Nanoscale MRAM Elements

S.J. Pearton⁽¹⁾ and J.R. Childress⁽²⁾

(1)Department of Materials Science and Engineering

University of Florida, Gainesville, FL 32611

(2)IBM Almaden Research Center, San Jose, CA

Table of Contents

Introduction

Magnetic Devices

(a) Theory of GMR

(b) Basic Mechanism of GMR

(c) GMR Read Heads

(d) Magnetic Random Access Memory

Patterning Processes

(a) The transition from wet etching to dry etching

(b) The ultimate goals of a dry etch process

(c) Basic mechanism of the dry etch process

(d) Dry etching techniques

(e) High density reactors

Dry Etching Mechanism of Cu and Magnetic Materials with UV Illumination

Introduction

Ferromagnetic thin films and multilayers have been researched intensively in recent years for application in various magnetic recording and nonvolatile memory devices [1]. Interest in these materials has increased dramatically since the discovery of giant magnetoresistance (GMR) in multilayers comprised of alternating ultrathin (10 to 50Å) ferromagnetic/noble metal layers. Briefly, the GMR effect can be understood in terms of spin-dependent scattering of conduction electrons at ferromagnetic/nonmagnetic interfaces. Conduction electrons with a spin direction parallel to a material’s magnetic orientation move freely, providing low resistance. Conversely, conduction electrons with spin direction opposite to the materials magnetic orientation are hampered by more frequent collisions with atoms in the material, producing higher resistance.

In computer and data processing systems the main form of data storage and retrieval is based on magnetic recording systems, either magnetic disks or tape drives [2-16]. Information is written and stored as magnetization patterns on a recording media, and can be transferred back and forth using a magnetic sense head. In addition Magnetic Random Access Memories (MRAM) are used for storage and processing of very high bit densities. These devices offer the advantage over semiconductor memories of being radiation-hard, high storage densities, fast access time, and infinitely rewritable. Finally there are magnetic sensors, transducers and actuators which are used in automobiles, aircraft, hydraulic equipment and defense applications (mine detection, perimeter defense) [1,9]. In all of these structures there is a need to pattern the magnetic layers [17] generally thin films of materials such as NiFe, NiFeCo, which are often incorporated into multilayers comprising magnetic and non-magnetic materials.

The fabrication of read/write heads and other magnetic storage elements requires methods for producing sub-micron features in multi-layer structures involving NiFe and NiFeCo [1,13-17]. Due to the relative involatility of the etch products of these materials in conventional plasma reactors, virtually all of the patterning is performed using ion beam etching or lift-off processes [17-20]. Both of these methods have drawbacks for producing very small features-ion beam etching has poor selectivity with respect to mask materials and can produce redeposition of the sputtered materials onto the feature sidewalls that degrades the magnetic performance of the devices [18,19] while lift-off processes typically suffer from poor yield relative to direct etch removal methods.

In a typical reactive plasma process, exposure of NiFe and NiFeCo to chlorine or other associated feed gases produces a reaction or selvedge layer consisting of involatile etch products (i.e. NiCl_X, FeCl_X, CoCl_X species) [18-22]. Since these do not leave the surface and their atomic volumes are typically larger than those of the elements they are formed from, there is actually net deposition, rather than etching. There are basically two methods to enhance desorption of the etch products. The first is to heat the sample during plasma exposure to increase etch product volatility, but typically temperatures above 300°C are required [18] and this exceeds the thermal “comfort-zone” of most magnetic multilayer structures. The second is to employ a high ion flux in conjunction with the reactive chlorine neutral flux incident on the sample. The ion energy, however, must be kept low under these conditions to avoid mask degradation and loss of etch selectivity. The etch rates are a strong function of ion flux, ion energy and plasma gas composition, all of which may be interpreted in terms of balancing formation of chloride etch products with concurrent ion-assisted desorption of these products. Thus, the role of ion bombardment is critical in this mechanism. Alternative plasma chemistries including bromine or iodine containing plasmas were also examined under high density plasma conditions. The inter-halogens ICl and IBr have been found to dissociated readily in high density plasma sources, producing high concentrations of reactive species.

However one concern with this process is the effect of residual chlorine or chlorinated etch residues remaining on the sidewalls of etched features. We report on the effectiveness of several different in-situ or ex-situ cleaning processes after Cl₂/Ar high density plasma etching of GMR elements. We find that Inductively Coupled Plasma (ICP) etching does not degrade the magnetic performance of single or multilayer structures, and subsequent water rinsing or in-situ H₂ or SF₆ plasma cleaning is efficient in preventing corrosion.

