New Interconnect Metals Need New Dielectrics

Just as circuit metallization must evolve to manage resistance as features shrink, so must the dielectric half of the interconnect stack. For quite some time, manufacturers have needed a dielectric constant (k) less than 4, which is the value for SiO₂, but they have struggled to find materials that combine a low dielectric constant with mechanical and chemical stability.

In work presented at the recent IEEE International Interconnect Technology Conference, Bo Xie and colleagues at Applied Materials explained that the Clausius-Mossotti equation ties a material’s dielectric constant to its molecular density and molecular polarity volume.

Where N is molecules per unit volume, α_V is molecular polarity volume, and ε (more commonly k) is the dielectric constant. The carbon-doped oxides currently used, such as Applied Materials’ Black Diamond material, generally rely on reduced density to reduce k.[⁠1]

At the same time, the dielectric must be able to tolerate etching, CMP, and other downstream processes. Balancing dielectric constant and stiffness (Young’s modulus, E) becomes more challenging as feature sizes shrink. Reducing k through porosity alone threatens the mechanical integrity of the device. The relationship between material structure and film properties remains somewhat unclear, though. It’s been difficult to identify promising candidate materials within the very large universe of Si-O-C compounds.

Advanced analytics identify promising precursors
Recently, the growing sophistication of machine learning tools has simplified identification of potential precursors. For example, the Applied Materials group trained a machine learning model on a set of six precursors, process parameters, and molecular fingerprints to determine what precursor characteristics best predicted the electrical and mechanical properties of deposited materials. The model identified linearity, branching, and other measures of polarizability and structural complexity, which were then co-optimized with process parameters. Their DFT studies showed that replacing Si-O bonds with Si-H or Si-CH₃ does not reduce the net polarizability density. Rather, as the porosity increases, the ionic contribution to the dielectric constant drops. It’s possible to optimize the dielectric constant separately from the material’s mechanical properties.

Similarly, researchers at EMD Electronic Materials analyzed data from more than 1,100 films deposited from 17 different precursors under a variety of process conditions. For each precursor, they extracted the Pareto Frontier, representing the films with the highest mechanical strength relative to the dielectric constant. Two proprietary precursors were especially promising. Next, they compared the distribution of carbon bonds in films deposited from those precursors relative to DMDMOS (dimethyldimethoxysilane), an industry standard precursor. As the density of Si-CH₃ groups increased, the mechanical strength of the film decreased, apparently due to disruption of the mechanically robust Si-O network. In contrast, SiCH₂Si groups did not appear to disrupt the Si-O network.[2]

A third group of researchers, Hanna Luusua and colleagues at PiBond Oy and TSMC, sought to avoid porosity altogether by incorporating flexible low polarity units into an organosiloxane polymer backbone. Because the resulting polymer is flexible, it becomes more dense during the curing step and pores do not form. Longer flexible units improve the dielectric constant, but also increase leakage current. While the physics of this relationship is not yet well understood, this work suggests that non-porous, non-halogenated organosiloxanes may be suitable for future dielectric applications.[3]

In damascene integration schemes, the dielectric has to be able to withstand the etching process used to create trenches and vias for later metallization. As feature sizes shrink, the etched surface accounts for a larger fraction of the overall material. Carbon depletion of this surface layer can substantially increase the effective dielectric constant. Xie’s DFT analysis showed that optimizing the dielectric constant is possible without depending on Si-CH₃ bonds, with the SiCOH network controlling the mechanical properties of the film. Increasing the modulus without Si-CH₃ bonding reduced process damage.

Still, the need to withstand etch damage is a severe limitation. Transitioning to subtractive metallization could substantially increase the library of available dielectrics. On the other hand, subtractive metallization requires dielectrics that can fill very small trenches uniformly. Back in 1998, when damascene integration schemes were introduced, features were much larger, but gap fill was already a difficult process challenge.

