Designing Porous Low-k Dielectrics

		At a Glance
	The 100 nm device generation calls for low-k dielectrics below 2.5, which should soon be lowered to 2.0 or less for future technology nodes. The dielectric constant, of a given material can be reduced by decreasing polarizability and film density, but the most powerful synthetic method involves the introduction of porosity. A variety of pore generation methods are being used to create ultralow-k spin-on dielectrics with k values as low as 1.3.

As minimum device features shrink below 180 nm, the increase in propagation delay, crosstalk noise and power dissipation of the interconnect become limiting factors in ultra-large-scale integration (ULSI) device performance.^1-4 To address resistance/capacitance delay,^3,4 copper wiring (R=1.7 µW/cm vs. 3.1 µW/cm for Al) and low-k dielectrics are being combined to replace the Al(Cu)/SiO₂ interconnect technology. Although a large variety of dielectric candidates — both spin-on (SOD) and CVD including purely inorganic, organic and inorganic-organic hybrids — have been examined over the past six years,^5-8 most candidates have been eliminated by triage schemes based on their inability to meet optimized and rigorous integration criteria.^5,6 Because device integration proved to be more difficult than predicted, the transition from SiO₂ (CVD k= 4.1-4.2) to new dielectric materials withk<3.0 was delayed by about three years, relative to initial SIA Roadmap targets.^2,7

One significant breakthrough in the implementation of ak<3 SOD occurred in June 2000, when IBM successfully integrated Dow Chemical Co.'s (Midland, Mich.) SiLK polyaromatic thermoset material.⁹ The device architecture was crafted using via pillars (sea-of-vias) to support a highly thermally resistant(>400°C) low-k material (k=2.65) that exhibits a significantly lower modulus and hardness when compared with SiO₂.^8,9

Since this announcement, Altis Semiconductor (Corbeil-Essonnes, France), an IBM-Infineon joint venture, and UMC (Hsinchu, Taiwan) have adopted the technology and expect to produce copper/low-k chips in late 2001. In response, CVD toolmakers such as Applied Materials (Santa Clara, Calif.), Novellus Systems (San Jose) and Trikon Technologies (Newport, UK) have introduced organosilicate films (Si_wC_xO_yH_z, C = CH₃ group, x ~10-25%) with k values ranging from 2.6 to 2.8. These OSG films are intended for use with existing or modified circuit designs, while leveraging existing toolsets and infrastructure.

Though the successful integration of low-k (2.7) SOD and CVD films for the 130 nm generation has largely been accomplished, companies face significant materials and integration challenges at the 100 nm node and beyond. Because few fully dense spin-on or CVD materials withk<2.5 satisfy the diverse requirements for successful integration, recent research efforts have focused on the preparation of porous materials, where the dielectric constant, k, scales with the host matrix density. Two main factors that affect the host matrix k are electronic and nuclear polarizability and density.

Polarizability

Dielectric constant depends on the summation of the polarizability from the electronic, ionic and polar components of the material. The electronic component refers to the oscillation of electrons in chemical bonds and/or in the extended supramolecular structure. The latter two components constitute the nuclear response and are important at lower frequencies(<10¹³ Hz), while the electronic response dominates at higher frequencies (4.74 × 10¹⁴ Hz). At typical device operating frequencies, currently <10⁹ Hz, all three components contribute to the dielectric constant and must be minimized for optimum performance.

The choice of a host matrix with low polarizability is aided by examining polarizabilities and the associated bond enthalpies for the atomic building blocks in the host material.⁵ For instance, single C-C and C-F bonds have the lowest electronic polarizability, making fluorinated and nonfluorinated aliphatic macromolecules potential candidates for low-k applications. Conversely, electronic polarizability is higher for materials containing a large number of carbon-carbon double and triple bonds because of the increased mobility of p electrons. Importantly, in practice, there is a trade-off in achieving a low dielectric constant and high bond strength (which leads to thermal stability). Low-polarizability single bonds are among the weakest, while fluorinated aliphatic structures display poor adhesion, low hardness, and reactivity toward Ti- and Ta-based liners. In contrast, aromatic p bonding configurations have significantly higher bond strengths. Dow Chemical's SiLK material (Fig. 1) is an example of a crosslinked polyaromatic matrix that largely eliminates the nuclear polarizability component and features high bond strength and, thus, excellent thermal resistance.^10,11 Permanent and transition dipoles in the low-k material contribute to its nuclear polarizability. The response is dominated by polar functionality, such as hydroxyl and carbonyl groups. Unfortunately, these groups can attract water (k=78.5) via hydrogen bonding interactions, which can drastically elevate the dielectric constant. This suggests that an optimal low-k material should be devoid of polar functionality containing oxygen or nitrogen that attracts and binds water.

