`
`Advances in Research and Development
`
`PLASMA SOURCES FOR THIN FILM
`DEPOSITION AND ETCHING
`
`Edited by
`
`Maurice H. Francombe
`
`Department of Physics
`The University of Pittsburgh
`Pittsburgh, Pennsylvania
`
`John L. Vossen
`
`John Vossen Associates
`Technical and Scientific Consulting
`Bridgewater, New Jersey
`
`Academic Press
`San Diego New York Boston
`London Sydney Tokyo Toronto
`
`'Rnna?a 7uabm
`
`Page 1 of 122
`
`Samsung Exhibit 1006
`
`Page 1 of 122
`
`Samsung Exhibit 1006
`
`
`
`
`
`i RUDE?E 7H3£l
`
`Page 2 of 122
`
`Page 2 of 122
`
`
`
`Contents
`
`Contributors .
`Prefurc .
`.
`.
`
`.
`
`_
`
`.
`
`.
`
`This book is primed on acid-free paper Gt)
`Common-r .0 I994 av museum rinses, no;
`Au. moms mum
`
`ho Milt or ‘I'Ills I'L'BLICATION my rte REPRODUCED or.
`nassnrrrro IN ANT mm on av ANY Minus urmomc
`IRNY l3.
`on mmntocaL. INCLUDING Filo-rooon FLEmRDING on
`TORMR‘I’ION' STORAGE AND RETRIEVAL SYSTEM WITH
`PERMISSIDN ITS WRITING FROM TH}. FUEL-“HER
`I
`OUT
`
`fiSnDI-IMIC PRESS. INC.
`525 atfeign” Brace & Company
`San Diego. California 92 ml «1495
`
`I
`United Kingdom Ed'
`'
`I
`rtton
`igADEMIC PRESS LIlf/lllf'ifggd by
`.
`Oral ImLonmrnvi to):
`
`‘
`
`L
`
`ll
`
`I
`
`l”-
`
`w,
`
`Design of High-Density Plasma Sources for Materials Procossing
`Miciioei A. Lieberman and Richard A. Gortscho
`Introduction .
`.
`.
`.
`.
`.
`.
`.
`.
`_
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`
`.
`
`.
`
`.
`
`.
`.
`
`.
`.
`.
`.
`
`.
`,
`
`.
`.
`.
`.
`
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`
`.
`.
`.
`.
`
`.
`_
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`i
`
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`
`_
`.
`.
`.
`
`.
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`,
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`_
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`,
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`ix
`‘lLi
`
`our-4
`
`13
`19
`23
`25
`26
`3]
`3-1
`40
`41
`42
`
`Lihrary of Con
`gress Catalog Ca :1
`-
`ISBN: 0-12-533013-9
`I Number. 63—46561
`
`notation!» 9:155:32]
`"twin: [8 m: L‘xrtm sure: or AMERICA
`
`Page 3 of 122
`
`V.
`
`VI.
`V“.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`,
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`.
`.
`.
`.
`.
`
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`
`.
`
`.
`.
`
`.
`.
`
`.
`
`.
`.
`
`.
`
`.
`.
`
`.
`.
`
`.
`
`.
`.
`.
`
`.
`
`.
`.
`
`.
`.
`
`.
`
`.
`
`.
`
`r
`
`.
`
`_
`
`.
`
`.
`
`.
`
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`
`.
`
`.
`
`.
`.
`.
`A. Capacitiver Coupled Radio Frequency Discharge Sources
`B. Limitations of Capacitiver Coupled Radio Frequency Discharges
`Cl Overview of High-Eificiency Sources
`Principles of Low-Pressure,High-Efficiency Source Design .
`A. Unified Analysis of Source Operation .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`B. Discharge Heating .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Electron Cyclotron Resonance {ECR} Discharges
`.
`A. Source Configurations .
`.
`.
`B. Electron Heating .
`.
`.
`C. Resonant Wave Absorption .
`Helicon Discharges
`.
`A. Helicon Configurations
`B. Helicon Modes
`.
`r
`.
`.
`.
`C. Antenna Coupling .
`D. Helicon Mode Absorption .
`.
`.
