Low-k Materials by Design

Current and future generations of ICs require non-traditional materials and processes for both the conductor and insulator. While copper has emerged as the preferred interconnect material, final decisions have yet to be made regarding the choice of interlayer dielectric (ILD) material.^1-3 Many new materials and processes have been proposed,^4-9 but issues encountered during integration have hindered their implementation. A major challenge arises from the materials' diminished mechanical properties with reducing dielectric constant, k (Fig. 1). As a result, more emphasis is being placed on identifying candidate materials that not only possess the required electrical properties, but also physical and chemical attributes that more easily enable integration into current processing schemes (e.g. superior mechanical and adhesive strength, as well as excellent chemical and thermal stability).¹⁰

The first-generation low-k material, fluorosilicate glass (FSG), is being used as the ILD for 180 and 130 nm technology nodes. However, FSG's k value is typically limited to ~3.6 because of chemical instability when fluorine loadings exceed 5 atomic % (at%). For the 100 nm technology node, solutions capable of delivering k=2.7 are largely focused on organosilicate glass (OSG) films. Extending OSG materials to technology nodes of m100 nm is desirable because it would alleviate re-addressing many of the integration challenges now being addressed at k of 2.7. While lowering the k of OSG is possible by engineering porosity into the structure, porosity degrades mechanical properties. Optimizing the relationship between film properties and precursor structure will enable development of materials suitable for integration and use in multiple ILD generations.

1. Reducing dielectric constant is typically accompanied by reduced mechanical properties including hardness.

The traditional dielectric material, silica or silicon oxide, ideally has a networked structure in which each silicon atom is bonded to four oxygen atoms, and each oxygen atom is bonded to two silicon atoms, described by the formula SiO₂. This structure gives SiO₂ excellent mechanical properties and k~4. Reducing the dielectric constant of a silica-based material can be done through compositional changes, the addition of free volume via terminal groups, the introduction of porosity, or combinations thereof.

A leading method to break up the Si-O network and provide materials with reduced density is through the addition of terminal groups. In FSG and OSG, k is reduced by substituting some of the oxygen for -F and -CH₃, respectively, which connect to the rest of the structure through a single chemical bond. Fluorine in FSG lowers the electrical polarizability per volume of silica, while reductions in ionic contributions by the replacement of oxygen atoms with organic groups is the major factor lowering k in OSG.¹¹ In both cases the reduced density and reduced network bonding contribute negatively to the film's mechanical properties.

2. Model-based analysis shows significant degradation in film hardness with the introduction of porosity.

We explored an alternative route for the transition from FSG to OSG. To gain potential synergy from the contributions of inorganic fluorine and organic groups, we developed a process to introduce ~5 at% fluorine into an OSG. The resulting material is an organic-doped fluorosilicate glass (OFSG) that incorporates fluorine as Si-F bonds without introducing C-F bonds to the structure. The OFSG materials possess a k value that can be tuned from 3.6 to 2.7 and mechanical properties that greatly exceed those of undoped OSG.

3. By adding <5% fluorine to OSG films, one can produce organic-doped FSG films that are 60% harder (at k=2.8-3.0) than OSG films.

The structure and composition of organosilicon molecules being offered as OSG precursors vary widely. It has been shown that addition of organic groups can lower the dielectric constant of silica to k of 2.7.¹⁴ It is widely considered that extending OSG materials to k=2.4 requires the addition of porosity.¹⁵ However, the negative impact porosity has on mechanical strength is a critical challenge for this approach. Using structural modeling to assist our efforts, we inferred that the highest mechanical strength for an ILD material that achieves a k of 2.4 from porosity can be derived from the strongest structural network possible for a non-porous material at a k of 2.7.

4. Films with the best electrical and mechanical properties contained two  Si-O bonds and a single methyl group per Si atom. The presence of more than one methyl bond per Si atom (as in DMDOSH and DMDMOS) tended to reduce hardness without decreasing k value. The DEMS precursor performed best, delivering k=2.75 with low carbon incorporation rates that allow high retention of the Si-O network, high density and good mechanical properties.

Z3MS is a leading OSG precursor for the k=2.7 generation. It is also the only precursor we tested without oxygen in its chemical structure. The electrical and mechanical properties for films deposited from Z3MS are in good agreement with recent published work.¹⁶ The molecular structure of Z3MS necessitates the use of an oxidant in the deposition process to provide low-k films based on a Si-O network. Network formation occurs through the replacement of Si-C bonds in Z3MS with Si-O bonds.

