`
`6
`ft,
`.
`i”
`:
`Bo ND
`ree pthpas
`gp tT aa
`¢
`if
`4
`nah
`
`i
`
`.
`
`+
`
`.
`
`eS
`
`oe
`TSMC 1012
`
`<¥
`
`At
`
`fh
`
`
`
`Physics World, Techno House, Redcliffe Way, Bristol
`BS1 6NX, UK
`Tel +44 (0)117 929 7481
`Fax +44 (0)117 925 1942
`Internet: pwld@ioppublishing.co.uk
`WWW: hitp:/Avww.iop.org
`
`Acting Editor: Peter Rodgers
`Features Editor: Kathleen Rosewarne
`Production Editor: Susan Curtis
`Associate Editor: Matin Durrani
`Production Assistant: Kate Boothby
`Diagram Artist: Alison Tovey
`Secretary: Rachael Anniss
`Advertisement Manager: Chris Manning
`Advertisement Sales: Chris Robinson
`Classified Sales: Sarah Brain
`Advertisement Production: Jayne Boulton
`European Sales Agent: Ric Bessford, Leopoldstr. 52,
`80802 Munich, Germany(tel. +49 89 39 00 55)
`
`Publisher: Mark Ware
`Studio/Production Manager: Andrew Slocombe
`Studio Designer: Mark Hay
`Circulation/Marketing Manager: Richard Roe
`Product Manager: Terry Hulbert
`Advisory Panel: Bill Bardo (GEC-Marconi), Claude
`Détraz (IN2P3), Danilo de Rossi (Universita degli Studi
`di Pisa), John Ellis (CERN), Colin Gough (University of
`Birmingham), Philip Gummett (University of
`Manchester), Jean-Pierre Hansen (ENS de Lyon), Peter
`Knight (imperial College), John Lewis (Consultant), Sir
`Martin Rees (University of Cambridge), Dietrich
`Stauffer (Universitit zu Kéln), Herbert Walther (Max-
`Planck-Institut fiir Quantenoptik), Frank Wilczek
`(institute for Advanced Study, Princeton), Peter
`Williams (Oxford Instruments)
`
`Email IOP@ULCC.AC.UK
`
`Published monthly as twelve issues per annual volumeby Institute
`of Physics Publishing Ltd, Techno House, Redcliffe Way, Bristol
`BSi 6NX, UK
`Subscription information 1995 volume
`Forall countries except the United States, Canada and Mexico, the
`subscription rates are: institutions £85 per annum, individuals £34
`in EC, £37 elsewhere per annum.Single issues £7.25. Orders to:
`IOP Circulation Centre, WDIS, Publishing House, Victoria Road,
`Ruislip, Middlesex HA4 05X, UK (tel. +44 (0)181 845 8545; fax
`+44 (0)181 845 7696). For the United States, Canada and Mexico,
`the subscription rates are:
`institutions US$ 160 per annum,
`individuals US$ 61 per annum. Orders to: American Institute of
`Physics, 500 Sunnyside Blvd, Woodbury, NY 11797, USA
`Copyright © 1995 by IOP Publishing Lid and individual
`contributors. All rights reserved. IOP Publishing Ltd permits
`single photocopying ofsingle articles for private study or research,
`irrespective of where the copying is done. Multiple copying of
`contents or parts thereof without permission is in breach of
`copyright, except in the UK under the terms of the agreement
`between the CVCP and the CLA. Authorization of photocopy
`itemsfor internal or personaluse,or the internal or personal use of
`specific clients, is granted by IOP Publishing Ltd forlibraries and
`other users registered with the Copyright Clearance Center
`(CCC) Transactional Reporting Service, provided that the base
`fee of $2.50 per copyis paid directly to CCC, 27 Congress Street,
`Salem, MA 01970, USA
`United States Postal Identification Statement
`Physics World (ISSN 0953-8585) is published monthly for $160
`per year within the United States in association with the American
`Institute of Physics, 500 Sunnyside Blvd, Woodbury, NY 11797.
`Second class postage paid at Woodbury, NY and additional
`mailing offices.
`POSTMASTER: Send address changes to Physics World,
`American Institute of Physics, 500 Sunnyside Blvd, Woodbury,
`NY 11797.
