Process Module Control for Advanced Gate Dielectrics

		At a Glance
	An effective process module control strategy has a variety of elements, ranging from high-precision measurements to optimal sampling plans and data-use algorithms. The foundation of such a strategy, described here, is the use of an advanced metrology tool set and methodology to generate a set of sensitive metrics that accurately reflect the yield-relevant optical and electrical properties of the gate dielectric material.

Advanced gate dielectrics having increasingly tight electrical performance requirements present a significant challenge in developing production-ready monitoring systems with measurements within industry-accepted process tolerance. One advanced class of gate dielectrics being developed is based on SiO₂ with starting oxide thicknesses in the 20-30 Å range. The controlled introduction of nitrogen into this base oxide is used to reduce boron diffusion from the doped polysilicon gate, through the oxide, and into the underlying gate channel. Nitridation also results in reduction of the gate dielectric's effective electrical thickness, though a physical dielectric thickness may be maintained that is consistent with acceptable levels of current leakage. This has led to growing interest in understanding not only the "optical thickness" of a film but also the electrical performance of the gate dielectric as early in the IC manufacturing process as possible, particularly in the area of advanced gate materials.

Currently, one popular, well-established method of measuring the thickness of SiO₂-based gate dielectrics is ellipsometry. As the industry migrates to nitrogen-doped gate dielectric processes, a variety of techniques are being employed to modulate nitrogen levels and spatial (depth) profiles, generating new parametric degrees of freedom for these films. Used alone, state-of-the-art optical techniques cannot completely resolve these degrees of freedom; instead, a composite optical and electrical scheme is required for effective material characterization. As detailed below, a combination of optical and electrical metrology has been shown to be sensitive to variation in the remote plasma nitrided oxide (RPNO) process.^1,2KLA-Tencor's ASET-F5 spectroscopic ellipsometer, an optically based thin-film measurement system, and the Quantox, an electrically based non-contact gate dielectric measurement system, are used in this study.

For a stack composed of a gate dielectric on silicon, the important optical parameters determined by ellipsometry are: thickness (t), refractive index (n) and extinction coefficient (k) of the dielectric, and n and k of the silicon substrate. The material dielectric constant (at optical frequencies), epsilon, is related to the refractive index, n, and extinction coefficient, k, by:

[1] epsilon = (n - ik)² = n² - k² - 2ink

where i is the imaginary unit.

The ASET-F5 has two subsystems, a spectroscopic ellipsometer (SE) and a dual-beam spectroreflectometer (DBS). SE measurements were taken in the spectral range of 240-800 nm. Absolute reflectivity data were taken between 195-205 nm using the standard DBS subsystem.

The Quantox uses charged corona ions, which are deposited onto any dielectric in a precisely controlled manner to produce an electric field. When measuring the dielectric thickness (T_ox), the known corona charge (Q) is applied to the dielectric incrementally, and the accumulated voltage (V) is measured. Since the parallel plate capacitor system is defined by the Q = CV relationship, knowing Q and V defines the system completely, and the effective electrical thickness, T_ox, is readily derived from equation 2:

 

 

where epsilon is the relative dielectric constant (at quasi-static electrical frequencies) and epsilon_o is the permittivity constant. Because the Quantox de-posits a calibrated charge per unit area, equation 2 results in a calculation that is independent of area.

3. Equivalent oxide thickness (EOT) measurements indicate distinctions between starting oxide film thickness and post-RPN/anneal thickness.

The net capacitance measured in a low-frequency C-V measurement trace will approach the true associated value of T_ox at both strong inversion and accumulation. Therefore, the system calculates a theoretical expected Q-V trace and fits the measured Q-V data to the theoretical trace using a non-linear least squares fit, with T_ox as one of the fitting parameters. Because epsilon for SiO₂ is used in the electrical measurements, the Quantox returns an equivalent oxide thickness (EOT) value.

Experiment

To test the process sensitivity of the ASET-F5 and Quantox, a set of eight, 200 mm p-type wafers was oxidized in a single furnace run. Starting oxide thickness was 26.5 Å. After oxidation, various combinations of the eight-wafer set were subjected to four different remote plasma nitridation (RPN1-4) and three different anneal (A1-3) processes. The process conditions used in this work are summarized in Table 1.