`.
`.
`Inductive Discharges
`_
`.
`.
`.
`.
`.
`.
`.
`.
`A.
`Inductive Source Configurations .
`.
`B. Power Absorption and Operating Regimes
`.
`.
`.
`C. Source Operation and Coupling .
`.
`.
`.
`D. Low-Density Operation and Source Efficiency .
`Helical Resonator Discharges
`.
`.
`.
`.
`.
`.
`_
`.
`.
`.
`Surface Wave Discharges .
`
`.
`.
`.
`
`.
`
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`
`.
`
`.
`,
`
`.
`.
`.
`.
`
`.
`.
`.
`,
`.
`
`.
`
`.
`
`.
`.
`
`.
`.
`.
`
`.
`
`.
`.
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`
`52
`52
`54
`56
`58
`60
`65
`
`Page 3 of 122
`
`
`
`Design of High-Density Plasma Sources
`for Materials Processing
`
`MICHAEL A. LIEBERMAN
`
`Department
`
`of Electrical Engineering and Computer
`
`Sciences,
`
`and
`
`RICHARD A. GOTTSCHO
`
`Laboratories,
`AT&T Bell
`Murray Hill, New
`Jersey
`
`.
`
`.
`
`..
`
`I. Introduction
`A. Capacitively Coupled Radio Frequency Discharge Sources
`B. Limitations of Capacitively Coupled Radio Frequency Discharges
`C. Overview of High-Efficiency Sources
`II. Principles of Low-Pressure, High-Efficiency Source Design
`A. Unified Analysis of Sources Operation
`1. Electron Temperature
`2. Ion Bombarding Energy
`3. Plasma Density and Ion Current Density
`B. Discharge Heating
`III. Electron Cyclotron Resonance (ECR) Discharges
`A. Source Configurations
`B. Electron Heating
`C. Resonant Wave Absorption
`IV. Helicon Discharges
`A. Helicon Configurations
`B. Helicon Modes
`C. Antenna Coupling
`D. Helicon Mode Absorption
`V. Inductive Discharges
`A. Inductive Source Configurations
`B. Power Absorption and Operating Regimes
`C. Source Operation and Coupling
`D. Low-Density Operation and Source Efficiency
`VI. Helical Resonator Discharges
`VII. Surface Wave Discharges
`
`2
`5
`9
`10
`13
`19
`19
`19
`22
`23
`25
`26
`31
`34
`40
`41
`42
`46
`50
`52
`52
`54
`56
`58
`60
`65
`
`1
`
`Copyright < 1994 by Academic Press, Inc.
`All rights of reproduction in any form reserved.
`ISBN 0-12-533018-9
`
`Page 4 of 122
`
`
`
`2
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`IX.
`
`VIII. Plasma Transport
`A. The Ion Energy Distribution Function
`1. Ion Transport and Etching Anisotropy
`B. Methods for Measuring Ion Energy Distribution Functions
`C. Methods for Measuring Plasma Potentials
`D. Measurements of Energy Distributions and Potentials
`1. Ion Acceleration Outside the Sheath
`2. Transverse Ion Energy
`E. Ion Energy Control
`1. Plasma Anodization
`Device Damage
`A. Atomic Displacement Damage
`B. Contamination
`C. Charging
`1. Plasma Uniformity
`
`2. Biasing
`
`D. Radiation
`X. Summary and Remaining Questions
`
`XI. Symbol Definitions
`
`Acknowledgments
`References
`
`I.
`
`Introduction
`
`69
`71
`73
`76
`80
`81
`81
`87
`90
`96
`96
`96
`98
`98
`99
`2
`
`4
`
`105
`8
`
`2
`
`1^
`
`The advent of sub-micron electronic device fabrication has brought
`unprecedented demands for process optimization and control (7,2)
`which, in turn, have led to improved plasma reactors for the etching and
`deposition of thin films. As a result, we have witnessed the introduction
`of a new generation of plasma systems based on electron cyclotron
`resonance (ECR) heating (3-6). ECR plasma etching of polycrystalline
`Si, single crystalline Si, suicides, Al, Mo, W, S i 0 2 , polymers, and III-V
`compound semiconductors have all been reported in recent years (7-33).
