Review Article: Tracing the recorded history of thin-film sputter deposition: From the 1800s to 2017

The use of thin films to enhance the physical and chemical properties of materials is ubiquitous in today's world. Examples are shown in Fig. 1: copper metallization layers for electronic communication among billions of transistors in a silicon integrated-circuit; coated architectural glass in office buildings for which the thin films are designed to enhance energy efficiency and comfort by, depending on the time of year and latitude, reflecting ultraviolet and infrared sunlight, while transmitting visible light, to minimize air conditioning usage, or reflecting infrared radiation from within offices to minimize heating; and coated cutting tools developed to reduce friction and wear during use and, hence, increase tool lifetimes. Other common examples include magnetic thin films for electronic data storage; transparent conductive oxide and absorber layers in solar cells; thin film resistors and dielectrics; catalytic layers for toxic-gas sensing; superconducting thin films for high-frequency devices, data storage, and magnetic circuitry; corrosion-, friction-, and wear-protective layers on automotive and airplane engine parts (spark-plug electrodes, pistons, cylinders, turbine blades, etc.); and multiple layers on eyeglasses to correct vision, minimize ultraviolet light transmission, and provide scratch resistance.

The adjective “thin” in the term “thin films” is ambiguous and poorly defined. It is used to describe, depending on the application, “coating” layers ranging in thickness from less than a single atomic layer (a partial monolayer) to films that are a significant fraction of a millimeter thick. The earliest documented purposefully made inorganic thin films were gold layers produced chemo-mechanically, for decorative (and later, optical) applications, by the Egyptians during the middle bronze age, more than 5000 years ago.^1,2 Gold films, with thicknesses <3000 Å (∼1465 atoms), have been found in ancient tombs, including the Pyramid of Djoser (actual name, Netjerykhet, second King of the Third Dynasty, Old Kingdom; ruled from ∼2667 to 2648 BC)³ in Saqqara,^1,3–5 southwest of Cairo, Egypt. The films were often gilded on copper and bronze statues, jewelry, and religious artifacts using mercury-based, compositionally graded, interfacial adhesion layers as discussed in Ref. 1.

The gold used to produce early thin films was mined by the Egyptians in the Eastern Desert, between the Nile River and the Red Sea. Ancient mining sites in Wadi Hammamat, along the trade route from Thebes (modern-day Luxor) to the Red Sea port of Al-Quseir, are accurately located on a papyrus map drawn by a scribe of Ramses IV during a quarrying expedition in approximately 1160 BC (Refs. 7 and 8) and now on display in the Museo Egizio, Turin, Italy.

Gold ore was purified by melting it in a mixture of “alum” [the mineral alunite: KAl₃(SO₄)(OH)₆], salt (NaCl), and chalcopyrites (e.g., CuFeS₂). The process produces sulfuric and hydrochloric acids (H₂SO₄ and HCl) which dissolve base metal impurities.^9,10 The purified gold still had several to a few tens of atomic percent of silver, copper, or both, depending upon where it was mined, thus giving rise to variations in color. Flattening of purified bulk gold was initiated by beating with a rounded stone and mechanical rolling, followed by many stages of thinning and sectioning of composite structures consisting of Au leaf sandwiched between layers of animal skins, parchment, and vellum.⁶ Figure 2 shows an image of a fresco from a tomb (∼2500 BC) in Saqqara, illustrating melting and purification of Au during which the temperature is adjusted by craftsmen using blowpipes (reeds) with clay tips.

Today, gold leaf can be beaten to ∼500 Å thick (partially transparent to visible light) by highly skilled craftsmen.⁵ In fact, the production of gold leaf, primarily for decorative purposes, remained a viable industry for craftsmen until the development, in the mid-1930s, of roll-to-roll sputter and evaporative coating technologies (see Sec. IV K).

Highly skilled ancient Egyptian craftsmen mastered the art of gold sheathing—the direct application of thin gold layers onto wooden and plaster objects (mostly for noble families) to provide the impression that the object is solid gold—at least as early as 2600 BC.^11,12 Striking examples were found in the tomb of Queen Hetepheres (wife, and half-sister, of Pharaoh Sneferu, Fourth Dynasty, Old Kingdom, ∼2613–2589 BC). Other spectacular specimens of early thin-film technology were found in the tomb of Pharaoh Tutankhamun (“King Tut,” 18th Dynasty, ruled ∼1332–1323 BC). Gold sheets were beaten into position over carved wooden structures to provide embossed hieroglyphic texts and decorations as shown in Fig. 3.

An autocatalytic solution-growth technique involving oxidation/reduction reactions was developed by the Moche Indians utilizing minerals available in the local area, the northern highlands of Peru, beginning ∼100 BC to deposit gold (as well as silver) films on copper and bronze artifacts.¹³ The technique is still in use today, although carried out in a more efficient manner, and referred to as electroless plating.¹⁴ Moche artisans first dissolved gold in a hot aqueous solution of equal parts potassium aluminum sulfate [KAl(SO₄)₂], potassium nitrate (KNO₃), and salt (NaCl), a process that took several days. The resulting mixture was then buffered with sodium bicarbonate (NaHCO₃) to form a weakly alkaline solution (pH ∼ 9), which was allowed to boil for several minutes before immersing the copper artifact to be plated. The overall reaction is

Metallographic studies of Moche artifacts, coated with gold films whose thicknesses ranged from ∼2000 Å to 1 μm, exhibit evidence of post-deposition heat treatment (annealing) to obtain a film/substrate interdiffusion zone, presumably for better adhesion. An excellent example of craftsmanship is depicted in Fig. 4.

The first thin films grown from the vapor phase, as discussed in Ref. 1, were likely metal layers deposited accidently on the ceramic pots and rocks surrounding hot charcoal fires used to reduce metal ores, a process which can be traced back more than 7000 years.¹⁵ Based upon both archeological evidence and metallurgical analyses, copper smelting (extraction from ore) and metal working originated independently in the Balkans (Serbia and Bulgaria) by ∼5500 BC and in Anatolia by at least 5000 BC.^16–19 The Roman philosopher Pliny the Elder (23–79 AD) discussed this process in his 37-book Naturalis Historia, the first encyclopedia, published in ∼79 AD.²⁰ 

The earliest reported purposeful growth of metal films from the vapor phase was in 1649 when Johann Schroeder, a German pharmacist, described a method for reducing arsenic oxide [As₂O₃] with charcoal^21,22 through the overall endothermic reaction

This is an example of a film-growth methodology which today is termed chemical vapor deposition.^1,23 As discussed in a 1966 review article by Rolsten,²⁴ carbon reduction of oxides was an important method for obtaining relatively pure metals, in order to investigate their fundamental physical properties, during the 1700s and the early 1800s.

The physical-vapor deposition (PVD) techniques of sputtering and evaporation were developed in the middle to late 1800s following the evolution of vacuum and electrical-power technologies beginning in the mid-1600s.¹ The first publication focused on sputter deposition of thin films was in 1852,²⁵ as discussed in Sec. IV A, while the first documented evaporation experiments were carried out in the early 1870s by Josef Stefan (1835–1893),^26–28 an Austrian physicist best known for his research in thermal conductivity of gases and blackbody radiation of solids,²⁹ and the early 1880s by Heinrich Hertz (1857–1897),^30,31 a German physicist well known for his work on electromagnetics³² and contact mechanics.³³ Reliable early measurements of the vapor pressures of solids (sublimation) and liquids (evaporation) were made by Martin Knudsen (1871–1949, Danish physicist) using what is now termed a Knudsen cell,^34,35 an isothermal enclosure with a very small orifice.

Sputter deposition, in its simplest configuration (Fig. 5), is carried out in an evacuated chamber which is backfilled with a low pressure of a rare gas such as argon. Argon is often the rare gas of choice for sputtering, primarily due to two reasons: (1) argon comprises approximately 1% of the earth's atmosphere and, hence, is relatively inexpensive and (2) the mass of argon (39.95 amu) is a reasonable match, resulting in significant collisional momentum transfer, to a wide range of metals in the middle of the periodic table (see discussion in Sec. IV H). A dc voltage is then applied between a metal target (the source of the film atoms) such as Cu and an electrically conducting substrate upon which the film is deposited. The voltage breaks down the gas to form a glow discharge consisting of Ar⁺ (with a much smaller fraction of Ar²⁺) ions and electrons. The positively charged ions are accelerated to bombard the target and, via momentum transfer, sputter-eject target atoms, some of which are deposited on the substrate. A reactive gas can also be added in order to form compound films (see Sec. IV J). Important commercial examples are the reactive sputtering of titanium and titanium-aluminum alloys in Ar/N₂ mixtures to deposit hard ceramic TiN and TiAlN coatings^38–40 on cutting tools, drill bits [see Fig. 1 (right panel)], gear hobs, and dies.

Kenneth Kingdon and Irving Langmuir (American surface chemist, 1932 Nobel Laureate in Chemistry for his research in surface science), from the General Electric Research Laboratory, dropped the “l” in favor of the word sputtering in their 1923 Physical Review paper on “The removal of thorium from the surface of a thoriated-tungsten (light bulb) filament by positive ion bombardment.”⁴⁷ Nevertheless, in the same year, in an article on the sputtering of tungsten published in the Philosophical Magazine by the “Research Staff of the General Electric Company and communicated by the Laboratory Director,” the term “cathode disintegration” was used in place of sputtering.⁴⁸ With time, however, the term sputtering prevailed and is now universal.

The rapid development of sputter deposition in the middle to late 1800s required the evolution of vacuum technology, beginning in the 1600s, and the invention of dc power supplies (batteries) in the late 1700s and the early 1800s. For a more detailed discussion of early vacuum technology and power supplies, see Ref. 1.

Progress in vacuum technology (the word vacuum is derived from the Latin vacuus, meaning empty space) was essential for providing cleaner deposition environments necessary for the advancement of thin-film science. In 1652, Otto von Guericke (1602–1686) of Magdeburg, Germany, a scientist, inventor, and politician, developed a mechanical piston pump that achieved a vacuum of 2 Torr (∼0.003 atm).^49,50 (For comparison, a typical household vacuum cleaner produces enough suction to reduce standard atmospheric pressure, 760 Torr, to ∼610 Torr).⁵¹ von Guericke's third-generation vacuum system, a model of which is shown in Fig. 6,⁵² consisted of a bell jar separated from the piston pump by a cylinder with a stopcock. The pump was equipped with two valves near the entrance to the nozzle at the bottom of the bell jar; the first valve was located between the nozzle and the cylinder and the second valve between the cylinder and the atmosphere. During the piston down-stroke, valve one is closed to stop air from entering the nozzle and the bell jar, while valve two is forced to open by the air displaced from the cylinder. During the piston return stroke, valve two is closed and valve one is forced to open by the pressure of the remaining air in the bell jar and the nozzle. The percentage of pressure decrease per complete piston stroke diminishes continuously as the bell-jar pressure is reduced toward the base pressure.

Much better vacuum was required in order for scientists in the 1800s to obtain longer gas-phase mean-free paths, higher deposition rates, and increased purity in films grown from the vapor-phase. This was solved by a German chemist, Herman Sprengel (1834–1906), who developed a practical mercury momentum-transfer pump in 1865.⁵³ The pump is related to a trombe in which water falls from an upper reservoir, while trapping air, into a pipe and deposits the air in a lower reservoir at higher pressure (a type of air compressor),^54,55 which had been known for “some hundreds of years.”⁵⁶ The base pressure claimed by Sprengel in his initial publication was ∼6 × 10⁻⁴Torr and limited by leaks in vulcanized rubber joints connecting the glass tubes (the rubber connectors were cemented to the glass tubes, and the joints were bound with copper wire). While lower pressures were achieved with later versions of the pump,⁵⁷ pressures of 10⁻³–10⁻⁴Torr were sufficient to provide ballistic environments (i.e., gas-atom mean-free paths of the order of, or larger than, system dimensions) for investigating gas discharges and sputter deposition in the small evacuated chambers of that era.

An initial prototype of the Sprengel mercury pump is shown in Fig. 7(a).⁵³ Droplets of mercury (a heavy metal which is liquid at room temperature), falling through a small-diameter (2.50–2.75 mm) glass tube, trap and compress air by momentum transfer. The tube, labeled CD in Fig. 7(a), was ∼76 cm long and extended from funnel A to enter glass bulb B through a vulcanized-rubber stopper. The bulb has a spout several mm above the lower end of tube CD.

In operation, mercury was added to funnel A and the stopcock at C opened, allowing mercury droplets to fall, trap air, and reduce the pressure in chamber R. Air and mercury were exhausted through the spout of bulb B. The mercury collected in basin H was poured back into funnel A for continuous pumping. The second version of the Sprengel pump, described in the same paper,⁵³ is shown in Fig. 7(b). It was approximately 1.8 m tall, and Sprengel reported using 4.5–6.8 kg of mercury during operation. The pump contained a mercury pressure gauge (similar to the one described below) attached to the evacuated chamber and a mechanical-piston backing pump S. Later versions incorporated continuous mercury recycling. With the combination of the mechanical and mercury pumps, a 0.5 l chamber could be evacuated in ∼20 min. The importance of Sprengel's work was recognized by the Royal Society of London which elected him as a Fellow in 1878. A later version of the pump, presently housed in the Dr. Guislain Museum, Ghent, Belgium, is shown in Fig. 7(c).

