Determining the limits of strain techniques in scaled CMOS devices

Koen Snoeckx, Peter Verheyen, and Geert Eneman, IMEC

 As traditional CMOS scaling
gasps for breath, every instantly applicable solution that boosts transistor performance is more than welcome. Strain techniques, which introduce compressive or tensile stress in the transistor channel, provide such an elegant method of improving charge mobility. Several strain techniques can be integrated into standard CMOS processing fairly easily. Moreover, combining different approaches improves the final result.

The trickiest part of using strain techniques is to find a simple solution that works for both NMOS and PMOS transistors. Together with its partners in the sub-45-nm CMOS program, IMEC, the Leuven, Belgium, research center for nanoelectronics and nanotechnology, is participating in the quest for the Holy Grail of strain engineering.

The most rudimentary way to introduce stress is by applying global strain—in other words, stress that is distributed equally over the entire wafer. By epitaxially growing silicon on a relaxed silicon germanium (SiGe) layer, the crystal lattice of the silicon can be stretched, resulting in higher carrier mobility and better performance. However, this approach has drawbacks. Several adjustments in the process parameters are required because of the introduction of the new substrate. What’s worse, the beneficial effect of global strain weakens as device dimensions are scaled.

A more durable solution appears to be the implementation of local-strain techniques, which are used only on the respective PMOS and NMOS transistors. The use of local-strain techniques has several requirements. The first is the direction in which the stress is introduced: parallel, vertical, or perpendicular. Each direction has a different sensitivity to how stress affects charge mobility. The second consideration is that NMOS and PMOS transistors can have an opposite reaction to local strain, which reduces the possibility of finding a simple and unified approach. Finally, each substrate and notch direction has its own characteristic transistor behavior, which can be a curse or a blessing when strain is applied.

To avoid working in the dark, it is possible to roughly predict the success of a local-strain technique theoretically by using the piezoelectric coefficients as basic parameters. This type of simulation determines the change in the electrical resistance of a material when stress is applied to it. The result provides an indication of the effect that stress has on charge mobility and, therefore, device performance.

For standard (100) surface wafers with a <110> notch, calculations indicate that NMOS transistors prefer a combination of tensile stress in the parallel direction along the channel and compressive stress in the vertical direction perpendicular to the wafer surface. In contrast, PMOS transistors prefer compressive stress in the parallel direction. While tensile stress in the in-plane direction perpendicular to the direction of the current is theoretically beneficial for both NMOS and PMOS transistors, that effect is difficult to obtain using common local-strain techniques. On the other hand, this stress is an important consideration in shallow-trench isolation (STI).

Several Options for Inducing Strain in CMOS Devices

The most popular local-strain technique exploits the strain induced by the contact-etch stop layer (CESL). Using simulation methods, the effect of tensile or compressive CESL on the stress in the channel region can be determined. Calculations show that the stress in the parallel and vertical directions is relatively equal in magnitude, although both types have opposite signs, as illustrated in Figure 1. For example, a tensile CESL results in tensile stress in the parallel direction and compressive stress in the vertical direction. While this outcome is suitable for NMOS transistors, it is problematic for PMOS transistors.

Figure 1: Simulation of the effect of (a) a tensile CESL with an intrinsic stress of 1 GPa and (b) an Si0.8Ge0.2 S/D on different stress directions in the center of the transistor channel. The gate length is 40 nm. Positive stress values are tensile, and negative values are compressive. Using common strain techniques, the resulting stress contribution in the perpendicular direction remains relatively small.

Fortunately, because stress theory is linear, variations can be accommodated. Hence, while a compressive CESL is beneficial for PMOS transistors, it has a deleterious effect on NMOS ones. IMEC has found that by manipulating the stress levels in the CESL, an Ion/Ioff performance gain of 10–15% can be obtained in NMOS transistors, while a gain of 25–30% can be obtained in PMOS transistors with high-k/metal gate stacks, as shown in Figure 2.¹

Figure 2: Measurements from L-arrays of PMOS transistors showing the additive effect of local-strain techniques on Ion/Ioff. Compared with the silicon reference, improvements of 30, 50, and 65% are obtained for silicon with a compressive barrier, 15% SiGe S/D with a compressive barrier, and 25% SiGE S/D with a compressive barrier, respectively.