Subtractive metallization eliminates dielectric etch, adds opportunities
Air gaps not only offer the lowest possible dielectric constant, but largely avoid the gap fill issue. At imec, researchers integrated air gaps alongside fully self-aligned vias. After etching ruthenium lines, they used a cap layer to close the exposed trenches, leaving air underneath. Because ruthenium metal is not prone to oxidation, no additional liner was required. This scheme achieved a 40% capacitance reduction relative to their oxide dieletric reference process.[⁠4]

An alternative class of materials, metal-organic frameworks (MOFs), offers a microporous crystalline structure with an open fraction as high as 70%. MOFs are widely used in catalysts, biomedical, and other applications that exploit their porous nature. Most MOF synthesis methods depend on solution chemistry, though, which has limited interest from the semiconductor industry.

In 2019, Mikhail Krishtab and colleagues at KU Leuven demonstrated an integrated circuit-compatible MOF-CVD process.[5] MOF-CVD schemes start with a metal oxide, which is consumed in a reaction with an organic linker compound. If the interconnect metal is suitable and does not require a barrier layer, its own native oxide can be used, or a precursor oxide can be deposited by ALD or CVD. MOF formation is inherently selective, occurring only on the precursor oxide layer. As the linker compound consumes the metal oxide, the film expands to as much as 10 to 15X the original thickness. This expansion allows the film to conform to small interconnect features.

Fig. 1: Conversion of CoO_x and ZnO to ZIF-67 and ZIF-8, respectively. Source: See Ref. 5

Two MOFs of interest in the semiconductor industry are ZIF-8 and ZIF-67. (ZIF stands for zeolitic imidazolate frameworks.) ZIF-8 forms by reacting ZnO with 2-methylimidizole, giving a film with the composition C₄H₆N₂Zn. ZIF-67 forms by reaction with cobalt oxide. According to Jacob Watson, who studied ZIF-8 for his Master of Science thesis at UC San Diego, it has a dielectric constant of 2.2 and a Young’s modulus of 3.4GPa, both of which are comparable to existing interconnect dielectrics.[6] At a ZnO seed layer thickness of about 5 nm, Watson said, ZIF-8 grains form a coherent layer, which forms a barrier to further diffusion of the linker compound.

Further work at UC San Diego showed that, if the seed layer is too thick, some unconverted metal oxide may remain. Higher process temperatures lead to more complete conversion, or a series of precursor/linker/conversion supercycles can be used to build a nanolaminate of the desired final thickness. Gap fill in these frameworks appears to occur via capillary condensation. If the linker precursor is at saturated vapor pressure, it adsorbs into high aspect ratio features, which appear to fill from the corners in and the bottom up.[7]

As with copper metallization, carbon-doped oxide dielectrics still have more to offer the industry. But the demands of smaller feature sizes are unrelenting. MOFs in particular offer an interesting balance between modulus and dielectric constant.

Fig. 2: Map of organosiloxane glasses, MOFs, and other candidate dielectrics. Source: See Ref. 5

References

1. Bo Xie, et al., “Advanced Black Diamond® for <2nm BEOL Low k Integration,” International Interconnect Technology Conference, San Jose, 2024, paper 4.2.
2. William R. Entley, Jennifer Achtyl, and Robert Ridgeway, “Data Analytics to Identify Improved Low k Films,” International Interconnect Technology Conference, San Jose, 2024, paper P11.
3. Hanna Luusua, et al., “A design approach for ultra-low-k dielectric organosiloxane polymers,” International Interconnect Technology Conference, San Jose, 2024, paper 9.2.
4. Gilles Delie, et al., “Airgap Integration in MP18 Two-Level Semi- damascene Interconnects with Fully Self-aligned Vias,” International Interconnect Technology Conference, San Jose, 2024, paper 2.3.
5. Mikhail Krishtab, et al., “Vapor-deposited zeolitic imidazolate frameworks as gap-filling ultra-low-k dielectrics.” Nat Commun 10, 3729 (2019). https://doi.org/10.1038/s41467-019-11703-x
6. Jacob Watson, “Metal-Organic Frameworks as Low-k Gap Fill for Logic and RF,” Thesis, Materials Science and Engineering, UC San Diego, 2023. https://escholarship.org/uc/item/3bd2q94s
7. Dipayan Pal, et al., “A Rapid Process for Vapor-deposited MOF as low-k dielectrics seamless gap fill for M1 and M2 Interconnects,” International Interconnect Technology Conference, San Jose, 2024, paper 6.2.

Related Reading
Ruthenium Interconnects On Tap
Chipmakers will stall as long as possible, but copper’s days are numbered.