Improvements in vacuum technology required better gauging in order to measure the increasingly lower pressures produced. In 1874, Herbert McLeod (1841–1923), a British chemist, developed what today is termed the McLeod mercury gauge,^58,59 which operates based upon Boyle's law. Boyle (1627–1691), another British chemist, showed in 1662 that for a closed system at constant temperature, the product of pressure P and volume V remains constant.⁶⁰ In operation, the gauge compresses a known volume V₁ of gas at the unknown system pressure P₁ to a much smaller known volume V₂ in a mercury manometer with which the pressure P₂ is measured.^61,62 Thus, by Boyle's law, the system pressure P₁ is given by the expression P₂V₂/V₁. Liquid mercury wets glass and thus forms the required glass/metal seals in the gauge.

In addition to vacuum, electrical power was necessary for initiating early experiments in thin-film sputter deposition. von Guericke also played an important role in this field through his development in 1663 of a crude friction-based electrostatic generator which transformed mechanical work into electrical energy.^78,79 The generator was based on the triboelectric effect (although the term did not exist at the time), in which a material becomes electrically charged (“static electricity”) through friction.

In 1745, the Dutch scientist Pieter van Musschenbroek (1692–1761) of Leiden University [mathematics, philosophy, medicine, and astrology (the latter is closer to theology than science!)] and Ewald Georg von Kleist (1700–1748), a German lawyer, cleric (Bishop of Pomerania, Prussia), and physicist, are credited with independently inventing what today is known as the Leiden jar (Leyden jar),^84,85 an early form of the modern capacitor, to provide pulsed power. Both Kleist and van Musschenbroek studied at Leiden University, and it is likely that Kleist developed his interest in electricity from lecture demonstrations in the Physics Department. However, it appears that van Musschenbroek obtained the idea for his research from Andreas Cunaeus (1712–1788), a lawyer who often visited van Musschenbroek's laboratory and had learned Kleist's experiments.⁸⁶ Cunaeus carried out the initial experiments that led to the Leyden jar while attempting to reproduce even earlier results by Andreas Gordon (1712–1751), a Professor at Erfurt, Germany, and Georg Mattias Bose (1710–1761) at the University of Wittenberg, Germany.⁸⁷ The device accumulates static electricity between electrodes on the inside and outside of a glass jar. The first mention in the literature of the Leyden jar experiments was by Trembley in February, 1745 (there is some controversy regarding the exact publication date).⁸⁸ 

In order to store charge in Leyden jars, the glass cylinder of an electrostatic generator was rotated, via a hand crank, against a leather (or wool) strip pressing on the glass. The friction resulted in positive charge accumulating on the leather and negative charge (electrons) on the glass. The electrons were collected by an insulated (perhaps comb-shaped) metal electrode. When sufficient charge accumulated, a spark jumped from the generator collector to the central collector electrode of a nearby Leyden jar, where the charge was stored. Originally, the capacitance of the device was measured in units of the number of “jars” of a given size or by the total area covered with metal. A typical 0.5-l Leyden jar had a maximum capacitance of approximately 1 nF.⁸⁹ 

The earliest devices were hand-held glass jars partly filled with water (the inner conductor) in contact with a nail inserted through a cork stopper in the top of the jar. Both Kleist and van Musschenbroek reported receiving significant shocks when they touched both the nail and the outside of the jar! Eventually, van Musschenbroek realized that adding a conductor (other than his own body) such as a metal foil to the outside of the jar was far more practical.

Benjamin Franklin (American scientist and inventor, 1706–1790) was the first to understand that charge is stored in the glass dielectric, not the electrodes. He realized that the water merely served as an electrode and was not essential. To prove it, he produced flat capacitors consisting of a sheet of glass between metal-foil electrodes.⁹⁰ Franklin discussed his findings in a letter dated 1749.⁹¹ The Leyden jar was also used by Franklin in his famous kite experiment in a thunderstorm to “capture lightning in a bottle.”⁹² (Note, however, that it is disputed whether the experiment was actually performed.)⁹³ 

Daniel Gralath (1708–1767), physicist (founder of the Danzig Research Society) and Mayor of Danzig, Poland, repeated the Leyden jar experiments and was the first to combine several jars, connected in parallel (see Ref. 1), to increase the total stored charge.⁹⁴ The term “battery” was reputedly coined by Benjamin Franklin,^95,96 who likened the group of jars to a battery of cannon.

Gentlemen's Magazine reported the following in 1745.⁹⁸ “Could one believe that a lady's finger, that her whalebone petticoat, should send forth flashes of true lightening, and that such charming lips could set on fire a house? The ladies were sensible of this new privilege of kindling fires without any poetical figure, or hyperbole, and resorted from all parts to the public lectures of natural philosophy, which by that means became brilliant assemblies.”

The invention of the electrochemical battery to provide low-voltage dc power is generally attributed to Count Alessandro Volta (1745–1827),¹ Professor of Natural Philosophy at the University of Pavia, Italy, based upon his work in the 1790s resulting in a classic paper published first in French⁹⁹ and then in English¹⁰⁰ in 1800.

Volta's interest in electrochemistry led him to discover that voltage can be obtained from stacks consisting of several pairs of different metal disks, each pair separated by an electrolyte (initially pieces of cloth saturated in brine), connected in series to form a “voltaic pile.”¹⁰¹ The first metals used were copper and zinc, but Volta found, based upon electrometer measurements, that silver and zinc produce a larger electromotive force, a term Volta introduced in 1796.¹⁰² Figures 8(a) and 8(b) show an illustration and a photograph, respectively, of an early voltaic pile. Such devices could only provide a few volts; obtaining larger potentials required a series (i.e., a battery) of many large voltaic piles. An example of a small double voltaic pile is shown in Fig. 8(c).¹⁰⁰ 

A practical problem with voltaic piles, especially with larger ones used to obtain higher voltages, is that the weight of the disks squeezes the electrolyte out of the cloths. In 1801, William Cruickshank (1745–1810), a surgeon and Professor of Chemistry at the Royal Military Academy, Woolwich (southeast London), solved this problem and designed the first electric battery for mass production.¹⁰⁴ In the initial version, Cruickshank arranged 60 pairs of equal-sized zinc and silver sheets cemented together with rosin and beeswax in a long resin-insulated rectangular wooden box such that all zinc sheets faced one direction and all silver sheets the other. Grooves in the box held the metal plates in position, and the sealed box was filled with an electrolyte of brine or dilute ammonium chloride (NH₄Cl), which has higher conductivity.

From the mid- to late-1800s, several papers were published on the use of optical emission to investigate the transition between a continuous glow discharge and an arc, the electrical structure and configuration of dc glow discharges, and the nature of metal atoms sputter ejected from the cathode (target). Michael Faraday (1791–1867), an English scientist, reported on the electrical and optical characterization of glow discharges in 1838.¹⁰⁵ His glass discharge tube contained brass electrodes and was operated in air, nitrogen, oxygen, hydrogen, and other gases at a pressure of 4.4 inches of mercury (∼112 Torr). While Faraday must certainly have deposited films during these experiments, this was not the objective. His later work on the optical properties of vacuum-arc-deposited thin metal films was discussed in his fifth Bakerian Lecture¹⁰⁶ in 1857 and published in the same year.¹⁰⁷ 

Heinrich Geissler, in 1857, used his mercury pump (see the Historical Footnote in Sec. III A) to evacuate small glass enclosures and develop the “Geissler tube” to study the optical and electrical properties of glow discharges in rare gasses, air, mercury, etc.⁷⁵ He reported observing a wide variety of discharge colors due to optical emission resulting from the decay of excited gas atoms. This gave rise to the production, beginning in the 1880s, of the first gas-discharge lamps which were sold primarily as novelty and artistic items.¹¹¹ Julius Plücker,¹¹² also at the University of Bonn, and his ex-graduate student Johan Wilhelm Hittorf¹¹³ used Geissler tubes to study gaseous electronic effects resulting from ion bombardment of metal targets (see Sec. IV D).

In 1852, William Robert Grove (1811–1896), a Welsh lawyer (later, judge) and physicist, published the earliest recorded description of sputter deposition and ion etching experiments.²⁵ A sketch of his equipment is shown in Fig. 9. Vacuum was achieved with a mechanical piston pump, similar to that developed by von Guericke as described in Sec. III A, with power supplied by Grove's version of a trough-style dc voltaic pile (following Cruickshank, Sec. III B)¹⁰⁴ combined with an “induction coil” step-up transformer supplied by Heinrich Ruhmkorff (1803–1877),¹¹⁴ a famous German instrument maker living at the time in Paris. The electrodes consisted of a copper plate, with a polished electroplated silver surface, and a rod, which passed through a leather stopper in the top of the glass vacuum chamber, with a steel needle attached to its end. The gas used to sustain the discharge was stored in a bladder.

The experiments were carried out at rather high pressures, ranging from ∼100 to 500 mTorr, with the steel needle target quite close to the silver-plated substrate (generally a separation of 0.25 cm, “but this may be considerably varied”). When using a mixture of hydrogen and air with the silver plate positive and the steel needle serving as the (negative) cathode, Grove observed thin-film deposition on the silver substrate. The layer was primarily iron oxide, i.e., reactive sputtering (although the term did not come into existence until more than a century later, Sec. IV J). The color of the oxide film “presented in succession yellow, orange, and blue tints” with increasing thickness (longer deposition time). Grove reported optical interference effects (as he noted later in the paper), which for a given substrate/film combination can be calibrated to provide film thickness versus color as is commonly done today for SiO₂ and Si₃N₄ dielectric layers on Si(001) wafers used in microelectronic device fabrication.

When Grove switched polarity such that the silver plate became the cathode (negative), he reported that the iron oxide film was removed by ion etching. Continuing the experiment, a polished region “occasioned by molecular disintegration” remained. Thus, Grove had not only removed the original oxide film, but also sputter-etched into the silver substrate layer. The word “sputter” did not yet exist (as discussed in Sec. II), and Grove described the process throughout the paper as “molecular disintegration.”

Grove repeated the above experiment by sputtering the steel target in an “air vacuum” (a term Grove attributed to Faraday) to produce a more fully oxidized film on the silver-plated copper substrate and then switched gases and electrode polarity to “sputter clean” the silver plate in a nitrogen discharge. In actuality, there must have been a thin silver-oxide layer remaining due to the competition between the rates of silver oxidation from the discharge, arising from the relatively poor vacuum, and sputter etching. However, this layer, a few tens of Å in thickness,¹¹⁵ would have been too thin for Grove to observe. Several more experiments in which he substituted different metals for the target needle and changed discharge gases were also reported. The results were similar, but he described observing differences depending upon the atomic masses of the target material and the gas, the ionization potential of the gas, and the oxidation tendency of the metals.

Interestingly, Grove realized that oxygen can form negative ions which are accelerated by the applied target voltage toward the substrate (anode). The significance of the fact that oxygen has a high electron-attachment probability¹¹⁶ was not fully appreciated until almost 130 years later. Researchers investigating the growth of piezoelectric and transparent conducting oxides (TCOs) in the early 1980s^117–119 and high-temperature oxide superconductors in the late 1980s,¹²⁰ all of which are typically deposited today by reactive magnetron sputtering (see Secs. IV F and IV J), were confronted with the deleterious effects of O⁻ and $O 2 –$ irradiation.

Negative ions, accelerated by the same potential used to produce sputtering by positive ions incident at the target, bombard the growing film with energies which can produce residual defects, change the preferred orientation of polycrystalline layers, degrade film properties, and decrease deposition rates by resputtering.^{115,117–121} Irradiation of the growing film by fast neutral O and O₂ species can also occur, as attached electrons are stripped from the corresponding ions in the plasma.^118,120,122 Early solutions involved increasing the discharge pressure in order to decrease particle mean-free paths and hence lower the average energies of ions and fast atoms via collisions,¹²¹ while later solutions focused on off-axis deposition and facing-target sputtering (Sec. IV F 1).^{119,123–127}

A few years after Grove's seminal paper, John Gassiot (1797–1877), a highly successful English businessman and gentleman scientist, presented his Bakerian Lecture (March 4, 1858) “On the stratifications and dark bands in electrical discharges as observed in Torricellian vacuums,” which was published in the Proceedings of the Royal Society of London¹²⁸ (also see Ref. 129). Most of the lecture was concerned with his observations of alternating bright and dark bands formed in rarefied-air discharges contained in glass tubes partially filled with clean boiled mercury and evacuated with a mechanical pump [as air is extracted, the mercury level sinks in the tube and a “Torricelli vacuum” is formed (see the Historical footnote, Sec. III A)].^1,72,73 During these experiments, Gassiot noted that the luminous discharge regions move under the influence of a magnetic field. However, he conceded in the conclusion of Ref. 128 that “I refrain for the present from any observations as to the action of the magnet on the discharge.”

