`
`SILICON PROCESSING
`
`FOR
`
`THE VLSI ERA
`
`VOLUMEI:
`
`PROGESS TEEHNDLOGY
`
`Second Edition
`
`STANLEY WOLF PM].
`
`RICHARD N. TAUBEB Ph.D.
`
`LATTICE PRESS
`
`Sunsel Beach, California
`
`2?
`
`
`\l
`ii
`
`
`
`IPR2015—01087
`
`Micron V. MIT
`
`MIT EXHIBIT 2016
`
`
`
`
`
`
`
`This publication is based on sources and information believed to be reliable, but the
`authors and Lattice Press disclaim any warranty or liability based on or relating to the
`contents of this publication.
`
`Published by:
`
`LATTICE PRESS
`Post Office Box 340
`
`Sunset Beach, California 90742, USA.
`
`Cover design by Roy Montibon, New Archetype Publishing, Los Angeles, CA.
`
`Copyright © 2000 by Lattice Press.
`All rights reserved. No part of this book may be reproduced or transmitted in any form
`or by any means, electronic or mechanical, including photocopying, recording or by any
`information storage and retrieval system without written permission from the publisher,
`except for the inclusion of brief quotations in a review.
`
`Library of Congress Cataloging in Publication Data
`Wolf, Stanley and Tauber, Richard N.
`
` DISCLAIMER
`
`Silicon Processing for the VLSI Era
`Volume 1: Process Technology
`
`Includes Index
`
`1. Integrated circuits—Very large scale
`integration. 2. Silicon. I. Title
`
`ISBN 0—9616721—6~1
`
`9 8 7 6 5 4 3 2
`
`PRINTED IN THE UNITED STATES OF AMERICA
`
`
`
`
`
`CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`207
`
`
`5.7 DVD 0F METALS, SILICIDES, AND NITRIDES FOB ULSI APPLICATIUNS
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`vvzwuuwzt—LLLW
`
`.—
`
`CVD has also been pursued as a thin-film deposition technology for a number of metals used as
`interconnects in ULSI, including tungsten, aluminum, titanium, and copper. Of this group, only
`CVD of tungsten has found wide acceptance as a production process in multilevel interconnect
`structures for technologies with feature sizes below 1 pm. As of 1999, CVD of the other metals
`has not been able to displace PVD (sputtering) as the main deposition technology. But the
`potential advantages of CVD (mainly with respect to step coverage and gap filling) continue to
`drive efforts to develop successful CVD techniques for the others. Here the recent status of the
`following conductor materials deposited by CVD:
`tungsten, tungsten silicide, and titanium
`nitride is discussed. The CVD of copper is covered in Chap. 15.
`
`5.7.1 CVD oi Tungsten (W)
`
`Refractory metals (i.e., W, Ti, Mo, and Ta) have been investigated for various applications in
`the interconnect systems of silicon ICs.3’78 Their resistivities are higher than those of Al and its
`alloys, but lower than those of the refractory metal silicides and nitrides. Of these metals,
`tungsten (W) ended up being adopted for several interconnect applications, although not as a
`stand-alone gate material nor as a global interconnect material. Instead it was selected to
`perform two other roles in IC interconnect systems. The most important of these is that of a plug
`(i.e., a material that can completely fill vias between aluminum films, as well as contact holes).
`It was chosen as a plug material because CVD—W provided better via—filling capabilities than
`did PVD aluminum at the time they were first implemented. That is, if contact holes and vias
`have minimum dimensions that exceed ~1.0 ,um, they can be adequately covered with Al films.