	1. The formation of the crosslinked SiLK material is driven by the extrusion of CO (entropic contribution) and the aromatization/formation of arylene functionality (enthalpic contribution). The polyaromatic structure does not contain polar chemical functionality, thereby eliminating the nuclear polarization component of the dielectric constant.
	2. The ladder structure in the MSQ oligomer is common to silsesquioxane-based materials. In practice, the structure is more randomized, further lowering packing density and, the k value relative to SiO2 films.
	3. The relationship between k value and porosity is the result of the Bruggeman effective medium approximation (EMA) model. To obtain a k=2.0 in an oxide material, ~50% porosity is needed, whereas only ~22% porosity is needed in a k=2.5 material. These predicted results are in reasonable agreement with recent experimental measurements on MSQ and oxide porous films.5 Differences may be related to differences in surface chemistry, including terminating OH groups and absorbed water, as well as pore geometries.

By eliminating points of chemical attack in the matrix by oxygen and nucleophiles, a chemist can increase the host's stability during curing and post-cure processes such as CMP. For example, hydrogen silsesquioxanes (HSQ, HSiO_1.5) condense at elevated temperatures to form random ladder networks. Because water is evolved during the cure, care must be taken during annealing to protect labile hydrogen atoms from moisture as well as oxygen, which could lead to the formation of silanol groups that raise the k-value. Methyl silsesquioxanes (MSQ, RSiO_1.5 where R=CH₃) and similar materials avoid this problem by the substitution of the labile H atoms by hydrophobic organic R groups (Fig. 2). However, when portions of the organic groups are integrated into a mostly inorganic resin matrix, they can segregate at the surface to lower their free energy, possibly leading to adhesion challenges.

Managing density

The matrix dielectric constant can also be lowered by decreasing its density. Researchers achieve lower density by incorporating low molecular weight atoms and/or adding more free volume to the structure. For example, the lower dielectric constant of non-polar organic polymers (~2.0-2.8) relative to SiO₂ (3.9-4.2) is partly due to the presence of lighter C and H atoms vs. Si and O, as well as the low packing density (d~1.0 g/cm³) relative to the dense tetrahedral silica network (d~2.4 g/cm³). The incorporation of non-polar and space-occupying groups such as methyl, ethyl and phenyl (R groups) will also increase free volume in a silica network. Optimized loading of R groups in a random silsesquioxane or silica network results in steric congestion, which interferes with polymer chain packing and attractive forces, which could lead to crystallization or areas of higher density. The result is a reduction in dielectric constant from 2.9 for HSQ to 2.5-2.8 for MSQ and other R-substituted silsesquioxane-based SODs.

A natural extension of the strategy is the introduction of nanometer-sized pores into the material. Porosity's effect on dielectric constant can be predicted using the Bruggeman effective medium approximation model¹²:

where f_1,2 represents the fraction of the two components, k_1,2 the dielectric constant of the components, and k_e the material's effective k. The simple EMA model assumes two components to the film — the solid wall material and voids — and provides the basis for ellipsometric measurement of film thickness, density, and index of refraction on porous or multicomponent thin films. Using this model, one can predict dielectric constant as a function of porosity for SiO₂ (k=4.0) and a k=2.5 host materials (Fig. 3). These results infer that, by initially choosing a host matrix with a lower k value than SiO₂, significantly less porosity is needed to achieve the lowest target k value.

Level of porosity

Advantages to minimizing the porosity needed to obtain a low k value include greater mechanical strength and greater thermal conductivity of the interlevel dielectric (ILD). To decrease potential for cracking and delamination — and thus ensure device structure integrity — pore sizes should be as small as possible, though some sources recommend no greater than 20% of the smallest feature size. Engineers prefer to work with a closed-cell architecture because of increased mechanical strength and resistance to absorption and diffusion of process chemicals. Studies show that Young's modulus decreases with increasing porosity, an important concern in CMP processing.⁵ For example, Dow Corning's range of XLK products displays a decrease in moduli with increasing porosity, from 7.5 to 2.0 GPa (k=2.5-2.0).¹³

Porous materials obtained by micelle templating and vitrification processes are expected to display higher mechanical strengths because of their engineered periodic structure. One example is Schumacher's (Carlsbad, Calif.) MesoElk, a silica-based material with a cubic closed-pack structure.¹⁴

	4. The relationships between thermal conductivity, porosity and dielectric constant for one example of DendriGlass SOD. Less-dense low-k materials typically display lower thermal conductivity than, for instance, TEOS oxide (1.4 W/m°C).