Gassiot also reported that for discharges formed between two platinum wire electrodes hermetically sealed about 4 inches (∼10 cm) apart in a discharge tube: “a black deposit takes place on the sides of the tube nearest the negative terminal. This deposit is platinum (analyzed by Michael Faraday)¹²⁹ in a state of minute division emanating from the wire, which becomes black and rough as if corroded. The minute particles of platinum are deposited in a lateral direction from the negative wire, and consequently in a different manner from what is described as occurring in the voltaic arc.”¹²⁸ “The platinum coating is deposited on the portion of the tube surrounding the negative wire, but none at or near the positive.”¹²⁹ Gassiot described sputter deposition from the platinum target. He noted that “…when this deposit is examined by transmitted light, it is translucent, presenting to the eye an extremely thin bluish-black film; but by reflected light, either on the outside or inside (i.e., viewed either from the glass or the film side), it has the appearance of highly polished silver, reflecting the light as from the finest mirror.”¹²⁹ 

Practical applications of sputter-deposited single and multilayer metal films as mirrors and optical coatings on telescope lenses and eyepieces were discussed in papers published in 1877 by Arthur Wright (American physicist, 1836–1915, Yale University).^131,132 The first of the two articles reported the growth of adherent noble-metal films sputter-deposited from wire targets onto glass microscope slides.¹³¹ Unfortunately, the films had large lateral thickness variations since the target was the tip of a wire whose length was encased in a glass tube. However, an ingenious solution was presented in Wright's next paper.¹³² He designed a deposition system, evacuated with a Sprengel mercury pump (see Sec. III A), in which the substrate was mounted on a pendulum to provide motion in two orthogonal directions with respect to the target such that films of uniform thickness could be “painted” onto the substrate (Fig. 10). In Wright's words: “The perfect control of the process obtained by the use of the movable electrode will even make it possible to apply the method of local correction for the improvement of a defective figure, or to parabolize a spherical mirror by depositing the metal in a layer increasing in thickness toward the center.”

Wright characterized the sputtering process spectroscopically using optical emission from gas, and ejected target atoms, which were excited in the discharge (following earlier work by Faraday).¹⁰⁵ As-deposited platinum films, some with thicknesses <350 Å (estimated using a combination of weight change, to within 10 μg, for thicker films, deposition rate calibrations, and optical interference rings for thinner layers), were analyzed using optical transmission as a function of wavelength. Mirror-like sputter-deposited films were found to be more adherent than solution-grown layers and less sensitive to local delamination caused by water penetration to the film/glass interface. By the late 1800s, sputter-deposition was routinely used in manufacturing commercial mirrors.

Wright described his films as “…surfaces of exquisite perfection and the most brilliant polish. They can only be compared to the surface of clean liquid mercury, far surpassing in luster anything that can be obtained by the ordinary methods of polishing.” Wright tuned the reflectivity of his mirrors based upon interference effects to obtain brilliant “white light” by depositing multilayer films with predetermined layer thicknesses.

In the late 1860s, William Crookes (1832–1919), a British chemist, developed what are now referred to as Crookes discharge tubes [Fig. 11(a)]¹³⁶ based on the earlier Geissler tubes.¹¹¹ Crookes also took advantage of the Sprengel mercury pump to obtain better vacuum, and hence longer mean-free paths, in his tubes which he used to promote research on gas-discharge electronics. Crookes tubes were instrumental in the discoveries of x-rays¹³⁷ (Wilhelm Röntgen, 1896, Nobel Prize in Physics, 1901), electrons¹³⁸ (J. J. Thomson, 1897, Nobel Prize in Physics, 1906), and thermionic emission^139,140 (Owen Richardson, 1901, Nobel Prize in Physics, 1928), which enabled vacuum-tube electronics.^141,142

Crookes published a very significant paper on sputter deposition of thin films in 1891.¹⁴³ During a long series of experiments, he employed a Sprengel-type mercury pump⁵³ to evacuate his discharge tube to pressures of the order of 7 × 10⁻⁴Torr. He then used the residual air to sputter Ag, Al, Au, Cd, Cu, Fe, Ir, Mg, Ni, Pb, Pd, Pt, and Sn targets as well as AlAu and CuZn (brass) alloys and measure the erosion rates via target weight loss.

Although the experimental details (sputtering pressure, voltage, and ion current densities) were not well specified, Crookes, in order to obtain comparable results, designed a multitarget sputtering system with indexed motorized external electrical contacts as illustrated in Fig. 12.¹⁴³ Every experiment was carried out using four wire targets, 0.8 mm in diameter and 20 mm in length, in which one of the four was always a gold reference electrode. Power was alternately applied to each target in succession, using a revolving commutator, for the same length of time (typically 6 s) over periods of several hours. By this means, variations in current and sputtering pressure were accounted for in order to obtain a set of metal sputtering rates, all referenced to that of gold. Aluminum and magnesium targets were reported to be “practically nonvolatile.” Today, we know that this was due to the formation of strongly-bonded oxynitride (primarily oxide) dielectric layers at the target surfaces due to the use of air (approximately 78% nitrogen and 21% oxygen by volume) as the sputtering gas.

The “rich purple color” of the aluminum-gold alloy target turned to the “dull white color of aluminum” as gold was preferentially sputter removed. This was the first mention in the literature of preferential sputtering from alloy targets. While this occurs for all alloys, the target surface composition during sputtering in rare gases rapidly (depending on the ion energy, current density, and relative ion and target-atom masses) reaches a steady-state value as the surface coverage of the low-sputtering-rate component increases to compensate for the difference in elemental removal rates as shown experimentally by Tarng and Wehner¹⁴⁴ and theoretically by Eltoukhy and Greene.¹⁴⁵ In Crookes' aluminum-gold alloy experiments, oxidation further decreased the aluminum sputtering rate.

Since the targets in Crookes' experiments were uncooled, low-melting-point metals such as tin, cadmium, and lead quickly melted. For these materials, he devised a holder to sputter liquid metals. It is likely that a significant part of their measured weight loss, especially for the high-vapor-pressure element cadmium, was due to evaporation, in addition to sputtering.

In his April 4, 1879 Bakerian Lecture,¹⁴⁶ Crookes, like Faraday¹⁰⁵ and Gassiot^128,129 before him, discussed the use of optical spectroscopy to characterize plasmas, in many discharge-tube configurations, at working pressures from “0.08 mm Hg (80 mTorr) down to a low of 0.00001 mm Hg (0.01 mTorr).”¹⁴⁷ He often specified the sputtering pressure not in pressure units but as the thickness of the cathode dark space, which came to be known as the Crookes dark space, adjacent to the very bright glowing region in front of the target. He noted that during long sputtering runs, it was necessary to periodically bleed some air into the discharge tube “to reduce the vacuum.” After examining metal films deposited on the inside of the discharge tube and finding them to be porous with rough surfaces, he concluded that sputtering gas (residual air in these experiments) was “occluded” in the growing films. While some gas was likely trapped in the growing films, most was captured (adsorbed) via reactions with the fresh metal layers, deposited on surfaces throughout the system, which acted as a getter pump.

Francis Aston (1877–1945, English chemist and physicist, Nobel Prize in Physics, 1922, for discovery of isotopes in nonradioactive elements) also made important contributions to the modern understanding of glow discharges. After three years working as a chemist in a brewery, Aston joined Birmingham University with a scholarship in 1803 to study glow discharges. In 1907, he discovered a very narrow region in gas discharges, now called the Aston dark space,¹⁴⁸ immediately adjacent to the target and preceding the bright cathode-glow region. The area is “dark” (less luminous) since the average energy of electrons emitted from the target due to ion bombardment (see discussion in Sec. IV D) is less than that required for the excitation of gas atoms. However, the electrons are rapidly accelerated by the applied electric field to produce the cathode glow.

Aston carried out his own glass blowing to fabricate gas-discharge tubes with movable aluminum electrodes in order to measure the distance from the target to the end of the “Crookes dark space” and, hence, provide an estimate of the width of the “cathode fall” over which ions are accelerated to the target. He showed, for a wide range of gases, that the cathode fall distance d is given by

in which P is the gas pressure, J_T is the target current density, and A and B are functions of the gas and target material.^149–151 This empirical relationship is now referred to as the Aston equation. Aston was appointed lecturer at the University of Birmingham in 1909 but moved to the Cavendish Laboratory in Cambridge in 1910, on the invitation of Thomson, to continue working on gaseous electronics.

In 1902, Thomas Edison (1847–1931), an American inventor and businessman, patented a very early forerunner to copper contact technology in modern microelectronic device fabrication.^152,153 Edison's U.S. patent 713,863 (“Process of coating phonograph records”) describes the use of dc sputtering for the deposition of metal films on wax phonograph masters as “seed” (and adhesion) layers for electroplated overlayers.¹⁵⁴ This follows an earlier 1892 Edison patent in which the seed layers were deposited by vacuum-arc deposition.¹⁵⁵ Edison claimed in the 1902 patent that the arc process was too slow and that sputter-deposited films had much more uniform thickness distributions.

Grove's and Crookes' research on sputtering attracted the attention of scientists worldwide. A review paper, entitled “Cathode Sputtering, a Commercial Application,” published in 1932 by Fruth,¹⁶¹ of Western Electric Company (Chicago), lists 113 references published in the field between the time of Grove's 1852 pioneering article²⁵ and 1930. Fruth described commercial equipment (Fig. 13, left panel) and procedures for sputter-depositing gold electrodes, from six gold cathodes, onto multiple radio-broadcasting microphone diaphragms. A photograph of the deposition chamber, which contains a rotating McLeod gauge⁵⁸ and a “bleeder” valve in order to maintain constant pressure, with diaphragms ready to be coated, is shown in the right panel of Fig. 13. Fruth described the system operation as follows.

“In order to maintain a constant residual gas pressure, the pump is operated continuously and air is allowed to leak in slowly through the bleeder valve which is located near the pump. This practice was found necessary in order to overcome variations in pressure due to the early evolution of gases and the later cleanup usually accompanying electrical discharges in vacuo. A pressure of 0.100 mm (100 mTorr) is readily maintained by this method. After a new charge has been placed in the bell jar, the bleeder valve is temporarily cut off by closing a stopcock so that the required vacuum can be more quickly obtained. By this means, sputtering can be started in about 4 min after the bell jar has been placed in position.”

Fruth demonstrated that dc sputter-deposited gold films, <1 μm thick, offer substantial lifetime advantages over previous electroplated films which developed “blisters,” peeling, and pinholes after three months of continuous use, while the sputter-deposited films exhibited no sign of wear or degradation.

An early ion-beam source was developed by Louis Maxwell in 1931.¹⁶² A hydrogen discharge at pressures of 3.5–120 mTorr, with a liquid-air cold trap to remove water vapor and minimize mercury contamination due to back-streaming from the pump, was established in a small brass vacuum vessel. The ion current was controlled by thermionic electron emission from a hot, low-work-function filament. Large magnetic fields, ∼12–17 × 10³ G, parallel to the positive ion beam were used to minimize ion losses to the wall and provide 0.1–3 mA through a 1-mm-diameter circular extraction electrode to a collector electrode in a small attached chamber maintained at 3.3 × 10⁻³ to 1.5 × 10⁻⁴Torr.

The first recorded description of a dc glow discharge ion-beam sputtering system was given by Seeliger and Sommermeyer in 1935.¹⁶³ They drilled a 2-mm-diameter hole in the cathode of their discharge tube to “collimate” an Ar⁺ ion beam (the beam was actually divergent) to strike solid silver or liquid gallium targets at energies of 5–10 keV and observed that sputtered-atom emission can be approximated by a cosine distribution.

In 1960, Wehner and Rosenberg,¹⁶⁴ using a mercury-pool-supported glow discharge (see Sec. IV E) to sputter polycrystalline metal targets with normally incident 100 to 1000 eV Hg⁺ ions, showed that sputtered-atom angular ejection distributions ranged from under-cosine at lower energies toward cosine at higher energies and noted that the crystalline orientation (texture) of the target was important. Much later (1986), Matsuda et al.¹⁶⁵ reported, based on normally incident Ar⁺ ion-beam experiments, that the angular distribution of sputtered iron atoms varied from cosine at 600 eV to slightly over-cosine at 1000 eV and increasingly over-cosine at 2 and 3 keV. With a simple cosine emission distribution, often used as a first approximation in sputter deposition (see Sec. IV F 2), the sputtered flux ejected from a given point on the target surface along any angle φ is just the flux at normal incidence multiplied by the cosine of φ.¹⁶⁶ 

Crookes, in his classic 1891 paper on metal sputtering rates,¹⁴³ compared the sputtering process with evaporation and described differences in the two processes as arising from coupling electrical versus thermal energy to the source material. That is, sputtering in his terminology was “electrical evaporation.” Charles Townes (1915–2015, Physics Nobel Laureate in 1964, sharing the prize for the invention of the maser) published a theory in 1944 based on ion-irradiation-induced local evaporation in which the sputtering yield (the average number of atoms ejected per incident ion, a measure of the process efficiency) depended on the ion-energy but not on the ratio of the ion-to-target atom masses.¹⁶⁷ Even though contradictory evidence had been accumulating since the late 1890s, the concept that sputter ejection of target atoms occurs by local “hot spot” evaporation persisted well into the 1900s. In fact, a review article published as late as 1962 (Ref. 168) followed Adolph Güntherschulze (1878–1967, German physicist, Technical University of Dresden) in attempting to popularize the term “impact evaporation.”¹⁶⁹ 

A half century after Grove's initial sputter-deposition and ion-etching experiments,²⁵ Eugen Goldstein (German physicist,1850–1930) in 1902 provided additional (although, by this time, unnecessary) evidence that sputtering is initiated by positive-ion bombardment. In his experiments, he used a perforated negatively-biased target and demonstrated sputter etching of a gold film on the wall of the discharge tube behind the target.¹⁷⁰ Goldstein called the positively charged beams “canal rays.” In 1908, Johannes Stark (1874–1951, Physics Nobel Laureate in 1919 for the discovery of electric-field-induced splitting of atomic spectral lines) argued strongly in favor of sputter-ejection occurring by ion-impact-initiated collision cascades.^171,172 There was abundant evidence to support these claims, including the relative insensitivity of sputtering rates to the thermal conductivity of the target and to target temperature.^173,174 Stark and Wendt,¹⁷⁵ in 1912, reported experimental evidence for sputter-rate variations with grain orientation in polycrystalline targets.