`Bibliographic codes ISSN: 0953-8585 CODEN: PHWOEW
`Printed in the UK by Warners (Midlands) plc, The Maltings,
`West Street, Bourne, Lincolnshire PE1O 9PH
`Institute of Physics, 47 Belgrave Square, London SW1X
`8QX.
`Tel +44 (0)171 235 6111.
`Fax +44 (0)171 259 6002.
`
`io)
`
`Corporate Affiliates of IOP Alcatel, Amersham
`International, British Gas, BICC Cables Ltd, British
`Nuclear Fuels, BT, Coates Electrographics, Defence
`Evaluation and Research Agency, GEC, GPT Ltd,
`Hewlett-Packard, IBM, ICI, The National Grid
`Company, National Physi-cal Laboratory, National
`Power, Northern Telecom/BNR, Nuclear Electric,
`Oxford Instruments, Particle Physics and Astronomy
`Research Council, Pilkington
`Technology, PowerGen plc,
`Thorn-EMI, UK AEA Fusion
`
`Physics World November 1995
`
`3
`
`PHYSICS
`world
`
`
`
`What future for X-rays?
`
`Why is Physics World not celebrating the centenary of the discovery of X-rays?
`Surely the discovery that wonthe first Nobel Prize in Physics — a mere six years
`after R6ntgen stunned the world with photographs of human bones and other
`“invisible” objects — should be feted in these pages. And it is certainly cause for
`celebration that, thanks to synchrotron radiation sources, the world now has a
`copious supply of X-ray photons for a wide range of experiments in physics,
`chemistry, biology, materials and beyond.
`But it is also worth highlighting, as this issue does, the relentless increases in
`computer power madepossible by advancesin silicon technology, and the physics
`challenges facing that industry (pp15—16 and 29-51). Indeed the X-ray andsilicon
`worlds already overlap: X-ray lithography is a possible replacement for optical/
`ultraviolet systemsas the feature sizes in integrated circuits approach 0.1 ym. And
`silicon components are playing an important role in X-ray optics as scientists
`struggle to cope with the heat-loads associated with the brightestX-ray sources.
`The silicon industry has a clear idea of where it has to go in the next few
`decades. What do wesee if we gaze into an X-ray crystal ball? The traditional X-
`ray tube is now rarely used in research although at least one can be found in most
`airports, hospitals and dental surgeries. Synchrotron radiation sources obviously
`dominate the picture but the intense competition for beam-time meansthat there
`is also a demandfor lab-sized sources.
`Plasma-based sources look the most promising. A laser-produced plasma can
`generate X-rays in a variety of ways: it is possible, for example, to set up a
`population inversion — the main prerequisite for laser behaviour — at X-ray
`wavelengths in a plasma. However, the take-up of X-ray lasers has been slow, with
`applications largely limited to probing other laser-produced plasmas(e.g. in laser
`fusion experiments). Nonlinear processes such as harmonic generation — in which
`odd numbers of photons are bundled into a single photon with a proportionally
`shorter wavelength — are just beginning to access the X-ray region. As it becomes
`possible to pack more laser energy into shorter pulses, nonlinear effects will
`increase. Exotic schemes involving the damping of plasma waves could produce
`photons at hard X-ray wavelengths.
`But the most promising ofall the laser-plasma sources, from the applications
`point of view, are thermal plasmas. The black-body radiation from a million-
`degree plasmastretches into the X-ray region of the spectrum, and the strongest
`emission lines from the plasma are competitive with synchrotron sources in some
`types of experiment. Create a hundred of these mini-plasmas every second and
`you have a high-average-power source that can be used in lithography,
`microscopy and microfabrication. All you need is a table-top laser and a supply
`of low-Z material (such as the plastic on the back of audio cassette tapes!)
`All these X-rays are of little use without the associated optics and detectors. The
`count rates in X-ray experiments are enormous, and expensive materials like
`diamond mayhave to be pressed into service to cope with them. And X-ray optics
`is notoriously difficult: ordinary mirrors don’t work at normal incidence and give
`large aberrations that are hard to reduce when used at grazing incidence.
