DC Parallel Ribbon Ion Beams for High-Dose Processes

Abstract

Ribbon beams of heavy ions have advantages over cylindrical beams, including higher space-charge limits. History of use goes from Calutrons, Freeman and Bernas ion sources, to the first ion implanters in the 1970s. In the 1990s, 300 mm uniform parallel mass-analyzed ribbon beams were developed to enable precise doping by mechanically scanning a substrate through the ion beam in one dimension. Ion species included the primary dopants boron, phosphorus and arsenic, but many others are also used. Such sources can produce currents of heavy ions with linear current densities at the source of the order of 10 mA/cm, but these sources are limited in the beam breadth they can produce. Broader beams are used for flat-panel display manufacture. A new linear source design combines a modified Penning trap with magnetic cusp confinement system, allowing extension of linear sources to meter scale beam breadths, maintaining around 10 mA/cm linear current density. Magnetic analysis of such beams has required new developments because the weight of conventional dipole magnets increases very steeply as the pole gap is increased. A new magnetic configuration has been developed to address this issue, reducing the potential weight of meter-scale analyzed systems by an order of magnitude.

1. Introduction

Ion implantation of semiconductors is the largest application of ion beam processing. In integrated circuits, flat-panel displays, and other devices, precise regions are implanted through masks with n-type and p-type dopants and other materials. Additional applications of ion beams exist, typically requiring high currents and fairly precise control of the ion dose. Ribbon beams are also still used in isotope separator systems, where pure isotopes are required for medical and other purposes. This chapter focusses on parallel DC ribbon ion beams, i.e. beams in which the breadth greatly exceeds its thickness; in this work I shall use the term beam breadth to indicate the major dimension, and thickness to indicate the minor dimension, regardless of system orientation in space. This is consistent with the ribbon analogy.

An excellent review of ion beam technology through its first century was given by Freeman [1]. The first isotope separators were developed by E.O. Lawrence based on his experiments in the 37-inch cyclotron at Berkeley, and used for uranium separation in the Manhattan project in the 1940s [2], using 16-inch broad ribbon beams. The ion beam was generated in an ion source immersed in the same uniform magnetic field used to separate the isotopes; there was therefore a strong and uniform magnetic field in the ion source oriented in the direction of the beam breadth. Electrons from a filament, inside a chamber with a single slot-shaped exit, were trapped by this magnetic field in a tight cylindrical region, where they ionized the uranium-containing vapor, from which the ribbon beam was extracted.

In the 1950s Freeman worked on applying aspects of this technology to peaceful purposes. His ion source used a magnetic field generated by a dedicated magnet which was separate from the analyzer magnet used for isotope separation. He ran a straight tungsten filament down the center of the region where it was desired to ionize the ion source vapor, and electrons were confined in complex pattens is a zone surrounding this filament. René Bernas [3] devised a similar ion source, with a coiled filament at one end of a similar arc chamber.

Freeman at Harwell then collaborated with Lintott Engineering in Horsham, UK, represented in the US by High Voltage Engineering in Massachusetts, to develop ion beam systems operating at tens of keV with mass resolving power around 60–100 to create pure beams of dopants for ion implantation [4]. Lintott’s early implanters were not known for reliability. Implanters were then commercialized by several companies in the 1970s including Extrion Corp, founded by Rose and others. The first commercially successful implanters tended to use lower currents than 1 mA, and scanned their ion beams across the surface of small silicon wafers; while usable, this scanning introduced variations in the incident angle, which resulted in variations in the range of the implanted ions, since the ions at certain incident directions could undergo planar or axial channeling in the monocrystalline silicon substrates. These systems often used a cold-cathode Penning ion source, producing a cylindrical beam of a few hundred microamps.

In 1988 Peter Rose founded Nova Associates Inc., with Ryding and Wittkower, to develop high-current implanters, starting with the Nova NV10–80. These took full account of the need for space-charge neutralization to transport currents of 10 mA or more of heavy ions at energies from 10 to 80 keV through a mass analyzer. These used a Freeman ion source, generating a ribbon beam about 44 mm tall and 3 mm thick at the ion source exit. Since a space-charge neutralized beam cannot be scanned electrostatically, they used two-dimensional mechanical scanning to uniformly raster the beam across the wafer surface. A batch of wafers was loaded into a circle onto a disk, which was spun to provide one direction of motion, and translated slowly in a radial direction to provide the other.