`Because PVD films until recently could not completely fill contact holes and vias, such PVD—
`filled structures were called non—filled-contacts (and ~vias). For technologies in which the
`minimum feature size is smaller than 1 pm the aspect ratios of contact holes and vias become so
`large that non-filled contact holes are no longer acceptable (due to excessive thinning of the
`PVD-Al films as they run down the sides of these steep and deep holes). A method to
`completely fill the contact holes was therefore sought. If it would have been possible to
`completely fill these holes with Al as well as was possible with CVD—W at that time, Al would
`have been used. But, since it was not, CVD-W prevailed, at least for several generations of
`technology (i.e., down to 0.18 pm). The second, somewhat lesser role of CVD-W is to serve as
`a local interconnect (for many of the same reasons listed above). The lower conductivity of W
`films compared to those of Al (or Cu), however, limits their use to short interconnect paths, and
`Al or Cu are retained for use as global interconnect materials. Here we describe the details of
`CVD—W film deposition. Note that although processes for forming CVD-W films both in
`selective79 and blanket deposition80 modes have been developed, most W applications in IC
`production use the latter.78 Selective~W, while appearing to possess many advantages, has not
`been widely adopted. This is because problems with loss of selectivity and substrate damage
`have not been completely overcome. As such, we will focus mainly on the blanket-W deposition
`process.
`
`CVD tungsten emerged as the most widely used of the refractory metals for interconnect
`applications for several reasons. First, it exhibits lower bulk resistivity than Ti 0r Ta, and about
`the same resistivity as Mo. (Note that the resistivity of tungsten films deposited by the hydrogen
`reduction of WF6 is in the range of 7-12 [JO—cm.) Second, it exhibits high thermal stability,
`
`
`
`
`
`208 SILICON PROCESSING FOR THE VLSI ERA
`
`having the highest melting point of all metals (3410°C). Third, it has low stress (<5x109
`dyn/cmz), excellent conformal step coverage, and its thermal expansion coefficient closely
`matches that of silicon. Finally, it has excellent electromigration and corrosion resistance but
`none of the stoichiometry control problems that often plague silicides. Some of its
`disadvantages include: a) its resistivity, although 200 times lower than that of heavily-doped
`polysilicon, is still about twice as high as that of Al—alloy films; b) W films adhere poorly to
`oxides and nitrides; c) oxides form on W films when temperatures exceed 400°C (and thus care
`must be exercised to prevent oxidation, especially during subsequent dielectric deposition); and
`d) silicidation of the tungsten occurs if it
`is in contact with silicon and is exposed to
`temperatures greater than 600°C.
`
`2 WF6 (gas) + 3 Si (solid) —> 2 W (solid) 4- 3 SiF4 (gas)
`
`This reaction is normally produced by allowing the WF6 gas to react with regions of exposed
`CVD tungsten
`Process
`
`(8.29]
`
`CVD Tungsten
`A
`
`
`
`
`
`
`
`
`
`
`
`5.7.1.1 CVD Tungsten Chemistry: The chemical vapor deposition of tungsten is generally
`performed in cold-wall, low~pressure systems (an example is shown in Fig. 6-46). Although
`tungsten can be deposited either from WF6 or WC16, tungsten hexafluoride (WF6) is better
`suited as the W source gas, since it is a liquid that boils below room temperature (17°C). On the
`other hand, WC16 is a solid that melts at 275°C. The low boiling point makes WF6 much easier
`to meter into process chambers in a reproducible way. The WF6 compound is produced by the
`reaction between tungsten and fluorine, and after several refining steps a very pure product can
`be routinely obtained (99.999%). The main drawback of WF6 is its high cost. In fact, it accounts
`for about 50% of the total cost of the blanket CVD—W process. The components within which
`the WF6 is flowed from its container to the reaction chamber must also be heated, to prevent
`WF6 condensation. Nevertheless, WF6 is the W source typically employed in all three of the
`reactions used in CVD-W, namely reduction of WF6 by: 1) silicon; 2) hydrogen; and 3) silane,
`since it can be reduced by all of these materials. The silicon reduction is given by:78
`
`Novellus Systems, Inc.