`Multilayers mirrors are used for near-normal incidence, but layers only a few
`atoms thick are needed, which leads to problems with interface roughness.
`Diffractive elements are best for high spatial resolution but currently have their
`own problems — mostly inefficiency.
`One remarkable aspect of the discovery of X-rays was the speed with which
`applications emerged. If X-rays continue to infiltrate applications other than
`radiography and research — IC fabrication for example — the consequences could
`be just as important.
`
`The contents of this magazine, including the views expressed above, are the responsibility
`of the editor. They do not represent the views or policies of the Institute ofPhysics except
`where explicitly identified as such.
`
`
`
`
`
`(cid:3) (cid:55)(cid:75)(cid:76)(cid:86)(cid:3)(cid:80)(cid:68)(cid:87)(cid:72)(cid:85)(cid:76)(cid:68)(cid:79)(cid:3)(cid:80)(cid:68)(cid:92)(cid:3)(cid:69)(cid:72)(cid:3)(cid:83)(cid:85)(cid:82)(cid:87)(cid:72)(cid:70)(cid:87)(cid:72)(cid:71)(cid:3)(cid:69)(cid:92)(cid:3)(cid:38)(cid:82)(cid:83)(cid:92)(cid:85)(cid:76)(cid:74)(cid:75)(cid:87)(cid:3)(cid:79)(cid:68)(cid:90)(cid:3)(cid:11)(cid:55)(cid:76)(cid:87)(cid:79)(cid:72)(cid:3)(cid:20)(cid:26)(cid:3)(cid:56)(cid:17)(cid:54)(cid:17)(cid:3)(cid:38)(cid:82)(cid:71)(cid:72)(cid:12)(cid:3)
`This material may be protected by Copyright law (Title 17 U.S. Code)
`
`Physics World November 1995
`
`35
`
`Silicides are a simple family of materials combining metal with
`silicon, but the latest generations of microchips have promoted
`these materials from the relative obscurity of inorganic chemistry
`to the forefront of semiconductor technology
`
`Simply irresistible
`silicides
`
`KAREN MAEX
`
`
`
`chips connecting transistors and other devices together)
`are the important factors for microprocessors. In both
`cases, memory bits are interconnected with long, narrow
`conductors known as “‘word lines”’, and it turns out that
`interconnects made from silicides can speed up signal
`propagation times by lowering the electrical resistance of
`these lines. Indeed, a ‘‘polycide”’ technology, mainly based
`on WSig, introduced into industrial production lines more
`
`1 A metal oxide silicon field effect transistor (MOSFET) shown in cross-section
`(top) and viewed from above (bottom). Current flows from the source to the
`drain. The presence ofthe silicide helps to lower the resistance of the source
`and drain areas, a crucial factor in industry’s quest to manufacture ever smaller
`devices. Typical dimensions are indicated for 0.25 um fabrication technology.
`
`isolation,oan,
`(silicon dioxide)"
`
`“interlevel dielectric®
`
`silicide
`
`runner (polysilicon)
`
`
`
` Junction a eS
`depth,
`7
`{silicon dioxide)
`.
`implanted
`source(silicide)
`drain (silicide)
`junction
`
`If THE microelectronics industry is to continueto satisfy the
`demand for ever increasing amounts of memory from
`computers, it needs to manufacture devices on ever smaller
`scales. However, each new generation of logic chips
`depends on having materials with the right characteristics
`to make fabrication both economical and practical. In the
`latest generations:of logic circuits, one group of materials,
`thesilicides, has proved highly beneficial.
`The growing interest in the application of silicides
`to microelectronics has widened the scopeofsilicide
`research and opened up many new avenues of
`investigation. The focal point of much of this work
`has been in improving our understanding of how
`silicides behave in combination with the other
`materials found in devices. No new material would
`ever be considered for device implementation unless
`industry was convinced that
`the material was
`compatible with current manufacturing processes
`and easily integrated into conventional
`integrated
`circuit (IC) technologies.