Thermal conductivity also decreases with increasing porosity,⁵ as the propagation of thermal energy flow through the matrix is interrupted by the nanoscopic voids. This suggests that the ILD material should have sufficient thermal conductivity or a circuit architecture that is optimized to avoid joule heating. Figure 4 shows the dependence of thermal conductivity on porogen loading and thus porosity.

Pore generation

The synthetic chemist can take a variety of approaches in the preparation of ultra-low-k SODs with controlled incorporation of porosity. The initial and most widely used approach relies on the incorporation of a thermally degradable material (porogen) within a host thermosetting matrix.¹⁵ Upon heating, the matrix material crosslinks, and the porogen undergoes phase separation from the matrix to form nanoscopic domains. Subsequent heating leads to porogen decomposition and diffusion of the volatile by-products out of the matrix. Under optimized processing conditions, a porous network results in which the pore size directly correlates with the original phase-separated morphology. This templated-porogen-based approach forms the basis for the National Institute of Standards and Technology (NIST)/IBM/Dow Chemical joint project on porous dielectrics, which began in 1999. The two most prominent materials derived from the ultra-low-k portion of this venture are Dow Chemical's porous SiLK and IBM's DendriGlass materials.

The templating strategy offers versatility and extendability of the host matrix from one generation to the next, since the matrix material remains constant and only the loading of porogen varies. The approach is applicable to a wide range of materials, both inorganic to organic, and to a range of porogen structures. For example, porous silsesquioxanes, such as DendriGlass, are based on the blending of organosilicates, such as phenylsilsesquioxane, with a polymeric porogen.^15-18 As shown in Figure 5, the process is easily transferred to manufacturing since it relies only on spin-coating followed by a thermal ramping profile to ~450°C. In the field emission scanning electron micrograph (FESEM) images, one can see the DendriGlass film obtained after vitrification at two different porogen loadings (Fig. 6).

The critical feature of a templating approach is the design and structure of the polymeric porogen. On the nanometer scale, the porogen must phase separate from the thermosetting matrix and decompose entirely, leaving no residue. A variety of thermally decomposable porogen structures is available by conventional chemical synthesis including Fréchet-type dendritic, hyperbranched and linear polymetric structures.^15-18 SOD suppliers such as JSR (Tokyo), Asahi Chemical (Tokyo) and Shipley Co. (Marlborough, Mass.) have used this approach, leading to materials with very small pore structures (~5 nm), a wide range of porosities and dielectric constants of 2.5-1.3.

5. The reactive process blends an inorganic silsesquioxane with an organic polymer porogen, which are solubilized in a common solvent. Spin-coating is followed by an initial heating step to vitrify the inorganic matrix and template the nanoscopic phase separation. Subsequent high-temperature heating affects thermal decomposition of the polymer leading to porosity. Many macromolecular porogens may be used, from dendritic to highly branched to linear polymers.

For organic matrices, suppliers such as Dow Chemical, Schumacher and Honeywell (Sunnyvale, Calif.) have used a similar blending approach to develop porous SiLK, PolyELK and porous FLARE. Dow recently announced the preparation of porous SiLK featuring a k value of 2.0 and pore sizes <20 nm. As with inorganic and hybrid systems, the porogens should be homogeneously dispersed within the crosslinked matrix without gross phase separation, which leads to large pores and open-cell architectures after porogen decomposition. With organic hosts, porogen decomposition must occur below the host's glass transition temperature (T_g). Otherwise, the void structure may collapse in the softer and more rubber-like environment.

An alternative to the blending approach involves the covalent attachment of a thermally decomposable organic polymer to a high T_g or thermosetting organic matrix to form a block copolymer. Self-assembly of the block or graft copolymer then leads to nanoscopic phase separation. A similar thermolysis strategy yields the desired nanoporous organic SOD. The original patent for the production of porous dielectric materials using a degradable porogen utilizes this approach.¹⁹ Other approaches involve the covalent attachment of a pendant thermally decomposable moiety to the polymer backbone. This approach has been recently extended to inorganic-organic hybrid materials as well as purely organic hosts.^15,20

A templated, self-assembly approach has been developed by Schumacher in collaboration with Sandia National Laboratories (Albuquerque, N.M.), and involves the surfactant-directed polymerization and crosslinking of a silicate oligomer.¹⁴ The resulting gelled or partially vitrified silicate matrix contains surfactant templated periodic liquid crystalline domains, which are then degraded by thermolysis to yield volatile fragments that diffuse out of the matrix, yielding a mesoporous cubic closed-packed structure. Claims for MesoELK include tunable k, narrow pore size distribution, and increased moduli relative to aerogels and xerogels.¹⁴

