Angle control in high-current ion implanters
10/01/2002
overview In serial implanters, ion beam incident angle variations are dominated by beam-steering issues, and in multiwafer implanters, by geometric effects. These variations are usually in the range of 0.5–0.7° for serial high-current implanters, but are often significantly below 0.5° for multiwafer implanters with dual-axis tilt. They almost always exceed 1° for multiwafer implanters with a single-axis tilt endstation. The process consequences of these differences for low- and high-energy implants are such that neither architecture maintains a consistent advantage in all cases.
The successful fabrication of advanced devices requires maximum reproducibility in the placement of the dopant profile in the silicon. A typical as-implanted implant profile has an extremely rapid rise and fall in the dopant concentration as a function of depth into the wafer, which is why these profiles are almost always plotted on logarithmic scales. Consequently, small changes in implant energy result in significant changes in doping concentration at a point on the rising edge of the profile. This can result in variations in threshold voltages and other undesirable effects. Implants into advanced devices must therefore be performed at an absolutely reproducible energy to prevent this. Variations in the total implanted dose will also affect dopant profiles and device characteristics, making dose control another crucial requirement of modern implanter systems.
The importance of minimizing variations in energy and dosing has long been recognized as the primary factor in achieving reproducible device performance by equipment and IC manufacturers alike. Since ion implanter manufacturers have been quite successful in recent years at minimizing energy and dose variations in most circumstances, attention has focused on variations in ion beam incident angle and their effect on devices.
No ion implanter ever has perfect control of ion beam incident angle across the wafer. Angle control is just like energy and dose control in this regard. In a serial high-current implanter with asymmetric magnetic angle correction, across-wafer variations in beam incidence angle always arise from errors in beam steering. In multiwafer implanter systems with spinning disks, variations arise from geometric effects in the endstation. This paper discusses the sources of these beam angle errors, and their potential effects on advanced devices.
Serial implanters
Most modern serial implanters are of the hybrid-scan type. The wafer is mechanically scanned in one direction, and the beam is electrically or magnetically scanned in the orthogonal direction. A variation of this is to spread the beam out into an unscanned "ribbon" beam that exceeds the diameter of the wafer. In either case, reproducible angle control is a significant technological challenge for beam transport design in this type of implanter. Because the wafer is mechanically scanned in only one direction (typically vertically), the beam can be modeled as a series of horizontal "beamlets," each of which is entirely responsible for dosing the corresponding horizontal position on the wafer. In order to achieve uniform doping, each beamlet must have the exact same intensity for dose control, and incident angle for angle control.
The beamlines of serial implanters are designed with an optimum path for each beamlet from the source through the analyzer magnet, through the scanning mechanism, and to the wafer. If the ion beam angle deviates by any amount from this optimum path as it leaves the source, or as it passes through any of the steering components of the beamline, it will arrive at the wafer at the wrong incident angle. Since the steering components of the beamline are by necessity designed for a perfectly aligned beam, any beam or beamlet that arrives at the wrong incident angle cannot be deflected back onto the optimum path. While it may be possible to design the beamline to minimize the sensitivity to these issues, they continue to be a concern on existing serial high-current implanters. This is important because the beam steering must be readjusted every time a new ion beam is set up on the implanter. Incident angle errors may therefore be different with each beam setup, leading to lot-to-lot variations in incident angle control.
null
The simultaneous, continuous achievement of both uniform doping and uniform angle may be difficult, especially on ribbon beam high-current implanters with asymmetric magnetic angle correction. It is difficult to fan out extracted ions into a ribbon beam that is a significant fraction of a meter wide without inducing variations in intensity, as shown in Fig. 1a. Once these variations are detected by a sliding faraday cup, the only way to minimize them is to use the correction magnets to redirect ions from beamlets of higher intensity to those of lower intensity. This requires that the trajectories of some ions be altered relative to the path of the ribbon beam (Fig. 1b), inducing across-wafer variations in ion beam incident angle. Because dose uniformity is more important for devices than beam angle uniformity, in practice, this unavoidable trade-off is always resolved in favor of maximizing dose uniformity at the expense of angle uniformity. The result is that serial implanters with magnetic angle correction specify the reproducibility of the beam angle on the wafer only to within 0.5–0.7°.
The use of symmetrical electrostatic beam scanning and angle correction greatly reduces this ribbon beam system problem, because the uniformity correction is made using a high-bandwidth scan waveform that allows modification of the flux profile independently of the angle correction. Also, the symmetric electrostatic correction lens allows parallel beams to be generated without bending the beam through an angle of some uncertainty.