`
`
`
`
`
`null End ("mu
`blot
`
`Dual Luadlack
`Cassette Module
`
`
`
`Fig. 6-46 Schematic drawing of a Novellus Concept Two CVD tungsten deposition system. Courtesy of
`
`
`
` CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`
`
`209
`
`
`solid silicon on a wafer surface at a temperature of about 300°C. The silicon surface must be
`
`quite clean (i.e., covered with less than 1 nm of native or chemical oxide) to permit initiation of
`
`the reaction. About two volumes of Si are consumed (and volatilized as SiF4) for each volume
`
`of W formed. However, the reaction is self—limiting when the film reaches a thickness of 10w15
`
`nm,9'10 since the W film serves as a diffusion barrier between the Si and the WP6 once this
`thickness is reached. No deposit occurs on regions of the wafer covered with SiO2 during this
`reaction.
`
`The overall hydrogen reduction reaction is given by:
`
`WF6 (gas) + 3 H2 (gas) —> W(solid) + 6HF (gas)
`
`(6.30]
`
`The hydrogen reduction may result in either selective or non-selective (blanket) deposition of
`W. The process is carried out at reduced pressures, usually at temperatures below 450°C.
`Typically the process is carried out in a large excess of hydrogen. The change in the Gibbs free-
`energy for the reaction is -278 kcal/mole, and the film growth rate is reaction-rate limited up to
`a temperature of about 450°C. As noted above, the resistivity of W-films deposited by the
`hydrogen reduction is in the 7—12 ,uQ—cm range. The selective deposition reaction requires good
`nucleating surfaces. Silicon, metal, and silicide surfaces provide good sites, while SiO2 and
`Si3N4 (especially at low temperatures), do not. On a silicon surface the deposition starts by the
`Si reduction, but once the W thickness becomes self-limiting, the H2 reduction takes over. At
`the outset of the deposition the carrier gas used is Ar. After the Si reduces WF6, H2 is added to
`the gas flow, and the Ar flow is stopped. This deposition process is not self-limiting in
`thickness. A practical blanket-W process is more complex than the selective one because W
`does not adhere well to SiOz. Thus, an adhesion layer is first deposited onto the SiOz, and the W
`is then deposited onto it.
`The overall silane reduction reaction is given by:
`
`2 WF6 (gas) + 3 SiH4 (gas) w) 2 W (solid) + 3 SiF4 (gas) + 6 H2 (gas)
`
`[5.31]
`
`This reaction (LPCVD at ~300°C) is widely used to produce a W nucleation layer for the
`hydrogen reaction. Better nucleation is consistently obtained with the silane reduction on most
`surfaces, including TiN. Note that in the silane reduction, if the gas phase mixture has excess
`WF6, W films are formed by the reaction, but if there is a silane excess, WSix films are
`deposited. In addition,
`the silane reduction of Eq. 6—31 appears to be in conflict with
`thermodynamic predictions. That is, the hydrogen formed in this reaction is predicted to react
`more readily with the WF6 in the gas phase (to form HF) than is WF6 to react with the SiH4.
`However, experimental data indicates the reaction proceeds according to Eq. 6-31. This implies
`the SiH4 reduction proceeds far from equilibrium, and the formation of HF via reaction with
`WF6 is kinetically blocked (i.e., it is slow compared to the silane reaction with WF6 to form W
`and SiF4).
`As noted earlier, the chemical vapor deposition (CVD) of tungsten is performed in cold—
`wall, low-pressure CVD (LPCVD) reactors. The wafer is held on a heated chuck opposite a
`showerhead through which a premixed flow of WF6 and one of the reducing agent gases (H2, or
`SiH4) is injected. Note that hot—wall systems are not used for several reasons. First, in hot—wall
`systems W would also deposit on the quartz furnace-tube walls. Since W doesn't adhere to Si02
`such films would soon delaminate from the walls and create particles. Frequent cleaning would
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`6.7.1.2 Blanket CVD W and Etchback: Tungsten can be deposited by CVD using either a
`selective or blanket process. Only blanket~W deposition has emerged as a production—proven
`process, despite the fact that it
`is more complex and expensive than selective CVD—W.