`Similarly, ECR plasmas have been used to deposit amorphous Si, silicon
`nitride, boron carbide, and S i 0 2 , to name just a few materials
`(34-40).
`Applications of ECR plasmas beyond etching and deposition have also
`been reported and include ion implantation (41-45), surface cleaning
`(46-59),
`surface passivation (60), and oxidation (53,61-63). Besides
`ECR, many other "novel" plasma generation schemes are now being
`offered to satisfy manufacturers' needs in these materials processing
`areas. All these schemes purport to offer advantages over conventional
`approaches such as the capacitively coupled radio frequency discharge
`now used in many factories for etching and deposition of thin films
`during integrated circuit manufacturing.
`
`Page 5 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`3
`
`But which scheme is best? What are the key aspects to plasma source
`design that affect materials processing? And why are the conventional
`approaches inadequate? While the answers to these questions remain
`elusive and are the subject of much current research, one can clearly
`identify commonalities and differences between the novel sources, whose
`most distinctive characteristic is higher efficiency than their conventional
`counterparts operated at low pressure. The purpose of this review is to
`(1) develop a unified framework from which all "high-efficiency" sources
`may be viewed and compared; (2) outline key elements of source design
`that affect processing results; and (3) highlight areas where additional
`research and development is needed. In so doing, we hope to assist those
`who use plasma for materials processing to make wise choices in
`constructing or purchasing sources, to guide vendors of high-efficiency
`sources in choosing designs that can best meet their customers' expecta-
`tions, and to inspire the research community to focus on problems of
`technological interest.
`Before such a review can be begun, several disclaimers must be made.
`First, the literature on applications, diagnostics, and modeling of high-
`efficiency sources is now so voluminous that we are not able to review
`or reference every paper. Rather, we have opted for highlighting key
`results in line with our objectives stated earlier. Second, we restrict our
`focus to those aspects of plasma processing that are uniquely affected by
`the use of high-efficiency plasmas. For example, we discuss aspects of
`source design that affect plasma-induced electrical damage in microelec-
`tronic circuits, but a comprehensive discussion of damage mechanisms
`is the subject of its own review and clearly beyond the scope of this
`work. Third, there are pertinent areas that while important are not yet
`ready for review. Foremost amongst these is the field of numerical
`simulation. While impressive results have been reported recently and we
`will draw on some of these, little has appeared in print and it is
`premature to review the field. Similarly, the stability of high-efficiency
`sources is a matter of some concern, and recent work illustrates that
`sudden mode changes and bistability may adversely affect materials
`properties, but too little has been reported and analyzed to make a
`thorough discussion meaningful. Finally, any review reflects the biases
`of the authors, and this work is no exception. Based on our interests and
`experience, we focus on applications of plasmas
`to microelectronics
`fabrication and, in particular, etching. Heavy emphasis is placed on
`simple, analytical, unifying theories and quantitative diagnostic measure-
`ments.
`Why new sources? In plasma etching, the shrinking dimensions of
`
`Page 6 of 122
`
`
`
`4
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`micro-electronic devices have placed unprecedented demands on process
`control. Consider critical dimension (CD) control where the width of the
`transistor gate is specified to better than 10%. For yesterday's CD of
`1 μηι, this means a linewidth variation of 0.1 μτη can be tolerated, but by
`the end of the 20th century when the CD should be only 0.25 /im,
`variations in CD must be less than 0.025 ^m. This requires unprece-
`dented anisotropy in the plasma etching of gate electrodes, contact
`windows, and metallic interconnections. To achieve such control, we
`need to increase the anisotropy of ion transport to the device wafer from
`what it is in the conventional capacitively coupled rf reactor. This means
`operating plasmas at lower pressures. But conventional rf sources are
`inefficient at low pressure, so that high powers must be used to achieve
`the high rates of ionization and dissociation necessary for high-through-
`put, low-cost manufacturing. Unfortunately, excessive power input to a
`capacitively coupled system leads to high ion bombarding energies that
`can degrade selectivity in etching and produce electrical damage that
`reduces device yield. Thus, new sources are needed to operate at lower
`pressure and higher efficiency.