	6. DendriGlass with 20% porosity (k=2.1) and 50% porosity (k=1.5).

A related templating technique uses a high boiling point solvent as the porogen, the so-called "solvent-as-porogen" approach.^{13, 21} In this arena, the SOD mixture contains a low and a high boiling point solvent. After initial thermal ramping, the low boiling point solvent is released, and vitrification is induced. The high boiling point solvent, which is a poor solvent for the vitrified matrix, undergoes phase separation, thus forming nanometer-sized domains. Further heating leads to volatilization of the solvent and generation of templated porosity. Dow Corning's XLK series of resins, with tunable k, uses this approach.

Honeywell's Nanoglass porous silica films are prepared by the sol-gel technique, in which a silicate solution is induced to form a wet gel arranged in an open pore structure.^{22, 23}

The gel must be dried in a controlled manner to produce a porous structure that will not collapse. Although this technology has been developed by a number of semiconductor manufacturers, its wide acceptance has been hampered by elaborate processing conditions. The hydrophilic nature of the open-cell structure also leads to absorption of aqueous process chemicals and poor mechanical properties of the films. High porosities (60-70%) may also be required to achieve dielectric constants below 2.0.

Conclusions

The concept of porosity as the preferred route to ultra-low-k materials is not restricted to SODs. A number of CVD suppliers are actively conducting research in this area in efforts to provide CVD-based low-k materials as an alternative to SODs fork<2.6. Applied Materials has disclosed the deposition of silicon-based materials containing labile organic groups with a peroxide compound followed by thermal annealing.²⁴ Novellus has also developed a method for the deposition of organosilicate materials followed by thermal oxidation of the organic side chains to give a porous structure.²⁵ Achieving a controlled closed-cell architecture with a narrow pore size distribution remains a major challenge for CVD methods.

In summary, modern chemical synthesis and nanoscopic engineering provide pathways to well-defined porosity-based ultra-low-k dielectrics. These technologies may give the edge to molecularly engineered SODs over CVD-deposited films. As the implementation of low-k dielectrics accelerates, the overall market is predicted to expand from $20M in 2001 to $285M in 2004.²⁶Although extendibility of existing processes, toolsets and infrastructure remain important practical issues, the successful integration of ultra-low-k dielectrics ultimately depends on the resulting film's chemical and physical properties. Traditionally, the preparation of photoresists has been the premier polymeric materials challenge in the manufacture of integrated circuits. It appears that current and future challenges may now reside with the design and integration of low-k dielectric materials.

Josh H. Golden, director of process technology at Microbar Inc., is involved in the development of dispense and handling technologies for spin-on low-k dielectrics, photoresists, copper chemistries and technologies for CMP wastewater treatment. He received his Ph.D. in chemistry from Cornell University (Ithaca, N.Y.)
Phone: 1-408-542-9069
e-mail: jgolden@microbar.com

Craig Hawker is a research staff member and an investigator in the NSF Center for Polymer Interfaces and Macromolecular Assemblies at the IBM Almaden Research Center. He received his bachelor's degree in chemistry from the University of Queensland (Brisbane, Australia) in 1984 and his Ph.D. in bio-organic chemistry from the University of Cambridge (Cambridge, England) in 1988. He recently was awarded the 2001 Carl S. Marvel Award in Creative Polymer Science by the American Chemical Society.
Phone: 1-408-927-2377
e-mail: hawker@almaden.ibm.com

Paul S. Ho is director of the Laboratory for Interconnect and Packaging at the University of Texas. He received his bachelor's degree in mechanical engineering from National Cheng Kung University (Tainan City, Taiwan), his master's degree in physics from National Tsing Hua University (Hsinchu, Taiwan), and Ph.D. in physics from Rensselaer Polytechnic Institute (Troy, N.Y.). He is a Fellow of the American Physical Society and the American Vacuum Society.
Phone: 1- 512-471-8961
e-mail: paulho@mail.utexas.edu

---

REFERENCES

1. P.V. Zant, Microchip Fabrication: A Practical Guide to Semiconductor Processing 4th ed. (New York, McGraw-Hill), 2000, p. 1.
   
2. International Technology Roadmap for Semiconductors, Semiconductor Industry Association, 1999, p. 172. For updates see: http://public.itrs.net.
   
3. T.N. Theis, "The Future of Interconnection Technology," IBM Journal of Materials Research and Development, Vol. 44, No. 3, 2000, p. 379.
   