In 1931, Mayer¹⁷⁶ and later Sporn¹⁷⁷ (1939) provided early indications that the ejection energies of sputtered atoms are of the order of several electron-volts, much higher than typical evaporated-atom energies of a few tenths of an electron-volt. The authors investigated optical emission due to the decay of atoms sputter-ejected from thin, low-work-function, alkaline-earth-oxide and alkaline-metal layers on metal targets and excited in the discharge. Ejection velocities were estimated from measurements of the thickness of the luminous discharge region together with known excited-state lifetimes. Güntherschulze¹⁷⁸ contributed supporting data in 1942 showing that sputtered atoms have much longer mean-free paths (and hence higher kinetic energies) than would be expected for thermal atoms.

Fetz added further evidence in favor of sputtering via collision cascades in 1942 when he showed that sputtering yields (determined by dividing the measured target weight loss by the ion current to ground) increase with increasingly oblique angles of ion incidence due to more effective momentum transfer in the near-surface region of the target.¹⁷⁹ In 1954, Fred Wehner, often referred to as the father of modern sputtering, found that atoms tend to be sputter ejected in a specular direction when subjected to oblique low-energy (<1 keV) ion bombardment.¹⁸⁰ 

None of the above results are compatible with target “disintegration” by an evaporation process. Wehner¹⁸¹ added yet another point in favor of momentum-transfer sputtering in 1955, while confirming the previous observations of Stark and Wendt.¹⁷⁵ He showed that during low-energy (150 eV Hg⁺) sputtering from single-crystal targets under conditions (see Sec. IV D) for which ion energies are monoenergetic and incident orthogonal to the target surface, sputtered atoms are ejected along close-packed crystalline directions.¹⁸¹ Examples are shown in Fig. 14 for sputtering of single-crystal (111)-, (110)-, and (001)-oriented face-centered-cubic (fcc) silver targets; the “spots” correspond to preferential sputtering along close-packed directions in all three cases. These results captured the imagination of scientists working in the sputtering field; several papers confirmed Wehner's “spot pattern” observations for fcc metals^182,183 and showed similar effects for body-centered^184,185 and hexagonal¹⁸⁶ metals as well as diamond-^182,183 and zinc-blende-structure¹⁸⁷ semiconductors. The latter authors noted that preferred ejection directions are not observed for sublimation from single crystals.¹⁸⁷ 

The existence of spot patterns as a function of target orientation indicates that the original lattice structure is at least partially intact during the violent sputtering events, further evidence against fully developed cascade focusons for which the lattice would lose orientational information. Overall, it is clear that the probability of sputtering an atom from the surface layer (i.e., providing sufficient energy to overcome the surface binding energy) will be higher due to a central collision initiated by a close lower-lying neighbor atom. The width (broadening) of sputtered-atom spots is determined by a variety of factors including thermal vibrations and deviations from the ideal surface structure due to previous ion impacts.

Stuart and Wehner^191,192 also provided additional definitive evidence that the earlier indications by Mayer¹⁷⁶ and Sporn¹⁷⁷ were correct: atoms sputtered by low-energy ions are ejected with average energies that are of the order of ten times larger than that for evaporation at the same rate. Wehner and colleagues used time-of-flight optical spectroscopy in low-pressure Hg-pool-supported discharges (see Sec. IV E) for which the sputtered-atom mean-free path was larger than the distance from the target to the measurement sampling region. In these experiments, the target was sputtered during sequential 1–μs pulses and the sputtered atoms were excited by electrons in the discharge. As each group of sputtered atoms moves away from the target, it becomes spatially dispersed due to the distribution in ejection velocities. The dispersion is detected as a time distribution of characteristic photons emitted by the sputter-ejected atoms as they pass through the observation volume, defined by a narrow slit, at a known distance from the target. The measured time distribution is then converted into an ejection velocity and/or an energy distribution. Time-of-flight results were verified by determining the Doppler shift for light emitted by sputtered atoms in the direction of travel.¹⁹² 

Figure 15 shows a plot of the average energies E_at of atoms ejected orthogonal to 22 different polycrystalline metal targets due to bombardment by 1.2 keV Kr⁺ ions.¹⁹² The data conclusively establish that sputtering is not a local evaporation process. Average ejection energies for elements in the middle of the periodic table are ∼10 eV, with considerably higher E_at values for heavy elements such as Ta, Pt, W, Re, and U which have atomic masses between 180.95 and 238.03 amu. The heavy-target-atom effect is directly related to the relatively shallow penetration depths of the much lighter Kr⁺ (83.80 amu) ions. That is, kinetic energy is deposited closer to the surface of heavy-atom targets. However, there is clearly more to the story; as discussed in Sec. IV H, even though the mass of Au (196.97 amu) is higher than that of the other heavy atoms, except for U, shown in Fig. 15, Au exhibits a lower average ejection energy.

Kingdon and Langmuir⁴⁷ published, in 1923, argon, cesium, helium, hydrogen, neon, and mercury sputtering yields (average number of sputtered atoms per incident ion) for thoriated-tungsten wire targets (light-bulb filaments) as a function of target voltage V_T between −50 and ∼−400 V in a glass dc-diode system. Although the sputtering conditions were poorly defined, the authors discussed the process in terms of momentum transfer. In 1926, Eric Blechschmidt¹⁹³ carried out exhaustive studies of the sputtering yields of 19 metals by hydrogen, 13 metals by neon, and 15 metals by argon ion bombardment. Unfortunately, Blechschmidt's results suffered from severe target contamination due to poor vacuum.

In addition to contamination, another issue in early sputter-yield measurements is the unknown contribution to the total measured current of the ion (as well as the fast-neutral-atom and photon) bombardment induced secondary-electron flux (see the Historical footnote below) from the target. The secondary-electron yield from clean metals is typically ∼0.1 for Ar⁺ ion energies less than ∼1 keV.^194,195 Thus, the direct effect on reported low-energy metal sputtering yields is not large. However, the indirect effects turn out to be very important indeed. A secondary-electron yield of 0.1 requires that each ejected electron must, on average, produce 10 new ions to maintain the discharge. This, in turn, requires operating at relatively high pressures in dc-diode systems, often near, or above, 100 mTorr, depending on the target-to-substrate distance, in order to provide sufficient collisions since the extremely small size of electrons results in them having correspondingly small scattering cross-sections and, hence, long mean-free paths.

Sputtering at high pressures introduces, in turn, yet another uncertainty in the accurate determination of sputter yields. The mean-free path of sputter-ejected target atoms, which have much higher collision cross-sections than electrons, is very short at these pressures. In 1926, Artur von Hippel¹⁹⁶ estimated that in typical sputtering systems at the time, the short mean-free paths result in up to 90% of the sputtered atoms being reflected back to the target at an Ar pressure of 100 mTorr. This not only dramatically reduces film deposition rates, it means that sputter yields based upon target weight-loss measurements are greatly underestimated.

An additional issue in determining reliable sputter yields S as a function of ion energy E_i at such high pressures is the inherent assumption that E_i = eV_T. However, at high pressures, the ion mean-free path is much less than the distance over which the applied voltage V_T “falls” to the target (the cathode fall or “target sheath” distance). Thus, ions make many collisions while transiting the ion sheath, and hence, their average energy upon striking the target is much less than eV_T. This further reduces the sputtering yield (and the resultant film deposition rate). von Hippel¹⁹⁶ used a retarding field to measure the energy-dependent ion-current distribution arriving at a probe positioned a short distance behind a small hole in a metal cathode immersed in a 100 mTorr, 15 mA, V_T = −1000 V, argon discharge. He reported that the average ion energy E_i was ∼250 eV, with an energy spread ranging across the full cathode-fall potential. Most of the ions arrived at the cathode with kinetic energies near zero. That the average value of E_i is much less than eV_T was confirmed years later by Davis and Vanderslice¹⁹⁷ who showed that the primary energy-loss mechanism for ions in high-pressure dc glow discharges is due to quantum-mechanical charge-exchange processes which have higher cross-sections than hard-sphere collisions.¹⁹⁸ It should be noted that the decrease in the sputtering yield as a function of V_T, compared to expected values if E_i was equal to eV_T, is partially mitigated due to sputtering by fast neutral atoms (i.e., accelerated ions which undergo charge-exchange reactions near the target), a process not accounted for in the measured ion currents.

In 1858, during his investigation of electrical conduction through glow discharges,¹¹⁰ Julius Plücker (1801–1868), a German Professor of Mathematics (and later Physics) at the University of Bonn, observed a green fluorescent region in the glass discharge tube across the metal cathode [Fig. 11(b)].¹⁹⁹ He also found that the position of the luminescence could be deflected by a magnetic field [Fig. 11(c)]. Johan Hittorf (1824–1914), who studied under Plücker, reported similar results in 1869,¹¹³ leading him to infer that the glowing spot at the end of the glass discharge tube was due to what he termed “glow rays” which emanate from the cathode and proceeded in straight lines to cause the luminescence. In 1878, Goldstein demonstrated that the particles (he referred to them as “cathode rays”) did, in fact, originate at the cathode²⁰⁰ and had a charge opposite to that of the “canal rays” (positive-ions) he observed during sputtering of a perforated target (Sec. IV B).¹⁷⁰ Crookes¹⁴³ (Sec. IV A) used the term “radiant matter” and established that the rays were not due to fast sputtered atoms.²⁰¹ He proposed (incorrectly) that radiant matter was due to negatively-charged ions ejected with high velocity from the cathode.²⁰² 

The German-born British physicist Arthur Schuster (1851–1934), a friend and competitor of Thomson,^203,204 expanded upon Crookes' experiments by placing metal plates parallel to the cathode rays and applying a potential between the plates. He observed that the rays were deflected toward the positive plate, thus proving (again!) that they are negatively charged.^204,205 This was confirmed, using a different approach, by Jean Baptiste Perrin (1870–1942), a French physical chemist, and discussed at a lecture before the Paris Academy of Sciences on December 30, 1895 (English translation, Ref. 206). In his experiments, he allowed a beam of cathode rays to enter a metal cylinder through a small hole and showed that the cylinder became negatively charged to a value which was determined using an electrometer. No negative charge was measured if the beam was deflected away from the cylinder hole by a magnetic field. Perrin, more famous for his studies of Brownian motion, received the 1926 Nobel Laureate in Physics “for his work on the discontinuous structure of matter, and especially for his discovery of sedimentation equilibrium (the atomic nature of matter).”

Schuster, in his 1884 Bakerian Lecture,²⁰⁵ made an important contribution to cathode-ray research by noting that “a particle of charge e and mass m, moving at speed v at right angles to a magnetic field $B ¯$⁠, would be deflected such that it moves in a circle of radius r = mv/Be, so that (the cathode-ray charge-to-mass ratio) e/m = v/Br.” Unfortunately, the pressure in Schuster's cathode tube was too high, 300 mTorr, for accurate measurements of v.

In his second Bakerian Lecture,²⁰⁷ 1889, Schuster combined the above equation with conservation of energy ½mv²= eV_T, in which V_T is the cathode-fall voltage, to yield the expression e/m = 2V_TB²r². He used this equation to provide an upper limit to e/m since it assumes that no energy is lost by collisions between cathode-ray particles and gas atoms (i.e., the use of low discharge pressures, which was not the case). He also provided a lower limit by assuming that the lowest cathode-ray particle speed v would be the root-mean-square speed of gas atoms at room temperature and the discharge-tube pressure. Again, however, he had experimental issues: “I have met with very considerable difficulties in the attempt to carry out the measurements in a satisfactory manner, and have only hitherto succeeded in fixing somewhat wide limits (which differed by a factor of ∼1000) between which the molecular charges must lie.”²⁰⁷ Moreover, Schuster's results were found to vary strongly with the gas composition²⁰³ due to the dependence of the cathode-ray collision frequency, in high-pressure discharges, on the size of the gas atoms.