`Acceptance of the selective-CVD process has been slowed because some of its problems have
`not been completely overcome, including those involving loss of selectivity of deposition and
`lateral encroachment and wormholes. On the other hand, blanket CVD-W and etchback (or
`
`CMP) has found widespread use for contact-hole and via filling applications in IC technologies
`below about 1 ,um. Both applications require adherent, low—cost films. The plug applications,
`however, call for high step-coverage and thickness uniformity, but can tolerate higher resistivity
`than is needed for W-films used as interconnects. For filling contact holes this W—plug-
`
`formation process has six steps:
`
`1. In situ surface pre-clean;
`2. Deposition of a contact forming layer (typically 3 Ti film formed by sputtering or CVD);
`3. Deposition of an adhesion/barrier layer (typically a TiN film formed by sputtering or
`CVD);
`
`4. Blanket-CVD of the W film (typically a two-step deposition process);
`5. Etchback of the W film:
`
`6. Etchback of the adhesion and contact—forming layers.
`
`210 SILICON PROCESSING FOR THE VLSI ERA
`
`be necessary to keep this problem under control. Furthermore, once the fused-silica furnace
`walls are coated with W they become opaque. IR radiation from the heating coils is no longer
`transmitted as efficiently through the walls as when the fused silica is transparent. Hence,
`temperature control of the wafers becomes a problem. Note that in cold-wall reactors, the wafers
`and their holders are the only hot objects in the chamber, with the wall temperatures being kept
`well below the temperature needed to drive the deposition reactions (<150°C). Thus,
`the
`problem of deposition on the chamber walls is curtailed (but note, not entirely eliminated).
`However, a large temperature gradient exists between the heated wafers and the chamber walls,
`which may also create temperature control difficulties in such reactors.
`
`to these insulators. Thus, a method which allows good adhesion of CVD—W to the substrate is
`
`The surface pre-cleaning step is designed to remove any native SiO2 material on the silicon in
`the contact holes or aluminum oxide on the aluminum in vias. Trends are moving toward
`making this an in situ cleaning step, so that the wafer surface is not exposed to atmosphere
`between the cleaning step and the deposition of the contact and adhesion layers. Most such in
`situ processes involve either an Ar sputter clean (discussed in Chap. 11) or a "soft" plasma
`clean. After the cleaning step, the contact and adhesion layers are deposited, either by CVD or
`sputter deposition. Currently, the most widely used materials for the contact and adhesion layers
`are Ti and TiN, respectively. The thin layer of Ti (30w50 nm thick) is used under the TiN
`adhesion layer because it provides lower contact resistance to the silicon substrate than is
`possible with TiN.100 The adhesion layer is needed because of the extremely poor adhesion of
`CVD W to such insulators as BPSG, thermal oxide, plasma—enhanced oxide and plasma—
`enhanced silicon nitride. Tungsten, however, adheres well to TiN, and TiN in turn, adheres well
`
`
`
`CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`
`211
`
`achieved.* A detailed discussion of depositing these layers is provided in later sections of this
`chapter, and etching of the W and TiN is described in Chap. 14.
`Next a layer of CVD-W is blanket deposited. For the W-plug applications the contact holes
`or vias must be completely filled. For this to occur the step coverage (in this case, defined as the
`ratio of the thickness at the sidewall at half depth of the hole to that of the nominal W thickness
`on the wafer surface) must be 100%. Otherwise, keyholes, or voids, will be formed (Fig. 6—47a),
`and these voids will become exposed during the subsequent W etchback step. If that happens,
`the contact holes or vias are no longer completely filled at the end of the W—plug formation
`process. In general, the hydrogen reduction gives better step coverage than the silane reduction,
`although the deposition rate of the former is significantly lower. However, W layers formed by
`the hydrogen reduction do not nucleate reliably on the TiN adhesion layers. So,
`in most
`commercial reactors a two—step blanket—W process is typically employed.”81 A thin layer of W
`is first nucleated using the silane reduction (several tens of nm thick). Then, the hydrogen
`reduction reaction is used to deposit the remainder of the blanket—W film. The silane reduction
`step is carried out at relatively low pressures (~1 torr), while the hydrogen reduction uses higher
`pressures (25-80 torr). Such higher chamber pressures during the hydrogen reduction process
`significantly improve the step coverage and produce void-free filled contact holes and vias (Fig.