2. Limits to high-current beam extraction

The maximum current that can be extracted in an ion beam is determined by the space-charge limit, which for a cylindrical beam is approximately:

$$\mathrm{\Pi }=I\sqrt{\frac{M}{2q}}=\left(\frac{4{\epsilon }_0}{9}\right)V^{\raisebox{1ex}{$3$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}\left(\frac{r^2}{g^2}\right)$$E1

where I is the ion current, M the ion mass, q the ion charge, ε₀ is the permittivity of free space, V the ion beam extraction voltage, g the gap across which this voltage is applied, and r the beam radius. The equation is exact for a beam originating on one plane surface and being accelerated to a second plane surface at a potential V. Since the beam passes through a hole in the second electrode, the potential at the center of the hole is less than V, so in in practice the beam radius r cannot exceed g. In this case, the equation is not exact, since r/g must be <1 in practice, and so it provides an upper limit to the maximum current. The quantity Π is usually called the beam poissance when used for heavy ions. However, for a ribbon beam this equation becomes:

$$\mathrm{\Pi }=I\sqrt{\frac{M}{2q}}=\left(\frac{4{\epsilon }_0}{9}\right)V^{\raisebox{1ex}{$3$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}\left(\frac{\mathit{ab}}{g^2}\right)$$E2

where a is the beam thickness and b the ribbon breadth at the source aperture. Now the maximum is set by the condition a/g < 1. It is now possible for b to greatly exceed a without the potential at the center of the electrode being significantly different from V, and as a result the perveance from a ribbon-shaped ion beam system can exceed that from a cylindrical system by a factor of the order of b/a. Thus a ribbon beam system in the case of the Nova NV-10 high-current system can carry ∼44/3 times more current than a cylindrical beam.

3. Space-charge neutralization

A high-current ion beam contains ions which mutually repel each other, and a beam of the maximum current will diverge very rapidly, unless steps are taken to control this space-charge blowup. This process was known from the first days of the Calutron; the blowup can be mitigated by trapping electrons in the positively charged ion beam so that the space-charge of the ions is largely counteracted by the electrons. However, for a few decades the details of this process were not widely understood. Bernas [3] showed that the necessary electrons could be generated by collisions with residual gas atoms in several tens of microseconds, depending on the pressure and other conditions. It was understood since the 1960s that the extraction electrodes for a high-current ion beam have to be a triode, as in Figure 1, with the intermediate electrode at a more negative potential than the final beam potential, and of course the ion source at a positive potential, defining the final beam energy. The negative electrode, known as the suppression electrode, should repel electrons that find their way into the ion beam, and prevent them from being accelerated toward the ion source.

Figure 1.

A high current source using a Penning Ionization Gauge Trap (HC PIG ion source). Note the convergent ribbon beam formation, triode electrode arrangement with negative electron suppression electrode, and the non-uniformity of the magnetic field. The exit slot is 3 mm wide.

Ideally the electrons trapped in the positive ion beam will reach thermal equilibrium through multiple elastic collisions, and an unusual plasma will form, comprising the fast beam ions, a component of ∼room-temperature positive ions from the residual gas, mainly generated through charge-exchange processes with the beam, and a population of electrons in thermal equilibrium, the density of which is roughly equal to the total ion density. Hiroyuki Ito, in his Ph. D. thesis [5], has modeled this population and shown that the positive ions typically occupy a zone in which the potential varies by 0.5 kT_e, where kT_e, the electron temperature, can be measured and is typically between about 3 and 8 eV in a beam 100% contained in conductive grounded walls, except for a negative suppression electrode as shown in Figure 1.

In commercial implanters it was found necessary to add two further features to most heavy-ion beamlines to achieve and preserve this neutralization: (1) a second negative electrode sandwiched between two grounded electrodes as a triode structure through which the beam passed, to isolate the bulk of the beam from any interactions with the beam target, which was often a silicon wafer covered in insulating photoresist, which could charge to a high potential, and (2) a source of low-energy electrons (such as a plasma flood gun) to suppress significant potentials from developing on insulating surfaces. In many systems, similar electron guns add electrons to the main beam, but these are unnecessary if the beam is fully surrounded by conductive grounded walls.

4. Beam hash and instability: the limit to analyzed beam current

It had been found early on in the Calutrons that there was a practical limit to the high beam currents that could be reliably analyzed, due to a phenomenon referred to as ‘hash’ on the beam—this was quasi-random high-frequency noise of several hundred kHz, which disrupted the space-charge neutralization of the ion beam, resulting in a loss of mass resolution and an inability to transmit higher currents [1, 2]. This phenomenon was not fully understood, but I had observed it when attempting to increase the transmitted current of arsenic beams above the specification. The beam could clearly be observed to blow up, and to exhibit hysteresis as certain source parameters were tuned. I later observed a similar phenomenon on a test stand I was working on in 1990 and 1991, where I had developed a PIG ion source delivering significantly more low-energy current than prior sources in a low-emittance ribbon beam, but when I tried to transmit it through an analyzing magnet, the beam blew up, even in the space before entering the analyzer magnet. The ribbon beam source I was using, like that in the NV10 implanter, had a slightly concave front aperture from which the beam was extracted, with the electrodes shaped to be parallel to this curve, so that the beam converged in the breadth dimension. The universally held belief (despite the Calutron experience) was that a magnetic field helped space-charge neutralization, but our observation was that the magnetic field could disrupt it. We could turn the magnet on and off, and watch the magnetic field cause the disruption of the beam exiting the ion source and entering the magnet. We made Langmuir probe measurements of electron temperature in the space very close to the beam, and we observed that applying the magnetic field to the beam could trigger a dramatic increase in electron temperature.