Walter Kaufman (1871–1947), a German physicist, used the approach suggested by Schuster in his second Bakerian lecture,²⁰⁷ to determine e/m in better vacuum.²⁰⁸ The result, published in November, 1897, was much closer to today's accepted value than that of Thomson.¹³⁸ In addition, Kaufmann's e/m value, in contrast to that of Schuster, was independent of the discharge gas. He later showed, in his best-known work, that the electron mass is velocity dependent (i.e., relativistic).²⁰⁹ In January of 1897, Emil Weichert (1861–1928) published results in which he obtained an e/m value for cathode-ray particles that was 2000–4000 times larger than that of hydrogen ions.²¹⁰ 

Thomson, generally acknowledged as the “discoverer” of the electron (which he called a “corpuscle”), first presented his cathode-ray results during a lecture at the Royal Institution in London on April 30, 1897.¹³⁸ Employing energy conservation to obtain e/m, he equated the kinetic energy of the cathode-ray particles to the heat generated by their collisions with the interior surface of a cylinder. Six months later, he reported more fully on the details of these experiments, together with the results of a second experimental approach based on the deflection of cathode-ray particles in electric and magnetic fields.²¹¹ When they were subjected to the electric field $E ¯$ alone, e/m = v²θ/EL, in which θ is the deflection angle and L the distance traveled by the particles at speed v in the field. The particles were next subjected to only a magnetic field $B ¯$⁠, which deflected them through an angle φ, yielding e/m = vφ/BL. The two equations were then solved for the two unknowns: v and the ratio e/m.

Thomson then adjusted $B ¯$ such that θ = φ, resulting in v = E/B and e/m = Eθ/B2L, from which he obtained, based upon seven independent measurements, an average e/m value of 0.77 × 10¹¹ C/kg, more than a factor of two lower than today's accepted value, 1.759 × 10¹¹ C/kg. Nevertheless, Thompson realized that his result was still nearly 1000× larger than e/m for a hydrogen ion. Further, he reasoned that the absolute value of the corpuscle charge was the same as that of a hydrogen ion (although of opposite sign). Hence, the corpuscle must have a mass approximately 1000 times smaller. Thus, in his Royal Institution Lecture, Thomson announced the discovery of particles that are small compared to atomic dimensions. “We have in the cathode rays matter in a new state… in which all matter—that is, matter derived from different sources such as hydrogen, oxygen, etc.—is of one and the same kind; this matter being the substance from which all the chemical elements are built up.”^138,211,212

While the term electron was well known at the time (see the Historical footnote below), Thomson nevertheless continued to use the term corpuscles “…until about 1910, during a period when electron theory flourished and was considered the most important topic of physical theory.”²¹² It is also interesting to note that: “In Thomson's Nobel-Prize citation there is no explicit mention of his discovery of the electron, but Thomson's (Nobel) lecture²¹³ was devoted almost exclusively to that topic.”²⁰³ The citation states: “in recognition of the great merits of his theoretical and experimental investigations on the conduction of electricity by gases.”

The next breakthrough in understanding ion-induced emission of “corpuscles” or cathode-rays (i.e., secondary-electrons) came much later. In 1954, Homer Hagstrum (1915–1994), an American surface physicist at Bell Laboratories, published a paper on the theory of ion-bombardment-induced secondary-electron emission, based primarily on Auger neutralization and, to a lesser extent, Auger de-excitation processes.²¹⁷ Hagstrum described Auger neutralization as “a process in which the interaction of two conduction electrons causes one electron to neutralize the (incident) ion and the other to be excited into the continuum above the filled band. The observed (secondary-electron yield) γ is determined by the probability that the excited electrons escape from the metal.” Hagstrum's theoretical predictions proved to be a good fit to his experimental results for 10–1000 eV rare-gas irradiation of polycrystalline tungsten in an ultrahigh vacuum system with a base pressure of ≤10⁻¹⁰Torr.²¹⁸ Secondary-electron ejection due to ion bombardment in this ion energy range, the upper half of which is commonly employed in modern glow-discharge sputter deposition, is today called “potential” electron emission. At ion bombardment energies E_i ≳ 1000 eV, (the “kinetic” electron emission range), γ increases with ion energy.^219,220

With E_i < 1000 eV, γ is not a strong function of ion energy and is found experimentally to be approximately 0.11 for 500 eV Ar⁺ irradiation of tungsten and 0.25, 0.06, and 0.02 for Ne⁺, Kr⁺, and Xe⁺ irradiation. Similar results were presented by Hagstrum for rare-gas bombardment of molybdenum (see Fig. 16).²²¹ Variations in γ with the choice of rare gas primarily arise from differences in first-ionization energies which range from 21.56 eV for Ne to 15.75 for Ar, 14.00 for Kr, to 12.13 eV for Xe. The observed decrease in the secondary-electron yield for Ne⁺ with E_i ≲ 200 eV (Fig. 16) results from the decreasing probability of Auger de-excitation processes involving short-lived excited states (thus the requirement for higher ion velocities) of the incident ion.^217,222

Several of the issues described in Sec. IV C which limited the accuracy of early sputter-yield S(E_i) measurements were solved by Güntherschulze and Meyer in 1930 using their newly developed triode glow-discharge sputtering system.²²³ Large electron currents were obtained by thermionic emission from a separate filament electrode in order to overcome the low secondary-electron yield of the target. This provided a sufficiently dense plasma to allow sustained discharge operation at pressures between 1 and 10 mTorr. Thus, the mean-free path of ions and sputtered atoms was of the order of the discharge-tube dimensions (ballistic transport) with minimal scattering and back-diffusion to the target. Güntherschulze and Meyer immersed planar 5-cm-diameter Ag and Cu sputtering targets, during separate experiments, in dense plasmas as a third independent electrode. The ion current density at the target was approximately 1 mA/cm² and the ion-sheath thickness at V_T = −1000 V was ∼5 mm such that the energy of ions (Ar⁺ or Ne⁺) incident at the target was ∼eV_T with a nearly monoenergetic distribution. The sputtering gas continually flowed and was spectroscopically monitored for purity.

Güntherschulze and Meyer sputter etched the target before initiating yield measurements and carried out separate experiments to determine the secondary-electron currents in order to obtain the actual ion current at the target. Figure 17 shows S(E_i) for Cu sputtered in Ar with and without accounting for the secondary-electron yield γ.²²³ The results are in reasonable agreement with modern sputtering yield data.²²⁴ The Cu target was uncooled and reported to reach a temperature of ∼530 °C during long sputtering runs. In a later publication,²²⁵ the same authors measured the sputtering rates of 16 different metals by Hg⁺ ions at a gas pressure of 5 mTorr, controlled by adjusting the temperature of an oil bath in which the glow-discharge tube, containing a liquid-mercury pool, was partially submerged.

In 1956, Wehner²²⁶ improved the design of an earlier (1940) version of a mercury-pool-supported glow-discharge tube developed by Fetz.²²⁷ An illustration of Wehner's initial mercury-pumped (see Sec. III A) glass sputtering system is shown in Fig. 18.²²⁶ While the base pressure was not specified, from comments in later papers, it was ∼10⁻⁶Torr.^228,229 The target in this system is immersed as a large Langmuir probe^226,230 in a low-pressure, high-density plasma created by a vacuum-arc discharge between the liquid-pool cathode and an anode. The Hg pool is in the bottom of the tube, and the vacuum-arc is ignited by a SiC rod inserted into the liquid pool.²²⁸ The cathode spot is anchored to a molybdenum strip that is partially submersed. An auxiliary (secondary) anode is required to maintain the cathode spot firmly anchored. The mercury pool is heated by an external water bath to 17 °C to provide a Hg gas pressure of ∼1 mTorr.

The system consisted of two parts; a stainless-steel ring, sealed with rubber O-rings, joins the upper tube, containing the primary anode and the target, to the lower rounded section. The plasma density was increased by inserting a fine-mesh grid in the ring to separate the upper anode space from the lower liquid-pool cathode space. Fetz²²⁷ had previously shown that an electrical double layer is formed in, and near, the grid holes and that electrons reaching this layer from the cathode side are accelerated toward the anode space. These energetic “beam” electrons (controlled by the potential applied to the grid) are more efficient in ionization and greatly increase the plasma density in the anode space. An additional enhancement in plasma density was achieved by placing the anode outside the direct path of the beam electrons and adding a repeller electrode with a negative potential to reflect beam electrons back into the intense region of the plasma. (Note that the glass walls of the tube became negatively charged and performed the same function.) Both the repeller and the grid were typically set to ∼−50 V relative to the anode.

The electrical conductivity of the resulting highly-ionized plasma was sufficiently high that the plasma and anode potentials were nearly identical and the applied target potential V_T determined the kinetic energy, eV_T, of the bombarding ions. The ion-current density at the target was approximately 10 mA/cm² and could be increased to 100 mA/cm² by the addition of a magnetic field near the target.²²⁹ Therefore, sputtering rates were large enough to minimize the effect of contamination on sputter-yield measurements. The targets in these experiments were metal rods which were heated to temperatures >300 °C, via a connection to an externally temperature-controlled sealed Kovar²³¹ tube, to prevent Hg accumulation prior to initiating sputtering experiments.^226,228 Heating was not necessary during sputtering since the targets reached much higher temperatures due to the large ion currents. For low melting-point targets, the Kovar tube was used to cool the target during sputtering.²²⁶ At an operating pressure of 1 mTorr with a target voltage of V_T = −200 V, the thickness of the cathode fall region in the discharge was less than 1 mm. Thus, the mean-free path of ions incident at the target was much larger than the width of the ion sheath, and sputtered-atom mean-free paths at 1 mTorr were larger than the distance from the target to the substrate.^226,228

Figure 19 shows Wehner's results for the Hg⁺ sputtering yield S/(1 + γ), uncorrected for the secondary electron yield γ, of polycrystalline platinum versus ion energy E_i.²²⁶ These, together with Refs. 223 and 225, are among the earliest reliable low-energy yield measurements and were carried out at target temperatures set to both 300 and 650 °C to show that Hg contamination of the target is negligible since the two data sets were within experimental uncertainty. Over the ion-energy range investigated, 100 to 410 eV, S/(1 + γ) increases essentially linearly with E_i.

The mercury-pool-supported discharge was also used by Wehner to measure low-energy rare-gas sputtering yields in an effort to determine “threshold” energies for different gas/metal combinations as discussed in Sec. IV G.

In order to measure rare-gas sputtering yields over an energy range more typical of that used in film deposition, Laegreid and Wehner,²³² as discussed in a previous paper,²³³ returned to the heated thermionic-cathode-supported triode discharge tube pioneered by Güntherschulze and Meyer.²²³ The upper part of the demountable tube, permanently connected to the vacuum pumping system, contains the thermionic cathode, while the target and the anode are in the lower section, with the two parts sealed by a copper gasket. The system base pressure, 10⁻⁷Torr, was measured using an ionization gauge. After baking the system for 14 h at 340 °C, the oxide cathode²³⁴ temperature was slowly increased, “never allowing the pressure to exceed 7 × 10⁻⁵mm Hg (7 × 10⁻⁵ Torr).” After filing the liquid-nitrogen trap, the background pressure with the cathode at temperature, but prior to introduction of the noble gas, did not exceed 8 × 10⁻⁷Torr.

The targets were uncooled, but the authors reported that “The influence of the target temperature (on measured sputtering yields) seems to be negligible (tested by varying ion current densities from 1 to 10 mA/cm²).” During sputtering, target temperatures ranged from 300 to 500 °C. Sputtering yields S(E_i), uncorrected for secondary-electron emission, were determined by combined target-current and weight-loss measurements. Example results are plotted in Fig. 20 for 15 different polycrystalline elements (of 28 studied) sputtered in argon with ion energies from 50 to 600 eV.²³⁵ The elements shown in Fig. 20 represent a significant fraction of the periodic table including p-block (Al), noble (Ag), base (Pb), ferromagnetic (Co and Ni), transition (Ti, V, Cu, and Y), lanthanide-series (Er), and actinide-series (U) metals, semiconductors (Si and Ge), and carbon.

It is interesting to note that while ion guns were available, and improving in quality,^236,237 during the prolific era of the 1950s and the early 1960s when supported glow discharges were being used to probe, and make enormous progress in understanding, the physics of sputtering over the energy range of interest for thin-film growth, they had little effect on the field. Although early ion guns were of poor vacuum design with highly divergent beams,¹⁶³ these issues were resolved over time.^236–238 However, the current densities provided by ion guns based upon dc-diode glow discharges remained low, and since the primary approach to determining absolute sputtering yields required measuring target weight loss, ion guns were not very practical for this purpose. The accuracy for measuring extremely small amounts of material loss due to low-energy ion bombardment was poor.

Conversely, at high ion energies, typically in the range of 10 to 100 keV, where sputtering yields are large, high-purity ion beams obtained from accelerators, which provide high current densities, were used in the early 1960s to measure S(E_i).^239,240 Almén and Bruce reported, for example, a sputtering yield of 48 atoms/ion for 45 keV Hg⁺ bombardment of polycrystalline silver.²⁴⁰ Figure 21 shows interesting results for a silver target sputtered by normally incident, extremely light, hydrogen and deuterium ion beams with energies in the low kilo-electron-volt range. The ions were obtained from an isotope separator and focused with an einzel lens.²⁴¹ The poor momentum transfer and large penetration depths due to the low mass of H⁺ (1.01 amu) and D⁺ (2.01 amu) ions, compared to that of the Ag target (107.87 amu), result in S(E_i) maxima at energies of a few kilo-electron-volt. For silver sputtered with Ar⁺ (39.95 amu), the S(E_i) maximum occurs at ∼40 keV.²⁴² 

An early investigation into the use of magnetic fields for supporting dc glow discharges was reported by Albert Hull in May 20, 1921, during a lecture delivered at the annual meeting of the American Institute of Electrical Engineers in New York City (published in September).²⁴³ Hull described a new vacuum-tube device, the magnetron, a member of the kenotron family of rectifiers and switches, which functions as an electronic valve controlled by a magnetic field. “If a constant voltage is impressed between cathode and anode, the current that flows through the tube is not affected by a magnetic field weaker than a certain critical value, but falls to zero if the field is increased beyond this value.”