4. X.W. Lin, D. Pramanik, "Future Interconnect Technologies and Copper Metallization," Solid State Technology, Oct. 2000, p. 63.
   
5. M. Morgan et al, "Low Dielectric Constant Materials for ULSI Interconnects," Annual Review of Materials Science, Vol. 30, 2000, p. 645.
   
6. L. Peters, "Solving the Integration Challenges of Low-k Dielectrics," Semiconductor International, Vol. 22, No. 13, 1999.
   
7. L. Peters, "Low-k Dielectrics: Will Spin-On or CVD Prevail?" Semiconductor International, Vol. 23, No. 6, 2000.
   
8. J.G. Ryan et al, "Copper and Low-k Dielectric Integration Challenges," Proceedings of the Low-k Dielectric Materials Technology Conference, SEMICON West, 2000, p. A1.
   
9. R. Goldblatt et al, "A High Performance 0.13 µm Copper BEOL Technology with Low-k Dielectric," Proceedings of the IEEE 2000 IITC, p. 261.
   
10. J.P. Godschalx et al, "Polyphenylene Oligomers and Polymers," U.S. Patent 5,965,679, 1999.
    
11. S.J. Martin et al, "Development of a Low-Dielectric-Constant Polymer for the Fabrication of Integrated Circuit Interconnect," Advanced Materials, Vol. 12, 2000, p. 1769.
    
12. R.M.A. Azzam, N.M. Bashara, Ellipsometry and Polarized Light, Amsterdam, Elsevier, 1977.
    
13. Data obtained from Dow Corning XLK literature. Also see: C. Jin, J. Wetzel, "Characterization and Integration of Porous Extra Low-k (XLK) Dielectrics," Proceedings of the IEEE 2000 IITC, p. 99.
    
14. C.J. Brinker, M.T. Anderson, R. Ganguli, Y. Lu, "Process to Form Mesostructured Films," U.S. Patent 5,858,457, 1999. Also see: www.schumacher.com/SolFall99.pdf.
    
15. C.J. Hawker, J.L. Hedrick, R. Miller, W. Volksen, "Supramolecular Approaches to Nanoscale Dielectric Foams for Advanced Microelectronic Devices," Materials Research Society Bulletin, April 2000, p. 54; J.L. Hedrick et al, "Templating Nanoporosity in Thin Film Dielectric Insulators," Adv. Mater., Vol. 10, 1998, p. 1049.
    
16. K.R. Carter et al, "Process for Manufacture of Integrated Circuit Device," U.S. Patent 5,895,263, 1999.
    
17. C.J. Hawker, J.L. Hedrick, R.D. Miller, "Integrated Circuit and Process for Its Manufacture," U.S. Patent 5,767,014, 1998.
    
18. W. Volksen et al, "Low Dielectric Constant, Nanoporous Silicates: Precursor Nature and its Effect on Porosity," Proceedings of the 7th International Conference on Dielectrics and Conductors for ULSI Multilevel Interconnection (DCMIC), March 2001, IMIC-777DC, p. 107.
    
19. J.L. Hedrick et al, "Foamed Polymer for use as Dielectric Material," U.S. Patent 5,776,990, filed March 11, 1993.
    
20. B. Zhong, "Method for Making Microporous Silicone Resins with Narrow Pore-size Distributions," U.S. Patent 6,197,913, 2001.
    
21. M.L. O'Neill, L.M. Robeson, W.F. Burgoyne, M. Langsam, "Nanoporous Polymer Films for Extreme Low and Interlayer Dielectrics," U.S. Patent 6,187,248, 2001.
    
22. B.E. Gnade, C-C Cho, D.M. Smith, "Method for Forming Porous Composites as a Low Dielectric Constant Layer with Varying Porosity Distribution in Electronics Applications," U.S. Patent 5,494,858, 1996.
    
23. D.M. Smith, W.C. Ackerman, R.A. Stoltz, "Polyol-based Method for Forming Thin Film Aerogels on Semiconductor Substrates," U.S. Patent 6,171,645, 2001.
    
24. R.P. Mandal, D. Cheung, W-F Yau, "CVD Nanoporous Silica Low Dielectric Constant Films," U.S. Patent 6,171,945, 2001.
    
25. T.W. Mountsier, "Chemical Vapor Deposition of Low Density Silicon Dioxide Films," U.S. Patent 6,054,206, 2000.
    
26. "The Global Outlook for Dielectric Materials in Semiconductor Devices, 1999 to 2004," Kline and Company Inc., 2000.
    

Acknowledgment

Thanks to J. Eric Carrubba for his aid in graphic design and implementation.