Implementing a batch atomic layer deposition approach for advanced DRAM dielectrics

Jon Owyang and Larry Bartholomew, Aviza Technology

The drive toward sub-100-nm feature sizes places extreme performance demands on semiconductor manufacturing processes and equipment. Chemistry and hardware challenges are being addressed through the development and implementation of novel processing techniques such as atomic layer deposition (ALD) and the incorporation of exotic materials using designer chemicals such as high-k dielectrics for films. The stringent control of film properties drives the development of advanced chemical-delivery techniques, control hardware, and the choice of materials.

To fabricate memory cell capacitors in advanced DRAM applications, it is critical to develop materials with a higher dielectric constant than that of Al₂O₃. At the 45-nm node, materials must be able to provide 100% conformal step coverage into deep-trench capacitors with
aspect ratios approaching 80:1. In addition, they must have high thermal stability and very low leakage. Finally, they must offer reasonably high throughput to achieve cost-effective manufacturing.

The study presented in this article focuses on efforts by Aviza Technology (Scotts Valley, CA) and a global device manufacturer to use a batch ALD tool to deposit high-dielectric-constant Hf_xSi_{1– x}O₂ films. Research at Aviza using a single-wafer platform has shown that Hf_xSi_1–xO₂ can meet the technical requirements of advanced DRAM applications. However, depositing the films using a single-wafer tool is not cost-effective because to minimize leakage, the films must be processed using a long recipe time to achieve step coverage and compositional control. To overcome this limitation, Aviza transferred its proprietary coinjected Hf_xSi_{1– x}O₂ process from the single-wafer platform to the Verano 5000 batch ALD system, improving throughput and overall cost-effectiveness.

Hf_xSi_1–xO₂ Films

As listed in Table I, several potential candidate materials can be used to manufacture DRAM capacitor films with higher dielectric constants than Al₂O₃. Of these films, Hf_xSi_{1 – x}O₂ is attractive because of its high thermal stability and chemical compatibility with silicon and polysilicon.¹ Although the dielectric constant of Hf_xSi_1–xO₂ at a 50% Hf:(Hf + Si) ratio is not much higher than that of Al₂O₃, the films can be made silicon- or hafnium-rich by changing the relative ratio of the hafnium and silicon precursors. Thus, their dielectric constant and thermal stability can be modified.

As shown in Figure 1, by increasing the hafnium content of Hf:(Hf + Si) films, a higher dielectric constant approaching that of pure HfO₂ can be obtained. By increasing their silicon content, the films' dielectric constant decreases and they become more like SiO₂.² With a higher silicon content, the films' thermal stability improves.

Figure 1: Estimated dielectric constant versus the concentration of hafnium in HfxSi1–xO2.

Because Hf_xSi_1–xO₂ films offer potential advantages for improving device electrical performance, they are being evaluated for production applications in advanced device technology nodes below 90 nm. Hence, hafnium silicate compounds may replace Al₂O₃ in next-generation DRAM dielectrics.

Single-Wafer versus Batch ALD Processing

Because of its low residual carbon content and high deposition rate, tetrakis-ethyl-methyl-amino hafnium (TEMAHf) is the preferred hafnium precursor for ALD hafnium silicate films. However, these hafnium liquid precursors have lower vapor pressure than trimethyl aluminum (TMA), the typical Al₂O₃ precursor.³ Because of their low vapor pressure, hafnium precursors require longer chemical pulse times in the process chamber to achieve sufficient chemical concentration and saturate the wafer surface. If surface saturation is not achieved, the thickness uniformity of the film is degraded, as illustrated in Figure 2.

Figure 2: Pulse times required to achieve depletion and saturation on monitor and patterned wafers processed on an ALD system dispensing TMA precursor. An extra chemical dose is required to saturate a patterned wafer surface with substantial topography.

In single-wafer tools, if the chemical-delivery system cannot compensate for the longer pulse times by injecting a higher-concentration dose of a precursor, the need to achieve film saturation and uniformity results in longer ALD process cycle and total recipe times, drastically reducing wafer throughput. Figure 2 shows the TMA pulse times required to achieve different amounts of surface saturation on two different types of wafers. The pulse time shown for flat monitor wafers also applies to patterned wafers with low-aspect-ratio structures, such as transistor gates. For DRAM deep-trench structures, a longer pulse time is required for the chemical precursor to reach the bottoms of deep trenches and achieve complete surface saturation. As indicated in Figure 2, it takes substantially more precursor pulse time to achieve either partial depletion or full conformal deposition to the bottoms of the trenches on patterned wafers than either partial depletion or full saturation on monitor wafers.