`6—47b). As described in Chap. 3, the total chamber pressure is increased by throttling the
`pumping speed. The process is run at temperatures around 450°C, where the deposition still
`operates in the surface—reaction~rate—limited regime. Details of the higher~pressure hydrogen-
`reduction deposition process are give in Ref. 78. To get complete filling of the contact holes or
`vias, the slope of the contact sidewalls should not exceed 90°. A blanket~W processes using
`SinF2 and WF6 as the reactants has also been studied and the results published, with some
`advantages claimed over the hydrogen reduction process.82
`
`blanket CVD W film deposited into a trench without void formation.
`
`* TiN has been reported to have the best set of properties for this adhesion/barrier-layer application, and
`the best resistance to the etch gases used to etchback the CVD W.
`
`Fig. 6-147 a) Blanket—CVD W films deposited into a trench with a process that results in voids.83 b) A
`
`
`
`212 SILICON PROCESSING FOR THE VLSI ERA
`
`Several other issues encountered in the blanket-W process should be addressed, including:
`1. Stress in CVD-W films.
`
`. Deposition of W on the wafer backside and edges.
`
`. Particle formation resulting from the CVD-W process.
`. Resistance of the W-plugs in contact holes and vias.
`
`. Failure mode of the blanket-W process ('volcanoes‘).
`
`Stress in CVD-Wfilms: Thick blanket-W films exhibit high stress, and such stressed films may
`cause wafer warpage. However, this is generally not a concern for the W-plug process, since the
`majority of the film is removed during the etchback step. If the W layer is to be used to form
`interconnect lines the stress issue must be considered. A two—step deposition process has been
`suggested by Clark that produces W film with stress levels below the values that will cause
`wafer warpage.84
`
`each wafer is pro— cessed, an in situ NF3 plasma clean is also performed to remove any W on the
`chamber walls.
`
`Concept One-W® system. Courtesy of Novellus, Inc.
`
`Backside W Deposition: If W deposits on the back or edges of the wafer where no adhesion
`layer exists (as a result of reactants getting under the susceptor or heater block), this material
`will delaminate and produce unwanted particulates in the deposition chamber. Because W
`exhibits such good conformal coverage, backside deposition is inevitable unless deliberate
`measures are implemented to prevent or counteract it. In one reactor design (the Novellus
`Concept-One-W®), the wafer is vacuum clamped at its center to the pedestal on which it sits.
`Inert gas is continuously flowed at the wafer edges from behind the wafer to prevent the WF6
`from reaching the wafer backside and edges (Fig. 6-48). In another reactor design,
`it
`is
`recognized that backside deposition may occur, despite taking measures to purge the region
`beneath the wafer with inert gas. However, to ensure that no W remains, a plasma etch step after
`W deposition is used to clean the backside of each wafer (Applied Materials 5000 WCVD®).
`This etch is carried out so that the W deposited on the wafer frontside remains untouched. After
`
`Particle Formation: Particles in the blanket-W process can also arise from two other sources.
`First, if the adhesion layer is sputter deposited, the wafer may be held during that sputtering
`procedure by clips at the wafer edges. Thus, the wafer area under the clips does not get covered
`Gas/RF
`:Showerhead
`l'fi‘i“t—t—t'“t“i‘t“l”l—t“t‘t“t—t*l*t
`
`Aluminum
`Pedestal
`
`Backside Gas
`
`Thermal
`Insulator
`
`Heater Element
`
`Vacuum
`
`Flu. 6-48 The clampless method used to prevent deposition of W on the wafer backside in the Novellus
`
`
`
`
`
`CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`213
`
`
`
`is later blanket deposited on such regions will flake off and
`by the adhesion layer, and W that
`
`produce particles (either in the W-deposition chamber or in a later process step). Second, the W
`
`
`deposited on the chamber walls can also build up after a number of runs until it spalls (even in
`
`
`cold-wall reactors), again producing particulates in the chamber.