Hull discussed two types of magnetron switches, whose geometries are illustrated schematically in Fig. 22, both having coaxial cylindrical symmetry.²⁴³ In the first one, Fig. 22(a) (referred to as the cylindrical-post configuration), the cathode is a thermionic filament surrounded by a cylindrical anode, while the second one [Fig. 22(b)], an inverted hollow-cathode magnetron, consists of a rod-shaped anode surrounded by a helical thermionic filament. With both types of magnetrons, the radial electric field is orthogonal to an axial magnetic field, parallel to the cathode surface, provided by a solenoid coil. In a previous publication (May of the same year), Hull²⁴⁴ presented analytical models describing the motion of thermionically-emitted electrons in the two magnetron geometries (although he did not use the term magnetron in the earlier paper). He also described results for the case of a parallel-plate diode with a magnetic field parallel to the electric field.

In the mid-1930s, Frans Penning (1894–1953) reported the first use of the cylindrical magnetron geometry, in which Hull's filament cathode²⁴³ was replaced by a rod, for sputter deposition.^245–247 There is no indication that Penning was aware of Hull's previous results. In 1936, Penning filed a U.S. patent (issued in February, 1939)²⁴⁵ describing a cylindrical-post magnetron (the word magnetron was not mentioned) as a competing approach to the supported glow discharge (Sec. IV E) for developing high ion currents at reduced gas pressures. Penning employed an axially symmetric magnetic field, provided by a Helmholtz coil, between a rod-shaped target and a concentric anode tube [Fig. 22(a)] in order to enhance the ion current by increasing the electron residence time in the discharge. More importantly, he produced (perhaps unknowingly) the conditions for magnetron sputtering, a closed electric- and magnetic-field trap which increases the plasma density at the target (cathode).

Secondary electrons emitted from the target respond to the $E ¯$ ×  $B ¯$ Lorenz force by initiating a cycloidal rotation around the cylindrical cathode. Those electrons which, during the first orbit, lose even a small amount of energy by making collisions or interacting with plasma oscillations^235,248 cannot return to the target against the radial $E ¯$ field. Instead, they continue their cycloidal motion in the closed-field trap. (Thus, varying the pressure in a magnetron discharge changes the fraction of secondary electrons returning to the target which, in turn, alters the plasma ionization rate and, hence, the target sputtering rate.) In order for the gyrating electrons to travel radially toward the anode, crossing axial $B ¯$-field lines, they must make collisions, thereby increasing the ionization probability near the target. However, as the electrons move away from the strong $E ¯$ field in the cathode-fall region, their $E ¯$ ×  $B ¯$ drift velocity decreases, they trace a more helical path, and their motion becomes primarily along $B ¯$-field lines, allowing them to escape from the intense region of the plasma. This gives rise, for the configuration shown in Fig. 22(a), to significant end losses, an issue which was solved a few years later by Penning and Moubis,²⁴⁷ who placed end caps on the post cathode [Fig. 22(c)] to form a more complete plasma trap.

Penning realized that the magnetic field strength in the plasma should not be so large as to affect the ions in the discharge, only the electrons.²⁴⁵ With the crossed electric and magnetic fields, “…the electrons are prevented from reaching the anode directly along the electric lines of force so that they traverse a materially longer path than in the absence of a magnetic field.” That is, “…the electrons describe, however, more or less helical paths around the magnetic lines of force…” With a metal target operated at 1000 V in Ar, a magnetic field of 300 G “brought about a 300-fold amplification of the current.” Penning noted that: “This leads to strong disintegration of the target and the cathode particles disintegrated readily pass though the gas at a reduced pressure.” The patent thus describes many of the key operational features of modern magnetron sputtering. It also proposes the use of the dc magnetron for what today is termed a sputter-ion vacuum adsorption pump²⁴⁹ in which sputter-deposited metal atoms “getter” (adsorb) reactive gases to reduce the background pressure in a vacuum system.

Penning's 1939 patent goes on to describe a parallel-plate geometry in which two planar targets (cathodes) face each other inside a cylindrical anode which is surrounded by a Helmholtz coil. In this case, the magnetic field is primarily parallel to the electric field, thus giving rise to magnetically enhanced, but not magnetron, sputtering (i.e., no closed-field trap at the target). However, Penning presaged, by almost 60 years, invention of the “facing-target” sputtering configuration which today is used to minimize bombardment of growing films by negative oxygen ions during reactive sputter-deposition of oxide films onto substrates arranged radially about the axis connecting the targets.²⁵⁰ 

Penning and Moubis,²⁴⁷ in 1940, described a system, with a base pressure of ∼1 × 10⁻⁵ Torr, consisting of a 10-cm-diameter glass discharge tube with a tubular water-cooled cathode (2 cm in diameter by 25 cm long) which had an opposing set of end plates for plasma confinement [thus, the shape of a wooden spool for storing thread or wire, Fig. 22(c)], and a coaxial anode. The magnetic field was parallel to the cathode cylinder. They showed that with this arrangement, the plasma is strongly confined near the target surface yielding increased ionization and resulting in a narrow, ∼1 mm thick, essentially collisionless target sheath. Sputter-deposition rates were estimated by measuring the increased weight of small mica substrates. Several mica disks were successively exposed, using a magnetic transfer rod, without opening the tube.

The current/voltage characteristics during sputtering of a copper target in a 14 mTorr flowing Ar gas discharge are plotted in Fig. 23 as a function of the magnetic field strength $B ¯$⁠.²⁴⁷ Penning and Moubis reported that with V_T = − 500 V, a discharge current of 3 A, $B ¯$ = 350 G, and a target-to-substrate distance of ∼4.5 cm, the film deposition rate was ∼0.3 μm/min, a very high rate for the time. In discussing their experimental results for aluminum targets, they noted that when the target is oxidized, a lower discharge voltage is obtained for the same discharge current, which they attributed (correctly) to a higher secondary-electron yield for aluminum oxide than for aluminum.^251,252 There are even earlier indications in the literature of the same effect during the sputtering of oxidized magnesium.²⁵³ 

W. Gill and Eric Kay, in 1965, described the construction of an inverted (hollow-cathode) magnetron sputter deposition system²⁵⁴ as well as its operational characteristics as a function of magnetic-field strength. Approximately a decade later, Alan Penfield and John Thornton at Telic Corporation developed large commercial cylindrical-post (>1 m in length) and inverted magnetron sputtering systems with closed-field $E ¯$ ×  $B ¯$ traps for thin-film deposition at the industrial scale.^248,255,256 An early application for cylindrical-post magnetron sputtering was the deposition of chromium layers on glass photomask plates used for metallization lithography during silicon device fabrication.²⁵⁷ 

In the parallel-plate dc-diode geometry (Fig. 24), the use of a transverse magnetic field, orthogonal to the axial electric field, increases the ionization efficiency (magnetically enhanced sputtering, as noted above). Unfortunately, it does not form a closed-field trap and simply sweeps the glow discharge to one side.²⁴⁸ In 1939, Rokhlin²⁵⁸ suggested the use of a quadrupole magnetic field that is radially symmetric, but with large transverse components, in a parallel-plate configuration. Kay,²⁵⁹ in 1963, investigated the use of such a system for thin-film deposition. He reported an increase in the ion current density at an aluminum target by more than an order of magnitude, which resulted in a corresponding enhancement in the film deposition rate. Mullaly, in 1969, also reported high deposition rates using a magnetron supported with a quadrupole magnetic field.²⁶⁰ In his case, the target was hemispherical. The crossed fields, in both Kay's and Mullaly's experiments, formed high-sputtering-rate rings at the target surface, yielding circular racetrack-shaped erosion profiles. The hemispherical target geometry of Mullaly resulted in a more uniform radial deposition rate at the substrate.

The earliest report of a parallel-plate sputtering geometry with a closed $E ¯$ ×  $B ¯$ field formed by a pair of permanent magnets having opposing poles was by Wolfgang Knauer of Hughes Research, Malibu, CA, in a patent filed on September 10, 1962, and issued on November 9, 1965.²⁶¹ Most of the patent discusses a closed-field configuration with a cylindrical magnetron (a “Penning cell”) for use as a sputter-ion vacuum pump, “which results in the removal of reactive gas due to adsorption by freshly deposited reactive sputter material as well as by ion burial.” However, Knauer also described a circular ring-shaped planar magnetron with an annular closed-field region of high sputtering rate at the target surface [Fig. 25(a)]. The magnets in Knauer's design were placed behind the targets as in modern magnetron sputtering systems.

In 1969, Wasa and Hyakawa described both a Penning-type cylindrical magnetron and a dc-diode parallel-plate closed-field magnetron [Fig. 25(b)].²⁶² In the latter device, the electrodes were circular disks facing each other in a sputter-down configuration. The solenoidal magnet was placed behind the substrate platen. However, no details were provided regarding operation of the planar magnetron, and all reported discharge characterization and film properties were obtained using the cylindrical magnetron.

The modern planar magnetron is often attributed to the patent of John Chapin, assigned to Airco, filed on January 31, 1974, issued on August 28, 1979,²⁶³ and discussed by Chapin in a 1974 industrial-magazine article.²⁶⁴ The device described in the patent was a water-cooled rectangular magnetron with a closed field, resulting in a continuous rectangular erosion profile into the target surface. However, Chapin pointed out: “Because of the planar cathode, the erosion region may be made in any desired configuration, depending upon the shape of the substrate…” The rectangular configuration was designed for commercial production in which substrates to be coated are continuously passed over the target (in a sputter-up configuration). “Thus, a plate having a width less than the length of the long parallel segments… will be provided with a highly uniform coating across its entire width. Accordingly, the apparatus of the invention not only achieves a high deposition rate, but may be utilized in high-production operations such as in connection with moving strip material.”

Figure 26 shows an illustration of Chapin's planar magnetron which he describes as follows:²⁶⁴ “…the magnetic field in the region just above the cathode surface or sputtering surface 12 forms a closed loop, indicated by 65 (the red shaded region defined by dashed lines) in Figs. 26(a) and 26(b). In this way, the electrons are confined in a sort of endless magnetic bottle adjacent the cathode and apart from the anode, further enhancing the rate of sputtering from the erosion region 15. In addition to trapping electrons, the magnetic bottle thus formed narrows the cathode dark space so that the space-charge-limited ion current is large. Because the cathode-anode circuit is entirely insulated from ground except through the ionized gas, there is little charge transfer to the substrate upon which the material is being deposited except for the small amount initially required to balance the cathode and anode at their natural potential, and except for a small amount due to the substrate contacting regions of the plasma at slightly different potentials.”

The first review article on modern planar magnetrons was published by Robert Waits in 1978.²⁶⁵ Figure 27, from Ref. 265, provides Illustrations of both circular and rectangular planar magnetron geometries, showing magnet placement and representative magnetic-field lines. The discharge electrons experience a complex trajectory with a general drift velocity, due to the closed $E ¯$ ×  $B ¯$ field, guiding them through the azimuthal (“racetrack”) magnetic tunnel.

An obvious disadvantage of the parallel-plate closed-field magnetron in the configurations shown in Fig. 27 is that the racetrack erosion profile consumes only a small fraction of target material giving rise to poor target utilization, originally ∼25%, which was increased to ∼33% by spreading out the magnetic field and opening up the racetrack. Since the permanent magnets of the parallel-plate magnetron are typically outside the vacuum system, further improvements in target utilization were obtained by rotating the magnet assembly, initially in a circular path and then in ever more complex hypotrochoidal paths allowing up to ∼60% of large rectangular targets to be sputtered.²⁶⁶ 

Another potential issue with magnetron sputter deposition is lateral film-thickness uniformity. This is relatively easy to achieve in the case of smaller substrates continuously translating under one or more larger rectangular targets for which deviations in the film thickness of ≤±1%–2% have been reported for reactive magnetron sputtering onto 2.5-m-wide architectural glass plates.²⁶⁷ Achieving this degree of thickness uniformity is more challenging for “single-wafer processing” on static substrates using circular targets as employed, for example, by the semiconductor industry during silicon-based device fabrication^152,268 and the optical storage industry for production of compact disks and digital video disks.^269,270 The measured and simulated film-thickness t results, published by Swan in 1988,²⁷¹ as a function of substrate radius r for copper deposition from a 10-cm-diameter circular target sputtered in Ar at 3.75 mTorr (target/substrate separation not reported) are shown in the left panel of Fig. 28. t(r) decreases from the maximum thickness, under the center of the target, to 59% at r = 5 cm, corresponding to the edge of the target.