Dual-CESL Approach. To exploit the benefits of both compressive and tensile stress on one wafer, a process flow must be found that integrates a tensile CESL on top of NMOS and a compressive CESL on PMOS. This dual approach appears to be the leading candidate for inducing stress in scaled CMOS devices.² As depicted in the schematic diagram in Figure 3, the method consists of eight steps:

Figure 3: Schematic diagram of the process flow for dual CESLs. The process involves seven more steps than the single-CESL process. A tensile liner is deposited on top of NMOS and a compressive liner on top of PMOS.

1. Depositing a tensile barrier.

2. Performing a lithography step to define the barrier on the PMOS.

3. Removing the etch barrier layer in the PMOS.

4. Postetch strip.

5. Depositing a compressive CESL layer.

6. Performing a final lithography step to define the compressive CESL on the NMOS.

7. Removing the etch barrier layer in the NMOS.

8. Postetch strip.

The entire process requires seven additional process steps, but experimental results indicate that despite its complexity, the dual approach results in better performance than a single- CESL approach.

Growing SiGe in the Source and Drain Regions. Another well-known local-strain approach is to integrate epitaxially grown SiGe source and drain (S/D) regions. Forcing SiGe, with its larger crystal lattice, to grow epitaxially on silicon results primarily in compressive stress in the parallel direction. Consequently, it can only be used for PMOS transistors.

The SiGe process flow involves eight steps, which include an etch step on the PMOS S/D regions followed by the epitaxial growth of SiGe inside the PMOS transistor. A performance gain of 20% is achieved when 15% germanium is applied in the source and drain, while a 40% gain is achieved for 25% germanium.

The SiGe method is not used for NMOS transistors. For NMOS, a similar effect may be achieved by epitaxially growing a material with a smaller crystal lattice, such as carbon-doped silicon. However, the current performance increase resulting from this technique is outweighed by the complexity of the deposition techniques required for this kind of material.

Another way to achieve stress is to combine SiGe S/D deposition with single or dual strained CESLs.³ The easiest option is to replace the neutral barrier with a tensile one, which results in NMOS performance gains while more or less maintaining the PMOS performance without additional process steps. Since the tensile barrier appears to have a less degrading effect in combination with SiGe, the need for an etch step on the PMOS is reduced. A plausible explanation for this phenomenon is that the topography of the SiGe S/D regions protects the PMOS channel from being influenced greatly by the CESL layer. This combined approach results in a 10–15% CESL improvement on NMOS and an estimated 18–38% improvement on PMOS, depending on the limited amount of degradation.

By pushing the limits, even higher performance gains are possible. The dual-CESL approach, combined with SiGe S/D, has the most beneficial effect on both PMOS and NMOS transistors: It results in a tensile liner on NMOS and an SiGe layer and a compressive liner on PMOS. The downside is that this method of inducing strain requires a total of 15 process steps (seven for the dual CESLs and eight for the SiGe).

Stress Memorization Technique. Another way to obtain local strain is the stress memorization technique.⁴ A possible process flow consists of a selective amorphizing implant on NMOS (with As+ or electrically neutral heavy atoms such as germanium), tensile liner deposition, highly doped–drain anneal, and liner removal before silicidation. It is hypothesized that after recrystallization, the poly gate preserves some of the stress condition, even when the tensile liner is removed.

The good news is that experiments using the stress memorization technique have resulted in performance gains of 6–8% for NMOS transistors. When combined with the dual- CESL technique, the performance gain is even greater. The bad news is that the technique has an extremely degrading effect on PMOS, the reason for which is not yet understood clearly. A major effect is that the stress-memorization technique results in an increase in the source-drain resistance of the PMOS. Other possibilities for inducing local strain are to manipulate the stress not only in the CESL but also in the entire premetal dielectric stack, or to exploit the stress introduced by STI oxide.

Turn and Repeat

The different methods for inducing local strain in CMOS devices can be used on substrates with different orientations.^5,6 The use of nonstandard orientations to improve PMOS performance was originally launched by IBM under the name of hybrid orientation technology. The technique starts with the basic understanding that a standard (100) substrate with a <110> notch is the worst possible combination for hole mobility but ideal for electrons. In contrast, a (110) substrate with <110> notch is bad for electrons but ideal for hole mobility, making it beneficial for PMOS transistors. As illustrated in Figure 4, IMEC has achieved improvements on the order of 40% for PMOS simply by using (110) wafers. However, unless measures are taken to prevent it, NMOS degradation is also approximately 40%, resulting in a net effect of zero.

Figure 4: Switching to the use of (110) wafers in place of standard (100) wafers results in a 42% performance increase for PMOS. Unless measures are taken to prevent it, NMOS degrades to the same degree that PMOS improves. (The device measured had a 1.5-nm HfO2 high-k layer; PFET VDD = –1.0 V.)

