Adapting Design of Experiments (DOE) to Enhance Device Performance in RIE Fabrication at SCD

Hello. My name is Guy Burstein, and I'm an R&D engineer at SCD in the process development group. At SCD, we specialize in production of infrared detectors, and one of the important steps in our production process is Reactive Ion Etching, or RIE for short.

In RIE, there are many parameters that can potentially affect the performance of our devices. Such as ICP and RF power, temperature, etch time, and many others. One of the important characteristics of infrared detectors is dark current.

In this work, we were focusing on optimizing RIE parameters to reduce dark current and enhance performance. Our RIE process includes two steps. With our knowledge and experience, we reduced the DOE into 2 factors, which are Stage 1 and Stage 2 etch time. The DOE was constrained to nine wafers, and we were also asked to test for non-linear response.

We started by creating a design space in JMP's custom DOE design platform, and then manually, we changed it to fit our needs.

We can see with the two continuous factors and the three levels, there are experiments with high and low levels of the factors and their combination at the corners. There are four midpoints and one center point.

However, we know that if Stage 1 is too short and the second stage is too long, there's high probability for increasing the dark current. We decided to take this top left corner experiment and use it for another center point.

As for the responses, we created test devices from our wafers. Then we measured their dark current, and we were able to produce IV curves that describe each test device. We're interested here in the value of the dark current at the stable region, which is called the plateau. In this region, the dark current doesn't change significantly. Our goal is to get the lowest value for the dark current at this plateau and the longest plateau as possible.

After completing the DOE, we planned a stage of validation where we would have three wafers that we can create actual infrared detectors with the best conditions that we would find from the DOE.

We started conducting the experiments, and as we got the initial results, we've seen that in experiments with 15 minutes of the second stage, the dark current was significantly higher than for 10 minutes experiments.

At this point, we decided to explore the lower range of etch time rather than continue with our original plan, with the 20 minutes experiments. We can see here how our design space evolved with new information. We continued conducting experiments and getting more results, and at this point, we could see how each factor affects our responses.

We can see that for the first stage, there's an optimal time at which the plateau is the longest. But for the second stage, as we reduce the etch time, we can see that the dark current continues to drop.

However, we can't eliminate this stage entirely, so we had to think more deeply how this etch stage affects the dark current. We came to conclusion that we had to alter the nature of our etch.

In RIE, there are 2 components that present at the same time. But varying the parameters and adjusting them, you could move the etch towards one or the other. There's a chemical etch component where radicals from the plasma arrive to the substrate. They react on the surface and then the products are evaporated and evacuated from the chamber. While in the physical component, there are ions that accelerated to the substrate, and with this bombardment, they progress the etch.

So in our case, we decided to move more towards the physical etch rather than the chemical component. At this stage, we had one more wafer left from the DOE, and we used it to create a test device under those conditions. Indeed, we've seen the best results from the entire DOE in this test device.

Moving towards the validation stage. For the first stage, we decided to take the optimal time that we found for the first stage. But for the second stage, we used three wafers where one would be 10 minutes, the second is 5 minutes, and the third wafer with the new conditions, the more physical etch. We can see here how reducing the etch time for the second stage reduces the dark current. But with the new conditions, with the more physical etch, we can see how the dark current drops even more.

To conclude, we've seen in this work that engineering experience and knowledge are very important to choose the correct factors and their levels for the DOE. We've seen that you should be open-minded and willing to change your plan as new information arrives. We've seen that DOE is a powerful tool not just for optimization of processes, but also to increase knowledge and gain more understanding of your processes. Finally, the focus of any DOE or experiment should be to gain more understanding rather than just fitting data to a plot.

I would like to thank my colleagues for helping me conducting these experiments, and to SCD for allowing me to present this work. If you have questions I would gladly meet you at the conference. Thank you.