`Almost all metals in the periodic table react with
`silicon to form silicides, which have the general
`chemical formula M,Si,. Most silicides are metallic,
`have low resistivity and can be divided into three main
`categories. There are the “refractory metal silicides’’,
`such as titanium silicide (TiSi,) and tungstensilicide
`(WSi), which generally have a high thermal stability;
`the “near-noble metal silicides”’, such as platinum
`silicide (PtSi) and cobalt silicide, whose main asset
`lies in their chemical reactivity; and the “rare earth
`metal silicides”,
`like erbium silicide (ErSi.), which
`are mainly investigated for their optical properties,
`such as their ability to absorb infrared light. Although
`the process of forming a silicide, known as
`“‘silicidation”’, is complex, the material properties of
`silicides have been widely studied and extensive
`reviews are available (see Further reading).
`Onedriving technology behind current and future
`developments in microelectronics is complementary
`metal oxide silicon (CMOS). CMOScircuits have
`two main applications: memories and microproces-
`sors. High bit capacity —.and hence large chip sizes — is
`the main feature of memory chips. Fast signal speeds
`and a high density of “interconnects” (the tracks on
`
`*
`
`@
`
`lat
`
`source area
`
`.
`
` silicide}
`
`ee
`
`—»|
`
`céntact (with tungsten
`and aluminium on top)
`
`.
`
`
`drain 4rea
`«(silicide}
`
`1 pm,”
`
`:
`
`|e0.1 em
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`oa.
`#spacer
`,Asiticor: dioxide)" pagq—>]
`'
`0.25 gin
`“
`“ * all 5 a ee a es
`
`
`
`€
`
`
`36
`
`Physics World November 1995
`
`
`
`best self-aligning properties are PtSi, NiSi, CoSi, and
`TiSiz because their reaction with silicon is
`the most
`controllable. PtSi has been used extensively in bipolar and
`infrared detector applications. Unfortunately it has poor
`thermal stability and is not the ideal choice for a CMOS
`circuit. NiSi has also been investigated in detail, but it is
`unstable in contact with silicon. TiSi, and CoSi, are
`therefore the two mainsilicides that have been ofinterest
`to MOS technology. TiSiz is now widely accepted by
`integrated circuit manufacturers, but recent studies have
`shownthat CoSi, has even better properties and could also
`be considered a serious candidate.
`
`Hotstuff
`
`
`
`
`
`than a decade ago, has improved propagation speeds by
`almost a factor of ten.
`However, when industry started manufacturing the
`0.5 um generation of CMOS circuits about
`five years
`ago, it becameclear that any further miniaturization would
`require technology changesto the transistoritself. This is
`wheresilicides have begun to play a key role. Figure 1
`shows a schematic cross-section of a typical metal oxide
`semiconductor field effect
`transistor
`(MOSFET)
`for
`0.25 um technology. The critical dimensionsare the gate
`length, which is used to define the “feature size’’ of the
`transistor quoted above, the depth of the p—n junctions
`forming the source and drain contacts, and the contact
`surfaces themselves. Electrical current mainly flows
`horizontally in the p—n junction of the contact, but the
`What happens when a metalreacts with silicon to form a
`resistance to the flow of current depends on the cross-
`silicide? In the case of titanium or cobalt, experiments on
`sectional area of the junction andits intrinsic resistivity.
`bulk reactants have given us a fairly good empirical idea of
`Because of the limit to the number of dopants that can be
`whatis going on. However, we need to know what happens
`solubilized in a given volumeofsilicon, thereis also a limit
`whenathin film of the metal reacts with thesilicon. This
`to the intrinsic resistivity of the junction. The unfortunate
`consequenceofthis is that as the junction depth decreases,
`type of system is
`a
`long way from thermodynamic
`equilibrium because the reaction takes place in the solid
`the total resistance of the contact goes up, jeopardizing the
`state at temperatures between 500 °C and 800 °C. Under
`The National Technology Roadmap for Semiconductors
`these conditions the elements are able to diffuse, and the
`system can lower its free energy by formingaseries of
`Feature
`Junction
`Bits/chip
`Bits/chip
`Year of
`different intermetallic phases.