Another method for preventing corrosion is applying non-corrosive gas chemistries such as CO/NH₃ or CO₂/NH₃. Nakatani [23] reported use of a non-corrosive CO/NH₃ plasma chemistry for NiFe, in which the etch products are expected to be carbonyls (e.g. Fe(CO)₅). The resultant etch rates were £300Å·min^-1 for Ni_0.8Fe_0.2, a factor of about three higher than purely physical Ar⁺ sputtering under the same conditions. The reactor employed was non-commercial, and might be classified as a magnetron-type, medium ion density system. We report on a parametric investigation of CO/NH₃ etching of NiFe, NiFeCo and related thin film materials using an Inductively Coupled Plasma (ICP) reactor. We find maximum etch rates for NiFe and NiFeCo in the range 350-400Å·min^-1, with the rates being a strong function of CO:NH₃ ratio, source and rf chuck power and pressure. We also compared use of CO₂ to replace CO, though this does not appear to offer any advantage in etch rates.

In this review, a comparison is given of high density plasma reactors such as ECR (Electron Cyclotron Resonance) and ICP (Inductively Coupled Plasma) based on either corrosive gas chemistries (Cl₂, IBr, ICl) or non-corrosive gas chemistries (CO/NH₃, CO₂/NH₃) will be described for the etching of NiFe, NiFeCo and other elements in magnetic multilayers. The influence of post-etch cleaning procedures on magnetic properties is also examined. Finally, the selectivity for etching the magnetic materials over common mask materials (photoresist, SiO₂) was measured over a broad range of conditions.

MAGNETIC DEVICES

(a) The Theory of GMR

In any material, some conduction electrons will be oriented with "up" spins, and some will be oriented with "down" spins. The material can be viewed as having two independent conduction channels, one for each electron spin orientation. The measured current will be the sum of these channels:

(1)

In a non-magnetic material, these two currents are equal. In a magnetic material, such as Fe or Co, the two currents are different: interaction of a conduction channel with its surroundings will be spin-dependent. First, the conduction electron energy bands will be different for majority spin electrons (spin parallel to the magnetization). Second, minority and majority spin electrons will have different scattering potentials when they encounter impurities and interfaces. Thus, in a bulk ferromagnetic material, the resistivity of the majority spin channel (r) is typically different from the minority spin channel (r¯) [24,25].

In a multilayer structure where the layers are antiparallel, however, the up conduction channel will be scattered as majority spin electrons in one layer and as minority spin electrons in the next layer, and vice versa. For this reason, the net resistivities of two channels will be the same, and approximately equal to the average resistivity of the two channels in the bulk:

(2)

Conductivities (s) of the two channels (s₁, s₂) add in parallel and the resistivity is the reciprocal of the conductivity. So:

(3)

In the absence of an applied field, r₁=r₂=r, and the total resistivity will be

(4)

At the saturation field, on the other hand, r₁=r and r₂=r¯. The layers will be oriented in parallel:

(5)

Clearly,

(6)

In other words, an applied magnetic field (for instance, a field stored in magnetic media) "short circuits" the magnetic layers, reducing the total resistivity. This short circuit is observed as the GMR effect.

(b) Basic Mechanism of GMR

Giant magnetoresistance is largely caused by a quantum effect called spin-dependent scattering, which results from the fact that the spin of an electron can point either up or down. Electrical resistance is caused by the scattering of electrons, but magnetic materials scatter spin-up and spin-down electrons differently (Figure 1) [26].

Given a difference in resistivity between spin-up and spin-down electrons, the low resistance of the multilayer is generated when conduction electrons with a spin direction parallel to a material’s magnetic orientation move freely. Conversely the higher resistance can be produced when conduction electrons with spin direction opposite to the material’s magnetic orientation are hampered by more frequent collisions with atoms in the material.

A structure consisting of two magnetic layers are separated by a non-magnetic spacer layer. If the layers are magnetized in the same direction, both will scatter electrons in the same way and the structure will have almost the same resistance as the bulk material. However, if the layers are magnetized in opposite directions, one layer will mainly scatter spin-up electrons, while the other will scatter spin-down electrons. The overall resistance will increase.

The key point is that only a weak magnetic field is needed to change the orientation of the magnetic fields in the layers. By using a strong antiferromagnet to keep the orientation in one of the layers constant, changes in resistance can be linked to the magnetic state of the other layer.

There are two significant advantages of GMR over competing devices. First, the large resistance change yields a strong signal, and second, the technology is compatible with integrated circuit technology, so that GMR devices can be included as part of a chip package to make smaller, faster, less expensive sensors and memory chips.