`In conventional rf systems, ion energy and flux are inexorably linked.
`But ion energy control is needed in plasma deposition to tailor film
`properties such as stress, composition, refractive index, crystallinity, and
`topography. Ion energy control is used in plasma etching to optimize
`selectivity and minimize atomic displacement damage while meeting
`linewidth and throughput specifications. Therefore, gaining superior
`control of ion energy and decoupling it from ion flux control is further
`motivation for developing new plasma sources and processing systems.
`In the remainder of this section, we review briefly the properties of
`capacitively coupled radio frequency plasmas and elaborate further on
`the advantages of high-efficiency sources. In the following sections, we
`first discuss the fundamental principles underlying high-efficiency plasma
`source design and, to compare one source with another, use a simple
`analysis in Section II that allows estimation of electron temperature, ion
`bombardment energy, and plasma density in terms of the gas phase
`cross-sections, gas density, absorbed power, and source dimensions. In
`this way, we provide an approximate but common framework with
`which one source can be compared to another. In Sections III-VII we
`discuss in greater detail ECR, helicon, inductive, helical resonator, and
`surface wave sources, respectively. Emphasis is placed on electron
`heating and power absorption, since these are the primary differences
`between one source and another. In Section VIII, we turn to the issue
`of plasma transport and independent control of ion energy and flux.
`
`Page 7 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`5
`
`Obtaining such control is largely independent of the electron heating
`mechanism but depends critically on source design parameters such as
`the magnetic field and power absorption profiles. We focus our attention
`in Section VIII on measurements of ion energy distributions, mostly in
`ECR systems since few data are available from other systems. In Sections
`VIII and IX, we relate ion energy and plasma uniformity, dictated by
`source design, to processing results such as etching anisotropy, atomic
`displacement damage, and charge-induced damage. In the final section,
`we highlight remaining issues and the areas where further investigation
`is needed.
`Throughout this paper we strive to be consistent with dimensional
`analysis despite not using a consistent set of units. Generally, magnetic
`field is expressed in gauss, distances in meters, centimeters, or mil-
`limeters, and the electron charge in coulombs. Energies are usually given
`in units of volts, not electron volts, so the value of e is explicitly written.
`Pressures are given in Torr or milli-Torr. While this does not conform
`to international convention, it does conform to common usage. We
`apologize to the purists.
`
`A.
`
`CAPACITIVELY C O U P L ED R A D IO FREQUENCY DISCHARGE SOURCES
`
`Capacitively driven rf discharges—so-called rf diodes—are the most
`common sources used for materials processing. An idealized source in
`plane parallel geometry, shown in Fig. la, consists of a discharge
`chamber containing two electrodes separated by a spacing / and driven
`by an rf power source. The substrates are placed on one electrode,
`feedstock gases are admitted to flow through the discharge, and effluent
`gases are removed by the vacuum pump. Coaxial discharge geometries,
`such as the "hexode" shown in Fig. lb, are also in widespread use. When
`operated at low pressure, with the wafer mounted on the powered
`electrode, and used to remove substrate material, such reactors are
`commonly called reactive ion etchers (RIEs)—a misnomer, since the
`etching is generally a chemical process enhanced by energetic ion
`bombardment of the substrate, rather than a removal process due to
`reactive ions. When operated at higher pressure with the wafer mounted
`on the grounded electrode, such reactors are commonly referred to as
`plasma etchers. In terms of the physical properties of these systems, this
`distinction is somewhat arbitrary.
`The physical operation of capacitively driven discharges is reasonably
`well understood. As shown in Fig. 2 for a symmetrically driven discharge
`
`Page 8 of 122
`
`
`
`6
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`GAS FEED
`
`SUBSTRATE —
`
`VACUUM
`PUMP
`
`RF
`SOURCE
`
`BLOCKING
`CAPACITOR
`
`(a)
`
`(b)
`
`FIG. 1. Capacitive rf discharges: (a) plane parallel geometry; (b) coaxial
`geometry.