In this case, excess saturation of a flat monitor wafer does not increase the deposited film thickness, since in true ALD processes, proper purging before the oxidizer pulse removes excess precursor, yielding no more than a monolayer of film with each process cycle. By the same token, if the chemical-delivery system deposits too much precursor, the increased time required to purge it decreases throughput. Furthermore, the use of very-high-concentration doses of precursor, especially those with high sticking coefficients, causes more than a single-monolayer ALD reaction on the wafer surface.

Without longer residence times to enable the precursor to reach the bottoms of the trenches at concentrations that do not overdose the top surface, it is impossible to achieve 100% step coverage. Consequently, using low-vapor-pressure hafnium precursors to meet the step-coverage requirements for DRAM deep-trench structures with aspect ratios of >50:1 requires pulse times lasting several seconds, not just milliseconds as in the case of TMA. For single-wafer ALD systems, such long pulse times are impractical for manufacturing. At a typical 7-nm film thickness, throughputs for planar low-aspect-ratio patterned wafers are in the double-digit range while throughputs for DRAM wafers with deep trenches having high aspect ratios are in the low-single-digit range.

Batch ALD systems can afford longer pulse times than single-wafer systems by processing many wafers at a time. To accommodate the increased surface area of multiple wafers in a larger-volume process chamber, pulse times are increased enough to satisfy the residence times required for the precursor to penetrate to the bottoms of deep trenches. The tool discussed here processes 50 wafers simultaneously, enabling the use of long pulse times so that the precursors can achieve full patterned-wafer surface saturation and step coverage while maintaining production-worthy throughputs. Figure 3 compares the throughput of a single-wafer system and a batch system used to process planar and deep-trench wafers.

Figure 3: Relative wafer throughputs for single-wafer and batch ALD tools as a function of wafer topography. The batch tool improved throughput by 300–400% for flat wafers and 200–300% for patterned wafers.

Another difficulty posed by the longer precursor pulse times used to deposit hafnium silicate compounds is that the ratio of hafnium to silicon must remain the same throughout the entire pulse to yield a conformal film of consistent composition at the top and bottom of the trench. Step coverage in deep-trench structures is a function of precursor dose and residence time. The scanning electron microscope (SEM) image in Figure 4 illustrates a deep-trench structure with an aspect ratio >50:1 in which an Hf_xSi_1–xO₂ film has achieved conformal step coverage.

Step coverage and compositional control of a compound film will not be attained if one precursor depletes relative to the other during the chemical pulse. For chemical-delivery systems that use an inert carrier gas through liquid bubblers, the vapor pressure of the TEMAHf precursor is insufficient to deliver a high-enough continuous dose because the consumption rate is higher than the vapor regeneration rate as a result of bubbler depletion. Thus, during an extended chemical pulse, the ratio of TEMAHf to TEMASi may decrease from the beginning to the end of the pulse.

Figure 4: SEM image showing conformal step coverage of HfxSi1–xO2 in a deep-trench structure with an aspect ratio >50:1. The deposited thickness was increased to highlight the step coverage.

Precursor depletion, an issue for single-wafer systems that attempt to saturate deep-trench structures, is magnified in batch Hf_xSi_1–xO₂ processing, where much more chemistry must be injected onto the wafer to cover much more surface area. While single-wafer systems may be able to prevent depletion of dual precursors at differing rates at the expense of throughput, batch ALD processing requires additional advanced chemical-delivery techniques to achieve uniform precursor injection and evacuation in the process chamber. Furthermore, rapid chamber evacuation and chemical purging are critical in ALD processing to maximize throughput and prevent CVD reactions.

By minimizing process chamber volume, the ALD system under investigation reduces the purge time between chemical pulses. Dual vertical injectors positioned at the wafer edge inject metal-organic precursor and oxidizer alternately while establishing the true cross-flow gas dynamics required to achieve uniform within-wafer and wafer-to-wafer film thickness, film composition, and conformal step coverage. Film-compositional control is achieved using independent dual liquid injectors for the hafnium and silicon precursors, a method that is not subject to the vapor-
pressure-depletion effects associated with bubblers or vapor-draw techniques used for long pulse times. The dual-liquid-injector technique allows the desired precursor dose to be injected during the desired pulse time, even if the pulse time is extended. Hence, the residence time required for the precursor to reach the bottom of high-aspect-ratio structures can be satisfied at an acceptable throughput.

Batch Repeatability

To demonstrate film-thickness uniformity within a 50-wafer batch and the repeatability of the process, four consecutive runs were performed on the ALD batch system under the same Hf_xSi_{1– x}O₂ process conditions. Figures 5a–d plot the spectroscopic ellipsometry results for 10 300-mm bare silicon monitor wafers spaced throughout each 50-wafer batch. Table II summarizes the thickness and refractive index (RI) uniformity achieved for each of the four runs using 49-point mapping with a 3-mm edge exclusion. Among all 40 monitor wafers measured, the worst individual within-wafer thickness uniformity was ±3.19% by range, or 1.62% σ/mean. The worst individual within-wafer refractive index uniformity was 0.87% σ/mean.

