Implanter, RTP system issues for ultrashallow junction formation
09/01/2001
Mike Ameen, Jeffrey Hebb, Axcelis Technologies Inc., Beverly, Massachusetts
overview
Only when both ion implantation and rapid thermal annealing are tightly controlled can successful ultrashallow junction production be performed. An implanter's sensitivities are to dose variation, energy, and elemental contamination. These can be controlled with advanced software-driven control systems, precision power supplies, and accurately designed beamline geometry. The rapid thermal processing system sensitivities are temperature calibration and repeatability of short-cycle spike anneals, again addressed in simplified control system architecture and advanced temperature measurement capability.
Recent evidence suggests that the International Technology Roadmap for Semiconductors (ITRS) specifications for sub-100nm shallow junctions may be too aggressive. Indications are that low-energy ion implantation and rapid thermal spike annealing, available on today's production systems, will satisfactorily meet device requirements for even the 70nm technology node. Modeling data [1] and device prototypes [2] have indicated that optimized implant energies and anneal conditions are sufficient for these aggressive nodes.
 Figure 1. SIMS profiles of low-energy B+ implants in drift and deceleration modes (300keV Ge pre-amorphised wafers). |
The challenge for chip manufacturers is to ensure high-volume process capability: the ability to reliably and reproducibly implant and anneal ultrashallow source-and-drain extensions that these devices require. These requirements place considerable demands and constraints on processing equipment. The precise control of implant dose and energy is paramount. Also, control of high-temperature, short-duration spike anneals to activate an implant is just as critical. These processes are not independent, and process interactions leading to a particular sheet-resistance junction-depth (Rs-Xj) condition must be characterized and controlled for success.
Energy accuracy and purity
Low-energy implants used to fabricate shallow junctions require precise control of final beam energy. Low-energy ion beams are generated using a deceleration-based beamline architecture that gives rise to energy contamination concerns caused when higher energy species are neutralized prior to deceleration and reach the wafer with undesirably high energies. For instance, to provide 500eV boron implants with production-capable beam currents, a beam must be transmitted through a beamguide at values of up to 2.5keV, then decelerated prior to implant.
We have found that energy purity must be managed using an optimized deceleration scheme. For example, a single high-precision power supply determines the final energy of ions. The deceleration voltage is set with an independent supply, and all supplies are referenced to ground potential. In this manner, the only variation of final ion beam energy that may be introduced is through a single power supply that is easily monitored. Other designs — with final energy and deceleration supplies wired in series, or dual deceleration designs — necessarily incur greater voltage inaccuracies, particularly if current leakage due to poor isolation occurs or if cross-talk between the supplies occurs.
Proper design of the beamguide minimizes energy contamination. This requires high conductance throughout the deceleration region to limit localized pressure buildup.
 Figure 2. Rs as a function of dose for 1keV implants. |
It is also important to control beam transmission efficiency. Recent work has demonstrated improved energy purity for high transmission beams [3], as well as the capability to judge energy contamination by measuring this efficiency. The transmission efficiency is related to beamguide design, proper mass resolution (m/Dm) , and control of space charge effects. Figure 1 shows an overlay of various boron (B+) implants using drift mode and four different deceleration ratios. In all cases, some energetic contamination is observed, as evidenced by the extended tails in the SIMS profiles, though levels are well below 1%. Of particular desirability are the results observed for the 4:1 deceleration ratio with and without photoresist coated wafers. In this experiment, end-station pressure was >2.0 x 10-4 torr when using photoresist. The identical signatures of the two 4:1 curves indicate higher pressures have introduced no adverse effects during photoresist processing.
Dose control
The interaction between implanted dose and sheet resistance (Rs) depends on several factors, including implant energy and specific annealing conditions. Figure 2 shows a family of curves for dose and Rs at various peak-annealing temperatures. Anneals for this experiment were done using fast ramp rate spike anneals. The sensitivity of Rs to dose is seen to be higher in the 1014 ions/cm2 region, which is typical for shallow implant dosages. These data indicate that a 1% deviation in dose can lead to a 0.5% deviation in Rs under typical ultrashallow junction implant-anneal conditions.