Multiwafer implanters
Modern multiwafer ion implanters employ a fixed ion beam and full mechanical scanning in two dimensions. The wafers are placed on a disk that is spun about its axis while simultaneously being moved back and forth in the direction orthogonal to the spinning axis at the location of the ion beam. This completely eliminates the across-wafer variations in dose and incident angle that arise in a serial implanter from having each horizontal portion of the wafer implanted only by a unique portion of the beam. Instead, each point on the wafer is implanted by the same fixed beam, leading to theoretically superior dose uniformity.
Although the fixed beam eliminates angle errors resulting from nonparallel beamlets, across-wafer variations in the ion beam incident angle do arise from geometric effects relating to the nonzero angle of the wafer pedestals relative to the plane of the disk ("pad angle"). The magnitude of the beam angle variations depends in a complex way on 1) the tilt angle of the disk (both tilt angles in the case of a dual-axis endstation); 2) the pad angle; and 3) the number of wafer sites on the disk (disk radius) [1, 2]. All three of these parameters must be given when calculating the angle control on a spinning disk implanter; otherwise, the result is meaningless. Because this effect is purely geometric, the magnitude of the variations can be precisely calculated prior to implanting any wafers, and they are completely independent of beam setup. Neither of these qualifications applies to beam incident angle variations on a serial implanter.
null
Furthermore, unlike on a serial implanter, there are many ways for the end user to minimize the across-wafer beam angle variations on a spinning disk implanter. If the disk is oriented so that its axis of rotation is parallel to the ion beam, all across-wafer variations in tilt and twist are completely eliminated, regardless of wafer orientation on the disk. This is analogous to the North Star appearing at the same location in the nighttime sky regardless of the location or orientation of the observer, because its light rays always remain parallel to the axis of the earth's rotation. This condition does not hold for any other stars, which seem to "change" their apparent position as the earth rotates.
The smaller the angle between the disk rotation axis and the ion beam, the smaller the across-wafer variations in tilt and twist will be. This implies a process control advantage for disks that can be tilted in two orthogonal axes over those that tilt in a single axis. Most spinning disk implanters with single-axis tilt are designed such that the disk rotation axis cannot be made parallel to the ion beam. The process implications of this are shown in Fig. 2. For a 3° source/drain (or retrograde well) implant on a single-axis tilt endstation, the large angle between the disk rotation axis and the ion beam leads to across-wafer variations exceeding 2°. On a dual-axis endstation, the extra flexibility allows these variations to be kept below 0.1° across the entire wafer [2]. This is also significantly better than the angle control performance of a serial implanter with magnetic angle correction, as shown in the shaded region of the figure.
Device-shadowing and ion-channeling considerations
null
Doping by ion implantation is always a statistical process. The ions are never placed at a single point; rather, they are distributed around a peak depth, with a known straggle in both the longitudinal and lateral directions that depends on the species and energy of the implant. Straggle phenomena are always present, even in pre-amorphized substrates. The dopant distribution is further broadened by thermal and transient enhanced diffusion that always occurs during subsequent activation steps. Because of this, device-shadowing effects are only significant if they cause lateral shifts that exceed the unavoidable lateral straggle of the ions.
Figure 3 shows the relative importance of the three factors that affect the alignment of the extension implants to the channel — lateral straggle, undercut, and gate edge shadowing. For both boron and arsenic extension implants with a shadow angle of 1.0° and a typical gate stack height of 150nm, the lateral straggle dominates at all but the lowest energies. That small variations in incident ion beam angle do not adversely affect devices has been experimentally confirmed on sensitive RF CMOS devices [3]. Jasper et al. of Motorola compared serial and multiwafer implanters for extension and threshold adjust implants, and concluded there was no difference in the devices between the two implanter architectures, and that the use of quad-repositioning implants provided the best device control [3].
null
Unless the silicon has been pre-amorphized prior to implanting the extensions, the effect of beam angle variations is further muted by channeling effects. Every axial and planar channel has a critical acceptance angle for channeling to occur. Ions that are incident within the critical angle will be directed into the channel due to electronic forces arising from the atomic potential of the ions in the channel wall. The lower the incident energy of the ion, the easier it can be steered into a channel, leading to a larger acceptance angle for channeling. The magnitude of the critical angle for channeling is proportional to (Z/E)1/2, where Z is the atomic number of the ion and E is the incident energy. For boron into the <001> axial channel of silicon, the critical angle is only 2° at 60keV, but grows to 11° at 2keV and 22°at 0.5keV [4]. At very low energies, ions arriving several degrees from the normal will be steered into the channel. The resulting doping profiles are identical to a 0° implant, meaning that small deviations in ion incident angle are of no consequence for the resulting depth profiles at low energies—a fact experimentally verified by Walther et al. of Varian [5]. Channeling is an additional reason why dose uniformity is more crucial to device performance than small variations in beam incident angle.