Figure 6: Mean film thickness, wafer-to-wafer thickness variation, within-wafer thickness uniformity, and refractive index values of HfxSi1–xO2 film for each batch of 50 wafers processed using the ALD batch system.

Wafer-to-wafer uniformity for all 40 wafers was ±2.50% for thickness and ±0.29% for refractive index. Batch-to-batch repeatability was excellent, as indicated by the minimal increase in wafer-to-wafer thickness variation over any single batch. Figure 6 plots the mean thickness, within-wafer thickness uniformity, and the mean refractive index for each batch. Also shown is wafer-to-wafer thickness uniformity, which was derived from the mean thicknesses of the 10 wafers measured in each batch. Mean within-wafer thickness uniformity was 1.12% σ/mean, or ±2.45% range. Batch-to-batch thickness repeatability was ±0.23% range, and batch-to-batch refractive index repeatability was ±0.05% range.

Figure 7: Relative wafer-per-hour throughput and per-wafer TEMAHf and TEMASi chemical consumption for batch versus a 100-Å single-wafer process. At the same 100 Å, the batch process improved throughput by ~450% over the single-wafer process while consuming one-third the amount of TEMAHf and one-eighth the amount of TEMASi.

In addition to the throughput improvement achieved by running the Hf_xSi_1–xO₂ process on a batch instead of a single-wafer tool, chemical consumption decreased significantly. Figure 7 shows relative 300-mm wafer-per-hour throughput and per-wafer chemical consumption for TEMAHf and TEMASi batch processing at various film thicknesses normalized to a 100-Å single-wafer process. The process conditions for both the single-wafer system and the batch system were set to obtain similar conformal step coverage results in advanced DRAM deep-trench structures. At the same 100-Å thickness, the batch process improved throughput by ~450% while consuming only 33% of the TEMAHf and 12% of the TEMASi per wafer compared with the single-wafer process.

Conclusion

Hf_xSi_1–xO₂ films are of interest to semiconductor device manufacturers because they can cover severe topographies such as deep-trench capacitors in advanced DRAM structures as well as more-planar gate-stack structures in low-leakage logic applications. These applications require materials with a higher dielectric constant than Al₂O₃.

Batch ALD processing has significant advantages over single-wafer processing because it enables higher throughput and significantly lower chemical consumption, reducing the cost of production manufacturing. The technical design of the batch ALD system delivers liquid precursors that yield good within-wafer film-thickness uniformity. The batch-to-batch repeatability studies described in this article demonstrate that the total wafer-to-wafer thickness variation of ±2.5% across four consecutive batches was only marginally higher than the worst within-batch wafer-to-wafer thickness variation of ±2.44% range. Finally, the use of independent direct liquid injection for dispensing TEMAHf and TEMASi precursors ensures that compositional control of the dielectric properties of the bulk film and the proper ratio of hafnium to silicon can be maintained.

Acknowledgments

The authors would like to acknowledge Carl Barelli, Yoshi Okuyama, S. G. Park, Chris Tousseau, Jay DeDontney, and Bryan Ford from Aviza Technology for their technical assistance, and Hood Chatham, also from Aviza, for his discussions on the topics discussed in this article.

References

1. GD Wilk, RM Wallace, and JM Anthony, "High-k Gate Dielectrics: Current Status and Materials Properties Considerations," Journal of Applied Physics 89, no. 10 (2001): 5243–5275.

2. Y Senzaki et al., "Atomic Layer Deposition of Hafnium Oxide and Hafnium Silicate Thin Films Using Liquid Precursors and Ozone," Journal of Vacuum Science and Technology A 22, no. 4 (2004): 1175–1181.

3. DM Hausmann et al., "Atomic Layer Deposition of Hafnium and Zirconium Oxides Using Metal Amide Precursors," Chemistry of Materials 14, no. 10 (2002): 4350–4358.

Jon Owyang is director of ALD product management at Aviza Technology (Scotts Valley, CA). Before joining the company, he served as a senior strategic enabling engineer at Intel, where he introduced high-k materials into development. Before his tenure at Intel, Owyang held technology development positions at LSI Logic and Philips Semiconductor. He holds five patents related to process technology. He received a BS in chemistry from the University of California, Berkeley. (Owyang can be reached at 831/439-6405 or jon.owyang@avizatechnology.com.)

Larry Bartholomew is a principal process engineer at Aviza Technology. After beginning his career at Watkins-Johnson, he worked for Silicon Valley Group's thermal systems division and ASML's thermal division. Bartholomew has 26 years of experience in atmospheric pressure CVD and ALD process development for production applications. He holds six patents and has two patents pending. He received a BA in physics from the University of California, Santa Cruz. (Bartholomew can be reached at 831/439-6313 or larry.bartholomew@avizatechnology.com.)