`size (um) depth (um)
`(DRAM)
`(SRAM)
`production
`in the
`The so-called “‘phase formation sequence’’
`0.50
`0.20
`16M
`4M
`1992
`reaction between thin metal films and silicon has been
`0.35
`0.15
`64M
`16M
`1995
`investigated extensively since the late 1960s by F M
`0.25
`0.12
`256 M
`64M
`1998
`0.18
`0.10
`1G
`256 M
`2001
`d’Heurle at the IBM Yorktown Heights Laboratory in
`0.12
`0.08
`4G
`1G
`2004
`New York, M-A Nicolet at California Institute of Tech-
`As the memory requirement from computers continues to grow everlarger,
`nology in Pasadena and J Mayerat
`each successive generation of logic circuits needs smaller and smaller
`Cornell University, among many
`features — including the p—n junction depth in transistors.
`others. It turns out that the pres-
`Source: Semiconductor Industries Association Roadmap (1994).
`ence of small amounts of contami-
`nants, such as oxygen or carbon —
`either in the metal or at the metal/
`relative improvementof the device speed that one expects
`silicon interface — have a big impact
`during scaling to smaller dimensions.
`on the solid-state reaction. How-
`To illustrate this point, table 1 gives typical junction
`ever, if the reaction occurs in a well
`depths for various generations of logic circuits. The
`controlled environment, as is man-
`expectation is
`that as
`the feature size (i.e.
`the gate
`datory for microelectronic pro-
`length) of circuits falls from today’s 0.5 to 0.12 um by
`cesses, the phase sequences shown
`the year 2004, the junction depth will have to shrink from
`in figure 3 are observed for
`the
`0.2 to 0.08 ym. In other words, as the gate length is scaled
`cobalt/silicon and the titanium/sili-
`down,the othercritical dimensionsof the transistors have
`con reactions.
`to be reduced to keep performance at an acceptable level.
`The reaction of a thin cobalt film
`It was to overcome these problems that many manu-
`with silicon was worked out theor-
`facturers began to introduceasilicide fabrication step into
`etically by Ulrich Gésele and King
`the production of the 0.5 um generation ofcircuits. In this
`Ning Tu back in 1982. The cobalt
`process, known as “‘self-aligned silicide’’ or “‘salicide’’, a
`diffuses into the silicon to form a
`silicide is simultaneously formed on the gate and source/
`cobalt-rich (Co,Si) phase (figure
`drain areas through a reaction betweenthesilicon and a
`3a). This then turns into the
`metal, such as titanium or cobalt. The presence of the
`monosilicide CoSi. Finally, when
`silicide lowers the resistivity of the junction. Although
`there is no free cobalt left, nuclea-
`some metallic elements can have an even lowerresistivity
`tion of CoSi, begins.
`(figure 2), silicides are preferred because they form more
`For silicidation of titanium a
`stable contacts with silicon. Silicides also have the
`different process takes place. Here
`advantage that they oxidize in air, a process known as
`the TiSi, phase forms immediately.
`‘‘self-passivation’’, to produce a protective silicon dioxide
`However,
`there are two allotropic
`(SiO) surface barrier that prevents thesilicide from being
`TiSi, phases, each with a different
`attacked during further processing.
`orthorhombic crystal structure (fig-
`Thelack of reactivity between metal and SiO,is a crucial
`ure 3b). One phase, referred to as
`factor because SiO,
`is also used as a dielectric layer
`C49, is full of defects and forms at
`isolating one component from the other. In addition,
`it
`temperatures below 600° C. The
`bestows the unique property of self-alignment. This is a
`other phase, C54, forms at temper-
`fabrication procedure in which two different materials, in
`atures above about 600° C. The
`this case the silicon and the silicide, can be accurately
`latter is preferred because of its
`positioned one on top ofthe other. Thesilicides with the
`
`2 The specific resisti-
`vity of various silicides
`and some metals. In
`comparison, the speci-
`fic resistivity of highly
`doped silicon is
`~500 pQQem.
`
`
`100 pps Resistivity
`
`
`TiSi
`ee CoSisne NiSi
`foe ee
`
`(uQ/cm)
`
`
`
`
`
`
`
`Physics World
`
`November 1995
`
`
`
`
`
`
`
`
`
`@ Co/Si reaction
`
`D Ti/Si reaction
`
`lowerresistivity (figure 2).
`the
`In both cases, however,
`silicide phase that ends up in
`thermal equilibrium with the
`silicon is
`the one with the
`lowest electrical
`resistivity.