`
`("hexode")
`
`operated at frequencies between the ion and electron plasma frequencies,
`the mobile plasma electrons, responding to the instantaneous electric
`fields produced by the rf (13.6 MHz) driving voltage, oscillate back and
`forth within the positive space charge cloud of the ions. At 13.6 MHz,
`the massive ions respond only to the time-averaged electric fields.
`Oscillation of the electron cloud creates sheath regions near each
`electrode that contain net positive charge when averaged over an
`
`Page 9 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`7
`
`oscillation period; i.e., the positive charge exceeds the negative charge in
`the system, with the excess appearing within the sheaths. This excess
`produces a strong
`time-averaged electric field within each sheath
`directed from the plasma to the electrode. Ions flowing out of the bulk
`plasma near the center of the discharge can be accelerated by the sheath
`fields to high energies as they flow to the substrate, leading to energetic-
`ion bombardment, which can enhance, inhibit, or otherwise modify
`surface reactions. Typical ion bombarding energies ε { can be as high as
`Vr{/2 for symmetric systems (Fig. 2) and as high as V T{ at the powered
`the rf voltage
`electrode for asymmetric systems (Fig. lb), where V T{9
`amplitude (peak rf voltage) between the two electrodes, might typically
`vary between 100 V and 1 kV.
`We note that positive ions continuously bombard the electrode over
`an rf cycle. In contrast, electrons are lost to the electrode only when the
`
`Page 10 of 122
`
`
`
`8
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`oscillating cloud closely approaches the electrode. During that time, the
`instantaneous sheath potential collapses to near-zero, allowing a suffi-
`cient number of electrons to escape to balance the ion charge delivered
`to the electrode. Except for such brief moments, the
`instantaneous
`potential of the discharge must always be positive with respect to any
`large electrode and wall surfaces; 1 otherwise the mobile electrons would
`quickly leak out. Electron confinement is ensured by the presence of
`positive space charge sheaths near all surfaces.
`The separation of the discharge into bulk and sheath regions is an
`important paradigm that applies to all discharges. The bulk region is
`quasi- neutral, and both instantaneous and time-averaged fields are low.
`The bulk plasma dynamics are described by ambipolar diffusion at high
`pressures and by free-fall ion loss at low pressures. In the positive space
`charge sheaths, high fields exist, leading to dynamics that are described
`by various ion space charge sheath laws, including low-voltage sheaths
`(for high density sources) and various high-voltage sheath models (for rf
`diodes), such as collisionless and collisional Child
`laws and
`their
`modifications (66-73). The plasma and sheath models must be joined at
`their interface. The usual joining condition is to require that the mean
`ion velocity at the plasma-sheath edge be equal to the ion-sound (Böhm)
`where e and M are the charge and mass of the
`velocity uB = (eTe/M)l/2,
`ion and T e is the electron temperature in units of volts.
`In the second column of Table I, typical rf diode source and plasma
`parameters are given. For anisotropic etching, pressures are in the range
`10-lOOmTorr, power densities are 0.1-1 W/cm 2, the driving frequency
`is typically 13.6 MHz, and multiple wafer systems are common. Plasma
`densities are relatively low, ~ 1 0 1 0c m ~ 3, and mean electron energies are
`of order 5 V, corresponding to Maxwellian electron temperatures of
`order 3 V. However, non-Maxwellian electron distributions (e.g., two-
`temperature) are also observed, with the bulk electron
`temperature
`sometimes much less than 1 V (74, 75). Ion acceleration energies (sheath
`voltages) are high, > 200 V, and fractional ionization is low. The degree
`of dissociation can range widely from less than 0.1% to nearly 100%
`depending on gas composition and plasma conditions (76,77). For
`deposition and isotropic etch applications, pressures tend to be higher
`
`low-frequency electronegative and dc
`in
`this rule are also possible
`to
`e x c e p t i o ns
`discharges. In the former, the buildup of negative ions can reduce the plasma potential
`below that of large surfaces in contact with the plasma (64). In the latter, the plasma
`potential can lie between the two electrode potentials if sufficient current is drawn from
`the plasma
`(65).