(c) GMR Read Heads

In the past couple of years, production of read heads for hard-disk drives has been largely converted from thin-file inductive heads to magnetoresistance heads using thin-film 80:20 nickel/iron magnetoresistors. This conversion was needed in order to get larger signals from smaller storage bits and keep pace with improvements in hard drives.

A read head detects a transition in magnetization stored in the disk as the head passes over it. The magnetic field produced by the transition is either up (head-to-head magnetization transition) or down (tail-to-tail magnetization transition).

The GMR read head uses a feature called a spin valve, which is etched into a material whose edge is oriented along the direction of the disk surface. Spin valve material normally composes of four thin films: a sensing layer, a conducting spacer, a pinned layer, and an exchange layer. The first three films are very thin, allowing conduction electrons to frequently move back and forth between the sensing and pinned layers via the conducting spacer (Figure 2) [26]. The magnetic orientation of the pinned layer is fixed and held in place by the adjacent exchange layer, while the magnetic orientation of sensing layer changes in response to the magnetic field from the disk. A change in the magnetic orientation of the sensing layer will cause a change in the resistance of the combined sensing and pinned layers.

In the GMR read head, the magnetization of the pinned spin-valve layer is directed vertical to the disk surface, and the soft layer lies parallel to the surface in the absence of a field from the disk. Up and down magnetic field, which create smaller or larger resistance in the spin valve are generated by stored magnetic data on a rotating disk.

(d) Magnetic Random Access Memory

Nonvolatility, the ability to store data when electricity is off, is a much desired memory property in applications where data retention is critical. Semiconductor nonvolatile random access memory technologies, such as electrically erasable programmable read-only memory (EEPROM) and flash memory, suffer from slow write times and wear out after data are stored more than 1 million times. Memory that uses magnetic materials can have fast write times and can store data indefinitely.

Working memory systems using these magnetic RAM (MRAM) chips have used for space and missile application in which resistance to radiation damage is critical. Nonvolatility and radiation hardness are important for space missions, and magnet storage is intrinsically radiation-hard. Recently, GMR materials have been used to make MRAMs with faster read access times.

With proper design of the areal geometry and the right thickness of magnetic films, submicrometer-size MRAM cells (single memory bits) have been made to operate like a spin valve, and they have been called "pseudo spin valve" devices. They promise very high density, high speed and nonvolatile memory applications.

A newer high-mangetoresistive innovation, called a spin-dependent tunneling (SDT) device, or magnetic tunnel junction (MTJ), uses tunneling current through a thin dielectric between two ferromagnetic films. STD devices work in a similar way to GMR devices. When two magnetic layers are magnetized in the same direction, the tunneling current is generally higher than when the two magnetic devices are magnetized in opposite directions.

PATTERNING PROCESSES

(a) The Transition from Wet Etching to Dry Etching

Etch processes may be classified by their rate, selectivity, uniformity, directionality (isotropy or anisotropy), surface quality, and reproducibility. All etching processes involve three basic events: (1) movement of the etching species to the surface to be etched, (2) chemical reaction to form a compound that is soluble in the surrounding medium, and (3) movement of the by-products away from the etched region, allowing fresh etchant to reach the surface. Both (1) and (3) usually are referred to as diffusion, although convection may be present. The slowest of these processes primarily determines the etch rate, which may be diffusion or chemical-reaction limited. There are two different etching methods by using two quite different media: liquid chemicals (wet etching) and reactive gas plasmas (dry etching). Wet etching is performed by immersing the wafers in an appropriate solution or by spraying the wafer with the etchant solution [27]. Wet-chemical etching is superior to dry etching in terms of effectiveness, simplicity, low cost, low damage to the wafer, high selectivity and high throughput. However, the main limitations of wet etching include its isotropic nature which results in roughly equal removal of material in all directions, making it incapable of patterning sub-micron features, and the need for disposal of large amounts of corrosive and toxic materials. As the requirements developed for increasing circuit density and narrower line-width in the manufacture of VLSI (very large-scale integrated circuit)/ULSI (ultra large-scale integrated circuit) devices, it became necessary to have new etching methods to replace the wet etching. Dry etching methods became favorable etching processes for integrated circuit manufacture. Plasma-driven chemical reactions and /or energetic ion beams are used to remove materials in dry etching system. The most significant advantage of dry over wet etching is that it provides higher resolution potential by overcoming the problem of isotropy. Other benefits are the reduced chemical hazard and waste treatment problems, and the ease of process automation and tool clustering.

(b) The Ultimate Goals of a Dry Etch Process

The success of a etch process must be measured by the nine parameters listed below. The greatest challenge is that each parameter can usually only be optimized at the expense of at least one of the others.

