The search for a replacement metal at advanced nodes appears to have a winner — at least for now.
Molybdenum is looking increasingly promising as a replacement for a variety of metals commonly used in semiconductor manufacturing today, especially at leading-edge nodes.
One by one, chipmakers are crossing metals off the list at advanced nodes. While ruthenium liners are nearly ready for production, the metal is not ready to replace copper in highly scaled interconnects. Ruthenium is very expensive, and current manufacturing processes don’t help. In addition, the amount of waste generated by the “over-deposit and polish back” steps in damascene schemes is a serious concern, according to Zsolt Tőkei, an imec fellow. And while subtractive metallization reduces the amount of waste, it requires more significant and costly changes to the overall process.
Copper isn’t the only metal with a short runway. Transistor contacts, wordlines in memory, and similar applications typically use tungsten, cobalt, and other metals — not copper. Still, they face many of the same scaling issues that copper does. Like copper, tungsten suffers from increasing resistivity as feature sizes shrink. It also requires a barrier layer to avoid dielectric contamination. In 3D NAND devices, researchers at Kioxia reported that fluorine residue from the WF6 precursor generally used for tungsten deposition can get trapped in voids, ultimately attacking the surrounding dielectric material. [1] Tungsten also faces electromigration concerns as features shrink and current density increases.
So what’s next? For these applications, an increasingly attractive option — at least for now — appears to be molybdenum. It offers multiple advantages relative to both the incumbent materials and alternatives like ruthenium, according to Tőkei. It has better resistivity than tungsten, does not require a barrier layer, and compared with ruthenium, it’s less expensive and offers better adhesion to dielectrics.
Less barrier, less resistance
Molybdenum is especially attractive as a barrierless contact metal in hybrid metallization schemes, where pre-filled vias are followed by copper damascene lines. Because the barrier at the bottom of a via or other vertical feature places an additional resistance in series, bottom barriers dominate contact and via resistance.
TaeYeon Oh, senior semiconductor process and integration engineer at Lam Research, and his fellow researchers showed that a barrierless hybrid molybdenum scheme can reduce overall resistance by about 56% relative to a conventional copper dual damascene design. Their work was shown at the recent IEEE Interconnect Technology Conference. [2]

Fig. 1: Lam Research’s ALTUS tool, which combines CVD and ALD to improve molybdenum deposition in advanced metallization applications. Source: Lam Research
Integrating molybdenum with a process flow like this would likely require few changes beyond the metal deposition module itself, Tőkei said. Molybdenum oxidizes more readily than ruthenium, making it more easily removable by CMP.
However, a thorough analysis by Jean-Philippe Soulié and colleagues at imec warned that the bulk properties of a metal are of limited value in evaluating its performance in real devices. For molybdenum — as for other nanowires — electrical, thermal, and electromigration properties all depend on the grain size and grain boundary structure of the deposited film. These, in turn, depend on the precursors, the process parameters, the surface characteristics of the underlying dielectric, and so on. [3]
Managing electromigration
Interfaces and grain boundaries are leading pathways for electromigration, while also causing electron scattering and degrading resistivity. For molybdenum integration, the metal deposition module will need to be able to handle solid precursors like MoO2Cl2 and MoCl5. Solid precursors are becoming more common in semiconductor manufacturing, in general, with chlorine-based examples for hafnium, aluminum, and tungsten in addition to molybdenum. Relative to gas or even liquid precursors, though, solids tend to be less thermally stable and give a less uniform flux of material.
Researchers at Lam Research, in work presented by David Mandia, staff process engineer, said they achieved precise grain-size control through cyclic deposition techniques, blending thermal- and plasma-based processes as needed to achieve the desired results. They showed that large-grain molybdenum films are essential for successful integration. In their work, the resistivity of small grain molybdenum showed a comparable thickness dependence to tungsten. The resistivity of large-grain molybdenum, in contrast, was much less dependent on thickness and was superior to tungsten, ruthenium, and even copper at thicknesses below about 7nm. [4]

Fig. 2: Resistivity of different metals at different thicknesses. Source: imec
When grain boundaries do exist in molybdenum, doping with elements like cobalt can help reduce scattering, as shown in a simulation study by Yeongjun Lim and Mincheol Shin, researchers at the Korea Advanced Institute of Science and Technology. At low concentrations, resistivity decreased due to charge compensation effects. At higher concentrations, though, resistivity increased sharply. The additional impurity states led to carrier localization, disrupting electron transport. [5]
Predicting metal behavior is especially challenging in backside power applications. Backside power designs are intended to reduce the standard cell size. As a result, though, backside power networks increase current density, and therefore the risk of electromigration. They also are prone to hot spots for similar reasons.
While electromigration and heat dissipation in backside power configurations have yet to be thoroughly analyzed, molybdenum has some clear advantages. As a refractory metal, it is mechanically stable even at very high temperatures. Better adhesion to dielectrics makes it less likely to form voids. It’s also a better thermal conductor than ruthenium. For all these reasons, molybdenum is likely to resist electromigration more successfully than ruthenium, though both metals should give acceptable results. Linlin Cai, a researcher at Sun Yat-Sen University, explained that better electromigration resistance allows designers to pack transistors more closely, reducing the overall device area. [6]
While more experimental results are needed, early molybdenum integration studies have been quite promising. The Kioxia group found that molybdenum’s lower resistivity relative to tungsten allowed them to reduce wordline pitch by 7.3% while keeping RC constant. Memory hole pitch shrank by more than 3.7%, for an overall 16.3% increase in bit density.
Overall, Tőkei said, molybdenum is a very natural fit for contact and wordline applications. It fits well in existing integration schemes. For the long-term, though, ruthenium may be extendible to smaller devices.
References
Related Articles
The End Of Copper Interconnects
Ruthenium plays a key role at advanced nodes, but it’s not a wholesale replacement.
Powering CFETs From The Backside
Next-gen transistors will benefit from the new power delivery scheme, but there are plenty of challenges ahead.
GE Semiconductor had molybdenum Met1 / Met1.5 on their AVLSI1 technology, in production when we (Harris Semi) bought them in the ’80s. Robust then. Of course grain size < linewidth was maybe more forgiving than modern times.
Disagree about the significance of contact barrier resistivity, as the barrier should be infinitesimal height.
barrier metal thickness would be significant if the overall film thickness is in the range of single digit nm or even tens of nm.
curious that the Cu sample in the study is TaN/Cu/TaN while the tungsten does not have barrier film as part of the study. It would be more of a fair comparison if the tungsten sample is with the barrier layer.