Analysis of the beam size, current density, perveance etc., and the application of some standard plasma science, led us to hypothesize that under certain conditions, unstable ion sound waves could be transmitted through the plasma within the ion beam. We never saw the beam blowup below a certain threshold current, but blowup would occur suddenly as the magnetic field was raised, only when certain quantifiable conditions were met, namely that the ion plasma frequency, Ω_pi exceeded the ion cyclotron frequency, Ω_c, by a significant factor, of the order of 2.

$${\mathrm{\Omega }}_c=\frac{\mathit{qB}}{M}$$E3

$${\mathrm{\Omega }}_{\mathit{pi}}=q\sqrt{\frac{n_i}{{\epsilon }_{0}M}}$$E4

The cyclotron frequency in ion implanters is typically in the range from 200 kHz to 1 MHz, and when beam blowup is observed it is accompanied by chaotic hash on the beam at frequencies above this threshold. Note, however, that n_i in Eq. (4), the density of ions, is not a well-defined quantity, as there are several ion populations present, whose interaction is not straightforward.

$$n_i=n_b+n_r$$E5

where n_b is the density of ions in the beam (not uniform) and n_r is the density of slow ions generated from the residual gas.

$$n_b=J_b\sqrt{\frac{M}{2\mathit{qV}}}$$E6

The slow ion density n_r is very difficult to evaluate, as it is proportional to the pressure, the beam current density, and the local potential distribution, since these positive ions will be weakly accelerated away from the beam center; the potential distribution is affected by the electron temperature, so n_r will drop as soon as the beam neutralization is perturbed. The pressure is usually related to the beam current and to geometric factors, and is rarely well-known in the center of a magnet.

$$n_r=\frac{I_b{n}_0\sigma }{2\left(t_b+b_b\right)v_{i}q}$$E7

where t_b and b_b represent the beam thickness and breadth within the magnet, v_i is the velocity of the slow ions as they are repelled from the beam by its plasma potential, n₀ is the residual gas density, and σ is the total cross section for slow ion generation by the beam. To further complicate the picture, the beam ions contain multiple species, and Alexeff [6] showed that this can create additional instabilities. The beam ions are traveling at supersonic speed in the plasma. White and Gray Morgan presented a paper [7] in the form of a poster at the IEEE Plasma Science conference of 1991, in which scaling with the background gas ion density was demonstrated, and discussions with other participants including Igor Alexeff, founder of IEEE Plasma Science division, established that our scaling law was consistent with his models [6], and further that the geometry we were using could function in a similar manner to a klystron, and amplify the instabilities with self-feedback. When the ion density fell below the threshold we describe, the beam was stable.

5. The first DC ribbon-beam architecture

Moore’s Law predicted rapid scaling up of the number of devices on a single chip, and this required several developments: higher currents at lower energies (as devices became smaller), larger silicon wafers, moving to 200 mm diameter in the 1980s, better control of beam angles, and a strong preference for serial wafer processing to avoid the inefficiencies of processing in fixed-size batches. Serial processing was used in most semiconductor manufacturing processes, but high-current implantation still processed wafers in batches, resulting in higher costs. Applied Materials had acquired Lintott Engineering Ltd. In 1980, and in ‘85 was manufacturing a somewhat successful batch high current implanter, the PI 9000, but as larger wafer sizes were considered, the future looked uncertain. Applied’s other semiconductor tools all used serial processing.

The ASM 220 medium-current ion implanter used a horizontally scanned beam which was rendered parallel by means of a non-uniform magnet bending on average about 10 degrees [8, 9]. A single wafer was scanned mechanically through this beam in the vertical direction, to achieve uniform doping. This technique provided unprecedented angle control between beam and the wafer, excellent uniformity, and high efficiency (Figure 2).

Figure 2.

The ASM 220 ion implanter in which a beam is scanned horizontally, then collimated by a dipole lens magnet, to generate a parallel uniform scanned beam.

In conjunction with marketing and technical colleagues at Applied Materials, White in 1989 proposed (a) the acquisition of the ASM implanter division, and (b) the development of a serial-process high-current implanter using a wide DC ribbon beam [10], with a similar mechanical wafer scan mechanism to the ASM 220; and the two implanters would be complementary for medium and high-dose applications, at least on 200 mm wafers. We anticipated that the next silicon wafer size would be 300 mm, and we sized our high-current proposal for a 330 mm wide beam. Applied Materials did not agree to these plans, and as a result, Diamond Semiconductor Group (DSG) was founded in 1991, with a mission of developing such an implanter.

In that intervening two-year interval, as described above, White and Gray Morgan had experimentally confirmed the scaling laws of the beam instability problem [7]. DSG concluded that the use of a convergent slit on the front of the ion source raised the ion density within the analyzing magnet above the threshold for beam instability at those beam currents and energies which the market desired. DSG’s goals now included maximizing the beam size within the analyzing magnet, to keep the ion density low enough that Ω_pi remained below Ω_c, at beam currents of 5 mA at 5 keV or less.