The presence of an oxidant, however, will impact bonds prone to oxidation such as C-H (i.e. hydrocarbon). As a result, it is difficult to control the balance between structure-forming Si-O bonds and incorporation of Si-CH₃ groups. Optimized Z3MS films were produced with O₂ :Z3MS flow ratios of ~1:6. Increases in oxygen flow rate result in a film with higher k and hardness as well as higher deposition rates. Films contain ~20 at% carbon as measured by XPS, and a H:C of ~2:1 by RBS/HFS. The lack of significant residual Si-H in these films suggests a structure comprised of a Si-O network with both bridging and terminal organic groups. In contrast, precursors that contain Si-O bonds in their structure generally provide low-k films without needing additional oxidant. This suggests that organosiloxane precursors may facilitate the formation of a Si-O network with the retention of Si-CH₃ groups.

Our evaluation indicates a strong relationship between Si-O:Si ratio present in the precursor molecular structure and mechanical strength of resulting films. As this ratio varies, we observe the interplay between k and mechanical properties. For example, MTES (methyltriethoxylsilane), with a Si-O:Si ratio of 3:1, produces films with very good mechanical properties but k~2.95. On the other hand, TOMCATS (tetramethylcyclotetrasiloxane) with a Si-O:Si ratio of 1:1, produces films with relatively poor mechanical strength but k~2.75. Interestingly, small additions of oxygen to the TOMCATS processes moderately improved mechanical properties.

The optimum Si-O:Si ratio in the precursor appears to be 2:1 based upon an SiO₂ network in which a portion of the oxygen atoms are replaced by organic groups. A number of precursors tested with this composition provide films with low k but widely ranging mechanical properties. We find the other two substituents on the molecule also play a critical role in the mechanical properties. Previous work by MacWilliams et al suggests that a preferred film composition is one with about one methyl group bound to every Si atom. The introduction of additional methyl groups serves only to disrupt the silica network and diminish the mechanical properties without lowering k.¹⁴ This tendency is exemplified by poly (dimethylsiloxane), a polymeric compound and a liquid at ambient conditions that has two Si-CH₃ and two Si-O bonds per Si atom and a k~2.75.¹⁷ We expected that precursors providing more than one organic functional group per Si atom might compromise film mechanical properties without benefit to electrical properties.

At our test conditions, precursors with Si-CH₃ :Si ratios of 2:1 such as DMDOSH (dimethyldioxysilylcyclohexane) and DMDMOS (dimethyldimethoxysilane) produce films with marginal mechanical properties. As with Z3MS, which has three methyl groups per Si atom, addition of oxygen improves mechanical properties at the cost of increased k. It is expected that a significant number of Si-CH₃ bonds present in the precursor remain intact during deposition, potentially limiting network formation. Therefore, we propose the optimum OSG precursor structure to be one containing two Si-O bonds and a single Si-CH₃ bond per Si atom. The final bond to Si should be a reactive site that can be used to form network bonds.

Molecules with the proposed composition and structure were found to provide the best overall electrical and mechanical properties. The DEMS (diethoxymethylsilane) and DMOMS (dimethoxymethylsilane) precursors provide two alkoxy groups, a single methyl group, and a labile Si-H bond. The labile hydride bond is easily broken upon plasma ionization, providing a molecular ion (M-H)⁺ that possesses the key ingredients to building a highly networked OSG material. While both precursors provide excellent mechanical properties, structural differences in these two molecules result in slightly higher k for DMOMS.

We used compositional data to deduce a structure for films produced from the DEMS precursor. Films contain ~10 at% carbon with a H:C of slightly greater than 3:1. Chemical analysis indicates few residual Si-H bonds. The DEMS film structure is composed of a Si-O network with carbon incorporated as a single methyl group on about every other Si atom. While this suggests not all Si-CH₃ bonds in the precursor are retained within the film, the presence of one methyl group on every other Si atom is sufficient to provide k=2.75. Furthermore, the low carbon incorporation provides high retention of the Si-O network, high density, and good mechanical properties.

The DEMS precursor appears to provide all the necessary ingredients to maximize mechanical properties while delivering k=2.75. Films deposited from DEMS incorporate the minimum amount of methyl groups needed to achieve desired k. This, along with excellent structure-forming groups, endows DEMS films with the best balance of electrical and mechanical properties of any precursor we tested. The ability to provide OSG materials with enhanced mechanical properties should enable extendibility of CVD technology to k of at least 2.2 through incorporation of porosity.

By gradually reducing the dielectric constant over successive generations, integration processes can evolve to handle the increasingly more fragile materials. Staying with one network former and reducing the dielectric constant by introducing fluorine or porosity is a scheme to extend some of the integration learning from one generation to the next.

There is a clear relationship between precursor composition/structure and film properties. OSG films typically involve a trade-off between mechanical properties and k value. However, by complementing our experimental efforts with modeling we are able to recognize trends in film and precursor structure and composition. Precursors and processes that permit control over the amount and type of organic functional groups and facilitate network formation are essential for optimizing k value and mechanical strength. By tailoring precursor chemistry, we can develop superior dielectrics for the k=2.7 generation. Finally, developing optimum materials at this generation will permit a smoother transition to porous low-k materials and eventually, ultralow-k films at future technology nodes.