(1) Critical dimension uniformity

Uniformity across the wafer-including densely populated areas and large open spaces, and within high aspect ratio features-is critical to maintain consistent device performance. Aspect ratio dependent etching also known as "micro-loading" is a common non-uniform problem.

(2) Selectivity

Defined as the ratio of the etch rate of one material versus that of another, the selectivity of the material being etched to the overlying masking layer (typically photoresist) is usually of the most concern, since this impact critical dimension and profile control, and the thickness of resist required (thinner photoresist is required to adequately resolve smaller feature sizes, so selectivity must increase as geometry shrinks). Also of concern is the selectivity to the underlying material upon which the etch stops. Different selectivity specifications may be given for edges and flat areas, since edges tend to etch faster.

(3) Etch rate

High etch rate is needed to keep the throughput of the system or process module high (usually measured in Å/min). There is usually a tradeoff between etch rate and other parameters, such as selectivity and damage.

(4) Etch profile control

It's usually desirable to have an anisotropic profile, that is one where the etched feature edges are close to vertical, to maximize packing density on the chip. But it's also desirable to have a flare out at the top of the feature to enable good step coverage in subsequent deposition steps.

(5) Low damage

Damage is an obvious concern. The high energy of the plasma can create currents on the wafer surface that cause electrical damage and energetic ions can cause mechanical damage to the films' crystalline structure.

(6) Residue

Residue which coats the interior of the etch chamber is a difficult problem to avoid. In addition to requiring more frequent cleaning, residue is also a source of contamination. The most significant factors in controlling residue are temperature, bottom rf power, backside cooling and process pressure.

(7) Corrosion

Corrosion is mainly a problem in metal etch. Upon exposure to water vapor (i.e. air), chlorine will immediately attack metals. Integrated post-etch treatments help eliminate this problem.

(8) Particle control

Particle control is another critical measurement of etch system performance. Today, fewer than 0.05 particles/cm² that are >0.35mm in size are required.

(9) Sidewall passivation

Sidewall passivation is important both during and after the etch. Carbon from the photoresist mask typically combines with etching gases and etch byproducts to form a polymer-like material on the sidewall of the feature. This is usually a requirement in creating anisotropic profiles. The biggest challenge is that after the etch, this polymer must be removed.

(c) Basic Mechanism of the Etch Process

The optimization of etch of these parameters-uniformity, selectivity, etch rate, profile control, damage, and residue control-requires an understanding and fine-tuning of the two very different mechanisms through which etching occurs.

As shown in Figure 3, one mechanism is purely chemical. Reactive species generated in plasma react with the wafer surface and create volatile etch products that swept away. By careful selection of the gases that are flowed into the plasma (typically chlorine and /or fluorine containing gases), it's possible to achieve very high selectivity through this process. However, since films tend to etch in all directions at once, the result is an isotropic etch. The other mechanism is purely physical. Energetic ions crossing the sheath transfer large amounts of energy and momentum to the substrate. The force of these ions can be strong enough to physically remove material. At low pressure where the mean free path is long, the ejected sputtered material can cross the reactor vessel and reach opposing walls. The main benefit of this etch mechanism is that it provides some directionality to the etch, making it possible to achieve highly anisotropic profiles. However, it is the least selective mechanism and also suffers from the disadvantages of low etch rate, damage and trenching [28].

(1) Sputtering

In sputtering, impinging particles (usually positive ions accelerated across the sheath) strike the surface with high kinetic energy. Some of the energy is transferred to surface atoms which then are ejected, leading to a net removal of material. This process is distinguished from other etching mechanisms in that the interaction is mechanical. It is sensitive to the magnitude of bonding forces and structure of a surface, rather than its chemical nature and quite different materials can sputter at similar rates. In a way this is symptomatic of using ion bombardment with energy far higher than the surface binding energy.

(2) Chemical etching

Chemical etching comes about when active species from the gas phase encounter a surface and react with it to form a volatile product. Product volatility is necessary for chemical etching since involatile products would coat the surface and protect it from further attack. In this type of the plasma reactor converts the feed into reactive chemical species, which are usually free radicals. There is usually no directionality and the etching can be specific (high selectivity) since it its governed by the relative chemical affinities between the etchant species and exposed materials. Because of this lack of directionality, chemical etching is commonly called isotropic etching

(3) Energy-driven ion-enhanced etching

There is usually little of no etching when the substrate surface is exposed to neutral chemical species alone in the absence of ion bombardment. Impinging ions damage the substrate material by virtue of their impact energy, and thereby render the solid substrate more reactive toward incident neutral radicals.