`Because the solid solubility of
`titanium and cobaltin silicon is
`extremely low,
`titanium and
`cobalt impurities in silicon do
`not
`interfere with the almost
`ideal semiconducting proper-
`ties of silicon.
`Although the reaction
`between metal
`andsilicon is
`important, so is the possibility
`of an interaction between the
`metal and any remaining SiO,
`on the surface of the silicon that
`has not been properly removed.
`Cobalt has nosignificant inter-
`action with SiO, but titanium
`can reduce SiO2, becauseofits
`high reactivity with oxygen.
`Thus, when titanium is depos-
`ited on the surface ofthe silicon
`it is less critical for the surface to be perfectly clean. This
`has practical advantages because surface-cleaning pro-
`cedures, whicharestill somewhat problematic and poorly
`controlled, do not have to be so stringent in this case.
`
`3 Thin films of cobalt or titanium can form silicides by reacting with silicon. The reaction takes place in the
`solid state at temperatures of 500-800 °C, with each stagetypically lasting 30 s. In both cases, various
`phases are encountered. (a) Cobalt first diffuses into the silicon to form a cobalt-rich (Co2Si) phase. This
`then turns into CoSi. Finally, when there is no morefree cobalt left, CoSiz starts to form. (b) The silicidation
`of titanium is different. The TiSiz phase forms immediately and there are two allotropic TiSiz phases — C49
`and C54 — each of which hasits own orthorhombic crystal structure.
`
`Silicides in runners and gates
`Ultralarge-scale integration (ULSI)
`is a term used to
`describe the grouping of many electronic components in
`the form of large and complex integrated circuits onto a
`single chip. In these circuits the gate electrode and the
`local electrical connections consist of very narrow lines of
`silicon in polycrystalline form, known as “‘polysilicon’’.
`Surrounding these lines are areas of SiO,
`(figure 4a).
`When such circuits are manufactured, a layer of metal is
`deposited onto each transistor (figure 4b). The idea is for
`the metal to react with the polysilicon, converting it into a
`silicide, but not with the SiO, (figure 4c). Because the
`silicide should not form in surrounding regions,
`the
`reaction is said to be “‘laterally confined’’.
`Although lateral confinement does not fundamentally
`change the reaction kinetics, these new boundary condi-
`tions may influence the rate at which thesilicide lines are
`formed. This is particularly so for TiSiz. Here, the C49-—
`C54 transition between the two different allotropic forms
`of TiSi,, which normally takes place at above 600 °C,is
`actually slowed down on these narrow polysilicon lines
`because of a lack of nucleation centres. However,
`the
`transition can be accelerated using rapid thermal proces-
`sing. In this technique a single wafer is heated by radiation
`from a lamp to around 800 °C for a couple of seconds.
`The higher temperature helps to yield more nucleation
`sites for the C54 phase and the short reaction time
`postponesthedisintegration ofthesilicide film. For CoSi,,
`however, these nucleation problems are not encountered:
`the CoSi—CoSi, transformation (the final step in figure
`3a) occurs faster in thinnerfilms than in the thicker ones.
`The thermalstability of thin silicide films on silicon is a
`major concern,sinceit limits the ‘“‘thermal budget” allowed
`for further processing. This term refers to the fact that in
`integrated circuit processingit is importantto limit the total
`exposure of the wafer to heat. The aim is to carry out
`
`reactions as fast and at as low a temperatureas possible.
`Overexpose the wafer to heat, and the microstructure can
`fall apart
`through processes such as “grooving”? and
`“islanding”. Both CoSiz and TiSi, are polycrystalline and
`suffer from these problems during prolonged high-
`temperature treatments. Thermal grooving is influenced
`by both thermodynamic and kinetic factors. From a
`thermodynamic point of view,
`the driving force for the
`morphological
`transformation is
`the reduction of the
`interface/surface energies. The grooving process occurs
`because of local energy equilibrium at the intersection of a
`grain boundary and thefilm surface or interface, causing
`matter to diffuse away from the grain edges.