Figure 28 also shows that Swan's calculated results, based upon an analytic geometric model developed much earlier,^272,273 provide good agreement with the measurements. The model assumes that the racetrack on a pristine flat circular magnetron target is composed of a set of narrow rings in which every point on each ring emits with a cosine distribution (see Sec. IV A) and that the sputtered atoms undergo no gas-phase scattering between the target and the substrate (ballistic transport). Film thickness distributions t(r) are calculated for each ring and summed. All terms, other than the mass of the film material and the film density, are purely geometric. For operation at higher pressures or larger target/substrate separations, gas-phase scattering must be accounted for, typically with Monte Carlo (MC) simulations.^274,275 Even though the target/substrate separation was not provided in Ref. 271, the good agreement between measured and model results indicates that deposition was essentially ballistic. After the extended target use, measured erosion profiles, for which the highest sputtering rate is in the middle of the profile, must be used as weighting factors in the model.

The right panel in Fig. 28 displays the calculated t(r) results obtained from the analytic model used by Swan compared with t(r) curves acquired from an MC code, with the acronym simulation of metal transport (SIMTRA), developed by Diederik Depla and W. Leroy at the University of Ghent.²⁷⁶ In this case, copper was sputtered from a 5-cm-diam target onto a parallel substrate 10 cm from the target. The MC code was run for 3 mTorr Ar pressure and assuming purely ballistic deposition; both curves are in good agreement with the analytic model. Note that t(r) at the substrate position corresponding to the target edge is 91% of the maximum, compared to 59% in the 1988 results. This is, in large part, due to significant improvements in magnetron manufacturing and magnetic-field design during the period between 1988 and 2012.

For the semiconductor-device-fabrication industry, where high thickness uniformity is required, the solution was again, as discussed above for rectangular magnetrons, complex rotation of the magnet assembly. In this case, the magnet motion typically follows “heart-shaped” pathways which, in turn, are rotated circularly^277,278 behind targets approximately 50% larger than the size of the wafer being coated. This allows target utilization up to ∼75% with film-thickness uniformity >98%.²⁷⁹ There are many other rotating magnet designs for different applications, and examples are described in Refs. 280 and 281.

The “S-gun” magnetron, patented in 1977 (filed in 1975) by Peter Clarke (1931–2002)²⁸² and illustrated in Fig. 29, was an early competitor of the planar magnetron. The device is designed such that the target, anode, water cooling, and power feedthroughs are all mounted on a single flange. The target is annular with a triangular (or truncated triangular) cross-section, the anode is a circular plate, and the permanent magnet is toroidal in shape to encircle the two electrodes. The device geometry is such that a strong $E ¯$ ×  $B ¯$ closed-field trap is formed at the target, as in the case of the planar magnetron, yielding high sputtering rates. Since both the target and the anode are in the dense region of the plasma (and hence both require water cooling), electron and ion bombardment of the substrate, typically placed across from the S-gun, is minimized. This can be an advantage for deposition on thermally sensitive substrates such as polymers and low melting-point metals.

A primary disadvantage of S-guns stems from a combination of the complex crossing of the electric- and magnetic-field lines coupled with the shape of the target; together, these factors result in a complicated sputter-erosion profile, leading, in turn, to a continuously changing film thickness distribution at the substrate, which varies more rapidly than that of planar magnetrons (Sec. IV F 2). Thus, full planetary substrate motion, utilizing all three axes of rotation, is required to obtain film-thickness uniformity. As a consequence, s-guns today are primarily used in small-volume specialty applications. Nevertheless, s-gun targets are still commercially available. The earliest review article covering operation of s-guns was by Dave Fraser, Bell Laboratories, in 1978.²⁸³ 

Rotatable tubular magnetrons, available in lengths >4 m, are increasingly popular for industrial applications, especially for high-deposition-rate coating of larger substrates such as architectural glass, for which standard “jumbo” panes in large office buildings are presently 6 × 3.2 m² (Ref. 284). The permanent magnet array in a rotatable magnetron is similar to that used in planar rectangular targets, with a corresponding rectangular racetrack, as shown schematically in Fig. 30. The earliest rotatable magnetron was described in a series of patents in the early 1980s by McKelvey.^285–287 The original concept was to provide greater target utilization. After a local erosion profile was etched nearly through the target, the magnetron was rotated to sputter a fresh area.

The first continuously rotating magnetron was the subject of a patent issued to the Von Ardenne Laboratory, Dresden, Germany, in 1982.²⁸⁸ The device was described a year later by Robinson at a conference in the US (Ref. 289) and discussed in the open literature in 1986 by Wright and Beardow of BVT Ltd., UK.²⁹⁰ As summarized in an excellent review article²⁹¹ by De Gryse, who was directly involved in the development of the modern rotatable magnetron, and colleagues from Ghent University, the continuously rotating magnetron has many advantages, in addition to increased target utilization, over the flat rectangular magnetron. These include:

1. a larger material inventory compared to a planar magnetron of the same length;

2. better target cooling efficiency which allows the use of higher powers in order to provide higher deposition rates (water flows through the rotating tube; thus, most of the tube is being cooled, while narrow regions rotate though the intense plasma zones); and

3. reduced arcing during reactive sputtering (Sec. IV J) since the length of the transition zone (the region at the highest risk of arcing)²⁹² between the high-sputtering-rate area and the poisoned-target area is greatly reduced, due to target rotation, as illustrated schematically in Fig. 30.

The Ghent group developed the first laboratory-scale rotatable magnetron and used it to carry-out fundamental investigations of target poisoning and hysteresis effects as a function of rotation rate.²⁹³ 

Chapin correctly pointed out in his 1978 patent²⁶³ that the substrate in a magnetron sputtering system is typically not located near the intense region of the plasma, which is trapped very close to the target by the crossed-field Lorentz force. While this configuration yields high sputtering rates, the disadvantage is that it is difficult to provide low-energy, high-flux ion irradiation to the growing film surface in order to control the film composition, preferred orientation, density, surface roughness, and physical properties^294–297 or to allow the growth of metastable phases via near-surface ion-mixing.^298–301 An initial step toward resolving this issue was provided a decade later by Brian Window and Nick Savides, from CSIRO in Sydney, Australia, who “unbalanced” the magnetron magnetic circuit. In a standard circular magnetron (for example), the permanent magnets are designed such that the magnetic field of the central pole is stronger than that of the outer annular magnet (see the left panel of Fig. 31). Windows and Savides replaced the permanent magnets with a pair in which the outer annular magnet was stronger than the central pole, thus opening a leak in the plasma trapped at the target and allowing low-energy ion bombardment of the substrate.³⁰² 

A more useful approach leading to modern tunable unbalanced magnetrons was developed in 1992 by a University of Illinois group who added a pair of external Helmholtz coils with iron pole pieces to controllably unbalance the magnetic circuit through the addition of an adjustable axial magnetic field in the region between the target and the substrate (Fig. 31).²⁹⁶ The external field (“external” to that of the magnetron permanent magnets) provides the ability to controllably shape the discharge near the substrate to provide independent control of the energy and flux of ions incident at the growing film, while having a negligible effect on the target sputtering rate. This approach to film synthesis is now employed in a wide variety of commercial applications with magnetron designs ranging from circular to rectangular to rotating cylindrical tubes.

“Threshold” energies for sputtering have been discussed in the literature since the late 1800s. Wehner and Anderson⁴³ noted that until 1912, a value of 495 eV was thought to be valid for all gas/metal combinations.³⁰³ Based upon the results presented in Sec. IV E, including Figs. 17, 19, and 20, this is far too large. Later reports (1928) described sputter damage of thoriated-tungsten filaments in gas rectifier tubes when ion bombardment energies exceed 20–30 eV.³⁰⁴ In the limit, the threshold is the minimum incident ion energy, for a given ion/target combination, that provides sufficient energy transfer during a near-surface collision cascade to allow a target atom to overcome its surface binding energy and be sputter ejected.

From the discussion in Sec. IV E, it is clear that measuring the threshold energy for a given ion/target pair is exceptionally challenging. Sputtering rates near threshold are far too low for standard direct weight-loss measurements, which do not extend much below sputter yields of ∼0.1 atoms/ion. More sensitive techniques are required. Interest in threshold energies in the middle 20th century derived not just from their importance in understanding the physics of sputtering and testing theories, but also for practical reasons. The sputtering threshold is related to radiation-damage thresholds in devices ranging from vacuum-tube electronics to sensitive detectors on orbiting satellites.

In a 1962 paper, Stuart and Wehner²²⁹ summarized sputtering threshold measurements starting from the work by Kingdon and Langmuir⁴⁷ and Hull and Winter,³⁰⁵ both in 1923, who used the change in thermionic emission from thoriated-tungsten wires, as thorium was sputter removed, to determine relative yields versus ion energy, to experiments based upon the resistance change of sputtered metal wire targets (1935),³⁰⁶ measurements of the surface ionization energy of low work-function alkali metals under ion bombardment (1954),³⁰⁷ radioactive-tracer methods (1956),³⁰⁸ and the use of an in situ microbalance (1957).³⁰⁹ The results yielded thresholds E_th for different ion/target combinations ranging from E_th = 45–55 eV for Ne⁺, Ar⁺, and Hg⁺ irradiation of Th-W wires⁴⁷ to ∼8 eV at a sputtering yield of ∼10⁻⁴ for Hg⁺ and Ar⁺ bombardment of Ni.³⁰⁸ However, Stuart and Wehner²²⁹ provided arguments as to why each of these results had significant uncertainties: low ion current densities compared to background-impurity chemisorption rates, thin wire targets leading to oblique ion incidence, plasma-tube design resulting in bombardment by doubly charged ions, etc.

In 1954, Wehner and Medicus,³¹⁰ using a glow-discharge tube supported via thermionic emission from a tungsten filament (Sec. IV E), reported yet another method for measuring relative S(E_i) values at very low ion energies E_i, in this case for Xe⁺ irradiation of platinum with E_i between 30 and 200 eV. A Langmuir probe was inserted in the discharge, and the authors took advantage of the fact that small changes in the probe work function can be accurately determined by displacements of the probe current/voltage characteristic along the voltage axis. The shift was calibrated versus the relative amount of platinum sputter deposited onto the probe. McKeown^311,312 in 1961 used the frequency change of a Au-coated piezoelectric quartz oscillator to obtain thresholds of 20–30 eV for Ne⁺, Ar⁺, and Xe⁺ ion bombardment with sputtering yields of ∼10⁻³ atoms/ion.

Stuart and Wehner^229,313 recorded threshold sputtering energies of ∼15 to 35 eV for several metals, plus germanium, bombarded by rare-gas and mercury ions. In these experiments, they employed a mercury-pool-supported discharge similar to that shown in Fig. 18, but with a planar target arrangement and a ring magnet to focus the discharge to the center of the target, thereby avoiding edge effects. The system, which had a liquid-nitrogen cold trap in the upper (target/anode) section to freeze the mercury out of the primary discharge and allow sputtering with a noble gas, provides ion current densities at the target of up to 100 mA/cm² with a background gas pressure of <10⁻⁶Torr. Mass spectroscopy measurements showed that double-charged ions contributed less than 1% of the ion current.

Relative sputtering yields S(E_i) at low bombardment energies were determined using optical-emission spectroscopy calibrated, at higher ion energies, with absolute weight-loss measurements.^229,313 The spectroscopic yield measurement geometry is shown in Fig. 32. Optical emission data were obtained close to the target in order to average the spectral line intensity over all ejection directions and thus minimize errors due to the variation in sputtered-atom ejection angles.

Example S(E_i) results, with E_i between ∼20 and 300 eV, are presented in Fig. 33.³¹³ From the yield curves in Fig. 33(a), as well as additional results in Ref. 229, the agreement between relative and absolute yield measurements at lower ion energies is better for rare-gas than for Hg⁺ ion sputtering, perhaps due to mercury contamination in the latter case. Figure 33(b) shows yield data for five metals, from relatively light Ti (47.87 amu) to very heavy Th (232.04 amu) targets, sputtered in Ar (39.95 amu). Based upon the results for 23 metals, ranging in atomic mass from Be (9.01 amu) to U (238.04 amu), sputtered with four different rare gases (Ne⁺, Ar⁺, Kr⁺, and Xe⁺), Stuart and Wehner concluded that metal sputtering thresholds, 15 to 35 eV, can be crudely approximated as being four times the target heat of vaporization.²²⁹ This was consistent with reports for radiation-damage induced lattice-atom displacement thresholds in solids.³¹⁴ 

Results from the earliest glow-discharge sputtering experiments, beginning in the mid-1800s, overwhelmingly support the momentum-transfer collision-cascade model of sputtering (see discussion in Sec. IV B). This is clearly reflected in review articles published in the 1960s^315–317 as well as in theoretical developments.^318–322 During cascade sputtering, the sputtering yield is expected, as noted by Wehner and coworkers,^43,228 to be proportional to an energy-transfer factor 4m_im_T/(m_i + m_T)², in which m_i is the mass of the incident ion and m_T is the target-atom mass. This simple function has a maximum when m_i = m_T. The sputtering yield should also vary inversely with the target-atom surface binding energy, often approximated for elemental targets as the heat of vaporization U_vap.³²³ Finally, for incident ion energies used in glow-discharge sputter deposition, typically a few hundreds to < 1000 eV, S for a given metal is approximately proportional to E_i²²⁴ (also see Fig. 20).