(4) Inhibitor-driven ion-assisted etching

Inhibitor-protected sidewall ion-enhanced etching differs from energy-driven ion-enhanced etching is that the chemical etching reaction in spontaneous, even without ion bombardment. Neutral etchant species from the plasma spontaneously gasify the substrate, and ions play a role by interaction with another component-a 'protective' inhibitor film. The role of ions in the surface-inhibitor mechanism is to clear the inhibitor from horizontal surfaces that are bombarded by the flux of ions impinging in the vertical direction. The protective film is not removed from the vertical walls of masked features because these surfaces only intercept those few ions that are scattered as they cross the sheath. This protective film may originate from involatile etching products or from film-forming precursors that adsorb during the etching process.

(d) Dry Etching Techniques

(1) Plasma etching

A wafer is exposed to a reactive gas such as chlorine which is in some cases dissociated in a plasma to create highly reactive atoms. Etching is isotropic or crystallographic, and temperature and reactant flux are used to adjust etch character. During the purely chemical etching process, three steps occur: adsorption of the necessary species on the materials surface, chemical reaction, and desorption of the products. The advantage of the technique is rapid etch rates, but the drawbacks are isotropy, a tendency for strong loading effects, and release of heat.

(2) Ion Beam Etching

Ion beam etching (ion milling) uses a broad-area ion beam composed of a nonreactive gas such as Ar with high ion energy. This technique only relies on physical sputtering. The uniform ion beam bombards a wafer to cause etching. Since ion beam etching is very anisotropic and the wafer is not exposed to plasma, it is well suited for etching of certain patterns.

A major problem with ion beam etching is that the etch rate is very low because of the nature of physical sputtering, and it is also recognized that the etch rate of ion milling is very much dependent upon the incidence angle. It typically peaks at between 30° and 50°, and becomes very small when angle is larger than 80°. There are a number of factors that limit the application of ion beam etching. First of all, etch products, usually nonvolatile, can redeposit on the wafers, especially onto the sidewalls of etch mask that degrades the magnetic performance of the devices. When the film plane is normal to the ion beam, the etch rate of the sidewalls is very small since the incidence angle there is close to 90°. As a result, after the etch mask is removed, undesired fence and trenches are often left on the edges of the etched patterns.

(3) Reactive Ion Etching

Reactive ion etching uses radiofrequency power to maintain a plasma. Applied rf power makes electrons accelerate in the sheath region changing direction upward and downward. The accelerated electrons can lose a large fraction of their kinetic energy through dissociative collisions with molecules and atoms. Because electrons are light and have high energy, they diffuse fastest, leaving an excess of positive charge and a plasma potential that is positive relative to the electrode. Since charged particles are most abundant in the central glow of the plasma, most of the potential drop appears across the sheath. Positive ions are accelerated through the sheath and strike the samples, giving rise to a physical etch component. In addition, ion flux is coupled with ion energy. High plasma densities and ion fluxes are gained at the expense of extremely high applied voltages and damage levels.

(e) High-Density Plasma Reactor

Increasingly, the limitation of traditional plasma etch technologies-which has resulted in the trend to high density plasma sources-is primarily one of process pressure. In the pressure regime of a few hundred millitorr where it's relatively straightforward to create a plasma, it becomes difficult to get etchant in and reaction byproducts out of openings that are smaller than about 0.25mm. The problem is more sever with higher aspect ratio.

The solution is to go to lower pressures, where the mean free path lengths of gas molecules and ions are longer, which reduces scattering collisions that can cause loss of profile control.

However, this is not as simple as it may seem since it requires a switch to a different type of plasma source-a so-called high density source-that is capable of generating enough ions to achieve acceptable etch rates at reduced pressure. High density sources are designed to more efficiently couple input power with the plasma, resulting in greater dissociation of etch species.

Although a wide variety of high density source types have been developed, they generally fall into one of three categories: electron cyclotron resonance (ECR), helicon resonance or inductively coupled plasma (ICP) type sources. All three are in use on production equipment. The main difference is that ECR and helicon sources employ an external magnetic field to shape and contain the plasma, while ICP sources do not [29].