`In the Applied Materials
`single-wafer W tool, the etch process that removes backside W is used a second time for in situ
`
`
`cleaning of the chamber.85
`
`
`
`
`
`
`
`
`
`
`
`
`ay
`he
`
`aU
`
`SE
`
`
`
`
`Series Resistance of W~plugs2 For 1 ,um contacts the total resistance of W-plugs is about 0.5 9 if
`the resistivity of the W film is a typical 10 ,uQ-cm. When such plugs are used to fill contact
`holes, this resistance value is negligibly small compared to the contact resistance (which is of
`the order of 20 9). For vias, however, this plug resistance may be significant, as the via contact
`resistance is only of the order of 0.5 9. Furthermore, the resistance of the W—plug increases as
`the features shrink, becoming ~5 Qfor 0.3 pm contacts and vias.86 W plugs therefore become
`less desirable for deep-submicron technologies, and alternatives, such as Al or Cu plugs, are
`being pursued as replacements.
`
`Failure Mechanism Triggered During the Blanket-W Deposition: As noted, a Ti layer is usually
`used under a TiN layer. Besides serving as an adhesion layer, the TiN also acts as a diffusion
`barrier. That is, it also prevents the WF6 from reacting with the Ti beneath it.* However, if there
`is any defect in the TiN, or if the grain boundaries of the TiN are not properly "stuffed" (to
`prevent them from becoming high-speed diffusion paths for WF6 molecules), then the WF6 will
`penetrate the TiN layer and react with the Ti film (Fig. 6-49).87 Once this reaction begins, the Ti
`under the TiN is consumed by the formation of a volatile by-product Tin and the simultaneous
`nucleation of W on the oxide surfaces. The depositing W pushes away the segments of the TiN
`no longer supported by the Ti. More Ti
`is thus exposed and the reaction continues. The
`nucleating W grows into thick W mounds near the point of penetration, forming humps greater
`than 1 am in size. Since these humps or "volcanoes" are so large, they are not removed
`completely by the etchback process, and the residual conductive material leads to intralevel and
`interlevel shorts (see Fig. 6—50). The WF6 may now also attack the uncovered Si at the contact,
`causing the contact to fail.
`The columnar growth of PVD TiN films cause them to be more porous at the upper corners
`of contact holes and vias and thus prone to penetration there. The issue of volcanoes becomes
`more severe as the contact window size decreases, since the upper corners tend to be sharper.
`Rounding of the upper corners of contact holes and vias, or use of CVD TiN alleviates the
`problem of volcanoes in the blanket-W process.
`
`6.7.2 Chemical Vapor Deposition oi Tungsten Silieide [WSixl
`
`Various refractory metal silicides including WSix, TaSiz, and MoSi2 were explored in the 1980s
`as possible shunt materials in polycide structures. Tungsten silicide WSix films ended up being
`most widely adopted for this application, with CVD being the chosen method of forming WSiX
`thin films. Polycides with WSix films now find widespread use as word-line and bit—line
`interconnects in IC memory chips. WSix has also found some application as a stand-alone
`adhesion layer for blanket CVD-tungsten films. A process sequence for producing the WSiX
`
`** The chemical reaction of WF6 with Ti is as follows:
`
`2 WF6 (gas) + 3 Ti (solid) ~> 2 W (solid) + 3 TiF4 (gas)
`with a AG = -1037 kJ/mole at 500°C, which means that the reaction will proceed vigorously.
`
`
`
`
`
`
`
`214 SILICON PROCESSIVG FOR THE VLSI ERA
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`polycide gate structure is shown in Fig. 6—51. Note that WSiX is blanket-deposited onto the
`doped polysilicon film, and then this multilayer is etched to form the polycide gate structure.
`The CVD deposition of WSix won out over other candidate processes (i.e., evaporation and
`sputtering) for the following reasons: a) this process can produce high-purity WSix films
`without the need for high—vacuum deposition equipment; b) its throughput is acceptable; 0)
`better step coverage is obtained than with PVD; and (1) good wafer—to—wafer and run-to—run
`
`Reprinted with permission of Solid State Technology, published by PennWell.