The resulting design was the SHC-80 ion implanter [11] from which the VIISta-80 and other descendants under the VIISta tradename have manufactured and sold by Varian, and after acquisition by Applied Materials are still being produced and sold. The key design decision was to use a horizontally oriented convex ribbon beam ion source, and to produce a horizontally divergent ribbon beam to enter an analyzing magnet and be sharply refocused at a resolving aperture. This was loosely based on the design of the Applied/Lintott PI9000 ion implanter (See pages 241 and 242 of [1]). Then this analyzed beam was allowed to diverge again and was shaped by a 70-degree magnet into a parallel beam greater than 300 mm in width. This beam then implanted single 300 mm wafers which were mechanically vertically translated through it (Figure 3) [12].

Figure 3.

The SHC-80 beamline. The divergent beam from the HC PIG ion source is refocused through a resolving aperture, diverges again and is rendered parallel by the 70 deg. collimator magnet.

6. Ribbon beam flat-panel display implanters

Diamond Semiconductor Group then developed an implanter beamline with a ribbon beam width of >620 mm, for use in flat-panel display implanters made by Mitsui Zosen [13]. Subsequently White developed methods of changing the ion source orientation and achieving higher resolving power, by orienting the ribbon vertically, and bending it horizontally [14] with a magnet using ‘bedstead’ coils. This concept was used by Mitsui Zosen for producing a 1300 mm tall ribbon beam (Figures 4 and 5).

Figure 4.

A vertical ribbon beam 150 mm tall at the ion source expands vertically, is analyzed by a 90^o magnet with ‘bedstead’ coils, expands to 1300 mm, and is collimated by a U-multipole [15], which also adjusts the uniformity.

Figure 5.

The ion source magnet with optimized poles and yoke for a more uniform source magnetic field, as used with a 150 mm broad expanding beam in the implanter of Figure 4.

To operate an HC PIG ion source with a 150 mm long slot, and produce a uniform but expanding ion beam, the ion source magnet had to generate a substantially uniform field down the length of the ion source, the y-direction—yet the overall yoke height was only ∼500 mm, as it was in Figure 2. Usually this creates a minimum field at the center, and substantially greater field near the poles. But here the poles were shaped to cause a rapid increase in a direction transverse to the beam direction and normal to its breadth, which I label x, which improves the uniformity along the central axis. In this manner we were able to achieve +/− 2% uniformity in the arc chamber and permit substantially uniform ionization.

7. The control of parallelism and uniformity in ribbon beams

The approach pioneered in the ASM220 implanter [9] (Figure 2) for collimating scanned beams was easily adapted for DC ribbon beams. Varian had acquired the ASM implant division and now licensed the technology of the SHC-80 implanter of DSG, so the first prototype of the SHC-80 used the ASM-220 (now E220) wafer handler. The SHC-80 [12] design shown in Figure 3 used a uniform field 70-degree magnet, which provided the correct amount of horizontal focusing to render the strongly diverging beam from the resolving aperture parallel, but also provided some vertical focusing, thereby delivering a substantially parallel beam in two dimensions to the endstation of the implanter.

The control of uniformity was a more difficult problem. The first approach was crude and simple: to trim those regions of the ribbon beam where the current was too high with mechanically adjustable trimmers, nicknamed ‘magic fingers’. To minimize sputter contamination of the silicon, these fingers were located inside the first magnet, at the local maximum in the beam width. Unfortunately, it was found that these only rendered the beam less uniform, for as each finger neared the beam, a plasma sheath formed beside it and the electric field in the sheath deflected the beam ions that came close to it sufficiently to generate peaks and valleys on the current profile across the final beam.

However, this provided a hint for the successful development and use of linear multipole lenses to correct the uniformity [16]. To raise the linear current density (measured in mA/cm along the beam breadth dimension) it would be necessary to slightly deflect ions from neighboring regions toward the low-density zone. To lower it, the opposite would be true. We therefore used linear multipole lenses, being a horizontal array of pairs of magnetic poles above and below the ribbon beam, to successfully achieve this end (Figure 6).

Figure 6.

The multipole for uniformity control from the SHC-80 implanter.

An alternative mechanical multipole style consisted of a movable set of fingers made of magnetic steel at the edge of one of the magnet poles, thereby allowing the local field to be slightly raised or lowered. This can be seen in Figure 3.

Subsequently, White developed a multipole variation of the ‘Piccioni’ quadrupole lens [17], to simplify the manufacture and implementation of these multipoles, comprising two parallel steel bars, each wrapped with an array of coils. Energizing one pair of opposite coils on these bars creates a local quadrupole field component, which raises or lowers the linear current density of the region of the ribbon beam passing between that pair of coils. This arrangement produces a smoother control of the uniformity from a simpler structure, though the underlying theory is the same. Passing a similar current through all the coils in this multipole generates a uniform broad quadrupole field, which can be used for correcting the beam parallelism. A pair of such quadrupoles can be used to simultaneously fine tune the breadth, the parallelism and the uniformity of a ribbon beam (Figure 7).

Figure 7.

A multipole lens based on the Piccioni quadrupole lens concept.