(1) Electron Cyclotron Resonance (ECR) Plasma

Various methods have been developed for reducing ion energies in the discharge which trying to maintain anisotropic etching. One of the methods involves the addition of magnetic fields configured to reduce electron loss from the plasma and the sample. This method of magnetically enhancing the discharge is called ECR plasma etching. In ECR discharges, free electrons in the plasma are forced to orbit about magnetic field lines while absorbing microwave energy. At the cyclotron resonance condition, outer shell electrons from gas molecules in the discharge may also be liberated, leading to a very high degree of ionization in the plasma. Since the motion of the electrons is constrained by the external magnetic field, fewer are lost by collisions with the reactor walls than conventional RIE, and therefore the plasma potential relative to ground is much lower. The resultant energies of an ion reaching the sample to be etched are typically £15eV. ECR etching should lead to much lower levels of damage than conventional RIE processes. ECR sources operate at the relatively high ion density of 10¹¹~10¹²cm^-3 compared with RIE tools (~10⁹cm^-3). ECR tools can also provide independent control of the ion energy and ion flux. Ion energy is controlled by rf or dc biasing of the substrate holder while control of the ion and neutral flux is achieved by varying the microwave power and neutral gas pressure. In addition, ECR discharges are capable of low pressure operation due to its efficient dissociation of gases from the discharge.

(2) Inductively Coupled Plasma (ICP)

Another high density plasma, ICP, has become popular because of disadvantages of the ECR technology. ECR suffers from difficulties in uniformity. The power supply may also limit the scaling of the ECR approach without the development of higher power magnetrons. The primary disadvantages of the ECR technology are the limits due to a commercially available, automatic-tuning microwave power supply, and the physical limits of the magnets required to create a uniform magnetized plasma. ECR tools appear difficult to scale to process wafers larger than 200mm.

In the ICP geometry, a rf coil encircles the chamber. The important features of this coil are that it carries rf current and generates a magnetic field in the upward and downward directions. The time rate of change of the individual magnetic field generates an electric field. Acceleration of the electrons is determined by the magnitude of the electric field, confining them in a circular motion. The plasma, first formed in the shape of a ring following the path of electrons, will diffuse to the center of the chamber and then downward toward the sample. The electrons, in circular path, will have only a small chance to be lost to the chamber walls, resulting in low dc self bias. Ion energy, separated from the ion flux, can be controlled by applying another rf source at the chuck [30]. Unlike ECR plasma sources, there is no resonance between electron motion and the frequency of the driving fields in ICP sources. Therefore, ICP sources have advantages over ECR sources, including easier tuning, scaling up and lower cost.

The push to continually increase bit densities in magnetic storage devices places emphasis on techniques for patterning submicron metallic multilayer structures. The component materials within these multilayers may include NiFe, NiFeCo [both are used for structures based on the giant magnetoresistance (GMR) effect], NiMnSb (a Heusler alloy potentially useful in advanced spin-valve structures) or the col-lossal magnetoresistance (CMR) materials LaCaMnO₃, LaSrMnO₃, and PrBaCaMnO_3. A general problem when processing magneto-resistive materials is their relative invola-tility in conventional dry etching techniques such as reactive ion etching (RIE).

Practical etch rates may be achieved under high-density plasma (HDP) conditions, where the high ion flux is able to promote desorption of halogenated etch products. Both inductively coupled plasma (ICP) and electron cyclotron resonance (ECR) systems have proven capable of providing the necessary ion-to-reactive-neutral ratio (>0.02). We have completed an examination of the etch rates of the three basic classes of materials (i.e., NiFe, NiMnSb, and the perovskite-based CMR materials) in different plasma chemistries and also measured the long-term magnetic and structural stability of dry etched magnetic multilayer stacks. In the latter case, we used magnetic random access memory (MRAM) elements as our test vehicle.

The NiFe and NiMnSb layers were deposited on Si (100) substrates by direct current magnetron sputtering from composite targets. Typical layer thicknesses were 5000Å. Liquid delivery metalorganic chemical vapor deposition using 2,2,6,6-tetramethyl-3, 5-hepanedionato (TMHD) precursors [i.e., La(TMHD)₃ ,Sr(TMHD)₂ ,Mn(TMHD)₃ , and Ca(TMHD)₂] was employed to deposit films of La_0.41 Ca_0.59 MnO₃ on Al₂ O₃ (0001) single crystal substrates at 700°C. The precursors were transported by N₂ carrier gas, with direct injection of O₂ and N₂O as oxidants. Films of Pr_0.65 Ba_0.05Ca_0.3MnO₃ were deposited on Si (100) in a pulsed laser ablation system (248 nm KeF laser, 10 kHz, 2.5 J cm^-2) energy density) at a substrate temperature of 700°C and an O₂ partial pressure of 250 mTorr. All of the CMR films were in the range 1500–2500Å thick.

Finally, the MRAM structures consisted of the following layers deposited on 300Å of SiN_x on 8500Å of SiO₂ on Si: 80Å NiFeCo, 15Å CoFe, 35Å Cu, 15Å CoFe, 80Å NiFeCo, 200Å Ta, 550Å TaN, and 800Å CrSi. The deposition was performed by Ar ion-assisted sputtering. A

3000-Å-thick SiO₂ mask was patterned by SF₆/Ar RIE as the etch mask for subsequent high-density plasma patterning of the metal layers.