`
`Page 11 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`9
`
`TYPICAL PARAMETERS FOR HIGH-EFFICIENCY A ND CONVENTIONAL rf PLASMA SOURCES
`
`T A B LE I
`
`Parameter
`
`/
`
`Pressure ρ
`Power Ρ
`Frequency
`Volume V
`Cross-sectional area A
`Magnetic field Β
`Plasma η
`Electron temperature T e
`Ion acceleration energy ε {
`Fractional ionization Χ- Σ
`
`Units
`
`mTorr
`W
`MHz
`1
`2
`c m
`kG
`c m "
`V
`V
`—
`
`3
`
`rf Diode
`
`High-Density
`Source
`
`10-1,000
`50-2,000
`0.05-13.6
`1-10
`300-2,000
`0
`- 1 0
`1 0
`1-5
`200-1,000
`i o - 6- i o ~ 3
`
`n
`
`9
`
`0.5-50
`100-5,000
`0-2,450
`2 - 50
`3 0 0 - 5 00
`0 -1
`1 0
`1 0
`- 1 0
`2 -7
`2 0 - 5 00
`ι ο ^ - ι ο " 1
`
`1 2
`
`and frequencies sometimes lower than the commonly used standard of
`13.6 MHz. For example, silicon nitride deposition used for chip encap-
`sulation is ordinarily performed at frequencies between 50 and 500 kHz
`where relatively large ion bombardment energies are used to tailor film
`stress and stoichiometry (78).
`
`B.
`
`LIMITATIONS OF CAPACITIVELY C O U P L ED RADIO FREQUENCY
`DISCHARGES
`
`A crucial limiting feature of rf diodes is that the ion bombarding flux
`Tj = nuB and the ion acceleration energy ε ι can not be varied indepen-
`dently. The situation is analogous to the lack of independent voltage and
`current control in diode vacuum tubes or semiconductor pn junctions.
`Hence, for a reasonable (but relatively low) ion flux, as well as a
`reasonable dissociation of the feedstock gas, sheath voltages at the
`driven electrode are high. For wafers placed on the driven electrode, this
`can result in undesirable damage, or loss of linewidth control. Further-
`more, the combination of low ion flux and high ion energy leads to a
`relatively narrow window for many process applications. The
`low
`process rates resulting from the limited ion flux in rf diodes often
`mandate multiwafer or batch processing, with consequent loss of wafer-
`to-wafer reproducibility. Higher ion and neutral fluxes are generally
`required for single wafer processing in a clustered tool environment, in
`which a single wafer is moved by a robot through a series of process
`
`Page 12 of 122
`
`
`
`10
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`chambers. Clustered tools are used to control interface quality and are
`said to have the potential for significant cost savings in fabricating
`integrated circuits (79). Finally, low fractional ionization poses a signifi-
`cant problem for processes where the feedstock costs and disposal of
`effluents are issues.
`To meet the linewidth, selectivity and damage control demands for
`next-generation fabrication, the mean ion bombarding energy, and its
`energy distribution, should be controllable independently of the ion and
`neutral fluxes. Some control over
`ion bombarding energy can be
`achieved by putting the wafer on the undriven electrode and indepen-
`dently biasing this electrode with a second rf source. Although these
`so-called rf triode systems are in use, processing rates are still low at low
`pressures and sputtering contamination is an issue.
`Various magnetically enhanced rf diodes and triodes have also been
`developed to improve performance of the rf reactor. These include, for
`example, the Applied Materials AMT-5000 magnetically enhanced reac-
`tive ion etcher (MERIE) and the Microelectronics Center of North
`Carolina's split cathode rf magnetron. In the AMT MERIE, a dc
`magnetic field of 50-100 G is applied parallel to the powered electrode,
`on which the wafer sits. The magnetic field increases the efficiency of
`power transfer from the source to the plasma and also enhances plasma
`confinement. This results in a reduced sheath voltage and an increased
`plasma density when the magnetic field is applied (80,81). However, the
`plasma generated is strongly nonuniform both radially and azimuthally
`because of Ε χ Β drifts, where Ε and Β are the local electric and
`magnetic fields, respectively. To increase process uniformity (at least
`azimuthally), the magnetic field is rotated in the plane of the wafer at a
`frequency of 0.5 Hz. While this is an improvement, MERIE systems do
`not have good uniformity, which may limit their applicability to next-
`generation, sub-micron device fabrication. Indeed, the strongly nonuni-
`form plasma over the wafer can give rise to a lateral dc current that can
`damage thin gate oxide films (see Section IX.C).