Dose variation in ion implanters may typically be attributed to changes in pressure and composition of gases in the beamline, assuming other hardware — such as Faraday cups and ion gauges — are properly designed and maintained. In particular, low-energy beams have a high charge exchange cross section with residual species, and ions thus converted into neutral atoms will not be measured by a dosimetry system. This will lead to an overdose condition for constant background gas if not accounted for, and variations in dose if the pressure is not constant.
The background gas consists primarily of outgassing species from photoresist and gas from the plasma flood neutralizing system. The latter is a constant contribution to gas neutralization, while the former varies significantly during processing and must be compensated for in the control system. Both hardware and software control algorithms have been developed to manage current variation due to background gas variation.
From a hardware perspective, a real-time dose control system allows precise control of beam dwell time during ion implant and enables measurement of unavoidable variations in beam current. These variations may be caused by beam current drift, beam dropouts or glitches, or other beam mode changes that may occur during a slow scan.
 Figure 3. Use of dosimetry compensation to achieve repeatable performance with and without photoresist. Results are from a 1 x 1014 B+ ion/cm2 implant. |
For 300mm wafers without real-time dosimetry, a 25mm/sec slow scan rate will leave the beam unmeasured for 12 sec/scan if the beam is sampled only at top and bottom of a scan, and 24 secs/scan for single sampling schemes. A real-time dose control system should be capable of sampling the ion beam current during small intervals while the wafers are being implanted, and should be able to continually modify the scan speed during the implant dependent on the measured beam current. Implant uniformity and repeatability are ensured with this mode of control.
The contribution of gases in the beamline is also best accounted for in real time. Changes in flow controller rates, chamber outgassing, leaks, and pumping speed may lead to dose shifts that will not be measured if a constant offset, such as a trim factor, is used.
In particular, the pumping speed of a cryopump is known to vary significantly over a regeneration life cycle. This can affect the relative concentrations of gas species in the background.
Pressure variations as measured by precision ion gauges can be used to feed information back to a dose control system to simultaneously measure variations in background gas and in chamber pressure that occur when photoresist wafers are implanted.
Chamber volume does not play as significant a role in this scenario, nor does overall pumping speed and conductance to pumps.
Vacuum design considerations are to maximize pumping speed and minimize any conductance, limiting pathways to the pumps.
Automated dose control, designed with these factors in mind, is part of advanced low-energy ion implanters. For example, we use control system software with an algorithm to modify the measured beam current based on constant pressure and photoresist contributions. This method typically does not require setup runs or manual recipe parameters. The ideal operation of this type of system is that constant residual gas contributions to dose errors are separated from the photoresist effects, for example:
Im = Ic x exp [Kx x Px + Kpr x (P-Px)] (1)
where
Im = Ic x exp [Kx x Px + Kpr x (P-Px)] (1)
where
Im = measured current,
Ic = beam current in absence of any charge exchange,
Kx, Kpr = constants related to the charge exchange cross section of flood gas (x) and photoresist (pr),
P = pressure due to outgassing as measured at an ion gauge, and
Px = pressure prior to implant.
Since charge exchange is a function of ion species and energy and these are known quantities, an implant system can be automatically configured to determine and hence control these parameters. What results is straightforward and accurate control for pressure variations during ion implantation.
For example, when we used 0.5sccm of Xe in a plasma flood, we found the constant contribution of Xe as well as the variation in dose as a function of pressure were accurately accounted for in the wafer results (Fig. 3). Our real-time dose system also allows correction in scan speed as the implant progresses, improving uniformity of the implant.
Other software-related control features might be used to further fine-tune dosimetry. These include open loop schemes that are triggered at high-pressure thresholds, allowing operation in high-power, high-current regimes. Also, beam density determination and beam noise sampling allow for predictive and corrective modes of operation. All of these require high-speed electronics and advanced computing systems to obtain the sampling rate needed.