Table 1. RPDO Processing/Annealing Conditions
Remote plasma nitridation:
RPN1 = High power, Medium time (Std.) RPN3 = Low power, Medium time	RPN2 = Low power, Maximum time > RPN4 = High power, Minimum time
Post nitridation anneal:
A1 = High temperature, Gas 1 (Std.) A3 = High temperature, Gas 3	A2 = Low temperature, Gas 2

Nine-site optical measurements were performed after each separate process step (oxidation, nitridation and anneal). Five-site electrical measurements were performed after the oxidation and anneal steps only. The first split focused on the standard processing conditions of RPN 1 and anneal A1. Figure 1 shows the optical thickness measurements made on the eight-wafer set subjected to a single furnace firing and identical RPN nitridation and anneal. Careful attention was paid to metrology processing time to prevent buildup of airborne molecular contamination on the surface of the wafers. Unambiguous correlation of process step to optical thickness is observed.

Figure 2 shows optical reflectivity measurements made on the same eight-wafer set. Again, clear correlation of process step to reflectivity is observed. Figure 3 shows the equivalent oxide thickness (EOT) measurements made by the Quantox on the same eight-wafer set. The starting oxide film thickness and post-RPN treatment/anneal film thickness conditions are clearly distinguishable.

Analysis of metrology capability

A capability analysis was performed to determine whether these metrology instruments are capable of resolving differences in optical thickness, reflectivity and EOT at various steps in the RPNO process. Precision of the metrology tools (defined as 6s variation) was determined from a 30-day gauge study. Capability was estimated by computing the difference in mean values of each data set, divided by the precision (6s variation) of the metrology tool. If the resulting value is greater than or equal to 1, the mean values are distinguishable by the metrology tools, at a confidence level of 99.9%. The significance calculation is:

where Y1 and Y2 are mean values of parameters measured at various process steps.

Table 2 shows results of this calculation for each of the process steps. The modulus of the significance value is not shown; instead the sign of the measurement differential is included to indicate the effect of processing on each parameter. Data in Table 2 demonstrate the proposed optical and electrical met-rology solution is capable of resolving the process variations observed here.

Table 2. Oxide Metrology Results (26.5 Å Starting Thickness)
Parameter	Significance
Optical Thickness:
Initial Oxide (Y1) — RPN (Y2) RPN (Y1) — Anneal (Y2) Initial Oxide (Y1) — Anneal (Y2)	-2.63 1.13 -1.50
Reflectivity:
Initial Oxide (Y1) — RPN (Y2) RPN (Y1) — Anneal (Y2) Initial Oxide (Y1) — Anneal (Y2)	3.55 -1.80 1.75
Electrical Thickness:
Initial Oxide (Y1) — Anneal (Y2)	1.05

Correlation to end-of-line electrical data

To determine whether in-line measurements of the Quantox EOT correlated with end-of-line (EOL) polysilicon MOSCAP measurements of EOT, a range of RPNO processing and annealing conditions was investigated on a second series of 26.5 Å furnace oxides. One set of four wafers was processed through each type of RPN treatment (RPN 1-4) and subsequently exposed to only one anneal treatment (A1). A second set of three wafers was processed using the standard RPN1 process only, then subjected to each of the three available annealing treatments: A1, A2 and A3. Finally, a replicate wafer was produced for the standard process conditions of RPN1 with A1 anneal. Five-site electrical measurements and nine-site optical measurements were made on all wafers.

The wafers subsequently were processed to enable corresponding measurements to be made using conventional polysilicon MOSCAP structures. Six sites per wafer — two at the center and four at the edges — were probed to determine the EOL equivalent oxide thickness.

4. The oxide process using the RPN1 with A1 conditions exhibits the thinnest electrical thickness.

Figure 4 shows the correlation observed between in-line Quantox determination of EOT for each RPN process (1-4) and the corresponding polysilicon MOSCAP EOL results. The standard annealing condition, A1, was used for all wafers. Figure 4 shows the oxides processed using the RPN1 with A1 conditions exhibited the thinnest electrical thickness. The lower-power, shorter-duration RPN conditions yielded electrically thicker dielectrics.