A related option involves using wafers with a rotated notch or simply rotating a standard wafer by 45°. Experiments at IMEC indicate that an improvement of approximately 20% for PMOS transistors—with no NMOS degradation—can be achieved by switching from standard (100) wafers with a <110> notch to (100) wafers with a <100> rotated notch.

The entire range of local-strain techniques can be investigated on wafers with alternative orientations, providing process engineers with a vast array of experimental possibilities. Simulations can enable them to make educated guesses. On (100) wafers with a <100> rotated notch, for example, calculations indicate that strain techniques have virtually no effect on PMOS. Substantiated by experimental results, these calculations are bad news for those who want to use SiGe but a blessing for those who wish to combine rotated notch wafers with a tensile CESL. The CESL boosts NMOS performance without affecting PMOS performance. PMOS performance improves thanks to the rotated notch. While work on combining different substrate orientations and rotated notches with local strain techniques is ongoing, the complexity of the final processing will determine the most favored options for industrial applications.

Conclusion

A broad range of techniques is available to boost transistor performance in scaled devices. The additive effects of these methods enable engineers to employ them in numerous combinations depending on their requirements. Moreover, most of the techniques—and the combinations thereof—can be used on wafers with a variety of substrate orientations. While fast and simple solutions such as tensile CESL on NMOS using wafers with a rotated notch are available, they often have limited value. More-complicated solutions such as SiGe with dual CESLs generally result in higher performance. Whether a particular strain technique will be used in the next generation will depend on how well it performs under conditions of device scaling. For example, the compressive stress induced by STI oxide will have a larger impact when transistor dimensions decrease.

Test results indicate that while a 65% increase in drive current is obtainable for transistors with an SiGe S/D containing large active areas, this improvement may decrease dramatically when transistor dimensions—such as device width and the source-drain length—undergo further scaling.⁷ When the transistor width is scaled, compressive stress increases in the perpendicular direction. While tensile stress is desirable in this direction, compressive stress is not. When the source-drain length is scaled, the compressive stress resulting from STI increases in the parallel direction. At the same time, however, the amount of SiGe in the S/D regions decreases, which also reduces the compressive stress introduced in the channel. These examples demonstrate clearly that research on strain effects in scaled devices must be approached with caution.

References

1. P Verheyen et al., “Demonstration of Recessed SiGe S/D and Inserted Metal Gate on HfO2 for High Performance PFETs,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 2005), 907–910.

2. S Pidin et al., “A Novel Strain Enhanced CMOS Architecture Using Selectively Deposited High Tensile and High Compressive Silicon Nitride Films,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 2004), 213–216.

3. K Ota et al., “Novel Locally Strained Channel Technique for High Performance 55nm CMOS,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 2002), 27–30.

4. T Ghani et al., “A 90nm High Volume Manufacturing Logic Technology Featuring Novel 45nm Gate Length Strained Silicon CMOS Transistors,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 2003), 11.6.1–11.6.3.

5. M Yang et al., “High Performance CMOS Fabricated on Hybrid Substrate with Different Crystal Orientations,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 2003), 18.7.1–18.7.4.

6. H Sayama et al., “Effect of <100> Channel Direction for High Performance SCE immune PMOSFET with Less than 0.15 µm Gate length,” in Technical Digest of the International Electron Devices Meeting (Piscataway, NJ: IEEE, 1999), 657–660.

7. G Eneman et al., “Layout Impact on the Performance of a Locally Strained PMOSFET,” in Symposium on VLSI Technology, Digest of Technical Papers (Piscataway, NJ: IEEE, 2005), 22–23.

---

Koen Snoeckx is a scientific editor at IMEC (Leuven, Belgium), where he is responsible for authoring and editing the organization’s technical documents and publications. He received a master’s degree in biochemistry from the University of Antwerp in Belgium. (Snoeckx can be reached at +32 16 288245 or koen.snoeckx@imec.be.)

Peter Verheyen, PhD, is a researcher at IMEC, where he is responsible for introducing strain techniques in scaled-down technology nodes. He received an engineering degree and a PhD degree from the Katholieke Universiteit Leuven. (Verheyen can be reached at +32 16 281603 or peter.verheyen@imec.be.)

Geert Eneman is a researcher at IMEC, where he is working toward a PhD in strained-silicon transistors. A research assistant for the Fund for Scientific Research, Flanders (Belgium), he received an MS in electrical engineering from the Katholieke Universiteit Leuven (Eneman can be reached at +32 16 281982 or geert.eneman@imec.be.)