Code in practice

Modelling software has proved invaluable in one semiconductor companys bid to develop a faster and cheaper method for depositing copper contacts on flip-chip carrier substrates.

Swedish semiconductor company Replisaurus Technologies has developed an interesting alternative to conventional photolithography for depositing copper contacts on flip-chip carrier substrates. The company claims that its electrochemical pattern replication (ECPR) process not only deposits copper much quicker – anywhere from ten to a 100 times faster – but is also much more economical.

However, as deposition takes place inside a cavity where it is impossible to install monitoring instrumentation, Replisaurus has had to carry out lots of experimental work to optimise the technique. It was helped in this task by a modelling software package from technical software developer Comsol. Called Femlab, the software allows Replisaurus to simulate hundreds of different process variations before committing to the time and expense of clean room trials.

Flip-chip technology eliminates wire bonds between the silicon die and the package, and has become a cost-effective means of dealing with the packaging and thermal issues of high-density, high-power integrated circuits (ICs). Typically, the final wafer-processing step deposits solder beads on the chip pads, so the die package must itself have pads with positions that align with the beads. Creating these carrier substrates with photolithography can involve almost as many manufacturing steps as when creating the IC itself.

Replisaurus, however, employs a reusable patterned master electrode as a template and provides for direct metallisation on a variety of substrates. Compared to lithography-based metallisation, which takes as long as 120 minutes, its ECPR process requires between one and five minutes. It also achieves higher precision for the plating/etching reaction, and is claimed to be more economical – primarily because it requires less capital equipment.

The process starts with two elements: a flat cathode substrate with a thin metal seed layer on which the pads and traces are to be deposited, and a master anode consisting of an electrically conducting electrode layer and a patterned insulating material. In the patterns gaps, the operator pre-deposits an anode material, usually copper. Then the operator places an electrolyte between the two layers and squeezes them together. The sandwich goes into a pressure vessel to hold in the electrolyte, and in the presence of a voltage across the layers, the metal migrates to the cathode at a deposition rate of between one and four microns per minute. The final step etches away the metal seed layer from the cathode, leaving an exact pattern of metal traces.

To push the limits of the process, the research team needed a deep understanding of it. “Before we started working with Femlab, we derived our results and understanding experimentally,” explained Replisaurus research and development manager Mikael Fredenberg, “We wanted a model that would explain the theory behind the phenomena we observed.”

When he decided to build a model, Mikael investigated Femlab, as he had used the software during his studies at Swedens Lund Institute of Technology. Femlabs ready-made interface for electrochemical engineering offered a good starting point for the models construction.

The initial model assumed a constant current. Fredenberg then learned how to add variations to refine the simulation. The model determines flux with the Nernst-Planck equation in the material balance in combination with the electroneutrality condition. The Nernst-Planck equation describes mass transport of copper ions in the electrolyte that occurs due to diffusion and migration. The diffusion rate is determined by the concentration gradient that appears when the process produces copper ions at the anode and consumes them at the cathode.

The migration defines the ion transport potential gradient causing the positively charged copper ions to move towards the more negative cathode surface. Next, the cathode and anode boundaries give flux according to the Butler-Volmer equation. This equation describes the current density at the electrode as a function of the overpotential, which in turn is given by the difference between the electrodes surface potential (applied with an external power source) and the potential in the electrolyte closest to the electrode surface.

Fredenberg was able to enter these equations for the electrode kinetics directly into Femlabs graphical user interface.

“The model allows us to play with a large number of parameters such as different voltage levels, warpage or unevenness in the substrate or electrolyte properties. We can try out all sorts of ideas, no matter how wild, and get a first estimation of results, enough to let us know if theyre worth pursuing,” said Fredenberg.

One modelling problem was dealing with moving boundaries. As material grows on the cathode, a non-uniform current-density distribution can lead to changes in cell geometry. Areas on the cathode with higher deposition rates grow faster and get closer to the anode, causing the current density to increase further in these areas.

Fredenbergs initial models didnt account for such an effect, so the Femlab support team came to his aid, showing him how to solve a moving-boundary problem using the packages implementation of the arbitrary Lagrangian Eulerian (ALE) method – a powerful tool not found in many mathematical-modelling codes.

One effect Fredenberg wanted to investigate in particular was uneven surfaces such as imperfections on the cathode. The modelling allowed him to rule out some issues he thought might cause a non-uniform current-density distribution.

Such models have allowed Fredenberg to perform analytical research that helps to find the optimum process voltage – for best balance between deposition rate and quality. Too low a voltage can result in a slow deposition rate whereby only a few crystallisation sites have enough energy for the copper to crystallise, leading to irregularities in plating. Too high a voltage can result in deposits that are too porous, so fast plating can lead to poor quality.

“Femlab is helping us to explain the phenomena weve seen in the lab,” said Fredenberg. “The information has been invaluable in debugging the process and is now helping us refine the technique for commercial operations.”

Replisaurus expects to demonstrate its ECPR process publicly later this year.

Figure 1. Replisaurus Technologies’ founders (left to right): Patrik Möller, Mikael Fredenberg and Peter Wiwen-Nilsson.

Figure 2. The ECPR process: the operator predeposits an anode material in the gaps of a pattern that matches that of an IC (Step 1), and when the anode and cathode are put together (Step 2) enclosed cavities result. A voltage migrates the metal from the anode to the cathode (Step 3), and removing the seed layer leaves only a replica of the original pattern (Step 4).

Figure 3. One FEMLAB model allows Replisaurus to evaluate the effects of an uneven surface (shown here as a bump) on the current density that controls the rate of copper deposition.