Additional features considered standard in ion implanters are data collection of all pertinent implant performance parameters and statistical process packages, necessary for troubleshooting and maintaining process capability.
Metals, cross-species contamination
The ITRS roadmap specifies stringent contamination requirements that require raising the standard of implant systems. Of particular concern is the amount of cross-species contamination introduced onto the surface of a wafer during implant. Advanced device technologies are near-surface sensitive, and many traditional options for protecting the surface, such as screen oxides, are not viable in the process flow. The option of dedicating implanters to one or more restricted species is undesirable, since tool utilization is compromised and autodosing from existing species is still an issue.
Cross-species contamination is caused by sputtering exposed surfaces during implant, leading to re-deposition onto a wafer's surface. The sputtering yield (Y) is a strong function of the energy of the implant and the mass of the ion. For this reason, low-energy boron implants are not expected to experience the same level of contamination that higher-energy arsenic beams would. To control cross contamination for these beams, the hardware can be designed to minimize exposed surfaces. In batch implanters, reduced surface area wafer disks reduce near-wafer sputtering, but lead to higher beam strike in the Faraday.
Use of a variable strike plate Faraday design (where the beam strikes triple surfaces at angles) is one way to avoid uncontrolled counterdoping or deeper junctions due to enhanced diffusion in arsenic. In this design, the strike plates in the Faraday are rotated depending on the ion species being implanted. In this way, a particular dopant is always striking the same surface, thus greatly reducing the chance of cross contamination. All other strike surfaces in the near-wafer vicinity are silicon coated to prevent metal contamination, though this will not significantly impact cross-species contamination.
In our measurements of differences in cross-species contamination between an open-architecture disk to a triple-surface disk Faraday — implanting 1 x 1015 As ions/cm2 at 50keV in an implanter that we loaded with 1 x 1017 phosphorus ions /cm2 — we found that the latter design practically eliminated autodosing.
Hot-wall RTP
For future device generations, spike anneals will be required for source-drain extension activation; this will be necessary to achieve sufficient electrical activation while avoiding excessive diffusion. Achieving acceptable within wafer (WIW) uniformity and wafer-to-wafer (WTW) repeatability while maintaining a small thermal budget is a current challenge for rapid thermal processing (RTP) systems.
We have found that an axisymmetric (i.e., spatially continuous) silicon carbide, bell-jar, hot-wall RTP system [4] is capable of delivering a manufacturable spike anneal process. This system provides a stable multizone thermal gradient in the vertical direction inside the bell jar; the heating zones are always at temperature, much like a classic diffusion furnace. Wafer temperature is controlled by moving the wafer up or down inside the bell jar. The system's spike thermocouples are set so the top of the bell jar is ~100-400°C higher than the desired processing temperature. The wafer is typically processed in the top half of the bell jar so that it is in quasithermal equilibrium with the bell jar at the processing temperature. Wafer temperatures up to 1200°C and ramp-up rates of up to 150°C/sec and ramp-down rates of up to 100°C/sec are readily achieved in this system.
This system has inherent advantages for temperature uniformity and control of spike anneals. The quasithermal equilibrium-significant environment results in a highly uniform temperature distribution across the wafer. Since the system inherently delivers WIW temperature uniformity, only the temperature at the center of the wafer needs to be measured and controlled, using state-of-the-art pyrometry incorporating emissivity measurement and correction [5]. This reduces control complexity to single-input single-output compared to multivariable control used in cold-wall RTP systems. The only variable that needs to be controlled is elevator motion using micron-level position tracking and control.
With this system, measured wafer temperature is used in conjunction with an adaptive, model-based control algorithm to produce the desired time-temperature profile [6]. Small thermal budgets are achieved in the system by increasing the maximum bell jar temperature (increasing ramp-up rate), and increasing the maximum acceleration and velocity of the elevator (increasing ramp-down rate).