An offset was observed between the Quantox and MOSCAP measurement techniques. In order to make the MOSCAP EOL EOT measurements, the wafers had to be subjected to several additional processing steps to create the test structures compared to the in-line Quantox measurements. In addition, polysilicon depletion was not taken into account for the P-doped polysilicon capacitors. Despite the offset, an R2 correlation of 0.96 was observed between the Quantox and MOSCAP measurements of equivalent oxide thickness.

5. Replicate oxides processed using the RPN1 with high-temperature A1 anneal show the thickest EOT.

Figure 5 shows the correlation observed between in-line Quantox determination of EOT for each anneal process (1-3) and the corresponding polysilicon MOSCAP EOL results. The standard RPN condition, RPN1, was used for all wafers. The replicate wafer processed using RPN1 and A1 is also shown in Figure 5. In this set, the replicate oxides processed using the RPN1 along with the high-temperature A1 anneal showed the thickest EOT. The lower-temperature A2 anneal, and alternative annealing environment of anneal A3, led to increasingly thinner EOT values. As observed with the previous data, an offset was observed between the Quantox EOT and the corresponding MOSCAP EOL electrical measurements of thickness. Likewise a good correlation (R2 = 0.99) was calculated between the Quantox and MOSCAP measurements of equivalent oxide thickness.

Conclusions

A combined optical and non-contact electrical metrology approach using the ASET-F5 and Quantox instruments was shown to provide statistically significant process metrology for remote plasma nitrided gate dielectrics of 26.5 Å initial oxide thickness. The linear correlation values of Quantox EOT to polysilicon MOSCAP EOL EOT, despite the inherent changes to the gate dielectric in the MOSCAP process, clearly show that a less costly, more rapid, in-line metrology approach can be implemented successfully. •

Acknowledgments

We gratefully acknowledge fruitful discussions with Sunil Hattangady, Antonio Rotondaro, Haowen Bu, Kwame Eason, Indira Gupta, Duane Peterson and Stephanie Butler at Texas Instruments. We also thank the TI PEMT/ FEP process-engineering technicians for wafer processing. The significant contributions of Steven Weinzierl, Bao Vu, Brian Letherer, Torsten Kaack, Carlos Ygartua, Albert Bivas, Pat Maxton, Greg Horner, Tom Miller and Mike Slessor at KLA-Tencor Corporation are greatly appreciated. Portions of this paper originally were presented, in modified form, at the 1999 SEMICON Japan SEMI Technology Symposium, Japan, December 1999.

Clive Hayzelden holds a B.Sc. in materials science and a D.Phil. in physical metallurgy. Following a professorship in materials science at Harvard University and six years of collaborative work with the IBM, Clive joined KLA-Tencor in 1997 as the strategic marketing manager for FaST. He has worked for 10 years in the field of electronic materials and has authored more than 50 papers. e-mail: clive.hayzelden@kla-tencor.com

Patrick Stevens is a member of the strategic marketing group for the Film and Surface Technology Division of KLA-Tencor. He received his Ph.D. in physics from the University of Texas at Dallas in the field of optical characterization of silicon. e-mail: patrick.stevens@kla-tencor.com

W. David McCain, regional metrology manager for the FaST Division of KLA-Tencor, holds a B.S. in mechanical engineering from the United States Naval Academy. He is a registered professional engineer in the state of Texas and a captain in the U.S. Naval Reserves. e-mail: david.mccain@kla-tencor.com .

Daniel Iversen is a member of the group technical staff in the Silicon Technology Research Department of Texas Instruments. He holds a B.S. in industrial engineering from Iowa State University. Dan has worked in the semiconductor industry for 16 years, in various process and equipment engineering roles. e-mail: d-iversen@ti.com .

---

REFERENCES

1. S.V. Hattangady et al., "Remote Plasma Nitrided Oxides for Ultrathin Gate Dielectric Applications," SPIE 1998 Symp. Microelec. Manuf., Santa Clara, Calif., Sept. 1998, p. 1.
   
2. C. Hayzelden et al., "Process Control for Advanced Gate Dielectrics," Proc. STS Symposium on Monitoring and Control Technology, SEMICON Japan, Makuhari, Japan, December 1999, pp. 23-32.
   

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

There are no other articles written by this author.

SPONSORED LINKS