larger ones are elongated near the SiOJSi interface, their diameter decreasing towards the surface. In region 3 they are no longer visible. In the silicon layer after implantation, the oxygen concentration is higher than the solubility limit16.17 and a large number of point defects acting as nucleation centres are produced by irradiation. Thus all the conditions for the growth of many precipitates are present”.“. Nevertheless, the knowledge of their formation process is lacking and at the present time their size evolution and the tendency of alignment along the (100) axis are not understood. Under high temperature annealing, these precipitates are unstable”, and therefore they dissolve or coarsen, the higher the annealing temperature, the larger the diameter of the remaining ones*‘. An important consequence is that the redissolution occurs in the upper part of the silicon layer (region 3, Figure 4b). Even if the real mechanism is not well known, a reasonable hypothesis would be that once the precipitates are dissolved, the oxygen diffuses towards more stable SiO, regions to oxidize the silicon at the interface. These regions can be larger precipitates or the buried and capping (SiO,) layers. Using D=8.8x 10” cm2 s-l for the diffusion coefficient of oxygen in silicon at 1185”C*’ we find a diffusion length of 6 pm h ‘, much larger than the thickness of the layer (0.4 pm). Some precipitates remain just beneath the surface, but since this region is just 30 nm thick, from a technological point of view, it is not important since one of the first steps in integrated circuits (IC) fabrication is a surface oxidation. The most important and promising feature is the behaviour under high temperature treatment of Si/SiO, interface. The observed phenomenon of coalescence which occurs leads us to think that for longer (or better, higher temperature annealing) this interface will become sharp and flat.






Conclusion We have built and tested a system to control the wafer temperature during implantation. After implantation (O+ ions, 1.7 x 10 ” ions cm-*, 200 keV) at 600°C the top silicon layer is almost free of linear defects such as dislocation loops but contains a lot of small (

’ P L F Hemment, Mat I&S Sot Symp Proc, 33,41 (1984). ’ P L F Hemment, E Maydell-Ondrusz, K G Stephens, J Butcher. D Ioannu and J Alderman, Nucl Instrum Metk, 209/210, 157 (1983). 3 K Das, G Shorthouse. J Butcher and K V Anand. Mirroelectronic J. 14. 88 (1983). 4 S R Wilson and D Fathy, J Electronics Mats, 13, 127 (1984). ’ 0 W Holland, T P Sioreen. D Fathv and J Naravan. ADD/ . . Phvs , Lefts. 45. 1081 (1984). 6 M E Mack, Ion Implantation Equipment and techniques, Springer Series in Electrophysics, vol 11, p 221, Proceedings of the 4th Int Conf Bertchtesgaden, Fed Rep Germany (13-17 September 1982). ’ J C Bailer, H J Emeleus and R Nyholm (Eds), Comprehensive inorganic Chemistry, vol 2, p 746. Pergamon Press, Oxford (1973). * I V Nistiryuk and P P Seregin, Soviet Phys Solid State, 17, 768 11976). 9 Dijkion Data, 4, 345 (1970). “Y S Tovioukan, R W Powell, C Y MO and P G Klemens (Eds), Thermophysical Properties of Matter, vols 1, 2, The TRPC data series, IFI/Plenum, New York (1970). l1 Carslaw and Jaeger (Eds), Conduction of Heat in Solids, Oxford University Press (1959). I2 M Dubus J Margail and P Martin, 7th Int Conflon Beam Analysis. Berlin (RFA)‘July 1985. I3 P L F Hemment, Vacuum, 29,439 (1979). I4 J Ziegler (Ed), The Stopping and Ranges of Ions in Matter, vol 4, Pergamon Press, Oxford (1977). I5 C G Tuppen, M R Taylor, P L F Hemment and R P Arrowsmith, Appl f’hys Letts, 45, 57 (1984). I6 C Jaussaud, J Stoemenos, J Margail, M Dupuy, B Blanchard and M Bruel, Appl Phys Lefts, accepted for publication (1985). I7 P L F Hemment, E Maydell-Ondrusz, K G Stephens, J A Kilner and J Butcher, Vacuum, 34, 203 (1984). ” F Shimura and H Tsuya, J Electrochem Sot, 129, 2089 (1982). l9 R B Fair, J appl Phys, 54, 388 (1983). ” I H Wilson, Nucl Instrum Meth, Bl, 331 (1984). 21 M J Binns W P Brown, J G Wilkes, R C Newman, F M Livingston, S Messoloras’and R J Stewart, Appl Phys Lefts, 42, 525 (1983).