A further variation of this was developed by White [15], in which one multipole has the two steel bars linked by a piece of magnetic steel at one end, creating a U-shaped magnetic steel yoke. This version can develop an overall dipole field superimposed on quadrupole and multipole field components, and it can be useful for introducing an overall deflection of 5–15 degrees into the beam. This can be seen in Figure 4, as used in Mitsui Zosen flat-panel display implanters.

8. Intense broad ribbon beams without local dipole magnets

8.1 Background

During and after the 1990s, almost all companies used ion sources based on the Bernas source with enhancements by White of a passive anticathode electrode at the opposite end of the arc chamber from the hot filament cathode, both being identically biased at about −60 to −120 V. This was later improved by Horsky [18], who replaced the hot filament with an indirectly heated cathode, (IHC) a block of tungsten with a hollow interior, heated by electrons from an internal hot filament, biased several hundred volts negative. This has a much greater service life, typically 500 hours, than a White source (∼150 hours) or a Freeman source (∼60 hours). I refer to all these variants as HC PIG (Penning Ionization Gauge) discharge sources. Early reports attributed little benefit to this HC PIG arrangement, but White and Westner found that when running BF₃ gas, the fast electron density could be significantly higher, and since B⁺ ions are generated from BF₂⁺ ionic molecules, which in turn are produced from BF₃ gas, the higher electron density results in a much higher fraction of B⁺.

In an HC PIG source, the locally applied magnetic field confines the fast primary electrons from the hot cathode radially, so they cannot reach the chamber walls at anode potential, and the cathode/anticathode electrodes repel the electrons from the ends. The result is the efficient creation of Penning Trap, in which fast electrons with energies between about 75 and 140 eV, depending on the cathode voltage, are confined. These efficiently ionize gas or vapor molecules within this zone.

The cross section for elastic scattering rises dramatically as the energy of electrons is reduced; as a result, those electrons which excite or ionize the gas, losing significant kinetic energy, then interact more strongly with each other through elastic collisions, and form a relatively cool plasma. The cooler electrons can diffuse quite readily across the magnetic field—it is only effective at confining the faster electrons, and so a cool plasma forms in thermal equilibrium at a temperature of 4–8 eV, and following classical plasma laws reaches an equilibrium voltage positive with respect to the potential of the anode walls (Figure 8).

Figure 8.

A long arc chamber with a 3 mm slot can generate a broad ribbon beam; the uniformity and efficiency of ionization along the arc chamber is crucial.

Now consider the problems of developing much larger ribbon beams, with sizes from 450 mm to say 2 m. With a much longer ion source, much higher beam currents could be obtained. The linear current density at the ion source which HC PIG sources produce is over 10 mA per cm, but they cannot be scaled up in their present form. The 400 mm Calutron sources were immersed in a gigantic uniform magnetic field; this approach is possible but completely uneconomic.

8.2 Design goals

We require an ion source producing a uniform ribbon beam with a breadth unlimited by practical scaling laws, at least 1.5 m, but with the thickness of the beam from prior sources, typically 3 mm but sometimes about 5 mm, with no magnetic field component along the axis of the arc chamber. This gives an aspect ratio of 360–500, and by Eq. (2) this gives very high upper limits to beam current, well over 1 amp of ions at keV energies. We require the ion source to be efficient, which means trapping 60–140 eV electrons in a tight efficient zone behind the extraction slit, and then introducing a uniform distribution of gas or vapor to be ionized.

We also want an analyzing device to separate ion species with modest resolving power, useful for a number of light ions, which also has a breadth essentially unlimited by scaling laws.

8.3 The CusPIG confinement trap

One of our first thoughts was to use a multicusp arrangement, the most obvious simple arrangement being a quadrupole magnetic field, oriented transverse to the beam breadth dimension, and extending uniformly down the arc chamber. However, we knew from prior work with multicusp trapping that it was far less efficient than a Penning trap, and fast electron densities in multicusp confinement were significantly lower than in a Penning trap, because of significant electron losses occurring at the cusps. So we looked for a way to block the four magnetic cusps with electrodes at cathode potential. An initial success with this arrangement led to a simplification: make the chamber walls at cathode potential, and place the anodes in a location where the magnetic shielding would be most efficient:

The arc chamber was an almost square section extruded in our first prototype to a length of 350 mm, but this could have been longer without any practical limit. At one end is an indirectly heated cathode, and at the other end is a wall, all the walls being at cathode potential. Magnetic poles run along the length of the arc chamber. The version shown in Figure 9 uses permanent magnets, but during the development, we used electromagnets. In principle there are four poles, but one pole is cut away for the beam exit slit. This producs a distorted quadrupole magnetic field extending along the length of the arc chamber, with the field components entirely in the plane of the cross section shown. There is a field null located within the arc chamber, just behind the exit slit. Two rods of molybdenum run the length of the arc chamber and are biased positive at a voltage between about 60 and 150 V. These are the only surfaces at anode potential, and are located where fast electrons from the hot cathode cannot reach them without crossing magnetic field lines with an integrated strength of about 0.3 T.mm. The crucial feature is that the magnetic cusps do not leak electrons, because they are blocked by walls at cathode potential. This is a form of quadrupole Penning trap, where cathode potential is placed at the magnetic cusps, and anode potential only exists in the region away from the cusps. We coined the name CusPIG to describe this geometry [19].