Two other factors besides ion beam-steering issues also determine the control of the angle between the incident ion beam and the crystal planes of the wafer. (For channeling considerations at higher energies where the critical channeling angle is small, this is a better metric than the angle of the ion beam to the wafer normal.) These factors are mechanical misalignment in the endstation where the wafer is mounted, and the crystal cut error of the wafer. Both of these error sources clearly apply to both serial and multiwafer implanters. The magnitude of these errors is almost always < 0.5°, but is enough to have process and device consequences under certain conditions. Sudhama et al. of Motorola performed statistical modeling of implant angle error due to crystal cut error and endstation misalignment [6]. The results are plotted in Fig. 4 as the probability distribution of the true incident angle for a 0° implant, before any beam steering or geometric issues are considered.
Despite the desire for a 0° implant, the average effective tilt angle is more than 0.6°, with a considerable chance that the tilt angle will vary from 0–1.5°. For low-energy implants with a large critical angle for channeling, this is of no consequence for the devices [3]. For higher-energy implants, such as retrograde wells, the narrow critical angle for channeling suggests that 0° implants may produce larger process variations than off-axis quad implants. This is independent of whether the implanter is serial or multiwafer, and suggests avoiding 0° implants altogether, if possible, to maximize process robustness.
Conclusion
For virtually all implant applications, the most critical implanter parameters to control are dose and energy, followed by ion beam incident angle. Both serial and multiwafer implanter architectures have variations in ion beam incident angle. In serial implanters, these variations arise from beam-steering errors resulting from the complexity of the beamline. Serial high-current implanters using asymmetric magnetic angle correction are forced to trade-off beam parallelism for dose uniformity, and the situation changes with each beam setup.
 Figure 4. Probability distribution of actual tilt angle for a 0° implant arising from statistical variations in endstation alignment and crystal cut [6]. |
null
In multiwafer implanters, the variations arise from geometric effects due to the nonzero pad angle. These variations are completely constant lot-to-lot and can be minimized with proper tilt selections on a dual-axis endstation. Since neither implanter architecture maintains a consistent advantage in incident angle control for all cases, the choice of implanter type is determined by other considerations.
Serial implanters are chosen when high tilt capability is required, while multiwafer implanters are preferred for the superior productivity provided by simpler beamline designs. This is why both implanter types continue to remain popular in the marketplace, including purchases to date for 300mm fabs. For maximizing process control and productivity, the judicious use of quad implants and chaining, respectively, makes more difference than the choice of serial or multiwafer implanters.
Acknowledgments
The author thanks Tom Parrill of Axcelis for providing several of the figures in this paper, and Robert Rathmell, Andy Ray, and Kevin Wenzel of Axcelis for helpful discussions.
References
- A.M. Ray, J.P. Dykstra, "Beam Incidence Variations in Spinning Disk Ion Implanters," Nucl. Inst. Meth. Phys. Res., Vol. B55, p. 488, 1991.
- M.A. Jones et al., "Across-wafer Channeling Variations on Batch Implanters: A Graphical Technique to Analyze Spinning Disk Systems," Proc. XI Int'l. Conf. on Ion Implantation Technology, p. 264, 1996.
- C. Jasper et al., "Electrical Comparison of a Parallel Beam and Batch Implanters," Proc. of the XIII Int'l. Conf. on Ion Implantation Technology, p. 376, 2000.
- R.B. Simonton et al., "Channeling Effects in Ion Implantation into Silicon," in J. Ziegler, ed., Ion Implantation Science and Technology, p. 303, Ion Implantation Technology Co., Yorktown Heights, NY, 2000.
- S.R. Walther et al., "Dopant Channeling as a Function of Implant Angle for Low Energy Applications," Proc. of the XII Int'l. Conf. on Ion Implantation Technology, p. 126, 1998.
- C. Sudhama et al., "Robust Ion-Implantation Process Design Through Statistical Analysis," Proc. of the 3rd Intl. Conf. on Modeling and Simulation of Microsystems, San Diego, March 27–29, 2000.
Leonard Rubinreceived his SB in materials science and his SM and PhD in electronic materials from the Massachusetts Institute of Technology. He is senior scientist at Axcelis Technologies, 108 Cherry Hill Drive, Beverly, MA 01915; ph 978/787-4257, fax 978/787-4050, e-mail [email protected].