`
`C.
`
`OVERVIEW OF HIGH-EFFICIENCY SOURCES
`
`The limitations of rf diodes and their magnetically enhanced variants
`have led to the development of a new generation of low-pressure, high-
`efficiency plasma sources. A few examples are shown schematically in
`Fig. 3, and typical source and plasma parameters have been given in
`Table 1. In addition to high density and low pressure, a common feature
`
`Page 13 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`11
`
`MICROWAVES
`
`11
`•
`
`r—*»
`LU
`
`Ξ
`S
`
`RF ANTENNA
`
`ι
`
`-MULTIDIPOLES
`
`ECR
`
`HELICON
`
`•RF BIAS
`
`C
`C
`
`RF
`
`RF
`
`HELICAL RESONATOR
`
`INDUCTIVE
`
`FIG. 3. Some high-density remote sources.
`
`is that the rf or microwave power is coupled to the plasma across a
`dielectric window, rather than by direct connection to an electrode in
`the plasma, as for an rf diode. This non-capacitive power transfer is key
`to achieving low voltages across all plasma sheaths at electrode and wall
`surfaces. Dc voltages, and hence ion acceleration energies, are then
`typically 20-30 V at all surfaces. To control the ion energy, the electrode
`on which the wafer is placed can be independently driven by a capaci-
`tively coupled rf source. Hence, independent control of the ion/radical
`fluxes (through the source power) and the ion bombarding energy
`(through the wafer electrode power) is possible. This subject is discussed
`at greater length in Section VIII.
`
`Page 14 of 122
`
`
`
`12
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`The common features of power transfer across dielectric windows and
`separate bias supply at the wafer electrode are illustrated in Fig. 3.
`However, sources differ significantly in the means by which power is
`coupled to the plasma. For the electron cyclotron resonance (ECR)
`source shown in Fig. 3a, one or more electromagnet coils surrounding
`the cylindrical source chamber generate an axially varying dc magnetic
`field. Microwave power is injected axially through a dielectric window
`into the source plasma, where it excites a right-hand circularly polarized
`(RHP) wave that propagates to a resonance zone, for cold electrons, at
`is the applied radian
`ω = œce where the wave is absorbed. Here ω = Inf
`frequency and œ ce = eB/m is the electron gyration frequency at reson-
`ance. For the typical microwave frequency / = 2,450 MHz used, the
`resonant magnetic field is Β « 875 G. The plasma streams out of the
`source into the process chamber in which the wafer is located.
`A helicon source is shown in Fig. 3b. A weak (50-200 G) dc axial
`magnetic field along with an rf-driven antenna placed near the dielectric
`cylinder that forms the source chamber allows excitation of a helicon
`wave within the source plasma. Resonant wave-particle
`interaction
`(Landau damping) is believed to transfer the wave energy to the plasma
`(82-86)
`(Section IV.D). For the helical resonator source shown in Fig.
`3c, the external helix and conducting cylinder surrounding the dielectric
`discharge chamber form a slow wave structure, i.e., supporting an
`electromagnetic wave with phase velocity much less than the velocity of
`light. Efficient coupling of the rf power to the plasma is achieved by
`excitation of a resonant axial mode (Section VI). An inductive (or
`transformer) coupled source is shown in Fig. 3d. Here the plasma acts
`as a single-turn, lossy conductor that is coupled to a multiturn non-
`resonant rf coil across the dielectric discharge chamber; rf power is
`inductively coupled to the plasma by transformer action (Section V). In
`contrast to the ECR and helicon sources, a dc magnetic field is not
`required for efficient power coupling in the helical resonator or inductive
`sources.