`
`Doping contacts
`The creation ofa silicide on the source and drain regions
`in a MOS transistor reduces the total resistance to the
`current flow. This resistance consists of two parts:
`the
`intrinsic resistance in the channel region, which depends
`on the voltage applied to the gate, andthe series resistance
`of the source—drain contacts between the doped junction
`and the metal that is used to wire all of the transistors
`together, which is independent of the applied bias. The
`contact between the wiring metal (through thesilicide)
`and thesilicon is said to be “‘ohmic’’ and its formation
`requires the silicon to be doped.
`Typical dopants in silicon include arsenic and phos-
`phorusas electron donors and boron as electron acceptors.
`Because these dopant elementsare able to diffuse, they play
`an active role in the silicidation reaction. Sincesilicide films
`are polycrystalline, the diffusion of these dopants depends
`on the sum of two separate factors: diffusion in the bulk
`lattice and (muchfaster) along the grain boundary.
`For a TiSi, film on the source ordrain of a transistor, the
`diffusion of dopants can even lead to the formation of
`titanium/dopant compounds. This consumes mostof the
`dopants, reducing the dopinglevel at the interface between
`the silicide and the silicon. Since the concentration of
`dopants at the interface directly determines the quality of
`the ohmic contact, dopant depletion should be minimized.
`Cobalt silicidation works better in this respect
`than
`
`
`
`Physics World November 1995
`
`source
`
`gate (polysilicon)
`
`isolation (silicon dioxide)
`
` a
`
`tungsten
`
`
`
`
`
`
`drain
`
`source
`
`spacer
`drain
`
`metal (titanium or cobalt)
`
`silicide (TiSiz or CoSiz)
`
`
`
`4 Diagram ofthe self-aligned silicidation process in a transistor, by which two different materials, in this case the silicon and thesilicide, can be
`layered accurately on top of each other. (a) Source/drain and gate regions made from polycrystalline silicon are first formed. (b) Titanium or cobalt
`metal is then deposited over the whole structure and the structure is heated to ~ 600 °C. The metal reacts with thesilicon, but not with SiOz, to form a
`silicide on the three electrode areas. Silicon dioxide spacers on the side walls of the gate prevent shortages between the source/drain and the gate
`by acting as electrical isolation. (c) The unreacted metal is washed awayin a selective chemical etch. A second thermal treatment terminates the
`self-aligned silicidation process, bringing the silicide to its lowest resistivity state. (d) Finally, the separate transistors are joined together with
`aluminium interconnects. Here only the first level of interconnects is shown.
`
`titanium because cobalt is chemically less reactive.
`Fora laterally confined silicide the difference in lateral
`expansion ofsilicide and silicon during heat treatment
`creates stress fields in the silicon. The resulting force is
`almost entirely transmitted to thesilicon lattice near the
`silicide edge through a small interfacial area. When the
`yield strength of silicon is exceeded, extended lattice
`defects may be created. Defect formation alongsilicided
`contact edgesis therefore a potential problem.
`At temperatures above 600 °C,localized stress fields can
`cause plastic deformation of the silicon by generating
`dislocations, either heterogeneously by the capture and
`multiplication of existing dislocations or homogeneously
`by the condensation ofsilicon interstitials.
`
`Back at the end
`
`Silicidation is usually the last of the transistor fabrication
`steps. It is followed by the so-called ‘‘back-end”’ process,
`in which metal conductors are deposited on the chip,
`connecting the contact areas of the individual transistors to
`form an integrated circuit (figure 4d). Since both TiSiz
`and CoSiz are metallic,
`the electrical contact between
`these areas and the interconnect metal on top is expected
`
`to be of high quality. In other words, the contact should be
`so stable that
`it does not chemically, physically or
`electrically disintegrate during further processing or
`during the transistor’s use. This requires careful control
`of the interactions between the interconnecting metal and
`the silicide.
`The current generation of integrated circuits has an
`interconnect technology based on aluminium and tung-
`sten. Unfortunately, the direct deposition of these metals
`onto silicides is problematic because they can react. For
`example, aluminium can react with CoSi, at 400 °C,
`yielding large silicon precipitates at the sample surface and
`the Co,Al, compound at the CoSi,/Al interface. Alumi-
`nium can react with TiSi, at 400 °C to form large silicon
`precipitates and large aluminium pits in the underlying
`silicon substrate. At 550°C,
`the ternary compound
`Ti7Al;Si,;2 can be formed.