`
`Fig. 6-49 Tungsten volcano from barrier failure: a) high stress and weak TiN at a sharp comer is attacked
`by WF6; b) Ti reacts with the WF6 and the TiN is peeled back; c) W deposits on both sides of the peeled
`back TiN, forming the volcanoes; d) contact hole with rounded top corners stays intact.
`
`Fig. 6-50 “Volcanoes” on vias where the TiN barrier has failed, and W is deposited under the TiN layer.88
`
`
`
`CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`
`215
`
`
`
`
`
`
`
`
`
`
`Fig. 6-51 Process sequence for producing a WSix polycide structure. a) Gate oxide growth; b) Polysilicon
`
`and WSix deposition by CVD; c) Pattern the polycide.
`uniformity is attainable. The first chemical reaction used to deposit WSix was the following, and
`it is capable of producing films with the benefits listed above:
`[6.32]
`WF6 (gas) + ZSiH4 (gas)
`——> WSi2 (solid) + 6HF (gas) + H2 (gas)
`Tungsten silicide film deposition using this reaction is carried out at pressures between 50-300
`mtorr (LPCVD) and at deposition temperatures between 300-400°C.89 This reaction is similar to
`the silane reduction of WF6 used to produce CVD-W films. As mentioned earlier, if it is desired
`that WSiX be deposited, a higher flow rate of SiH4 is needed. This ensures that the deposited
`film is WSiX and not W. During film deposition, WSi2 forms with excess Si collecting in the
`grain boundaries. Since an excess of silicon exists in such films, the chemical formula to denote
`them is given as WSiX. However, unless x 2 2.0, WSix films formed with this reaction are prone
`to cracking and peeling away from the underlying poly during later high—temperature steps. That
`is, the excess Si in the as—deposited silicide film avoids the possibility of consuming some of the
`underlying poly material, which could cause the above problems. Thus, in practice a SiH4/WF6
`flow ratio of more than 10 is used to ensure a deposition with x ~2.2—2.6. As-deposited WSix
`films exhibit high resistivity (~500 pQ—cm), but this drops to about 50 ,uQ-cm after an RTP
`anneal at 900°C. Note that the resistivity of CVD WSix also depends on its stoichiometry as is
`shown in Fig. 6-52, where x varies from 2.2 to 2.6. The resistivity of the film increases as it
`becomes progressively richer in Si.
`It was found that WSix films formed by the reaction of Eq. 6—32 had high fluorine
`concentrations (~1020/cm3). This creates problems when such films are used with gate oxides
`thinner than about 20 nm, because low-field oxide breakdown and threshold-voltage shifts are
`observed.35 Such problems arise because some of the fluorine gets incorporated into the gate
`oxide during anneal of the polycide stack. As a result, an alternative chemistry using
`dichlorosilane (DCS) or (SinClz) instead of SiH4 was developed for CVD of WSix;90
`[6.33]
`WF6 (gas) + 3.SSiH2C12 (gas) —«a WSi2 (solid) + 1.551c14 (gas) + 6HF (gas) + HCl (gas)
`This DCS process is also LPCVD and is carried out at 570—600°C. The fluorine content is much
`lower than in films formed with the silane reduction of Eq. 6-32, and the chlorine content is also
`low. The resistivity is comparable in both types of films but the step coverage is better in DCS
`
`
`
`
`
`
`Q5.
`
`Vms—A:v0
`
`
`
`
`
`
`
`
`410
`
`430
`
`450
`
`390
`T (°C)
`Flu. 5-52 Resistivity of CVD-WSiX versus x, where x varies from 2.2 to 2.6. Courtesy of Genus Inc.
`films. Peeling and cracking is less severe in WSiX films formed from the reaction of WF6 and
`DCS, and DCS is thus replacing silane in CVD WSix.
`Exposure of the WSix films to high-temperature oxygen ambients (900°C) will cause a
`dense adherent layer of Si02 to grow on the silicide while the polycide structure underneath
`remains intact. The oxide forms by reaction of oxygen with the excess Si dissolved in the
`silicide film, and continues after this silicon is consumed by diffusion of additional silicon from
`the polysilicon to the surface of the WSiX. The oxide-formation reaction occurs at the
`oxide/silicide interface.