Results from a five-day spike anneal repeatability study with this system, where the thermal budget allowed is only 2.4 sec above Tpeak 50°C, have been previously reported [4]. The thermal profiles demonstrated excellent repeatability where the range of maximum temperatures is >±1°C. For the 500 wafers run over the course of the five-day study, the total temperature variation was 1°C (1s), demonstrating suitability for manufacturing of shallow junctions (Fig. 4).
In our tests, we achieved 0.38% (1s) WIW uniformity when annealing a 500eV 5 x 1014 B+ ions/cm2 implant at 1150°C for 30 sec, conditions where the sheet resistance is most sensitive to implant dose. Using a 1050°C spike anneal with a 1.8 sec thermal budget, we also determined that 1.9°C (3s) WIW uniformity is equivalent to the uniformity achieved in state-of-the-art conventional RTP systems for standard soak anneals.
Ultrashallow junctions
Finally, the data in Fig. 5 show a representative production simulation of an implant-anneal process sequence designed to produce sub-40nm junctions. Here implants were 5 x 1014 B+ ions/cm2 at 500eV and anneals were 1.8 sec in 1000ppm of oxygen in nitrogen at 1050°C. The results show good repeatability and indicate potential for excellent process capability in a manufacturing environment. Collected SIMS data indicate that the junction depth was 37nm with <0.5% deviation. Other metrology techniques, including carrier illumination and capacitance measurements, are currently under evaluation as nondestructive in-line tools for maintaining process integrity.
Conclusion
The reliable production of ultrashallow junctions requires control of two very demanding processes in tandem. The implanter must be capable of delivering high beam currents at low energy levels. Reliable systems must be in place to provide real-time dose control despite photoresist outgassing. Energy contamination must be minimized while decelerating the beam to the low energies required and care must be taken in the system design to minimize elemental contamination both from exposed surfaces and other dopant species. The RTP system must provide excellent WIW and WTW repeatability while maintaining a small thermal budget.
We have demonstrated that an axisymmetric, hot-wall RTP can provide this repeatability using a state-of-the-art pyrometry system. Temperature control is easily achieved by moving the wafer up and down in the vertical furnace. The inherently uniform heating in a hot-wall furnace design provides the requisite repeatability across the wafer and from wafer to wafer. Only when both processes are tightly controlled can successful junction production be performed.
We have demonstrated that new systems with advanced features will enable this challenging process requirement for the next generation of chips.
Acknowledgments
We acknowledge Christina Sohl, Adam Stevenson, Jim Willis, and Chris Hatem of Axcelis Technologies for implants and anneals, and we thank Tom Parrill, John Morgan, and Aditya Agarwal, Axcelis Technologies, for useful discussions.
References
- H.J. Gossman, C.S. Rafferty, P. Keys, "Junctions for deep sub-100nm NMOS: How far will ion implantation take us?" Proceeding of the MRS Spring Meeting, 2000.
- H. Wakabayashi, Technical Digest of IEDM, 2000.
- H. Murooka et al., "Energy contamination control in deceleration beam lLine," Proceedings of IIT, Alpach, Austria, 2000.
- J.P. Hebb, A. Shajii, M. Flynn, "Furnace-based rapid thermal processing," Solid State Technology, pp. 155-164, October 2000.
- J.P. Hebb, A. Shajii, "Wafer temperature measurement in a furnace-based RTP system," 7th International Conference on Advanced Thermal Processing, 1999.
- A. Shajii, J.Hebb, B. Matthews, "Temperature control and uniformity in a furnace-based RTP system," 7th International Conference on Advanced Thermal Processing, 1999.
Michael Ameen received his PhD in chemistry from the University of North Carolina at Chapel Hill. Ameen is manager of process technology at the Implant Systems Division of Axcelis Technologies, 108 Cherry Hill Dr., Beverly, MA 01950; ph 978/921-9580; fax 978/787-3652, e-mail [email protected].
Jeffrey Hebb received his B. Eng in mechanical engineering from the Technical University of Nova Scotia and his MSc and PhD in mechanical engineering from the Massachusetts Institute of Technology. Hebb is the process development manager of the Thermal Processing Systems group at Axcelis Technologies.