Modern sputtering theory is based upon a linear-transport model published in 1969 by Peter Sigmund from Aarhus University, Denmark.^324,325 He assumed that sputter ejection of target atoms by recoiled energetic ions or target atoms results from atomic collision cascades in an infinite medium. Edward Lamar and Karl Compton,³²⁶ in 1934, appear to be the first to suggest that sputtering can occur via momentum transfer as an ion penetrates the target and is reflected from a lower layer to collide with a surface atom. However, they neglected the important role of energetic displaced target atoms in collisional cascade processes.

Sigmund described collision cascades using a generalized Boltzmann transport equation with, for ion energies used in sputter deposition, a Born-Mayer type collision cross-section. He found that while lattice-atom scattering events take place over a thickness which is a sizable fraction of the ion range, depending on the ion mass and energy, target-atom mass, and ion/target geometry, the majority of sputtered particles originate from a very thin (∼5 Å) surface layer. For the simplest (although still quite complex) case of an amorphous target with a flat planar surface and normal ion incidence with E_i < 1 keV, the sputtering yield is given by

consistent with the experimental observations discussed above. The term α depends only on the ion and target-atom masses:

The fact that the above formulas provide a reasonable fit to sputter-yield data from polycrystalline and single-crystal targets in the few hundred electron-volt to 1 keV range is another indication that long-distance focusons (see Sec. IV B) do not play a significant role in determining sputtering rates. Related formulas developed by Sigmund, using a Thomas-Fermi collision cross-section, provide even better fits to high-energy, several kilo-electron-volt to mega-electron-volt, sputtering results.

Returning to Wehner's data in Fig. 15 (Sec. IV B) showing the average energies E_at of atoms which are sputter ejected, due to bombardment by Kr⁺ ions, orthogonal to the surface of a large number of elemental polycrystalline targets, the relative E_at results for the heavy elements Ta through U can now be understood based upon Eqs. (2) and (3). A “light” ion hitting a very heavy target does not penetrate deeply and has a high probability of being scattered sideways. Thus, more energy is deposited near the surface than would be the case for a collision of the same ion with a light-atom target; therefore, the product of the sputter yield S and the average ejection energy E_at will be high. However, as shown in Eq. (2), S ∝ 1/U_vap. Comparing the E_at results for Pt (195.08 amu) and Au (196.97 amu), whose masses are essentially the same with respect to the 83.80 amu mass of Kr⁺ (thus momentum transfer differences can, to first order, be ignored), S_Au is predicted to be larger than S_Pt, as observed experimentally, since U_vap,Pt = 5.8 eV while U_vap,Au is only 3.7 eV.³¹⁵ As a consequence, the average Au atom ejection energy is much lower than that of Pt, in agreement with the results in Fig. 15.

In the early 1980s, German physicists Bohdansky et al.³²⁷ developed an empirical formula for the sputtering yield in the near-threshold regime. Yamamura et al., from Nagoya University, Japan, proposed three different versions of a correction term to Sigmund's yield formula^324,325 to account for threshold-energy effects.^328–330 A fourth correction term was proposed by Bodansky.³³¹ In 1991, Ivan Petrov, Orlinov, and Grudeva,³³² of the Bulgarian Academy of Sciences concluded, based on comparisons of published experimental results with the predictions of Bohdansky et al.,³²⁷ Yamamura et al.,³²⁸ and Matsunami et al.,³²⁹ that for rare-gas sputtering (i.e., using Ne⁺, Ar⁺, Kr⁺, or Xe⁺) under conditions relevant to film deposition, the best fit is obtained with the latter two formulas using, in the case of heavy ions impinging on targets composed of lighter-mass atoms, for which m_T/m_i < 0.3, an additional small correction in the threshold-energy value.

Petrov et al.³³³ also showed that the energy efficiency of sputtering,

has a maximum η_max at an ion energy E_i ≃ 7E_th, in which E_th is the threshold energy (see Sec. IV G) For sputtering with argon (39.95 amu), η_max is between ∼0.4% and 4% for elemental targets from boron (atomic number Z_T = 5) to uranium (Z_T = 92), with periodic oscillations within each row of the periodic table as shown in Fig. 34. While the oscillations have been attributed to systematic changes in U_vap across the rows,³³⁴ Petrov et al. pointed out that this explanation is not sufficient; U_vap cancels out in Eq. (4) since from Eq. (2), S(E_i) ∝ 1/U_vap.

The use of neon (20.18 amu) as the sputtering gas for the same elements yields higher sputtering efficiency with an increase by up to ∼20%, compared to argon, for elements with Z_T ≲ 20; while krypton (83.80 amu) and xenon (131.30) are more efficient for targets with Z_T ≳ 40 and Z_T ≳ 50, respectively, by up to ∼65%. Argon is more efficient for sputtering elements in the middle of the periodic table.

Beginning in 1980, Jochen Biersack (Hahn-Meitner-Institut, Berlin Germany), Jim Ziegler (IBM Research, Yorktown Heights, NY), and colleagues developed transport of ions in matter (TRIM), a Monte Carlo computer code,^335–338 which now forms the core of a larger program, stopping and range of ions in matter (SRIM), designed to simulate the complex interactions associated with fast-ion bombardment of solids. SRIM is available as freeware on the worldwide web.³³⁹ The programs are continuously upgraded with major changes occurring every few years. The SRIM and TRIM codes are based upon the binary-collision approximation,³⁴⁰ with a randomly-selected impact parameter for each ion incident at the target. The required input parameters are the ion type, ion energy, and the target material (which is assumed to be amorphous). Outputs from the code include the average ion penetration depth (projected range), and standard deviation (straggle), into the target; the target sputtering yield; sputtered-atom energy and angular distributions; and the backscattering probability and backscattered energy distribution of incident ions. SRIM/TRIM is a very useful tool for estimating the magnitude of effects due to the change in the ion mass, ion energy, and/or the target material. TRIM can also be used to evaluate the role of ion and target choice on fast backscattered neutral-species bombardment of the growing film and to model the transport of sputter-ejected atoms through the gas phase.³⁴¹ 

MARLOWE, another code based on the binary collision approximation, was developed by Mark Robinson (Oak Ridge National Laboratory, TN)^340,342,343 to treat ion-bombardment of crystalline targets. Wolfhard Möller and Wolfgang Eckstein (Max-Planck-Institut für Plasmaphysik, Garching, Germany) created TRIDYN,^344,345 a dynamic version of static TRIM, which accounts for changes in the target composition, including implantation of incident ions, during sputtering of multicomponent systems.³⁴⁶ 

Peter Sigmund, the father of modern sputtering theory,^324,325 published a wonderful invited review article in 2012 that recounts, from a theorist's perspective, the recent (from ∼1955) evolution in the understanding of ion-bombardment-induced sputtering processes.³⁴⁷ The paper has the charming title: “Recollections of fifty years with sputtering.”

It was clear to those involved in the earliest studies of dc-glow-discharge sputtering that the technique cannot be used with insulating targets. The accumulation of positive charge at the target, which cannot be conducted away, will decrease, and eventually eliminate, the ion sheath (cathode fall), thus extinguishing the discharge if the process has not already been terminated due to arcing. An ac discharge is required in order that electron bombardment during the positive part of each cycle neutralizes the accumulated positive-ion charge from the previous negative half cycle. However, this, by itself, is still not sufficient to produce significant sputtering from a dielectric target.

Several researchers during the period from the mid-1950s to the mid-1960s recognized that the discharge frequency is a crucial parameter for optimizing dielectric sputtering. At low frequencies, discharges can be established, but there is no net dc potential at the target (i.e., there is no ion sheath) since the ion flux reaching the target during the negative half cycle is sufficient to neutralize the negative charge from the positive half cycle. However, electron mobilities, due to their low mass, are very much higher than ion mobilities; thus, as the discharge frequency is increased, at some point, there will not be enough time for the majority of ions to reach the target in a single half cycle. That is, the ions become inertially confined (their transit time across the ion sheath is much less than the ac period). At this point, the target surface accumulates negative charge and the net dc potential repels “excess” electrons to maintain a zero net dc current over each cycle. As the positive-ion sheath potential, which accelerates ions to the target, increases with increasing frequencies, sputtering begins to occur. The lowest frequency at which an ion sheath can form was estimated by Butler and Kino³⁴⁸ in 1962 to be ∼10 kHz, but simple estimates made by Wehner and colleagues³⁴⁹ in 1962 show that frequencies in the MHz range are more useful for sputtering since the net dc cathode-fall potential V_T increases with frequency. Both estimates are in the radio-frequency (rf) range which extends from approximately 3 kHz to 1 GHz.³⁵⁰ 

The earliest report on rf gas discharges was by J. J. Thomson in 1891.^1,352 He employed a Sprengel-type mercury pump (Sec. III A) and applied power inductively from a Leyden jar battery (Sec. III B) at frequencies in the MHz range (“…currents change their directions millions of times in a second…”). In his experiments, Thomson noted that it was important to have flowing gas: “It is essential to success that the gas in the bulbs or tubes should be quite dry and at a suitable pressure; there is a pressure at which the brilliancy of the discharge is a maximum, and as endeavoring to get to this pressure the exhaustion may be carried too far, it is convenient to use a form of mercury pump which will allow of the easy admission of a little gas.”³⁵² His experiments were carried out in air, carbonic acid (H₂CO₃), hydrogen, oxygen, coal gas, and acetylene (C₂H₂). In the same paper, Thomson also described the effect of the addition of a magnetic field, oriented both along and perpendicular to the discharge axis, on the distribution of discharge luminosity as observed by the naked eye and via spectroscopic analyses.

In 1933, Banerji and Ganguli performed experiments with 4 MHz inductive rf plasmas to deposit mercury films on the glass walls of a discharge tube.³⁵³ They realized that mercury atoms, evaporated from a heated source, were ionized in the discharge [an early example of ion plating,³⁵⁴ a term introduced by Don Mattox, Sandia National Laboratories, Albuquerque, NM, in 1964 (Ref. 355)]. In the same year, Robertson and Clapp reported the use of an inductive rf discharge for removing (ion-etching) metal layers deposited on Pyrex, soda glass, and quartz discharge tubes.³⁵⁶ “On applying a high-frequency field by means of external wire electrodes so that a luminous discharge was obtained [in air at a pressure of a few millimetres (a few Torr)], in a few minutes the deposit was removed in the neighborhood of the electrodes. By shifting the position of the electrodes, the whole tube was cleaned in a short time… A high-frequency potential (frequency not specified) applied to the electrodes caused no removal unless a gas or vapor was present to carry an actual discharge. Most of the work was done with air as ionized gas, but a few observations with hydrogen, nitrogen, and oxygen showed that with hydrogen and oxygen the action was extremely rapid, with nitrogen very much less so.”

A sketch of Robertson and Clapp's apparatus is shown in Fig. 35. The electrode positions are indicated by the letters E, and the cross-hatched region corresponds to the luminous area of the discharge. Hay, in 1938, reported similar results using frequencies of 3, 60, and 300 MHz in air, nitrogen, oxygen, and hydrogen discharges operated at pressures ranging from 70 mTorr to 3.5 Torr.³⁵⁷ 

In a 1955 review article on dc sputtering, Wehner³⁵⁸ described the concept of rf sputtering: “The positive charge on the insulator surface (due to ion bombardment) can be removed or the negative potential part time increased when a high-frequency potential is applied to a metal base underneath the insulator.” Jackson and Kane, in 1959,³⁵⁹ reported that the application of an rf field in a low-pressure environment (in their case, two collinear antennas, driven at 7.75 MHz, on a rocket passing through the ionosphere) results in a sheath accelerating positive ions to an insulator. This was demonstrated in 1961 by Butler using an inductively powered ring-electrode around a glass discharge tube; the results are described in a 1961 Stanford University Report.³⁶⁰ Two years later, Butler and Kino published both experimental and theoretical results in the open literature.³⁴⁸ 

Anderson, Mayer, and Wehner,³⁴⁹ in 1962, are credited with designing the first modern capacitively-coupled rf discharge for depositing dielectric thin films. “We describe a simple method for large-area sputtering of insulators which promises to be useful in many applications. The trick is to use a high-frequency potential at the target whereby the positive charge which accumulates on the target is periodically neutralized with plasma electrons during a portion of each cycle.”

The electrode configuration in the discharge tube used by Anderson et al.³⁴⁹ is shown in Fig. 36 and described as follows. “The dielectric target is immersed in the plasma of a low-pressure gas discharge. Immediately behind the dielectric target is a conducting plate to which a high-frequency potential [typically in the Mc (MHz) range] is applied with respect to the plasma. The potential of the dielectric surface, measured with respect to the plasma potential as a function of time, would be roughly as pictured in Fig. 37(a).” What the authors realized in sketching the figure was that the entire input waveform has to float down in potential such that the highly mobile electrons are captured only during a very small fraction of each cycle, while the rest of the cycle time is used to attract the slower ions. That is, the ions respond to the time-averaged electric field, while the much lighter electrons respond to the instantaneous field. Thus, although not mentioned explicitly in the article, the ions are accelerated on average by approximately half the peak-to-p