`
`CVD WSix films are deposited in cold-wall reactors. Batch processes were initially used but
`single wafer systems have also been introduced. A process that
`integrates polysilicon
`deposition, polysilicon doping, and WSix deposition in a single cluster tool is available.91
`Advantages of such a configuration include cleaner processing and less cleanroom floor space.
`6.7.3 CW] 0! Titanium Nill’ide (TiN)
`
`216
`
`SILICON PROCESSING FOR THE VLSI ERA
`
`
`
`GrowthRate(nm/min) AO
`
`MO
`
`\
`
`
`
`330
`
`350
`
`370
`
`also satisfactorily perform this role when deposited into deep-submicron high—aspect-ratio
`
`In ULSI global interconnects (which use Al-alloy and Cu materials for the interconnect lines,
`and possibly also for contact and via filling), as well as in blanket—W processes that form W
`plugs and W local interconnects, supporting-role films are inevitably fabricated beneath these
`metals. Such films (consisting of refractory metal silicides and nitrides) serve two major
`purposes, depending on the metal they assist. For Al-alloy films they act as diffusion barriers to
`prevent the formation of intermetallic compounds that would destroy the contact behavior (i.e.,
`by shorting the shallow junctions below the contact or increasing the series resistance between
`the metal and silicon). For blanket layers of W and Cu they serve not only as diffusion barriers
`but also as adhesion layers (i.e., films to which Cu and W adhere well, and which in turn adhere
`well to the oxide below). Note that their role as diffusion-barriers in blanket-W processes
`actually has two purposes: 1) to prevent reaction of the contact—resistance—enhancing—Ti layer
`and WF6, which would cause "volcanoes" (see Sect. 6.6.1.2); and 2) to protect the silicon
`contact from damage by reaction with WF6. When Cu interconnects are used, such banriers must
`prevent Cu diffusion into the underlying Si substrate. In any case, these barriers must retain their
`function over the full range of temperatures encountered after their deposition, and they must
`
`
`
`
`
`
`CHEMICAL VAPOR DEPOSITION OF AMORPHOUS AND POLYCRYSTALLINE FILMS
`217
`
`
`
`
`
`
`
`
`
`
`
`‘ and
`
`se a
`eath
`the
`tom
`
`the
`
`but
`:on
`
`'91
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`contact holes and vias. As mentioned earlier, films that are required to perform both adhesion
`and diffusion—barrier functions are also termed liners. The deposition of liner films is done by
`either PVD or CVD. PVD technology has been the traditional deposition method, and work to
`improve this method continues. However, as shall be discussed in detail in Chap. 11 (and briefly
`here), CVD and ion metal plasma (IMP) PVD offer potentially better step coverage over sharp
`upper edges of contacts and at the bottom corners of deep submicron contacts and vias. Here we
`discuss the CVD of the most widely used liner film titanium nitride (TiN).
`TiN is an attractive material as a diffusion barrier in silicon ICs because it behaves not only
`as an impermeable barrier to silicon, but also as a barrier to other substances attempting to
`diffuse through it. In the latter cases, the activation energy for the diffusion of other impurities
`in TiN is high (e.g., the activation energy for Cu diffusion into TiN thin films is 4.3 eV, whereas ,
`the normal value for diffusion of Cu into metals is only 1 to 2 eV). TiN is also chemically and
`thermodynamically very stable (its melting point is 2950°C), and when in thin film form it
`exhibist one of the lowest electrical resistivities (25-75 yQ-cm) of the transition metal carbides,
`borides, and nitrides.
`
`The specific contact resistivity of TiN films to Si is somewhat higher than that of Ti (~10
`,uQ-cmz), and as a result TiN is ordinarily not used to make direct contact to Si. As discussed
`earlier,
`it is commonly used in contact structures together with an underlying layer of Ti
`(TiN/Ti/Si). Such contact structures exhibit very low specific contact resistivities to Si and
`remarkably high thermal stability. However, if a conventional reactive sputtering process is used
`to