Figure 9.

Cross section of the CusPIG arc chamber, permanent magnet version. The magnetic field has a null running the length of the arc chamber, in line with the hot cathode.

There are other alternative implementations: the quadrupole magnetic field could be re-oriented by rotating 45 degrees, the magnetic poles being located near the corners. But the basic principle is a quadrupole magnetic field enhanced by Penning Trap potentials.

It was found that the efficiency of confinement varied very little with the strength of the field, so this permanent magnet arrangement became standard. The simplest construction provides two extended north poles each excited by an array of ¼” Neodymium-iron-boron magnets, one at each side, extending the length of the arc chamber, i.e. for the breadth of the beam. The arc chamber was graphite, and allowed to run hot, but was surrounded by a water-cooled aluminum case. A magnetic yoke surrounds three and a half sides of the arc chamber, provides a discrete south pole at the base, and provides two rudimentary weak south poles on either side of the beam exit slot. This creates a quadrupole field with a significant sextupole component; the fields are strongest in the back of the arc chamber and weaker near the exit, while the null line is located just behind the exit slot. Anode potential is confined to the two metal rods located where the magnetic field is strongest. This is the preferred construction.

Within a plasma magnetized by such a quadrupole field, fast electrons have a cycloidal path around magnetic field lines with a constantly changing curvature. Unless scattered by an ionizing event, the fast electrons have very low mobility normal to the field lines. But there will be a weak electric field normal to the field lines, giving rise to some electron cross-field drift at velocity v = ExB along the length of the arc chamber, in opposite directions in adjacent quadrants. Modeling also appears to reveal chaotic drifting along the arc chamber. Electrons are magnetically blocked from reaching the anodes, and are electrostatically blocked from reaching the walls. On reaching the end of the arc chamber there is a high chance of the electron hopping into another quadrant and returning in the opposite direction. By means of modeling the ballistic trajectories in the modeled magnetic field, and approximated electric field, using OPERA/TOSCA, we show that the whole arc chamber fills rapidly and uniformly with fast electrons except very close to the walls.

In Figure 10, the left view shows a transverse cross section of the electron trajectories as produced by the source of Figure 9. The Penning trap arrangement prevents any electron loss where the magnetic confinement would fail. The color represents the electron velocity; blue is low velocity, red is high—so the deceleration and reflection of electrons at the cusps can be seen. The quadrupole magnetic confinement shields the anode rods, while providing a zero field zone near the center of the chamber. Weak electric fields in and near the plasma sheath create ExB drift of the electrons, weakly emulating a magnetron racetrack, to help circulate the electrons around the discharge zone and enhance uniformity. The electron motion is very complex, consisting of cycloidal motion, but since the energy changes rapidly in the plasma sheath, this motion is somewhat chaotic. It extends rapidly down the length of the arc chamber, as shown.

Figure 10.

At left is a ray-trace of multiple electrons, viewed from the end of the arc chamber. At right is a side view of the trajectory of a single electron.

Gas is introduced through orifices shown in Figure 9, blocked by the anode rods from direct line-of-sight to the exit slot. This makes for very efficient ionization of the introduced gas. If the mass flow is precisely known, and the ion current is precisely known, one can calculate the ionization efficiency, in terms of the fraction of the gas that is introduced which exits as beam ions. This is a function of many factors, the arc voltage and current, which determine both the fast electron density and the cross section. Operating the ion source for a constant ion current, as the gas flow is reduced, a higher arc current is required. Using ∼6A of arc current and an anode voltage of 120 V, with argon gas, the ionization efficiency had a highest value of ∼30% at 8.9 sccm.

The cathode is at one end of the arc chamber. The density of ionizing electrons is observed to fall with distance from the cathode, since electrons which ionize an atom or molecule lose energy, are less effective, and rapidly diffuse out of the core zone. We found that this attenuation was a strong function of the anode voltage—the higher the voltage, the further the mean distance that the electrons traveled before being lost from the useful population. We have measured the attenuation of the ionizing electrons as a function of distance from the cathode by measuring the uniformity of the ion beam. The result is informative and useful. First, an arc voltage minimum of about 38 V is required to get a discharge, and the current density falls very rapidly with distance. At 70 V, the electrons travel 3 m before the ionization falls to 1/e of the initial value. (Incidentally, because their paths are not straight, this is equivalent to about 20 m of actual travel). At 120 V the attenuation was 20% at a distance of around 1.5 m from the cathode. It should be pointed out that because of the cycloidal electron motion, the true mean free path is many times greater than this. But as a practical matter, with 100–120 V anode voltage, the non-uniformity would be about 20% for a 1500 mm broad beam, and if a cathode were placed at each end (which is easily done), the intrinsic non-uniformity would fall to well under 1% (Figure 11).

Figure 11.

The measured drop in local ionization rate as a function of distance from the cathode.