`To prevent these unwantedreactions, a barrier layer can
`be used to help minimize the diffusion of aluminium and
`silicon throughthesilicide. Commonbarriers include TiN
`and TiW (figure 4d). The barriers also allow contacts to be
`formed on the silicon that are stable during heat
`treatments to temperatures as high as 550 °C. Another
`benefit of these barriers is that they promote adhesion of
`
`38
`
`
`
`
`
`
`
`
`
`
`
`
`5 Scanning electron micrographs of a
`0.1 um transistor after silicidation with
`CoSis, in cross-section (top) and from
`above (bottom). The top CoSi, layer on
`the polysilicon is 0.1 um wide.
`
`that minimize the amount of
`silicon that is consumed when
`silicide is formed in the junc-
`tion? One proposed material is
`NiSiz, because it uses only half
`of the silicon atoms. However,it
`is unclear how one would avoid
`forming the more stable NiSi
`during later processing steps.
`Drastic changes to the back-
`end process are also expected
`over the next few years. Metals
`with a better conductivity than
`aluminium, such as copper,
`have been considered as poss-
`ible conductors to be deposited
`on the chips during this process,
`as have dielectrics with lower
`permittivity than SiO,. Since the
`interconnect density is a driving
`force in the scaling of logic
`circuits, it is hoped that in this
`way RC delays and cross-talk
`problems between different
`wires can beeliminated.
`Changes in the back-end pro-
`cess will also have an impact on
`silicide integration. The intro-
`duction of new dielectric mate-
`rials often goes hand in hand
`with a reduction of the thermal
`budget, which is beneficial for
`the integrity of the already
`formed silicides. If copper is
`going to be a material
`for
`contacts and interconnects,
`new reaction barriers between
`the silicide and the copperwill have to be introduced.
`The decreasingcritical dimensions of MOS transistors
`will result in ever higher resistances. Therefore the use of
`pure metals, rather thansilicides, on source/drain and gate
`electrodes could be attractive. Both tungsten and
`aluminium can be deposited selectively on silicon, and
`these metals could be considered as future alternatives to
`silicides. However,
`the stability of the silicide/silicon
`contact andtheself-passivating nature ofthe silicide in a
`reactive environmentallows moreflexibility during further
`processing and ensures thatsilicides will remain a vital part
`of integrated circuit fabrication for a long time to come.
`
`.
`
`‘CoSia
`
`poly-Si
`
`CoSi, on mono-Si area
`
`CoSi. on poly-Si area
`
`Physics World November 1995
`
`39
`
`
`
`the tungsten interconnects to the silicide. Clearly,
`potential problems that would arise if tungsten were
`‘deposited directly onto thesilicide are largely avoided.
`
`Future perspectives
`In 1994 the US Semiconductor Industry Association, the
`US network of microelectronics companies, government
`and academia, published a road-map describing future
`integrated circuit processes (see Yoshio Nishi ‘‘Silicon
`faces the future’? p29). According to the report
`the
`0.25 um generation of chips is expected to enter the
`production lines within the next three years. In the latest
`0.35 um generation of circuits, which is currently on the
`verge of production, TiSi, is the most commonly used
`silicide. However, further scaling of processes based on
`TiSi, becomeincreasingly difficult.
`CoSi,, on the other hand, allows much moreflexibility
`in that respect. No limits for the implementation of CoSi
`have been encountered down to critical dimensions of
`0.1 ym (figure 5). This is expected to be reached in the
`first decade of the next century. The possibility of forming
`CoSi, epitaxially on silicon (i.e. with its orientation
`dependentonthe silicon substrate) is another advantage.
`This might produce more stable contacts. However,
`industry will first have to be convinced that it is economic
`to switch from titanium to cobalt. It would rather stick
`with TiSi, for the time being, even if it becomes ever more
`expensive and complicated to use.
`As ICs become smaller and the depth of the junction
`becomesshallower, what are the hopesoffinding materials
`
`Further reading
`S Coffa and F Priolo (eds) 1992 Crucial Issues in Semiconductor
`Materials and Processing