`Figure 3 also illustrates the use of high-density sources to feed plasma
`into a relatively distinct, separate process chamber in which the wafer is
`located. As shown in the figure, the process chamber can be surrounded
`by dc multidipole magnetic fields to enhance plasma confinement near
`the process chamber surfaces, while providing a magnetic
`field-free
`plasma environment at the wafer. Such configurations are often called
`"remote" sources, another misnomer since at low pressures considerable
`plasma and free radical production occurs within the process chamber
`near the wafer (see Section VIII.D). Hence, such sources are not actually
`
`Page 15 of 122
`
`
`
`D E S I GN OF H I G H - D E N S I TY P L A S MA S O U R C ES
`
`13
`
`remote. For reasons that are discussed further in Sections II.A.2, VIII.D,
`and IX.C, the source and process chambers are sometimes combined, or
`the wafer is placed very near to the source exit. Such configurations are
`useful for obtaining increased ion and radical fluxes, reducing the spread
`in ion energy, and improving process uniformity. But the wafer is
`exposed to higher levels of damaging radiation as well (Section IX).
`Although the need for low pressures, high fluxes and controllable ion
`energies has motivated high-density source development, there are many
`issues that need to be resolved. A critical issue is achieving the required
`process uniformity over 200-300 mm wafer diameters. In contrast to the
`nearly one-dimensional geometry of typical rf diodes (two closely spaced
`parallel electrodes), high-density sources are often cylindrical systems
`with length-to-diameter ratios of order or exceeding unity. Plasma
`formation and
`transport
`in such geometries
`is inherently
`radially
`nonuniform. Another critical issue is efficient power transfer (coupling)
`across dielectric windows over a wide operating range of plasma
`parameters. Degradation of and deposition on the window can also lead
`to irreproducible source behavior and the need for frequent, costly
`cleaning cycles (87). Low-pressure operation leads to severe pumping
`requirements for high deposition or etching rates and hence to the need
`for large, expensive vacuum pumps. Furthermore, plasma and radical
`concentrations become strongly sensitive to reactor surface conditions,
`leading to problems of reactor aging and process
`irreproducibility.
`Finally, dc magnetic fields are required for some source concepts. These
`can lead to magnetic field induced process nonuniformities and damage,
`as seen, for example, in MERIE systems (88).
`
`II. Principles of Low-Pressure, High-Efficiency Source Design
`
`For the pressures of interest (see Table I), the plasma is not in thermal
`equilibrium, and local ionization models (89), where the ionization rate
`is a function of the local field and gas density only, fail. For all sources, the
`electrical power is coupled most efficiently to plasma electrons. In the
`bulk plasma, energy is transferred inefficiently from electrons to ions and
`neutrals by weak collisional processes; for ions, energy can also be
`coupled by weak ambipolar electric fields. The fraction of energy
`transferred by elastic collision of an electron with a heavy ion or neutral
`is 2 m / M ~ 1 0 ~ 4, where m and M are the electron and heavy particle
`
`Page 16 of 122
`
`
`
`14
`
`M. A. L I E B E R M AN A ND R. A. G O T T S C HO
`
`masses. Hence, the electron temperature T e much exceeds the ion and
`neutral temperatures,
`and % respectively, in the bulk; typically,
`7^~5V, whereas 7J and Τ are a few times room temperature (90). A
`more complete discussion of the ion temperature is given in Section VIII.
`However, dissociation and excitation processes can create a subgroup of
`relatively high-energy heavy particles. Also, the ambipolar electric fields
`accelerate positive ions toward the sheath edge, and typically, the ions
`in the bulk acquire a directed energy at the sheath edge of order TJ2.
`At these low pressures, the mean free path for ionizing electrons,
`with energies of 10-15 V, is typically comparable to the source dimen-
`sions. Hence, even if the electric power is deposited in a small volume
`within an unmagnetized source, the electron-neutral ionization rate v i z
`is expected to be relatively uniform, since the ionization occurs on the
`distance scale of this mean free path. In magnetized plasmas, on the
`other hand, the ionization rate may be highly nonuniform as
`the
`magnetized electrons have trouble crossing field lines, so ionization
`along a magnetic flux tube might be uniform but significant radial
`nonuniformities may persist. In addition, the propagation and absorp-
`tion of the exciting electromagnetic fields depend on the charge density
`distribution. The coupling is nonlinear and c