Analysis Of BEOL Metal Schemes By Process Modeling

Benchmarking the performance of ruthenium, cobalt, and copper in a damascene vehicle with varying critical dimensions.


The semiconductor industry has been diligently searching for alternative metal line materials to replace the conventional copper dual damascene scheme, because as interconnect dimensions shrink, the barrier accounts for an increasing fraction of the total line volume. The barrier layer’s dimensions cannot be scaled down as quickly as the metal line width (figure 1). Popular barrier materials such as TaN have high resistivity and exhibit more electron scattering at the sidewalls. Therefore, the increase in the relative barrier size causes a more significant RC delay, which may reduce circuit performance and increase power consumption.

A composite image that displays the relative size and area of copper lines as the line width decreases from 20nm to 15 nm to 10 nm. The images show that the 2 nm TaN barrier around the lines make up a progressively larger proportion of the area as the line width decreases, with the Cu Area = 56% at a 15 nm Cu line width and the Cu Area = 25% at a 10 nm Cu line width.

Fig. 1: Image of scaling in copper and barrier line structure.

Recently, new alternative metal lines such as Ru and Co have been suggested and tested [1]. These materials can mitigate resistivity issues at narrow line widths and smaller areas [2]. Process modeling can be used to benchmark the performance of Ru, Co, and Cu in a damascene vehicle with varying CDs, for different trench depths and sidewall angles (figure 2). Average line resistance, two-line mutual capacitance, and RC values can be extracted for the total conductor cross-sectional area using modeling. Then, trends can be compared between the proposed Ru, Co, and Cu metal schemes.

3 images, from left: a) A simulated 3D structure of the damascene vehicle, displaying where resistance and capacitance are measured, b) A cross section of the trench, with CD, trench depth and sidewall angles identified, c) Cu, Co and Ru damascene cross sections showing the barrier width and liner measurements.

Fig. 2: 3D structure of two metal lines for R and C extractions (left) and 3 cases with different metal and barrier materials (right).

To systematically investigate design and material impact of using these different metals, a Design of Experiments containing 1000 virtual split experiments was executed using a Monte Carlo uniform distribution across three variables (CD, depth, and sidewall angle).

RC Simulation, Design of Experiment results (dots represent the predicted DoE data, lines display the trend curve). From left to right: Capacitance vs. Area, Resistance vs. Area, RC vs. Area.

Fig. 3: RC DoE results (dots : DoE data, line : trend curve). From left to right: Capacitance vs. Area, Resistance vs. Area, RC vs. Area.

Figure 3 highlights crossing points for R and RC for each metal, demonstrating that the barrier-less Ru scheme outperforms the other two metal materials at smaller dimensions. This occurs at a line CD value of ~20nm and an area value at ~400nm2, respectively. The results indicate that the barrierless Ru line resistance is lowest at a line CD < ~20nm. At line CD values less than 20nm, resistivity of the 2nm TaN barrier dominates the resistance of the Cu and Co lines, causing a sharp resistance increase. Additional scattering at the wall and grain boundaries also occur as the line CD shrinks, which contributes to higher resistance. Trench etch depth and SWA appear to change resistance linearly. Resistance is reversely proportional to the line’s cross-sectional area.

Impact of line edge roughness (LER) on resistance was also investigated.

From left to right: (a) Cu line model view with CD=20nm when LER Amplitude =1 and Correlation = 1 (b) Box plot of DoE results for Ru and Cu lines (CD=15,20,25nm cases).

Fig. 4: (a) Cu line model view with CD=20nm when LER Amplitude =1 and Correlation = 1 (b) Box plot of DoE results for Ru and Cu lines (CD=15,20,25nm cases).

In Fig 4(b), due to the barrierless structure, the Ru line’s RC values (at a line CD of 15nm) shows far less sensitivity to LER amplitude then Cu, whereas Cu is very vulnerable to RC value changes due to the highly resistive TaN barrier.

In conclusion, traditional scaling requires a 2-3 nm minimum barrier/liner thickness, leaving little space for Cu lines at modern advanced logic nodes. New metals, such as barrier-less Ru, are being considered as a potential replacement for Cu while maintaining the ability to satisfy EM reliability requirements. This study indicates that Ru may be a superior metal candidate at advanced nodes, with much lower RC delay. Typically, many wafer-based experiments would be required to perform this type of metal scheme pathfinding. Virtual semiconductor process modeling is an alternative method to investigate metal line design options at a much lower time and cost.


[1] Liang Gong Wen et al., “Ruthenium metallization for advanced interconnects,” 2016 IEEE International Interconnect Technology Conference / Advanced Metallization Conference (IITC/AMC), San Jose, CA, USA, 2016, pp. 34-36, doi: 10.1109/IITC-AMC.2016.7507651.

[2] M. H. van der Veen et al., “Damascene Benchmark of Ru, Co and Cu in Scaled Dimensions,” 2018 IEEE International Interconnect Technology Conference (IITC), Santa Clara, CA, USA, 2018, pp. 172-174, doi: 10.1109/IITC.2018.8430407

Leave a Reply

(Note: This name will be displayed publicly)