The etching was performed in either Plasma-Therm 790 ICP or Plasma-Therm SLR 770 ECR reactors. In both systems the samples were thermally bonded to a radio frequency powered (13.56 MHz) chuck which was He-backside cooled. The respective HDP sources were powered up to 1000 W at either 2 MHz (ICP) or 2.45 GHz (ECR). The gases were injected directly into the sources through electronic mass flow controllers at a typical load of 15–20 standard cubic centimeters per minute. We investigated halogen- (Cl_2, BI₃, BBr₃, ICl, IBr, SF₆), CH₄/H₂- and CO/NH₃-based mixtures since these cover the full range of possible etch products (i.e., metal chlorides, bromides, iodides or fluorides; metalorganics or carbonyls).

Magnetic properties before and after plasma etching were determined using superconducting quantum interference device magnetometry (Quantum Design MPMS-5S00) at 4.2 K. Scanning electron microscopy (SEM) was used to examine sidewall smoothness on etched features.

Under RIE conditions (i.e., zero watts HDP source power) we invariably saw net deposition on the samples upon exposure to halogenated mixtures, or essentially no etching with CH₄/H₂ and CO/NH₃ mixtures. Examination of the halogen-plasma exposed surfaces, by Auger electron spectroscopy revealed large concentrations of chlorinated residues. Since the halogenated etch products have larger lattice constants than their pure metal constituent and the products are essentially involatile under RIE conditions, then one observes a buildup of these species as shown schematically in Figure 6 [31-33].

Table I shows a compilation of results for NiFe etching in the different chemistries investigated. The highest rates were achieved with Cl₂/Ar, where the role of the inert gas additive

is to provide ion-assisted desorption of the chlorinated etch products. We found that the mass of these inert species also played a role, with Xe providing slightly faster rates than either Ar or He addition. The rates with these Cl₂ -based mixtures were approximately a factor of two faster than with pure Ar sputtering. Bromine or iodine-based plasma chemistries produced lower rates than with chlorine, and were close to Ar sputter rates. Both CH₄/H₂/Ar and SF₆/Ar led to extremely low etch rates, while the CO/NH₃ mixture had a slight degree (40%) of chemical enhancement. It has been suggested that the role of the NH₃ is to suppress dissociation of the CO so that carbonyl etch products can form, but an alternative explanation might be that atomic hydrogen scavenges surface carbon species and prevent carbonization of the NiFe surface. The fact that Cl2-based plasma chemistries produce the fastest rates for NiFe (and plasma chemistries produce the fastest rates for NiFe (and NiFeCo) is consistent with the higher vapor pressures of the chlorinated etch products relative to their brominated or iodidated counterparts.

Table II shows the corresponding comparisons for NiMnSb. In their cases the Cl₂ -based mixtures produce excellent etch rates (1500–5000Å min^-1 for both ICP and ECR tools), but the fastest rates were achieved with SF₆ /Ar mixtures. By sharp contrast, NF₃ /Ar showed net deposition rather than etching for source powers >100 W or at high NF₃ percentages. The surface under these conditions showed strong Mn enrichment and were oxidized, with an underlying Sb-deficient region. With all of the plasma chemistries, careful attention had to be paid to the removal of the native oxide prior to the commencement of etching to avoid the presence of a relatively long incubation time.

For the CMR materials, we did not observe any chemical enhancement in etch rate with any of the plasma chemistries discussed (Table III). The etching was dominated by physical sputtering under all conditions investigated, with etch yields typically <0.1 and relatively high ion energies (>150 eV) needed to initial removal of material.

A key issue with the use of corrosive gas mixtures for etching metallic multilayers is that of postetch stability of the patterned structures. Severe corrosion and delamination of the films is observed in the absence of preventive measures. We examined use of several different postetch treatments. The first was simply rinsing the samples in deionized water immediately upon opening the chamber (which is contained within a N₂ dry box). The samples were then thoroughly dried with filtered N₂. In the other three methods, various in-situ plasma cleaning procedures were examined. After Cl₂ /Ar etching was complete, the chamber was evacuated for 15 min, and then a 30 mTorr discharge of either H₂, O₂ or SF₆ (500 W source power, 5 W chuck power) was used to clean the residual chlorine for 10 min prior to removal of the samples from the reactor. In these cases, no H₂ O rinsing was performed. It should be pointed out that all of these cleaning procedures have been employed previously for removing etch residues after Cl₂-based plasma etching of Al interconnects in Si microelectronics.