The spacing and profiles of the electrodes, and the suppression voltage, are modeled with OPERA/ELECTRA, and can be optimized for voltages from ∼2 keV to 60 keV with ease. Figure 12 shows a 3 keV case. The beam is self space-charge neutralized, without the addition of neutralizing electrons from a gun. Great care was taken to avoid any unshielded connections to electrodes, etc., which should disrupt this neutralization. For the present we do not use a dynamically adjustable electrode gap, because the high precision required to optimize tuning along the full arc chamber length would be very mechanically challenging.

Figure 12.

Photograph from above a 350 mm broad vertical argon ribbon beam of 120 mA at 3 keV. The beam at left is emerging from the third electrode of the triode. Divergence is ∼+/− 20 mrad. Scale is about 50%.

Ionizing collisions are not the only inelastic collisions that can occur. Neutral excitation and resulting subsequent light emission also occur. For this reason, viewing the glow of the ion beam can be useful but misleading, for the glow is generated by electron transitions resulting from charge-exchange and other collisions with residual gas, so the emitted light is mainly from excited neutral atoms, so some of the halo around the beam may come from excited neutrals which travel outside the beam before decaying.

Discharge uniformity can be improved by two methods. The first is simple to implement. The gas flow to the different gas ports can be adjusted in real time to adjust the uniformity of the extracted beam. Crude adjustment is particularly easy: the current on the suppression electrode is proportional to ions which strike the electrode—ideally zero. The suppression current therefore measures beam de-tuning. Initially the suppression voltage can be adjusted to minimize this current, but then differential gas flow adjustments (made with the total beam current or arc current held constant) can be made to further reduce the unwanted suppression current, which is to effectively make the beam divergence—and hence its uniformity—more uniform. A second method, as discussed above, is to place a hot cathode at each end. If the ion source is well over 2 m long, it may be beneficial to add an additional hot cathode at a central location.

9. Extendable mass-Analyzer for meter-scale ribbon beams

Conventional magnetic mass analysis of ribbon beams broader than ∼500 mm is difficult and does not scale well. The gap between the poles of a conventional dipole magnet is much smaller than the bending radius of the trajectories in the magnet, to avoid enormous aberrations, stray fields and related difficulties. As the beam breadth is increased, the cost, weight, power of the magnet all increase more steeply than with the square of the beam breadth. The fringe field grows in proportion to the pole gap and is a source of aberrations and crosstalk with other system components, which also increase in magnitude with the square of the pole gap.

A number of partial solutions have been developed, as described above, but these have been incremental improvements with significant tradeoffs. Aitken proposed a radically new analyzer in 2002 [20] which used a new approach, from which the present design has taken inspiration; however, it was more complex and larger than the present device for given beam parameters. I attempted a design using an E-shaped magnetic yoke [21], which was successfully tested, but the present design is simpler and lighter, with better resolution. In a conventional magnet the focal length scales with the bending radius, and the bending radius of necessity scales with the gap between the poles, so for a beam 2 m broad, one can expect the focal length to be several meters, while the aberrations would be severe, and the weight to exceed a hundred tons.

The magnet design presented here addresses these issues, and differs from conventional dipole magnet configurations in the following ways:

• Ampere-turns are independent of beam breadth.

• Aberrations are independent of beam breadth.

• Overall dimensions and weight are a fraction of conventional magnets

• System path length does not increase with beam breadth

• Resolving power is adequate for boron and phosphorus, but not for medical isotope separation

• Beam current limit is proportional to beam breadth.

A system meeting these goals can have no component of magnetic field in the beam breadth direction, or it must violate these requirements. But if a field component normal to the breadth deflects the beam, then after a finite short distance the opposite field restores the beam direction, there must be a component of field in the gap between these zones in which a component of field is orthogonal to the s-shaped beam trajectories. Thus the solution is a magnet which deflects initially in the breadth direction, and produces a 3D bent S-shaped set of trajectories, hence the acronym U3DS [22].

9.1 The U3DS design

This magnet comprises a single U-shaped iron yoke as shown in Figures 13 and 14, providing two rectangular zones of uniform dipole field extending across the large dimension of the ribbon beam. Three optional iron bars labeled P1b, P2b, and P3 are shown placed on the opposite side of the beam. They are not mandatory, but serve the auxiliary functions of homogenizing the field, lowering the ampere-turn requirement, and thus the power requirement, and enabling some reduction in aberrations. Two simple identical rectangular coils provide the magnetic induction.

Figure 13.

Perspective view.

Figure 14.

Sectional view of the U3DS analyzer.

The U3DS magnet deflects the beam in an S-shaped path, bending in its major (breadth) x-direction and back, as illustrated in Figure 15, causing an offset in its path, and this deflection increases with the field strength. The net angular deflection in the breadth dimension is zero and is achromatic. However, unexpectedly, the beam is also deflected away from the exciting coils through an angle which is proportional to the square of the field strength, and is simultaneously strongly focused in this direction. The trajectories have 3D S-shaped paths, and this y-direction deflection allows the device to be used as a spectrometer in the conventional manner. The dispersion is surprisingly high, and therefore so is the resolving power. The ribbon beam is bent through a modest angle of 15–25 degrees in its thin direction, and strongly focused, while the peak deflection in the S-shaped path in the breadth direction is about twice this amount. The dispersion achieved is twice that of a simple dipole with the same bend angle, so the performance is as good as a magnet bending ∼50 degrees. The focal length is substantially shorter.