Figure 7 shows the magnetization of each of the samples over a period of approximately six months. In each case the samples were simply stored in air between the measurements and no special precautions were taken to prevent corrosion. Each of the cleaning procedures produces samples with extremely stable magnetic characteristics. This is also reflected in their appearance. Figure 8 shows SEM micrographs of patterned MRAM elements three months after Cl₂/Ar etching and postetch cleaning. There is no indication of corrosion on any of the samples and the sidewalls are smooth (to the resolution used in the photos). There is no indication of striations often observed on dry etched features. Note, however, that in the case of O₂ plasma cleaned samples there was a slight decrease in the magnetization per unit volume relative to the samples treated in water or H₂ or SF₆ plasmas. A possible reason for this is that the feature sidewalls become more oxidized than with other treatments, leading to a degradation in magnetic properties.

A comprehensive survey of etching results for magnetic materials in different plasma chemistries has produced the following conclusions:

(i) The optimum chemistry for NiFe is Cl₂/Ar, for NiMnSb is SF₆/Ar, while no chemical enhancement of etch rates for CMR oxides was observed.

(ii) Postetch rinsing in H₂ O or in-situ plasma cleaning with H₂, O₂ or SF₆ discharges are all effective treatments for removing chlorine etch residues. Of these, only O₂ plasma exposure appears to degrade the magnetic properties of MRAM stacks. Once the residues are removed, there is no change in magnetic or visual properties over a period of ~six months (extent of our study).

(iii) The CO/NH₃ chemistry, while being noncorrosive, produces relatively slow etch rates and is only suitable for patterning of thin (<1000Å) structures.

Table IV shows thermochemical data for the potential metal chloride or metal carbonyl etch products for NiFe and NiFeCo in Cl₂ or CO/NH₃ plasmas. From this data we can calculate the Gibbs free energies of reactions of Ni, Fe and Co with atomic or molecular chlorine, and with CO and CO₂. There are several important features of this data in Table V. First, the reaction of the metals is more favorable with atomic chlorine than with Cl₂, which emphasizes the need for efficient dissociation of the feedstock gas in the plasma source. Second, CO is more reactive with the metals than is CO₂, as we have previously reported in a comparison of the two gases. We emphasize that in a plasma etching environment there will be a strong ion-assisted component to the etch mechanism and the thermodynamic data provides only a guide to the reaction pathways [34-37].

DRY ETCHING MECHANISM OF COPPER AND MAGNETIC MATERIALS WITH UV ILLUMINATION

In recent years several research groups have studied dry etching of copper for the next generation of metallization in the semiconductor industry, focusing on development of new etch techniques to increase etch rate [38-47]. They used Cl₂ plasmas with or without photon sources using ultraviolet (UV) laser, UV lamp, illumination and IR light. In contrast to conventional dry etching that requires relatively high temperatures (> 200^oC) in order to produce practical etch rates, they all reported substantial enhancement of etch rates at low temperatures [48]. Among them Choi and Han first reported high etch rates of about 3000 Å/min at room temperature with Cl₂ discharges in an Inductively Coupled Plasma (ICP) system.

Magnetic materials such as NiFe and NiFeCo are widely used in sensors, magnetic random access memories (MRAMs) or read/write heads for data storage industry. Due to the relative inertness of these materials there is a strong interest in the development of high density plasma etching processes for them. There are two basic plasma chemistries for the etching of NiFe and NiFeCo under ICP conditions, namely Cl₂ and CO/NH₃. However, the etch rates are still low (< 500 Å/min) and are limited by desorption of the etch products such as NiCl_x, FeCl_x and CoCl_x [49-65]. Cho et al. first reported the effect of UV illumination on the etch rates of the NiFe-based magnetic materials

Since the etch mechanism with UV illumination has not been studied in detail, in this paper we propose an etch mechanism of copper and magnetic materials with UV irradiation based on subprocesses occurring in the Cl₂ -ICP etching system. We also carried out ICP etching of NiFe and NiFeCo in Cl₂/Ar discharges with or without UV illumination. We found that the chlorination of copper surface is enhanced with UV irradiation and the absorption of UV photons by metal chlorides is critical to enhance the removal rate of chlorides. The proposed etch mechanism of copper showed good agreement with observed data determined by mass spectrometry, taken from the literature.

There are likely five subprocesses involved in etching of copper with UV illumination: 1) photo-dissociation of Cl₂ in gas phase, 2) surface chlorination, 3) absorption of UV photons by reaction products, 4) photo-assisted removal of reaction products, and 5) Gas-phase reactions between desorbed species.