Figure 15.

Plan view: S-shaped path of reference trajectory.

The new magnet provides unperturbed uniformity along the beam breadth direction because there is no intrinsic variation of any field component along the breadth, and this avoids several of the worst aberrations of conventional dipole magnets. Nevertheless, there is a first-order aberration limiting resolving power, proportional to the intrinsic random divergence spread in the beam breadth direction. This limits the resolving power in practical situations to about 30.

Manufacture of IG6 flat panel displays requires a beam broader than 1.5 m, able to analyze an 80 keV P⁺ beam. A suitable U3DS magnet would weigh about 7 tons. I estimate that a suitable dipole magnet with a 1600 mm pole gap would weight about 100 tons. The beam current carrying capacity is many times greater, as discussed in detail below.

In Figure 15, three zones of magnetic field, q, r, and s, are labeled. Zone q is where the trajectory first passes between two poles (labeled pole P1a and P1b in Figure 14), where the field will be substantially uniform and orthogonal to the trajectory. The field will fall at the edges; the effective length of uniform field is denoted by dimension a in Figure 14, and the gap between poles by g. The magnetic field direction is reversed in the right-hand half. Magnetic field zone s in Figure 15 corresponds to this reversed reflection of zone q, adjacent to pole P2a. Central zone r is characterized by a predominant component of field from left to right in both views, but this field is non-uniform, and the field lines have some curvature. Poles P1b, P2b, and P3 are supplementary pole pieces which concentrate the H-field where it is required, thereby greatly reducing the required coil current and power. They can be used to control the field shape to minimize aberrations.

The deflection in the y-direction arises because in the central zone r the ions have a large component of motion in the x-direction, and the magnetic field is predominantly in the z-direction—hence the deflection is normal to both—and is in the y-direction. Being proportional both to the z-field and to the amount of motion in the x-direction, which is itself a function of B, the amount of y-deflection scales with B².

Optical elements are defined as illustrated in Figure 16. The focal length can be estimated from the geometry, given that the total deflection 2β = ∼2 h/f and 2 h = ∼3 g, and assume g = d, using symbols defined in Figure 16 This works well in practice. Greater accuracy requires full 3-dimensional trajectory modeling.

Figure 16.

Section with trajectories and optics.

The beam plasma instability in magnetic fields discussed above applies to any magnet design. Experimentally, as discussed above, the background ion plasma density seems to be the most important factor. Ion sound waves in this plasma can carry instabilities through the magnet and cause them to be amplified, if the threshold for this instability is significantly exceeded. For a rectangular beam of high aspect ratio A > > 1 whose width (narrow dimension) within the magnetic field B is t_b, and in an ambient pressure P, the dimensionless figure of demerit for instability in residual gas is.

$${\mathrm{\Omega }}_{\mathit{pi}}/{\mathrm{\Omega }}_c\cong \frac{M_i}{B}\sqrt{\frac{J_b{t}_b\mathit{P\sigma }}{{\epsilon }_0\mathit{qk}{T}_0\sqrt[]{M_{i}k{T}_e}}}$$E8

M_i is the ion mass, B is the magnetic field in the beam, J_b the current density, t_b is the beam thickness in the magnet in the narrow direction, and thus the quantity J_bt_b is the one-dimensional current density along the beam breadth direction, a quantity of direct interest. σ is the cross-section for slow ion production by charge exchange with beam ions, T₀ is the residual gas temperature, and T_e is the plasma electron temperature. This equation follows from Eq. (7), assuming a very broad beam whose thickness can be neglected in this context. As an example, for the magnet used as a worked example above, the figure of demerit for the background gas instability for a beam of 15 keV P⁺ at 0.43 T Tesla, in a pressure of 2 × 10⁻⁵ mbar, at a current of 500 mA per meter, is ∼1. This is a current well in excess of existing commercial requirements; the instability usually becomes problematic when this ratio is ∼2 or more; so for practical purposes we do not expect to see beam instability with this device.

10. Conclusions

The CusPIG ion source can generate current densities of ∼30 mA/sq. cm of many ion species, because of its efficient trapping of ∼100 eV primary electrons, and has a slot exit ∼3 mm wide, producing a ribbon beam, whose breadth has been demonstrated to 350 mm and can be extended to meter-scale dimensions; thus the linear current density has a maximum of around 1ampere per meter breadth.

The U3DS analyzer design can analyze meter-scale broad beams, while being less susceptible to the ‘hash’ instabilities of high-current beams in magnets. It allows the maximum current to increase in proportion to beam breadth. Paired with the CusPIG ion source, the U3DS magnet can analyze and transmit current densities approaching 1 ampere per meter, at an energy of 15 keV, from a 3 mm slit. With such an ion source it can achieve a modest resolving power of around 30.